MICROWAVE AND MILLIMETER WAVE SYSTEM INTEGRATION

WITH ADDITIVE MANUFACTURING

By

Michael Thomas Craton

A DISSERTATION

to Michigan State University

in partial fulfillment of the requirements

Submitted

for the degree of

Electrical Engineering – Doctor of Philosophy

2020

ABSTRACT

MICROWAVE AND MILLIMETER WAVE SYSTEM INTEGRATION

WITH ADDITIVE MANUFACTURING

By

Michael Thomas Craton

This dissertation demonstrates a collection of strategies to package and heterogeneously integrate
microwave and millimeter wave (mm-wave) electronics using additive manufacturing (AM). These
strategies not only facilitate heterogeneous integration and higher functional density of a system-
in-package (SiP)/system-on-package (SoP), but furthermore allow for structures and performance
that is difficult or impossible to match with conventional manufacturing strategies.

These strategies are implemented using aerosol jet printing, but may be applicable to other sorts
of additive manufacturing technologies such as ink-jet printing. The processes developed in order
to facilitate these procedures are described, including the use of multi-material aerosol jet printing
(MMAJP) for the manufacture of composites and functionally graded materials for use in mm-wave
circuits, and a chip-first approach to microwave and mm-wave circuit packaging. These processes
enable the production of application specific SiP/SoP electronics for use in radar, communications,
imaging, sensing, as well as the supporting circuitry such as filters and antennas which may be
difficult or expensive to implement with semiconductor processes. Included are demonstrations of
the integration of some of these strategies in complete packages.

Characterization of materials used including conductors and dielectrics are presented, as well
as simulations and measurements of package sub-components, and complete packages. Finally, the
future of this research is discussed.

Copyright by
MICHAEL THOMAS CRATON
2020

Dedicated to all of the teachers, mentors,

and friends I have been lucky enough to learn from.

iv

ACKNOWLEDGEMENTS

I would like to express my deepest appreciation to my committee, Dr. John Papapolymerou, Dr.
Premjeet Chahal, Dr. John D. Albrecht, Dr. Edward Rothwell, and Dr. Aljoscha Roch for their
instruction and guidance. My interest in how things work and my work ethic are owed in no
small part to the influence of my late grandfathers, Karl Larry Craton and Earle Eugene Fischer. I
would also like to extend my deepest gratitude to all of my other co-authors, and the other current
and former advising and student members of my research groups during my time at Michigan
State University, including the Electromagnetics Research Group (EMRG). Specifically, Dr. Leo
Kempel, Dr. Shanker Balasubramaniam, Dr. Jeffrey Nanzer, Dr. Jennifer Byford, Dr. Chris
Oakley, Dr. Yuxiao He, Dr. Jubaid Qayyum, Dr. Jakub Sorocki, Dr. Ilona Piekarz, Dr. Saranraj
Karuppuswami, Xenofon Konstantinou, Adamantia Chletsou, Nick Sturim, and Yining He, all of
whom have directly assisted with my work, and who I am extremely grateful to have been given
the opportunity to work with. I also wish to thank the instructors of all of my graduate coursework
not already mentioned, Dr. A. Cagri Ulusoy, Dr. Carlo Piermarocchi, Dr. Tim Hogan, Dr. Neeraj
Buch, Dr. Daina Briedis, and Dr. Stanley Flegler. I am also grateful to Brian Wright, Roxanne
Peacock, and Michelle Stewart for helping me find the resources needed to complete my work,
as well as Amy Albin and Carol Flegler with the Center for Advanced Microscopy at Michigan
State University. Finally, I would like to express my gratitude for the diverse academic community
which is composed of all races, gender identities, and religions. The unique experience, insight,
and perspective that diversity provides is what makes progress, scientific or otherwise, possible.

The work in this dissertation was supported by a MSU foundation professorship and Honeywell
Federal Manufacturing & Technologies, LLC which manages and operates the Department of
Energy’s National Security Campus under contract DE-NA-0002839.

v

TABLE OF CONTENTS

.
.

.
.

.
.

.
.
.

.
.
.
.
.
.
INTRODUCTION .

LIST OF TABLES .
.
.
LIST OF FIGURES .
.
.
KEY TO ABBREVIATIONS .
CHAPTER 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Current Packaging Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
1.2 Performance Bottlenecks .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 A Proposed Alternative Packaging Strategy . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.1
1.2.2 Materials
1.2.3

Interconnects .
.
.
. .

.
Flexibility .

.
.
.

. xv
1
1
6
6
8
8
9

ix

x

.

.

.

.

. .

. .

. .

2.1 Design .

2.2 Fabrication .

.
Silicon Package .

.
Silicon Package .

CHAPTER 2 A CHIP-FIRST APPROACH TO MILLIMETER WAVE PACKAGING . . . 12
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2 Gallium Arsenide Package . . . . . . . . . . . . . . . . . . . . . . . . . . 15
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Gallium Arsenide Package . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.3
Fabrication Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Gallium Arsenide Package . . . . . . . . . . . . . . . . . . . . . . . . . . 25
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

.
Silicon Package .

2.4 Conclusions .

2.3 Results .

. .

. .

. .

. .

.

.

.

.

.

.

.

.

.

.

.

.
.

.
.

.
.

. .

3.1

CHAPTER 3 MULTI-MATERIAL AEROSOL JET PRINTING (MMAJP)

. . . . . . . . 29
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Introduction .
.
3.1.1 Background .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.2 Comparison with Conventional Fabrication . . . . . . . . . . . . . . . . . 30
3.1.3 Applications . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Theory of Operation .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
.
3.3 Design of Characterization Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.1 Overview .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.2 Material Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Initial Ink Formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.4.1
3.4.2 Y-Fitting .
Printing .
3.4.3

3.4 Fabrication .

. .

.
.

.
.

.
.

.
.

.
.

.
.

.

.

.

.

.

.

.

vi

.

.

.

3.5

Initial Fabricated Device Characterization . . . . . . . . . . . . . . . . . . 54
3.4.4
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Initial Results .
3.5.1 Electrical Characterization . . . . . . . . . . . . . . . . . . . . . . . . .
. 55
3.5.2 Microscopy and Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Process Improvement for Lower Loss . . . . . . . . . . . . . . . . . . . . . . . . . 68
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.7 Conclusion . .

. .

.

.

.

.

.

Inductors

CHAPTER 4 APPLICATIONS OF MMAJP: OTHER COMPOSITES . . . . . . . . . . . 75
4.1 Magnetic Composite Circuit Design . . . . . . . . . . . . . . . . . . . . . . . .
. 77
4.2 Magnetic Composite Circuit Fabrication . . . . . . . . . . . . . . . . . . . . . . . 78
4.3 Magnetic Composite Measurement Results . . . . . . . . . . . . . . . . . . . . . . 81
4.3.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3.2 Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
.
4.3.3 Resonator .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.4 EDS .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.4 Resistive Composite Formulations . . . . . . . . . . . . . . . . . . . . . . . . .
. 92
4.5 Resistive Composite Measurement Results . . . . . . . . . . . . . . . . . . . . . . 94
4.6 Conclusions .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

. .

.
.

.
.

.
.

.

.

.

.

.

.

.

.

.

.

.

CHAPTER 5

5.3 Results .

5.1 Design .
5.2 Fabrication .

. .

.
.

.
.

.
.

. .
.
.

. .
.

5.2.1

INTEGRATING CHIP-FIRST PACKAGING AND MMAJP . . . . . . . . . 99
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Packaging Considerations for Active Components . . . . . . . . . . . . . . 109
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.3.1 Material Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.3.2
. . . . . . . . . . . . . . . . . . . . . 115
5.3.3 Large-Signal Measurements
. . . . . . . . . . . . . . . . . . . . . . . . . 120
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.3.4 Reliability .
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

Small-Signal Package Measurement

. .
.
.

. .

. .

. .

. .

.
.

.

.

.

5.4 Conclusions .

CHAPTER 6 EXTENDING THE CHIP-FIRST PACKAGING STRATEGY TO W-

. .

6.1 Design .
6.2 Fabrication .
6.3 Results .
6.4 Conclusion .

BAND FREQUENCIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
. 129
. 131
. 133
. 140

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. .
.
.
. .
.

. .
.
. .

. .

. .

.
.
.
.

.
.
.
.

.
.
.
.

CHAPTER 7 CONCLUSIONS AND FUTURE WORK . . . . . . . . . . . . . . . . . . . 141
7.1 Potential System Performance Gains from AM Integrated Microwave Circuits . . . 141
. . . . . . . . . . . . . . . . . . . . . 142
7.1.1 Heterogeneous Integration Benefits
Interconnect Performance Gains . . . . . . . . . . . . . . . . . . . . . . . 143
7.1.2
7.1.3
Processing Gains .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.1.4 Materials Performance Gains . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.2 Extension of this Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.2.1 Multilayer Packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

vii

. 148
7.2.2 Conformal Packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.3 High Power & High Reliability Packaging . . . . . . . . . . . . . . . . . . 149
7.2.4 Novel Microwave Structures with MMAJP . . . . . . . . . . . . . . . . .
. 149
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

.

.

.

.

.

.

7.3 Final Thoughts

.

.

.

APPENDICES .
APPENDIX A
APPENDIX B
APPENDIX C
APPENDIX D

BIBLIOGRAPHY .

.

.

.

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
.
Aerosol Jet Printing Notes and Procedures . . . . . . . . . . . . . . 153
Material Processing and Preparation . . . . . . . . . . . . . . . . . 160
AJP Chip-First Process
. . . . . . . . . . . . . . . . . . . . . . . . 164
. . . . . . . . . . . . . . . . . . . . . . . . . . 167
List of Publications
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

.

.

.

.

.

.

viii

LIST OF TABLES

Table 2.1 Comparison of AM High Frequency (Single) Interconnects . . . . . . . . . . . . 28

Table 3.1 Aerosol Mix Ratios for Combined Aerosol

. . . . . . . . . . . . . . . . . . . . 54

Table 3.2 Measured Thickness and Roughness of Circuits . . . . . . . . . . . . . . . . . . 60

Table 3.3 Comparison of Results with Modified Lichtenecker (ML) and Maxwell-Garnett

(MG) Equations . .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

Table 3.4 Unloaded Quality Factor of Measured and Simulated Resonances

. . . . . . . . 74

Table 4.1 Comparison with other planar inductors . . . . . . . . . . . . . . . . . . . . . . 85

Table 5.1 Measured Printed Material Thicknesses . . . . . . . . . . . . . . . . . . . . . . 112

Table 5.2 Comparison with current state-of-the-art AM capacitors

. . . . . . . . . . . . . 115

Table 5.3 Comparison with other AM Packages

. . . . . . . . . . . . . . . . . . . . . . . 119

Table 6.1 Comparison with other AM Packaged Semiconductor Devices . . . . . . . . . . 139

ix

LIST OF FIGURES

Figure 1.1 Example of a MATLAB-generated digital pseudo random bit sequence (PRBS)
signal spectrum with Gaussian white noise added: a) first 200 bits of NRZ
PRBS15 signal (out of 215 − 1), b) frequency spectrum of NRZ PRBS15 signal

Figure 1.2 Evolution of functional block innovations working towards higher functional

density in packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 1.3 Typical in-package microwave interconnect methodology: a) bond wire/rib-

bon, b) flip chip . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.

.

Figure 1.4 Aerosol jet printing enabled SiP/SoP component by chapter. . . . . . . . . . . .

2

3

6

7

Figure 2.1 Silicon die package design (in units of mm): a) package length = 4.7, b)
50 µm wide signal length (to edge of die) = 0.1, c) package width = 1.825,
d) probe ground spacing = 0.325, e) signal width = 0.225, f) probe ground
width = 0.25, g) printed signal length = 1.9, h) printed gap width = 0.05, i)
probe ground length = 0.5, j) printed package substrate height = 0.1, k) die
height = 0.15. ©2018 IEEE.

. . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Figure 2.2 GaAs package design (in units of mm): a) package length = 3.7, b) printed
signal length = 1.25, c) inductive signal length = 0.184, d) CPWG to MS
transition length = 0.1, e) probe ground width = 0.505, f) CPWG ground to
ground space = 0.107, g) inductive signal width = 0.048, h) MS width = 0.108,
i) CPWG signal width = 0.075, j) 2x through trace length = 2.4, k) package
height = 0.05, l) mounted die height = 0.125. ©2019 IEEE.

. . . . . . . . . . . 18

Figure 2.3 Si package before printing Ag.

. . . . . . . . . . . . . . . . . . . . . . . . . . 20

Figure 2.4 Finished Si package under measurement with inset detail of interconnect

region. ©2018 IEEE.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Figure 2.5 One of the two identically fabricated GaAs packages and 10x magnified detail

of interconnect region. ©2019 IEEE.

. . . . . . . . . . . . . . . . . . . . . .

. 22

Figure 2.6 Examples of package failures due to a high temperature gradient: a) cracked
PI which additionally pulled away from a die, in this case a GaAs amplifier,
b) wrinkled PI.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

.

.

.

Figure 2.7 Measured and simulated scattering parameters of the Si package compared to
the unpackaged bare die performance: a) transmission, b) reflection. ©2018
IEEE.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

.

.

.

.

.

.

.

.

.

x

Figure 2.8 Measured and simulated scattering parameters of the of the two identical
measured GaAs packages, the simulated performance of these packages,
the measured unpackaged attenuator, and the printed 2x through line: a)
transmission, b) reflection. ©2019 IEEE. . . . . . . . . . . . . . . . . . . . . . 26

Figure 2.9 Calculated loss of a single interconnect in the GaAs packages (0.234 mm

long). ©2019 IEEE.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

Figure 3.1 Representation of the aerosol mixing and deposition with critical metrics,
where P, Q and l are the pressure of aerosol A or B, the volumetric flow rate
of aerosol A, B or C and the length of the tube from atomizer A or B to the
Y-fitting or from the Y-fitting to the print head. ©2020 IEEE.

. . . . . . . . . . 34

Figure 3.2 Top down view of the characterization circuit with dimensions in millime-
ters and degrees showing both large and small ring resonators: a) PI nano-
composite, b) silver ink. ©2020 IEEE.

. . . . . . . . . . . . . . . . . . . . . . 38

Figure 3.3 Top down detail view of coupling structure (shown on small resonator) with

the location of the reference plane for measurements. ©2020 IEEE. . . . . . . . 39

Figure 3.4 Simulated S-parameters for ring resonator circuits using 
r = 3.5, tan δ =
0.008 for the PI nano-composite: a) larger ring resonator, b) smaller ring
resonator. ©2020 IEEE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Figure 3.5 Dispersant investigation for BaTiO3 nanoparticles. . . . . . . . . . . . . . . . . 46
Figure 3.6 Third from left MMAJP film using sub-optimal BaTiO3 seen de-laminating

from an Al carrier in initial feasibility study of MMAJP. . . . . . . . . . . . . . 48

Figure 3.7 Design of 3D printed Y fitting for MMAJP a) side, b) side, c) top. . . . . . . . . 49

Figure 3.8 MMAJP Fabrication setup a) 3D printed fitting, b) commercially available fitting. 50

Figure 3.9 Concept of alternate embodiment of mixing channel for more than two
aerosols including computer controlled valves to completely shut off or open
aerosol flows a) perspective, b) side, c) bottom. . . . . . . . . . . . . . . . . . . 52

Figure 3.10 S-Parameters for large ring resonators 0-67 GHz. Simulations use measured

and calculated material parameters a) Mix 1, b) Mix 6, c) Mix 11. ©2020 IEEE. 56

Figure 3.11 S-Parameters for small ring resonators 75-110 GHz. Simulations use mea-
sured and calculated material parameters. a) Mix 1, b) Mix 6, c) Mix 11.
©2020 IEEE.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

.

.

.

.

.

Figure 3.12 Calculations from measured MIM capacitors. Parasitic effects lead to error

increasing with frequency. a) capacitance, b) tan δ, c) 
r. ©2020 IEEE. . . . . . 58

xi

Figure 3.13 Calculations from ring resonator measurements a) 
r, b) tan δ. ©2020 IEEE. . . 59
Figure 3.14 The fabricated ring resonators and MIM capacitors for mix numbers 1, 6, and

11. ©2020 IEEE.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Figure 3.15 Fabricated 5-mm by 5-mm films, a) fabricated gradient film with mix 1-11
©2020 IEEE, b) six films labeled with their 
r fabricated with improved ink
formulation including a nearly continuous gradient.

. . . . . . . . . . . . . . . 63

Figure 3.16 SEM images for a) mix 1, b) mix 6, and c) mix 11.

In each image, the

magnification is 5000 at 5.0 kV beam energy. ©2020 IEEE.

. . . . . . . . . . . 64

Figure 3.17 EDS analysis for a) mix 1, b) mix 6, and c) mix 11, showing an increase in

the constituent elements of BaTiO3. ©2020 IEEE. . . . . . . . . . . . . . . . . 65

Figure 3.18 S21 of lower loss composites compared to simulations a) large ring resonator:
0-67 GHz, mix 1 b) large ring resonator: 0-67 GHz, mix 11. ©2020 IEEE.

. . . 71

Figure 4.1 Annotated design in millimeters with labeled material regions. Dimensions
in parenthesis are target material thicknesses. (a) Inductor design. (b) Trans-
mission line design. (c) Ring resonator design. . . . . . . . . . . . . . . . . . . 79

Figure 4.2 Result of composite printing without PI border.

. . . . . . . . . . . . . . . . . 81

Figure 4.3

(a) Fabricated inductors. (b) Fabricated transmission lines and ring resonator
on composite. Measurement reference planes are indicated by dotted red lines.

82

Figure 4.4

Inductor parameters extracted from de-embedded S-parameters. (a) Induc-
tance, (b) Quality factor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

Figure 4.5 Transmission line S-Parameter measurements on composite with and without

DC magnetic bias. (a) 2x transmission (b) 2x reflection.

. . . . . . . . . . . . . 87

Figure 4.6 Transmission line S-Parameter measurements on composite and PI and with
and without DC magnetic bias. (a) Line 1 transmission (b) Line 1 reflection.

. . 88

Figure 4.7 Transmission line S-Parameter measurements on composite and PI and with
and without DC magnetic bias. (a) Line 2 transmission (b) Line 2 reflection.

. . 89

Figure 4.8 Composite ring resonator S-parameters.

. . . . . . . . . . . . . . . . . . . . . 91

Figure 4.9

(a) Composition backscatter electron image of composite with region high-
lighted in red where EDS analysis was performed. (b) EDS analysis.

. . . . . . 93

Figure 4.10 Calculated resistivity from four-point-probe measurements of MMAJP resis-

tive composites.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

xii

Figure 4.11 (a) Samples printed with carbon ink composed 13.5 % by vol. graphene and
carbon black mixture + 1.9 % silver, (b) detail of 40% carbon ink MMAJP
mix, (c) detail of 50% carbon ink MMAJP mix, (d) detail of 60% carbon ink
MMAJP mix, (e) detail of 70% carbon ink MMAJP mix.

. . . . . . . . . . . . 96

Figure 5.1

(a) Exploded view of package illustrating fabrication procedure and (b) cross
section showing material layers. ©2020 IEEE.

. . . . . . . . . . . . . . . . . . 101

Figure 5.2 Package design for both fabricated packages (package 1 and package 2) with
dimensions in millimeters. Dimensions in parentheses are target material
thicknesses of PI (a) and Ag (b). ©2020 IEEE. . . . . . . . . . . . . . . . . . . 102

Figure 5.3 Capacitor layout for nanocomposite characterization with dimensions in mil-

limeters of PI (a) and Ag (b). ©2020 IEEE.

. . . . . . . . . . . . . . . . . . . 103

Figure 5.4

2x through design for AJP packages with dimensions in mm for PI (a) and
Ag (b). ©2020 IEEE.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

Figure 5.5 One of the two fabricated packages and material characterization structures:
(a) bypass capacitors, (b) packaged die, (c) perspectives of the package in
(a) and (b), (d) 2x through, and (e) capacitors for material characterization.
©2020 IEEE.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

.

.

.

.

Figure 5.6 A mechanically good package which failed prior to high temperature and ESD

process changes. .

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 111

Figure 5.7 Measurement setup.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

Figure 5.8 Measured capacitors and extracted dielectric properties. Parasitic effects lead
to error increasing with frequency: (a) capacitance, (b) 
r, and (c) tan δ.
©2020 IEEE.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

.

.

.

.

Figure 5.9 Measured small signal S-parameters of the AJP packages compared to the
published bare die [1] and COTS packaged die [2]: (a) S11, (b) S12, (c) S22,
and (d) S21. ©2020 IEEE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
. . . . . . . . . . . . . . . . . . . . . . . 118

Figure 5.10 Measured 2x through. ©2020 IEEE.

Figure 5.11 Packages with loss of 2x through removed compared to bare die performance

[1]. ©2020 IEEE.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 118

Figure 5.12 Large signal load pull measurements of the AJP package at 8 GHz:

(a)

package 1 and (b) package 2. ©2020 IEEE. . . . . . . . . . . . . . . . . . . . . 121

Figure 5.13 Large signal load pull measurements of the AJP package at 9 GHz:

(a)

package 1 and (b) package 2. ©2020 IEEE. . . . . . . . . . . . . . . . . . . . . 122

xiii

Figure 5.14 Large signal load pull measurements of the AJP package at 10 GHz: (a)

package 1 and (b) package 2. ©2020 IEEE. . . . . . . . . . . . . . . . . . . . . 123

Figure 5.15 Large signal load pull measurements of the AJP package at 11 GHz: (a)

package 1 and (b) package 2. ©2020 IEEE. . . . . . . . . . . . . . . . . . . . . 124

Figure 5.16 Deviation in performance with power and temperature cycles: (a) maximum
gain and (b) comparison between initial gain measurement and after 1,000
power cycles and after 1, 3 and 7 temperature cycles. ©2020 IEEE.

. . . . . . . 126

Figure 6.1 Annotated package design in millimeters. Dimensions in parenthesis are

target material thicknesses.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

Figure 6.2

(a) Exploded view of package illustrating fabrication procedure. (b) Illustra-
tion of the multi-leveled structure of the package design. (c) Illustration of
interconnect detail.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

Figure 6.3 Annotated fabricated package composite image.

. . . . . . . . . . . . . . . . . 134

Figure 6.4 Test setup.

.

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 136

Figure 6.5 Measured S-parameters (a) Transmission S21 (b) Reflection S11 and S22. . . . . 137
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Figure 6.6 Package loss estimate.

Figure 7.1

Initial samples of AJP microbumps. . . . . . . . . . . . . . . . . . . . . . . . . 148

Figure 7.2 AM microfluidic cooling: (a) cavity design for constant flow, (b) serpentine

design, (c) device under test. ©2017 IEEE. . . . . . . . . . . . . . . . . . . . . 150

Figure A.1 (a) Disassembled PA, (b) Assembled PA, (c) assembled print head.

. . . . . . . 157

Figure B.1 (a) SEM image of silver after sintering at 180°C, (b) EDS analysis of sintered

silver. .

.

.

.

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

xiv

KEY TO ABBREVIATIONS

AcD Aerosol co-Deposition
Ag Silver
AM Additive Manufacturing
AoP Antenna on Package
AJP Aerosol Jet Printing
BCB Benzocyclobutene
CoC Chip on Chip
COTS Commercial Off The Shelf
CPW Coplanar Waveguide
CPWG Coplanar Waveguide with Ground
CSP Chip Scale Package
CTE Coefficient of Thermal Expansion
Cu Copper
DMSO Dimethyl Sulfoxide
EDS Energy Dispersive Spectroscopy
ESD Electrostatic Discharge
FDM Fused Deposition Modeling
GaAs Gallium Arsenide
GaN Gallium Nitride
GSG Ground Signal Ground
HEMT High Electron Mobility Transistor
HMDS Hexamethyldisilazane
InP Indium Phosphide
LCP Liquid Crystal Polymer

xv

LNA Low Noise Amplifier
LRRM Line Reflect Reflect Match
LTCC Low temperature co-fired ceramic
McM Multi-chip Module
MG Maxwell-Garnett
MIM Metal Insulator Metal
ML Modified Lichtenecker
MMAJP Multi-material Aerosol Jet Printing
MMIC Monolithic Millimeter Wave Integrated Circuit
mm-wave Millimeter Wave
MoCu Molybdenum Copper
MPA Medium Power Amplifier
MS Microstrip
N2 Nitrogen
Ni0.5Zn0.5Fe2O4 Nickel Zinc Ferrite
NMP N-Methyl Pyrrolidnone
NRZ Non-Return to Zero
PA Power Amplifier
PCB Printed Circuit Board
PI Polyimide
PoP Package on Package
PRBS Pseudo Random Bit Stream
PVDF Polyvinylidene fluoride
PVP Polyvinylpyrrolidone
RF Radio Frequency
SEM Scanning Electron Microscope/Microscopy
Si Silicon

xvi

SiGe Silicon Germanium
SiP System in Package
SLA Stereo Lithography
SLM Selective Laser Melting
SLS Selective Laser Sintering
SoC System on chip
SOL Short Open Load
SOLT Short Open Load Through
SoP System on Package
SoW System on Wafer
TEM Transverse Electromagnetic
TRL Through-Reflect-Line

xvii

CHAPTER 1

INTRODUCTION

As demand for microwave and mm-wave electronics has risen, the need for novel packaging and
fabrication strategies has increased. Higher frequency electronics have important applications in
sensing and biomedical imaging as well as in high data rate digital electronics and communications
with the advent of 5G. Recent interest in mm-wave applications such as automotive radar and mm-
wave imaging and sensing have brought into focus new challenges in packaging and integration
of the required electronics for these applications. That is, a need for higher functional density,
heterogeneous integration (e.g. SiGe and GaAs integration), with low loss interconnects, and the
ability to monolithically integrate passive devices and circuitry. These new application spaces also
demand a manufacturing process that is adaptable and compatible with a wide arsenal of materials.
While digital electronics have historically followed very different design methodology from RF
electronics, data rates have become high enough in recent years that signal integrity concerns must
be addressed at every stage of the design process [3]. Prior to this more recent revelation, data rates
did not correlate to wavelengths short enough to merit more than a list of simple design rules which,
if followed, ensured a working channel. Furthermore, digital electronics require broad bandwidth
interconnects illustrated in Figure 1.1 and [4]. That is, digital and analog design converge as speeds
and frequencies increase.

1.1 Current Packaging Strategies

Since the advent of semiconductor technology, a major factor limiting performance of systems and
devices has been the packaging of the semiconductor technology. While a lot of focus has been
placed on the speed of circuits alone, especially for digital electronics, there has been a parallel
push in packaging research for increased functional density. Out of this focus has come a series of

1

a)

b)

Figure 1.1 Example of a MATLAB-generated digital pseudo random bit sequence (PRBS) signal
spectrum with Gaussian white noise added: a) first 200 bits of NRZ PRBS15 signal (out of
215 − 1), b) frequency spectrum of NRZ PRBS15 signal

packaging innovations which facilitate the integration of functional blocks of circuitry. The idea
of System-on-Chip (SoC) or, an entire functional block fabricated on a single semiconductor has
been part of this broader research focus. SoC demonstrations have been shown in many computing
applications, in particular combining memory and logic circuits as discussed in [5] and for data
processing as in [6]. There have also been demonstrations of microwave, mm-wave, and sub-
millimeter wave SoCs, which especially have application in THz circuitry, where there is a lack of
chip to chip or chip to package interconnect methodology which is suitable for those frequencies.
So, at THz frequencies, a designer has no option other than to monolithically integrate a front
end transmitter or receiver on a single chip. Such systems have been demonstrated by [7] who
showed a transmitter and receiver including on-chip antennas operating between 160 GHz and 1

2

System-on-

Chip
(SoC)

System-in-
Package
(SiP)

System-on-Wafer
(SoW):
Silicon carrier (Wafer
level CSP)

3D Integration
-Chip-on-Chip (CoC)
-TSV interconnects
-semiconductor pro-
cesses

Multi-Chip-Modules
(McM):
ceramic/organic
/laminate carrier
(Conventional CSP)

3D Integration
-Package-on-Package
(PoP)
-PCB processes

System-on-
Package
(SoP)

AM Enabled SiP/SoP
-Expanded materials
-3D processing
-Application Specific
Packages

Figure 1.2 Evolution of functional block innovations working towards higher functional density in
packaging

THz. Additionally, [8] demonstrated a 77 GHz SoC receiver. SoC limits a designer to use a single
semiconductor technology, limited by the most demanding sub-circuit, even if other parts of a
functional block do not require it. For many applications this makes SoC technology prohibitively
expensive.

For example, a radar front end which may include lower frequency amplifiers and oscillators
as well as higher frequency mixers and amplifiers, high power requirements for transmitting, low
noise requirements for receiving, and large areas for antennas, capacitors, and filters are less
practical to fabricate as a SoC when it is preferable to fabricate low frequency circuits on cheaper
silicon technology and higher frequency and power circuits on III-V semiconductors such as InP,
GaAs or GaN. Furthermore, it is often not practical to fabricate other microwave structures on
semiconductors such as antennas or filters. As an example, the works previously mentioned, [7, 8]

3

both demonstrated their work on SiGe BiCMOS technology. Other processes would provide better
performance, but are cost prohibitive. These limits are physical limits due to the materials used. The
electron mobility, which limits fmax of silicon is 1350 cm2/Vs at 300 K , and the electron mobility
of Ge is 3900 cm2/Vs at 300 K . GaAs however has an electron mobility of 8500 cm2/Vs at 300 K
[9]. Furthermore, while mobility of SiGe is higher than pure Si, the band gap of SiGe is reduced
relative to pure Si, leading to a higher noise figure of circuits. The caveat being that there have been
demonstrations of mm-wave SiGe technology for low noise applications using cryogenic cooling
as discussed in [10], and performance is also influenced by defects in manufacturing processes. If
improved technology for heterogeneous integration existed for mm-wave applications, then it would
not be necessary to go to such lengths as integrating cryogenic cooling to stretch the performance
of less expensive technologies.

A System-in-Package (SiP) is meant to address this challenge and increase functional density
of circuits. This has been accomplished primarily through two avenues: first using a ceram-
ic/laminate/organic carrier as in [11, 12, 13], and second with a semiconductor (usually silicon)
carrier as a way of connecting die within a package as in [14, 15, 16] and others. These carriers
replace a lead frame in a package and may act as a re-distribution layer, or die may have a separate
redistribution layer fabricated via a lithography processes on the die itself, often using benzocy-
clobutene (BCB) polyimide (PI) or similar materials. Some have referred to using a silicon carrier
as System-on-Wafer (SoW) as in [14]. These chip-scale-packages (CSP) evolved from a single die
to many, i.e., multi-chip-modules (MCM), a term which has been broadly applied to describe any
module containing multiple die. These SiP implementations additionally may include other passive
components on the carrier such as capacitors and resistors as in [12]. In the case of semiconductor
carriers, active elements may also be included on the carrier.

Each of these solutions have advantages and may be practical for different applications. SoW
is more limited in materials than using a conventional (ceramic, laminate, or organic substrate)
carrier which makes SoW less appealing for microwave/mm-wave circuits. SoW generally creates
a better coefficient of thermal expansion (CTE) match between the chip(s) and the carrier. For

4

this reason, conventional carriers are an ongoing area of research, investigating materials such
as Low Temperature Co-fired Ceramic (LTCC) which has a good CTE match to semiconductor
technologies but is difficult to reliably manufacture since the material shrinks unpredictably during
processing [12, 17]. Liquid Crystal Polymer (LCP) is another material which has received a lot of
research interest as a potentially low cost and low loss material [18, 19].

The next step in increased functional density from these categories of SiP contained on a
carrier is the stacking, or 3D integration, of die and/or packages. The 3D integration of die, often
using Si interposers/redistribution layers have been referred to as Chip-on-Chip (CoC) technology.
The corollary to CoC is Package-on-Package (PoP) technology in which packaged die are stacked
to create higher functional density and a more integrated system. Many of these strategies are
discussed in [20, 15, 21].

Variants, and hybrid processes of two or more of these strategies have also been proposed or
demonstrated for various applications. These state-of-the-art packaging technologies still leave
significant challenges for system designers and integrators to overcome, especially for microwave
and mm-wave circuits in which material selection and the size of circuits are particularly important.
These processes also all use sequences of lithography, etching, and varied semiconductor, printed
circuit, and clean room processes.

The cost of designing a sub-system integrated with these technologies can be prohibitive based
on available equipment or technology and requires communication and coordination across the
globe. These processes require infrastructure that is available to very few, and processes which
require the copious use of dangerous chemicals for etching and cleaning.

The term System-on-Package (SoP) is often used synonymously with SiP as in [19], and even
when it is distinguished from SiP, it requires a much narrower definition of SiP than is more
commonly used in literature. [22], who introduced the term SoP, limits the definition of SiP only
to “the vertical stacking of similar or dissimilar ICs”. Other uses of the terminology SoP have been
applied to packaging which integrate elements not conventionally included in a package such as an
antenna (i.e. an Antenna-on-Package (AoP)). Additionally, some use SoC to refer to 3D integration

5

a)

b)

Figure 1.3 Typical in-package microwave interconnect methodology: a) bond wire/ribbon, b) flip
chip

of chips with semiconductor interposers, i.e., CoC.

Current research in packaging for mm-wave applications consists of refining and adding to
conventional processing techniques, such as research in glass packaging technology [23, 24, 25,
26, 27] and LTCC packaging [28, 29, 30, 31, 32, 33], both of which aim to apply semiconductor
processes to glass or LTCC materials to integrate semiconductor devices.

The additive manufacturing packaging solutions proposed here use many of the same materials,
but with a process that is far more flexible than semiconductor processes so packages can be
application specific. These additive processes can also supplement these conventional techniques
since they may provide expanded functionality and materials not achievable with the aforementioned
processes.

1.2 Performance Bottlenecks

1.2.1

Interconnects

Internal to these packaging strategies, a primary bottleneck for performance as well as density
is interconnect technology. Conventional packaging technology typically relies on bond wire
interconnects to connect die to a lead-frame or chip carrier. Bond wire interconnects present
a significant design challenge for microwave engineers as they will typically introduce a large

6

Figure 1.4 Aerosol jet printing enabled SiP/SoP component by chapter.

inductive discontinuity in the signal path. Many strategies have been published on successful
implementation of bond wire interconnects through the W-band, but these solutions are for the
most part narrow band and require extra design work to tailor to the specific application such as in
[34, 35, 36]. Bond ribbon provides typically better performance [35] than bond wire, but fails to
entirely mitigate the problem.

Flip chip technology, as is more typically used in state of the art packaging, especially for 3D
integrated packages as discussed previously, is a wide band high frequency capable interconnect
methodology [37], but is inflexible and expensive to implement. Flip chip prevents effective
inspection of parts and makes it difficult to detect faults, making them prone to yield issues and
not conducive to rework. Fabricating micro-bumps to enable flip chip implementations also adds
extra steps in the packaging process. The surface planarity required for flip chip is difficult to
maintain on conventional, that is, not semiconductor, chip carriers which would often be preferred
for microwave applications. Thus, there is a need for novel interconnect methodologies that are
flexible and straightforward to implement for designers which maintain the functional density of
state-of-the-art packaging solutions. An illustration of these interconnects is shown in Fig. 1.3.

7

1.2.2 Materials

The availability of materials which can be integrated into state of the art packaging strategies is also a
major limiting factor in package performance. Materials must be deposited either by semiconductor
processes, that is, growing oxides or thin film deposition methods, and successive lithography and
wet or dry etching techniques. That is, for these subtractive methods, materials need to be photo-
definable or etched, and many thin film techniques require specialized equipment and operators.
Furthermore, these processes create a lot of material waste, and relatively large quantities of often
toxic and dangerous chemicals must be used. Since these processes are layer based, everything
is fabricated in two-dimensional space, making it difficult to fabricate more complicated (three
dimensional) structures without micro-machining, or many successive lithography steps. This also
can create redundant unused space in layers which have only one purpose.

Material gradients are also difficult or impossible to fabricate with conventional methods. The
range of materials compatible with these strategies is limited by what can be processed with
lithography. Therefore, composites can be difficult to integrate and selectively pattern.

1.2.3 Flexibility

Designs using lithography processes can easily scale fabrication quantities to large volumes, how-
ever changes in a design can be cost prohibitive, so application specific packages are not feasible.
This fact has given rise to MMICs which are designed to fit wide application spaces. That is, these
components are designed to be wide band, but are often used in narrow band systems which requires
additional filtering, and sacrificing performance for the significant cost of designing application
specific MMICs. This ultimately adds unnecessary complexity to these systems and makes it cost
prohibitive to produce small numbers of a particular system design. There is a need then not only for
rapid prototyping capability, but also flexible packaging strategies which allow for design changes
tailored to an application. These systems may be less expensive and have the potential to perform
better. This capability would also allow for much wider accessibility to packaging technology if

8

it is not prohibitively expensive, as it presently is using conventional methodology which requires
expensive clean rooms and equipment as well as dangerous chemical processing.

1.3 A Proposed Alternative Packaging Strategy

What this dissertation proposes is additive manufacturing as a means of integrating and packaging
systems as well as a supplement to conventional packaging techniques. Using AM, many of
the shortcomings of conventional packaging strategies can be resolved, and the capabilities and
flexibility of packaging expanded. This opens the door to application specific packaging where a
system can be easily adapted to its application.

Additive manufacturing has been demonstrated for the manufacture of basic mm-wave and
RF electronic components by [38, 39] and others. This previous work has focused mostly on
passive structures such as antennas, waveguides, and filters. AM is an attractive technology for
the manufacture of these structures because of its inherent flexibility and the ability to build up
structures in three dimensions with a wide range of materials. AM additionally allows for material
patterning without the use of conventional lithography, so processes require fewer of the chemical
and specialized processes associated with state-of-the-art printed circuit board and semiconductor
fabrication.

To the best of my knowledge, the first demonstration of additive manufacturing, specifically
three dimensional stereolithography, was by [38] who demonstrated a helical antenna mock up, a
high Q cavity resonator, as well as two and four pole filters, all operating at approximately 19 GHz.
They achieved an unloaded quality factor of over 3000 in their resonator. Since then there have
been many other demonstrations of passive RF components as well as some measure of integration
into larger systems.

Work since [38] has included many functional antennas such as [40, 41, 42, 43, 44, 45, 46] which
also include demonstrations of antennas which would be challenging to fabricate with conventional
methodology. Many other sub-components of RF and mm-wave systems have been demonstrated

9

including many varieties of waveguides and transmission lines, based on both planar and non-planar
structures as in [47, 48, 49, 50, 51, 52, 53, 54]. Filters, couplers, sensors and other basic microwave
circuit elements have been demonstrated up to THz frequencies [55, 56, 57, 58, 59, 60, 61].
Similarly, dielectric lenses have been shown as in [62, 63, 64] as well as passive components like
capacitors and inductors as in [65, 66, 39].

While there has been a lot of work published on these basic canonical building blocks of
microwave electronics, there has been relatively little work using AM as a means of integration, i.e.
including in these AM systems active components. Even less work has been shown for microwave
electronics using bare die (unpackaged) devices.
[61, 67] in the same year published the first
demonstrations of a fully additive packaging and interconnect technique with commercial-off-the-
shelf (COTS) die. In [68] I expanded upon my previous work in [61] to show improved interconnect
performance, also reported later in this dissertation.

Other similar demonstrations of AM enabled hybrid processes for die interconnects have been
shown by [69, 70] who both used processes similar to flip chip technology to create something like
a redistribution layer which was metalized by AM. [71] also showed a similar interconnect where a
die was embedded in a polymer and interconnects formed using AM. All of these demonstrations of
bare die integration with AM have been accomplished with so-called direct write technologies such
as aerosol jet printing or inkjet printing. More conventional AM technologies like stereolithography
(SLA), fused deposition modeling (FDM) and selective laser sintering (SLS) (sometimes selective
laser melting) are not presently capable of the printing resolution required to form such fine
interconnects on bare die components, and are also more limited in material compatibility compared
to their direct-write counterparts.

Material availability, as discussed above, is an important bottleneck for both conventional and
additive processes for packaging and integrating microwave systems. This dissertation will therefore
demonstrate a new process not previously applied to microwave electronics for creating composite
materials called multi-material aerosol jet printing (MMAJP) wherein multiple materials may be
mixed in place allowing fine control of material properties during manufacture. This allows for the

10

creation of material gradients which are impractical or impossible to fabricate using conventional
methodology. Prior demonstrations of material gradients for microwave circuits using AM have
relied on changing a fill factor of material which only allows for a lower dielectric constant than the
stock material as in [72, 64]. Demonstrated later in this dissertation as well as in [73] are material
gradients formed by mixing high dielectric ceramic materials like BaTiO3 to create high dielectric
constant composites.

The intent of this dissertation is to demonstrate the basic building blocks of a complete packaging
strategy. This body of work shows that AM may provide a viable alternative or supplement to
conventional packaging techniques for microwave electronics. These new processes may allow for
presently unexplored strategies for system integration as well as opening the door to new kinds
of microwave circuits not possible to fabricate using conventional techniques. Previous work
on the use of AM in microwave electronics has focused on passive structures such as antennas,
waveguides, and filters. This dissertation expands upon that work by reporting on strategies
for integrating active components and processes which expand the selection and functionality of
materials available for packaging. This toolbox of techniques allows for the printing of application
specific heterogeneous packages with performance that in many cases surpasses conventional
methodology. These strategies also carry with them all of the well-known benefits of AM, namely,
fewer chemical and specialized processes, and the ability to rapidly prototype components. All of
the work demonstrated in this dissertation uses only two major pieces of equipment for fabrication:
an aerosol jet printer and an inert gas oven for processing. After presenting the basic structure
and building blocks of this packaging process, several complete packages are demonstrated as well
as what the future of this line of research might look like. An illustration of the AJP chip-first
packaging process by chapter is shown in Fig. 1.4.

11

CHAPTER 2

A CHIP-FIRST APPROACH TO MILLIMETER WAVE PACKAGING

To demonstrate this packaging strategy in its most general and basic form, this chapter shows
packaged COTS Si [61] and GaAs 0-dB [68] attenuators. The packages are entirely aerosol jet
printed and demonstrate broadband interconnects operating up to mm-wave frequencies. This work
was some of the first fully additively manufactured packaging demonstrations for microwave and
mm-wave applications.

I chose 0-dB attenuators to allow the interconnect performance to be isolated. The first of these
packages consisted of a Si die and the second, two GaAs die. The silicon package is demonstrated up
to the operating frequency of the attenuator. The GaAs packages are demonstrated beyond the rated
operating frequency of the die, and improved on both the bandwidth and loss of the interconnect
shown with the Si package. These packages also served as a development vehicle for this packaging
process. The primary purpose of these packages is proving the efficacy of printed interconnects, as
this is one of the most significant performance bottlenecks in conventional packaging techniques.
This technique employs a chip-first approach to packaging, where die and other components
are positioned and a package and interconnects are subsequently built up. The rising demand for
mm-wave electronics for wide ranging applications including automotive radar, imaging, sensing,
and communications hold in common the need for high frequency and wide bandwidth interconnect
strategies.

Conventional interconnect strategies typically rely on wire bonds or ribbon bonds between die
placed in a cavity. As discussed in Chapter 1, these interconnects introduce a large inductive
discontinuity which must be compensated for, usually with capacitive stubs as in [36, 34, 74].
These narrow band compensation methods are not effective for applications which may require
wider bandwidth operation, such as imaging.

Flip chip technology is also problematic as it is not typically mechanically robust and requires

12

additional design work to prevent microwave circuits on the die from being loaded by the substrate.
Flip chip interconnects also require additional mechanical constraints like surface planarity and
typically underfill, which worsens detuning of the die. Die cannot be visually inspected after
fabrication, leaving them prone to yield issues.

At 20 GHz, the Si die interconnect loss was measured to be 0.22 dB. The Si die has a published
operating range to 26.5 GHz. The interconnects to the GaAs die achieve a worst case loss of
0.290 dB at 40 GHz. Beyond the rated operating range of the device, which is from DC to
43.5 GHz, the interconnects achieve a worst case loss of 0.490 dB at 60 GHz. The approach that
we present provides a flexible mm-wave capable and broadband packaging solution that enables
next generation System-In-Package (SiP)/System-on-Package (SoP) technology.

The prospect of packaging and integrating electronics using AM, and specifically so-called
direct write technologies such as aerosol jet printing (AJP) or inkjet printing, was demonstrated
by [75, 70]. As far as I am aware, prior to the work in this chapter, the only other demonstration
of an AM interconnect for microwave and mm-wave applications was by [76] up to 40 GHz on a
blank die. That interconnect however did not attach to a circuit on the die. Because of the unique
challenges in integrating microwave and mm-wave electronics, the distinction from other (low-
frequency) electronics is not trivial. The significance of the work presented in this chapter is that
these packages are some of the first demonstrations of fully AM microwave/mm-wave interconnects.
They also represent a significant improvement in the loss and frequency of prior work.

2.1 Design

While the design of both of these demonstration packages is simple, the design did evolve somewhat
from the packaged silicon die to the later gallium arsenide die. The basic structure common to both
is a printed low loss dielectric structure which supports a printed conductor, forming transmission
lines to the device and interconnects to the device. The intent of this strategy is to control
the characteristic impedance of the transmission line from the package to the die so there is no

13

impedance discontinuity which would limit the bandwidth of the interconnect.

With this strategy, for many COTS MMIC parts the largest impedance discontinuity is the bond
pad itself. Typically multiple wire bonds would be used to reduce the inductance they introduce, so
pads are made large to accommodate multiple bond wires. If the capacitance of the bond pad can
be estimated, then the interconnect region can be narrowed to compensate the capacitance of the
pad. Other design considerations are primarily mechanical. The adhesion of the printed material
to the substrate and to the die is critical.

Both of these designs use polyimide printed on a copper carrier. Later work used a molybdenum
copper alloy carrier to CTE match to semiconductor materials. Polyimide was used as the primary
dielectric for the following work, but as discussed later, a low CTE formulation of polyimide was
found to improve mechanical performance. In either case, polyimide was chosen because it can
withstand high temperatures and has a relatively low CTE compared to many other dielectrics used
in packaging such as BCB, or SU8. The CTE of polyimide is sufficiently matched to copper, but
I found that the low CTE of semiconductor materials leads to low yield problems due to cracking
around the die. The use of an adhesion promoter was also found to increase yield in future packages.
Of equal concern is the loss tangent of the dielectric material. PI has a typically low loss tangent
compared to SU8 in particular.

Another important consideration is the printed conductor thickness. Conductors should be thick
compared to the skin depth of the metal to prevent high losses. At 20 GHz, the skin depth of silver
(Ag), as was used for this work, is 0.448 µm. In both cases the copper carrier serves as the ground
reference for the microstrip transmission line. Both packages are designed to operate in a 50 Ω
environment and to be measured with GSG probes.

2.1.1 Silicon Package

The initial goal of the package strategy was to print polyimide to the height of the die. The Si 0
dB attenuator used was a YAT-0-D+ manufactured by Mini-Circuits, with an operational frequency
range of zero to 26.5 GHz. This die is 100 µm thick.

14

Figure 2.1 Silicon die package design (in units of mm): a) package length = 4.7, b) 50 µm wide
signal length (to edge of die) = 0.1, c) package width = 1.825, d) probe ground spacing = 0.325, e)
signal width = 0.225, f) probe ground width = 0.25, g) printed signal length = 1.9, h) printed gap
width = 0.05, i) probe ground length = 0.5, j) printed package substrate height = 0.1, k) die height
= 0.15. ©2018 IEEE.

For this design, a relative dielectric constant for the polyimide of 3.45 was used. A 50 Ω
microstrip transmission line is 0.225 mm wide on this substrate. In the interconnect region, 50 µm
from the edge of the device, the microstrip is narrowed to 50 µm. The microstrip extends 2 mm
beyond the edge of the die on each side, with the polyimide extending an additional 50 µm beyond
the Ag. The microstrip line is then 3.8 mm long and the printed structure is 4.7 mm long. Ground
straps from the surface of the PI to the Cu carrier are included to facilitate measurement with
GSG probes. The width of the printed structure is 1.1 mm wider than the die, for a total width of
1.825 mm. The designed structure is shown in Fig. 2.1.

2.1.2 Gallium Arsenide Package

While the Si package was successful, it was clear that there were several points that could be
improved on. First, the previous design did not showcase the frequency capability of this strategy,

15

so in this improved GaAs design a CPWG to MS transition was included to extend the frequency
range of the launch at the GSG probe landing locations. Also, the GaAs die was chosen because
it has a wider operational frequency range of zero to 43.5 GHz. We used a 0-dB GaAs attenuator,
KAT-0-DG+ manufactured by Mini-Circuits. Second, it was not necessary to print PI to the surface
of the die, so this design iteration targeted a 50 µm thick PI dielectric around the die, so a 50 Ω
microstrip transmission line is 108 µm wide using 
r ≈ 3.5 for PI. The GaAs die for this package
is the same thickness (100 µm) as the previous Si die. Finally, packaging a GaAs die demonstrates
heterogeneous compatibility of this general strategy

The CPWG to MS transition was designed as in [77]. A major difference between that prior
work and this is the inclusion of grounding “straps” to the Cu carrier, allowing operation from
DC to approximately 70 GHz. These ground connections take advantage of the natural slope that
forms on the edge of the printed polyimide from surface tension at the edge, which is sufficient to
print Ag at a normal incidence to make a connection between the Cu carrier ground reference and
the surface of the polyimide. The cutoff frequencies of the structure are determined by resonant
frequencies of the ground pads on either side of the signal. These cutoffs can be moved by adjusting
the shape (primarily the length and width on the polyimide surface) of the pads. With the grounding
straps, these dimensions do not significantly affect low frequency performance, but without those
DC connections to ground, the width and length of the ground pads would significantly impact
lower frequency performance since this transition, as outlined in [77], relies on creating an RF short
between the ground reference below the substrate and the grounds coplanar to the signal. Since
the operational frequency of the die is 43.5 GHz, when designing this launch it was not thought
to be necessary to extend the cutoff frequency beyond 70 GHz. As discussed later, it is possible
this structure could have performed adequately at higher frequencies extending into the W-band, at
least for the purpose of characterizing the interconnect strategy.

As before, the interconnect region is narrowed to compensate for the capacitance of the bond
pad. This discontinuity would typically be insignificant compared to the inductance introduced by
a wire bond. The length of the interconnect region is 0.06 λ long at 67 GHz, so it can be treated as

16

an inductive lumped element. This interconnect region was optimized in simulation using ANSYS
HFSS©, but the appropriate microstrip width can also be roughly calculated.

Treating the bond pad as a capacitive section of microstrip transmission line on GaAs, using

r = 12.9, and the pad width (125 µm) and length (100 µm), the approximate characteristic
impedance of this transmission line is Z0c = 38 Ω. At 40 GHz, the approximate capacitance of
this transmission line is C = ( βclc)/(ωZ0c, where β = 2π/λg. For this structure at 40 GHz
the guided wavelength, λg = 2.52mm. The printed transmission line, 48 µm wide on 50 µm
PI has a characteristic impedance of Z0i = 76 Ω, λg = 4.72mm and an approximate inductance
L = Z0i βili/ω. For a characteristic impedance of Z0 = 50 Ω, the appropriate length can be solved
for with

(cid:114)(cid:18)

(cid:114)

L
C

=

(cid:19)(cid:30)(cid:18)

(cid:19)

Z0i Z0c βili

βclc

Z0 =

which, solving for li simplifies to

Z2
0 βclc
Z0i Z0c βi

.

li =

(2.1)

(2.2)

This evaluates to 162 µm in this example. This is only slightly shorter than what was found to be
optimum in simulation (184 µm). This can only be a rough estimate since this does not account
for coupling into the die from the printed transmission line, and assumes the printed transmission
line is of constant impedance, when in fact this interconnect region lies on a ramp up to the die.
The exact capacitance of the pad cannot be known in most cases since internal circuits of the die
may not be known. This stepped impedance compensation is similar to how inductive bond wires
would be compensated for, but since the discontinuity being compensated for is so much smaller,
this approach allows for broader bandwidth performance.

An important observation about this strategy is that it does not require large bond pads on the
device, but COTS microwave and mm-wave components are not typically available with small bond
pads. The details of the design are shown in Fig. 2.2.

17

Figure 2.2 GaAs package design (in units of mm): a) package length = 3.7, b) printed signal
length = 1.25, c) inductive signal length = 0.184, d) CPWG to MS transition length = 0.1, e) probe
ground width = 0.505, f) CPWG ground to ground space = 0.107, g) inductive signal
width = 0.048, h) MS width = 0.108, i) CPWG signal width = 0.075, j) 2x through trace
length = 2.4, k) package height = 0.05, l) mounted die height = 0.125. ©2019 IEEE.

18

2.2 Fabrication

Rather than forming a cavity first, the die are placed first on a carrier and then the package and
interconnects are built up around them. This means that this approach is not sensitive to die
placement since the printer can align to the die to correct for poor die placement. Both the PI and
Ag were printed with an Optomec Aerosol Jet 5x printer for both packages. Because of the large
standoff distance of the printer, packages can be printed around devices even when their bonding
surfaces are at different relative heights. Because features smaller than 10 µm can be resolved with
AJP, this approach is also scalable to high density packaging.

For both package designs, the polyimide precursor ink for the dielectric structures is 5% wt.
polyamic acid in N-methyl-2-pyrrolidone (NMP) and the Ag ink is 25 % wt. Clariant Prelect TPS
50 + deionized water. The PI ink was aerosolized using a pneumatic atomizer and printed with
a 300 µm nozzle with an aerosol flow rate of 25 SCCM. The Ag ink was aerosolized with an
ultrasonic atomizer and printed with a 100 µm nozzle with an aerosol flow rate of 20 SCCM.

Both packages use a copper carrier. As is explored later in this dissertation, other substrates
would be acceptable provided there is a conductive path for signals (which may be printed). This
strategy is independent of the substrate/carrier so long as material compatibility is ensured.

2.2.1 Silicon Package

For the Si package, die attach was performed with 50 µm indium Ag film. This package was printed
by first printing PI with a 50 µm gap around the die with the print stage heated to 100° C. This PI
was printed in approximately 20 µm thick layers at a time and curing those layers before adding
more. Each layer is approximately 1 µm thick. Heating the print stage for the PI package allows
a thicker film to be printed since the polyamic acid dries as it is printed. Next, the gap was filled,
printing at 10° to normal with no print stage heating, allowing the polyimide to flow more freely
around the die.

Because of the additional thickness from the die attach, the surface of the die sits slightly above

19

Figure 2.3 Si package before printing Ag.

the surface of the PI. Filling the cavity then creates a small fillet or ramp around the device. Because
this device and many others include a metal seal ring around the device, PI was printed on the
surface of the die to cover this ring to prevent a short to the printed Ag. The package prior to
printing Ag is shown in Fig. 2.3.

Five layers of Ag were printed for a final conductor thickness of approximately 2 µm. On the
ramp to the surface of the die the stage was tilted to 10° to normal as before. The gap filling and
the last 20 layers of PI were cured simultaneously with the Ag. Layers were cured or imidized by
heating in nitrogen to 180° C for 40 min, then 280° C for 30 min, and finally 300° C for 10 min.
The polyimide surface roughness was measured to be 0.29 µm with a NanoMap-500LS Surface
Profilometer. The printed Ag sintered alongside the final layers of polyimide in nitrogen achieves
20% of the bulk conductivity of Ag (a resistivity of 8x10−6
Ω-cm). Conductivity was measured
with a Lucas Labs Pro4 four-point probe measurement. The packaged Si die can be seen in Fig.
2.4.

20

Figure 2.4 Finished Si package under measurement with inset detail of interconnect region.
©2018 IEEE.

2.2.2 Gallium Arsenide Package

Two GaAs packages of identical design were printed. Die attach was performed using Ablebond
84-1LMISR4 conductive epoxy. Unlike the previous package, polyimide was printed first around
the die at an angle 15° to normal, again covering the seal ring on the edge of the die surface on
an unheated stage. This was done first so that overspray would not interfere with the PI adhesion
to the die, and this PI was cured before proceeding with printing. The remainder of the package
PI was printed to a target thickness of 50 µm at an angle normal to the surface of the carrier with
the stage heated to 100 °C, and this PI was cured in one step before printing Ag. 23 layers of PI
were printed in total, averaging about 2.2 µm a layer. This was thicker than the PI for the previous
package because the ink used was replaced. The older ink used in the Si package was more viscous
due to solvent evaporating and the ink absorbing water, causing it to be atomized less efficiently.
This polyimide was cured at a temperature of 200 °C in nitrogen.

Three layers of Ag were printed on the cured polyimide. On a representative sample of cured
printed Ag ink on polyimide, we achieved 53% of the bulk conductivity of Ag (a resistivity

21

Figure 2.5 One of the two identically fabricated GaAs packages and 10x magnified detail of
interconnect region. ©2019 IEEE.

of 3x10−6
Ω-cm), measured with a Lucas Labs Pro4 four-point resistivity system. A higher
conductivity is achieved because the Ag ink used requires the presence of oxygen during sintering.
Since the Ag for the Si package was sintered in nitrogen, the oxygen was provided by the imidization
of the polyamic acid PI ink and so was limited and only present for part of the sintering process.
The silver for the GaAs packages was sintered in atmosphere so oxygen was present for the whole
sintering process. The thickness of the PI was measured to be 40.9 µm for package I and 51.04 µm for
package II. The difference is due to some amount of inconsistency (process drift) of the atomization
of the PI ink. The surface roughness of the PI was measured to be 0.5 µm and 0.3 µm, respectively.
The silver was measured to be 5.1 µm and 5.2 µm thick with a RMS surface roughness of 0.7 µm
and 0.4 µm, respectively. One of the two fabricated parts is shown in Fig. 2.5.

2.2.3 Fabrication Challenges

The primary failure mechanism observed during the fabrication of these packages was poor adhesion
of polyimide to the die. The stress on the PI film while drying and curing caused it to wrinkle,
crack, and pull away from the die as shown in other packages in Fig. 2.6. This is the reason
for curing sequentially 20 µm layers at a time in the Si package. In work presented later in this
dissertation, this process was improved by controlling the temperature gradient during the curing

22

a)

b)

Figure 2.6 Examples of package failures due to a high temperature gradient: a) cracked PI which
additionally pulled away from a die, in this case a GaAs amplifier, b) wrinkled PI.

process and by using an adhesion promoter. The details of these improvements are discussed in
future chapters.

2.3 Results

All measurements were taken with a MPI TS150-THZ Probe System and a Keysight N5227 PNA.
All simulations were performed with Ansys HFSS©.

2.3.1 Silicon Package

After line-reflect-reflect-match (LRRM) calibration, the reference plane of the measurements is at
the tips of the ground-signal-ground (GSG) probes. The measured and simulated S-parameters are
shown in Fig. 2.7.

These results show the simulation and measurement of the entire printed package including the
1.9 mm microstrip leading up to the die on each side. Within the published frequency range of the
attenuator the interconnect performs well and correlates well with simulations. Deviations between
simulations and measurements are due to mismatch in the microstrip impedance. The performance
of the device limits the interconnect as well as the performance of the probe launch. At 20 GHz, the

23

a)

b)

Figure 2.7 Measured and simulated scattering parameters of the Si package compared to the
unpackaged bare die performance: a) transmission, b) reflection. ©2018 IEEE.

measured loss of the package is 0.65 dB. The loss of the attenuator at this frequency is 0.26 dB, so
the loss of the printed components (transmission lines and interconnects) between the two reference
planes of measurement is 0.44 dB. This correlates to a line loss of 0.12 dB/mm at 20 GHz including
both of the interconnects, and 0.22 dB insertion loss for each individual interconnect plus 1.9 mm
transmission line leading to the device.

24

2.3.2 Gallium Arsenide Package

After short-open-load-through (SOLT) calibration, the reference plane of the measurements is at
the tips of the ground-signal-ground (GSG) probes. The measured and simulated S-parameters are
shown in Fig. 2.8. Both packages performed similarly with reflections ≤-10 dB to beyond 55 GHz
but it is clear that one of the two packages performs slightly better. The reason for this is a greater
impedance mismatch in Package I due to a slightly thinner dielectric than designed for. The surface
roughness of the printed conductors is likely the reason for higher losses with increasing frequency
than predicted by simulation.

21 = (Sm

21 − St

21, Sd

21 − Sd

Using the simultaneously fabricated 2x through calibration line and the measured unpackaged
die, the loss of the interconnects alone is determined by Si
21)/2. Where Si
21,
21, St
21 represent the loss of a single interconnect, the measured loss of the entire package,
Sm
the measured loss of the 2x through calibration line, and the measured loss of the die, respectively.
Fig. 2.9 shows the loss of the interconnects alone. The effective length of each interconnect is
0.234 mm, found by the printed signal length minus half of the 2x through length plus the inductive
signal length. The difference in loss between the two packages is less than 0.62 dB over the range
of measurement, and less than 0.28 dB over the operating range of the attenuator. At approximately
65 GHz, above the operating frequency of the device, non-passivity errors exist in the calculated
loss with magnitudes ≤0.07 dB.

These errors can be accounted for by differences in the magnitude of the reflection between the
calibration trace and the measured packaged device at these frequencies. The losses of the inter-
connects presented here are significantly better than previously published work. The interconnect
loss of this work at 20 GHz including the transmission line is comparable to the loss in [61] at 20
GHz, 0.289 dB and 0.209, but the package presented here is 1 mm shorter. This work extends the
usable frequency range of our previously published work from 20 GHz to 67 GHz.

These results improve on the Si package by shortening the interconnect region, improving the
probe launch, more carefully compensating for the bond pad capacitance, and with a thicker higher

25

a)

b)

Figure 2.8 Measured and simulated scattering parameters of the of the two identical measured
GaAs packages, the simulated performance of these packages, the measured unpackaged
attenuator, and the printed 2x through line: a) transmission, b) reflection. ©2019 IEEE.

conductivity silver layer. A thicker silver lowers loss from the skin effect. A comparison of reported
AM mm-wave, and Ku to Ka band interconnect performance is shown in Table 2.1.

26

Figure 2.9 Calculated loss of a single interconnect in the GaAs packages (0.234 mm long). ©2019
IEEE.

2.4 Conclusions

This chapter establishes the efficacy of aerosol jet printing for forming die interconnects which
outperform conventional wire bond interconnects and demonstrates the most basic form of this
chip-first packaging strategy with AM. This chapter also shows that this process is compatible with
heterogeneous integration, showing both Si and GaAs package examples. These results also show
interconnects operating at mm-wave frequencies with 200% fractional bandwidth. These package
demonstrations also significantly improve on prior similar work. Future chapters will address some
of the fabrication challenges mentioned in this chapter.

27

Table 2.1 Comparison of AM High Frequency (Single) Interconnects

Craton et al. Craton et al. Tehrani et al. Tehrani et al. Lomakin et al.

AM Tech.

Die

[68]
AJP

[61]
AJP

COTS Si
-0.22 dB

[67]
Inkjet

COTS GaAs

COTS GaAs
-0.146 dBa
-0.290 dBa
-0.490 dBa

S21 (20 GHz)
S21 (40 GHz)
S21 (60 GHz)
aWorst case of two samples, bMay include PCB loss of die stand-in
©2019 IEEE.

-0.5 dB
-4 dB
n/a

n/a
n/a

[76]
Inkjet
Blank Si
≈ -1.5 dB
≈ -1.6 dB

n/a

[78]
AJP
PCB

≈ -1.3 dBb

n/a
n/a

28

CHAPTER 3

MULTI-MATERIAL AEROSOL JET PRINTING (MMAJP)

3.1 Introduction

3.1.1 Background

To address the performance bottlenecks in microwave electronic packaging, we will need new
technologies and fabrication strategies for System-in-Package (SiP)/System-on-Package (SoP) that
allow for more integrated systems and higher functional density. Part of this challenge is integrating
a broad range of material properties serving various functional purposes in a package. Conventional
packaging materials are limited to those which can be deposited with thin film techniques and
typically need to be etched in order to be patterned. It is desirable to have access to materials both
with high dielectric constants as well as low dielectric constants that can be deposited next to each
other with both abrupt and smooth transitions between regions. Related to the need for so-called
tunable electronic materials, composites have been an active area of research. In particular, barium
titanate (BaTiO3) based materials are of interest due to its high relative dielectric constant (
r),
between 500 and 7000 [79, 80]. 
r of BaTiO3 can change as a function of crystal orientation,
preparation, and of the temperature it is measured at. High dielectric materials and composites
are important for fabricating components such as capacitors. It is preferable for many microwave
circuits to be designed on high dielectric substrates as well.

Beyond high dielectric composites, nanocomposites of other materials with useful properties
to microwave circuits such as barium strontium titanate (BST) [81, 82] or other materials whose
dielectric constant can be electrically tuned are important areas of research. Magnetic nanomaterials
based on iron, nickel, cobalt, MnFe2O4, or other ferrites as in [44, 83] also have applications in
microwave circuits.
In addition, conductive composites have many applications in sensing and

29

for fabricating resistors. There is therefore a need for methods of integrating and patterning these
materials in a package.

Many of these composites are fabricated in a polyimide (PI) based polymer matrix. PI is a
relatively low loss material and can withstand particularly high temperatures. Other researchers
have successfully demonstrated PI and BaTiO3 nanocomposite films with as high as 90 vol. %
ceramic loading [84, 85]. Other work has demonstrated these composites with materials such as
polyvinylidene fluoride (PVDF) [86], thermoplastics [82] as well as epoxy [87, 88].

3.1.2 Comparison with Conventional Fabrication

In these prior works, nanocomposites are fabricated by mixing components prior to deposition
and patterning via subtractive processes as in [88, 89]. These processes can be wasteful and time
consuming, particularly if many mix ratios are required. In this chapter, first presented in [73],
I present the development of an alternative to these processes for the fabrication of functional
electronic materials at microwave/mm-wave frequencies. This alternative, Multi-Material Aerosol
Jet printing (MMAJP), relies on dynamic mixing of polymer-ceramic composite materials during
the AJP process. By mixing composites in place, the ratio of materials can be tailored on-demand
during fabrication allowing for both smooth gradients and abrupt material changes. Since deposition
is accomplished with AJP, there is no additional need for patterning the composites. Additionally,
this method requires fewer active fabrication steps and pieces of equipment than would be required
with conventional methodology.

Using conventional manufacturing strategies, fabricating gradients in particular would require
diffusion processes or time and material intensive successive layering and patterning of gradient
steps. These processes are not efficient or as flexible as the strategy presented in this chapter which
allows for material gradients in three dimensions. The process I present in this chapter has many
unique applications in microwave and mm-wave electronics, optics, and electronics packaging.
Material gradients in microwave and mm-wave structures have not been thoroughly explored due
to the lack of methods to create them. Microwave circuits like filters or other periodic structures

30

could benefit from the ability to adjust material properties in such a flexible manner. Stepped
impedance filters which typically rely on conductor dimensions could be miniaturized using such a
MMAJP process where dielectrics can be tuned as well as conductors. Another possible application
of MMAJP is to tune the mechanical properties of materials such as the coefficient of thermal
expansion (CTE) as in [90]. CTE matching is particularly important in electronics packaging since
semiconductor materials have characteristically low CTE.

To demonstrate MMAJP in this chapter, I show PI and barium titanate (BaTiO3) nanocomposite
films with variable levels of ceramic loading. The intent of these nanocomposites is to increase the
relative dielectric constant of the material with an increase in the ceramic loading of the PI. This
characterization is done with fully AJP ring resonator circuits with patterned dielectrics as well as
conductors. Two ring resonator circuit designs are used–the first designed for a first resonance in
the X-band below 12 GHz, and the second for a first resonance in the W-band below 110 GHz. MIM
capacitors are used to characterize the materials electrical properties at lower frequencies. Both the
dielectrics and conductors for these circuits are deposited and patterned with AJP, showing three-
dimensional structures patterned without the use of photosensitive materials. The films presented
in this chapter show a range of relative dielectric constants from 3.1 to 8.9. To illustrate the
unique capabilities of this manufacturing process, I also show in this chapter examples of smooth
printed material gradients. Finally, I show processing strategies for improving the electrical loss
characteristics of the printed composites–this also provides insight into the loss mechanisms of
these materials. With these results, I estimate the ceramic loading in the composites. Additionally,
using scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS) I
characterize the dispersion of the nanoparticles as well as the ceramic content of the composite
films.

This flexibility in composite fabrication offered by MMAJP is not possible with conventional
processes and allows for structures not possible or not practical to fabricate otherwise. Aerosol
co-deposition (AcD), a modified form of Aerosol Deposition Method (AD or ADM), is a similar
strategy which has been used primarily to fabricate ceramic films. In these processes, composite

31

films are made from dry powders of pre-mixed components, and deposition is not selective. In
most previous demonstrations of AJP nano-composites, as in [65, 91], composites are pre-mixed
before deposition.

Prior to the work in this chapter, [92] published a similar strategy using two pneumatic atomizers
to fabricate yttria doped stabilized zirconia and lanthanum manganite for solid oxide fuel cell
electrolyte and cathode layers. After that, [91] showed a conductive polymer matrix nanocomposite
of PI and carbon nanotubes using a pneumatic and ultrasonic atomizer with a static mixer, however
no details of the mixer were given. To the best of my knowledge, the work in this chapter is the first
demonstration of any similar strategy for microwave applications, as well as the first demonstration
of a polymer matrix ceramic nanocomposite fabricated by MMAJP. Additionally, this is the first
microwave/mm-wave material characterization presented with MMAJP materials.

The polymer-matrix ceramic nanocomposites in this chapter are demonstrated with on-demand
mixing ratios during the AJP process. This demonstration shows a practical implementation of
MMAJP for functional patterned mm-wave structures and material characterization. Dielectric
films are patterned in the x-y and z dimensions without photosensitive materials as would be
required with conventional processes [88, 89].

3.1.3 Applications

This process could be applied to many microwave and mm-wave components which take advantage
of changes in the electrical characteristics of a material or wave guiding structure. For example,
thin film lenses or spatial filters made from layered media could use an approach like this to deposit
selectively and control material properties such as in [93, 94]. This method could be employed to
fabricate a Luneberg lens or any other lens which requires a gradient of 
r as described in [95]. [96]
describes a photo lithography process for the fabrication of thin film lenses which involves several
steps of masking and etching, but these same lenses could be fabricated as they have described,
or even embedded in another material, using the process we describe here. Dielectric loaded
antennas would also benefit from the use of a process such as this. Previously reported styles of

32

dielectric loaded antennas are described in [95] including strategies such as what was used by [97]
or [94]. There is also clear application in periodic structures such as [98, 99, 100]. There is clear
application of MMAJP in structures like stepped impedance filters or dielectric based substrate
integrated waveguides. In these potential application examples, conventional manufacturing tech-
niques usually require a series of photolithography processes, etching, diffusing, micro-machining
etc., whereas using the process we present here, similar structures may be fabricated in a monolithic
fashion with fewer steps and more control over material properties.

MMAJP can be used as a supplement to conventional fabrication as well as a part of a fully
printed packaging strategy like that presented in Chapter 2 and in later chapters. This method could
be expanded to more than two materials and is not limited to polymer matrix nanocomposites. Any
two or more materials compatible with each other and with the AJP process could conceivably be
simultaneously mixed and printed. Some examples of other functional composites are shown in
Chapter 4. With this flexibility in materials it is conceivable to print microwave material properties

− µ− ρ similar to red-green-blue color printing where base materials are mixed to form composites
with desired electrical characteristics at a point in 3D space.

3.2 Theory of Operation

The goal of this process is to mix materials in an aerosol form rather than in liquid form, precluding
the necessity to formulate new liquid mixtures for every concentration. By mixing aerosols,
composites can be mixed in place as they are being deposited on to a surface during the AJP
process. In this work, two materials are mixed together, but more materials could be mixed at the
same time with modifications to this process.

Two or more aerosols of liquids or suspensions are generated by atomizers and carried through
tubes to a combining section. Fig. 3.1 shows this arrangement with two aerosols with associated
pressures, Pa or Pb and volumetric flow rates Qa or Qb. The atomizers do not need to be of a
similar configuration and the pressures and flow rates do not need to be close in magnitude to each

33

Figure 3.1 Representation of the aerosol mixing and deposition with critical metrics, where P, Q
and l are the pressure of aerosol A or B, the volumetric flow rate of aerosol A, B or C and the
length of the tube from atomizer A or B to the Y-fitting or from the Y-fitting to the print head.
©2020 IEEE.

34

other, though these variables do impact the resulting mixture.

For an AJP application, the tubes used are small in diameter, on the order of 1.5 mm, and
a limiting factor in the printing process is maintaining a laminar aerosol flow free of clogs with
minimal overspray after printing. Because of this, mixing should not be performed by creating
turbulent flows in a mixing area with baffles, folding structures or similar as these will speed up
the formulation of clogs in the tubes. The process presented here relies on advective/convective
mixing of the aerosols.

The aerosol flows can be shown to be laminar by calculating Reynolds number as Re =
(Qd)/(v A) where v is the viscosity of the aerosol, d is the diameter of the tube, Q is the volumetric
flow rate, and A is the cross sectional area of the tube. Using typical printing parameters, the
volume fraction of the nitrogen carrier gas of the aerosol to the atomized ink is ∼ 106, so Re can be
approximated using the properties of the nitrogen carrier gas. For our configuration, Re is expected
to be on the order of 5 to 100, so the flow is laminar in nature. A Y-fitting is used to combine two
aerosol flows into a composite aerosol. At this point the aerosol is not well-mixed. The combined
aerosol with volumetric flow rate Qc = Qa + Qb then flows through a length of tube lc. Then, in
the print head of the aerosol jet printer, the combined aerosols are focused with a sheath gas of
N2. In this length of tube and in the focusing length, the aerosols mix through advection. When
the flow is confined during focusing the aerosols mix much more rapidly. This is analogous to a
hydrodynamic flow focusing mixer in the field of microfluidics as in [101, 102].

Advective mixing of the aerosols is expected to obey the relation J = −D∇c, where D is the
diffusivity of the aerosol and c is the concentration of the aerosol as a function of location. The
length over which mixing takes place is the striation length (in our case the length lc plus the length
of the focusing area) which is proportional to τ

D where τ is the mixing duration.

√

The Peclet number, Pe = (hQ)/D can be calculated for mass transfer, and the Stokes-Einstein
relation can be used to estimate diffusivity of the particles as D = kBT/(6πvr), where kB is
Boltzmann’s constant, T is temperature, v is viscosity, and r is the particle radius. It is clear that
mixing can be improved by heating the aerosol, which is not demonstrated in this paper, but could

35

be accomplished with heating coils around the combined aerosol tube. For a room temperature
aerosol of 1µm droplets in N2, D ≈ 2.84 × 10−11m2
/s. The required mixing length can then be
written as lmix = hPe.

Pe for an aerosol flow rate of 30 sccm and a tube diameter of 1.6 mm is ∼28, and so the mixing
length required is 45 mm. This estimate is fairly reasonable for the typical AJP setup, but aerosol
droplets can be as large as 5 µm in diameter; so for these larger particles, the required mixing length
can as much as 225 mm. This length is not as reasonable, and considering that particles of the
aerosol may recombine in the lengths la and lb forming even larger particles, this advective mixing
alone is not adequate. After the length of tube lc, in the print head of the aerosol jet printer the
combined aerosols are focused with a sheath gas of N2, and through this process the aerosols mix
much more rapidly. Physically, the effect is a decrease in the characteristic length the aerosol is
confined within, h, when it is focused by the sheath gas. In our application, we expect the variable
h to drop to approximately 10 to 70 µm from 1.595 mm. The focused and now well-mixed aerosol
can then be selectively deposited.

Two parameters that can be adjusted to improve the mixing of the aerosols are lc and the
focusing gas flow rate, Qs. Both parameters are limited however; lc is limited by the pressure drop
across the combination of tubes and fittings that can be tolerated by the atomizers. Qs is limited
by the maximum laminar flow rate through the nozzle which if exceeded will result in overspray of
the aerosolized ink; that is, poor printing resolution [103]. If Pa and Pb are close in magnitude to
each other than this is not as much of a concern. However, if Pa and Pb are different in magnitude,
then the pressure at the Y-fitting of the length of tube lc becomes much more critical. For example,
in this chapter, we use a pneumatic atomizer which results in a high pressure aerosol in order to
atomize more viscous liquids alongside an ultrasonic atomizer for which the rate of atomization
goes down as pressure goes up. In order to prevent the atomization rate of the second atomizer
from dropping too low, the pressure at the Y-fitting must not be too high. lc must be adjusted so
that sufficient mixing occurs before the print head, and so that the pressures at the atomizers are
within acceptable limits.

36

The resulting composite may not need to be very well-mixed depending on the application. For
example, a 100 GHz signal propagating in a dielectric 
r ∼ 9, with a gradient in dielectric constant
periodic every 100 µm may be acceptable since this is periodic only with λ/10. That is, electrically,
this material would behave as an isotropic media. The resulting film in this case may not be as
mechanically robust as a well-mixed composite however.

3.3 Design of Characterization Circuits

3.3.1 Overview

Each sample of the initial composites was characterized with two fully printed ring resonator circuits
and one parallel plate metal-insulator-metal (MIM) capacitor. Additionally, a single resonator was
fabricated for two additional samples which are described in more detail at the end of this chapter.
Larger ring resonators designed for a first resonance below 12 GHz, and smaller resonators for
a first resonance below 110 GHz. Higher order resonances allow for material characterization at
intervals of the first resonance, but this is limited by measurement capability and by the resonance
of the coupling structure. This collection of data points provides material characterization from
low frequencies with MIM capacitors up to the W-band. These circuits are shown in Fig. 3.2. Ring
resonators are the preferred method of material characterization in this demonstration because they
additionally serve as an implementation of a microwave circuit as well as an application of MMAJP
as a manufacturing strategy.

The ring resonator circuits include a conductor backed co-planar waveguide (CPWG) to mi-
crostrip (MS) transition similar to [77] but including grounding straps before the probe location as
in [68], Chapter 2, and in later chapters. This transition performs well from DC to W-band. The
substrate/carrier that parts were printed on serves as a ground reference for the transmission line
and capacitor structures. The ring resonator circuits are designed to operate in a 50 Ω environment.
The dielectric for the ring resonators was designed to be approximately 20 µm thick. The
closed-loop MS ring resonators are coupled into the circuit with an annular ring coupling arc

37

a)

b)

Figure 3.2 Top down view of the characterization circuit with dimensions in millimeters and
degrees showing both large and small ring resonators: a) PI nano-composite, b) silver ink. ©2020
IEEE.

38

Figure 3.3 Top down detail view of coupling structure (shown on small resonator) with the
location of the reference plane for measurements. ©2020 IEEE.

similar to what is described in [104]. Detail of this structure is shown for the W-band resonator
in Fig. 3.3. In material of 
r = 3.5, the wavelength at 110 GHz is λ = 1.468 mm. In 
r = 10,
the wavelength at 110 GHz is λ = 0.862 mm. At 110 GHz, the dielectric is thin relative to a
wavelength, so, a narrow coupling gap is required to couple enough of the signal into the resonator
to effectively measure. Without an annular coupling structure, the gap required would be too small
to fabricate using AJP. The minimum reliably manufacturable gap determined the length of the
annular coupling arc feeding the ring resonators.

Since both dielectric and metal layers are meant to be printed, the dielectric is patterned as
shown in Fig. 3.2. The dielectrics are deposited using the MMAJP process and the metal is

39

deposited using the standard AJP process with a single ink. It is possible however to use MMAJP
as a way to seamlessly switch between inks that are not meant to be mixed such as a conductor and
a dielectric.

The silver ink formulation is described in Section3.4.1. The silver ink can be printed to the
feature size limits of the equipment (10 µm line widths and 20 µm gaps). To determine an acceptable
level of coupling, resonators were simulated using 
r = 3.5 and tan δ = 0.008 as a stand-in for
the PI nanocomposite. Those properties approximate PI alone so it is expected that the resonance
of circuits with higher permittivity dielectrics would shift down. All simulations were performed
with ANSYS HFSS®. This nominal simulation is shown in Fig. 3.4.

3.3.2 Material Characterization

Material properties of the substrate are calculated from the measured resonant frequencies and
quality factors of the resonator circuits. The required equations for this calculation are given by
[105, 18, 106, 104]. The resonant frequencies supported by a closed loop ring resonator fed on two
sides are expressed in terms of the effective dielectric constant as

fn =

nc

2πr√


 e f f

.

(3.1)

This equation holds for a loosely coupled resonator so it applies in this case as shown in Fig. 3.4
which simulates at minimum 10 dB insertion loss for both resonator designs. In this equation, n
is the mode number corresponding to the resonance, and 
 e f f is the effective dielectric constant
of the microstrip structure around the resonator. From equation (3.1), 
 e f f is given for a given
resonant frequency and mode.

An expression for 
r of the substrate is derived from the calculated 
 e f f in [107] as

2
 e f f +

1 + 12h
We f f


r =

,

(3.2)

(cid:21)−1/2 − 1
(cid:21)−1/2

+ 1

(cid:20)

(cid:20)

1 + 12h
We f f

40

a)

b)

Figure 3.4 Simulated S-parameters for ring resonator circuits using 
r = 3.5, tan δ = 0.008 for the
PI nano-composite: a) larger ring resonator, b) smaller ring resonator. ©2020 IEEE.

41

is the effective microstrip width. An effective microstrip width accounts for the

where We f f
thickness of the conductors and is defined in [108] as

(cid:20)

(cid:19)(cid:21)

(cid:18)2h

t

We f f = W +

1.25t
π

1 + ln

,

(3.3)

where W, t and h are the width of the microstrip, thickness of the conductor, and height of the
substrate, respectively. Therefore, the relative dielectric constant of the substrate can be calculated
from the measured resonant frequencies and measured dimensions of the fabricated circuits.

The dielectric loss characteristics of the material are derived from the measured quality factor
of each resonance. The loaded quality factor is QL = fn/∆ f , where ∆ f is the 3-dB full-bandwidth
of the resonance. The unloaded quality factor is calculated from the loaded quality factor by
Q0 = QL/(1 − L) [105], where L is the measured S21 at resonance. The attenuation constant in
the resonator in Nepers per unit length is α = π/(Q0λg), where λg is the guided wavelength in
the microstrip. Attenuation is a function of losses from conduction, radiation and dielectric losses.
To isolate the dielectric losses, the losses due to radiation and conduction must be calculated. As
noted by [18], the radiation losses from the ring itself are negligible, but the radiation from the
open circuit transmission line weakly coupling to the ring in this work should be considered. The
attenuation due to radiation, αr, can be written as [109]

αr = 4(60)

F(
 e f f , 
r ),

(3.4)

where a factor of four appears to account for two open microstrip lines at the end of the two annular
coupling arcs feeding the resonator on either side. The term F, the radiation factor, is given by
[109] as

(cid:18)2πh

(cid:19)2

λ0

F(
 e f f , 
r ) =

1 + (


(cid:26)

1

2
 e f f

r

(cid:20)

·

+

2(
 e f f − 1)
(cid:20)
r − 

2

 e f f

2
e f f




−

(cid:21)
r − 4
r + 2
 e f f + 1)
2
1/2
+ 1
e f f
e f f − 1
1/2
ln

1
1/2
e f f
1

4


ln

+







2(
 e f f − 1)

1
1/2
e f f

4


42

(3.5)

(cid:21)(cid:27)

.

1/2
+ 1
e f f
e f f − 1
1/2







Conduction losses, αc, are calculated by [108, 18].

αc = 6.1 × 10−5
· Rs Z0
 e f f

h

h

We f f

2h
t

(cid:21)

1.25
π

ln
1 +
0.667We f f /h
We f f /h + 1.444

,

+

(cid:19)(cid:21)

(3.6)

(cid:20)
(cid:20)We f f

1 +

h

(cid:32)

(cid:18)

(cid:20)

where Rs is the sheet resistance of the conductor, and Z0 is the characteristic impedance of the
microstrip as calculated from the measured 
r. The term αc does not account for surface roughness
so a scaling factor is introduced to correct this error by

αrc = αc

1 +

2
π

tan−1

1.4

(cid:19)2(cid:21)(cid:33)

(cid:18) ∆

δ

(3.7)

where αc is the conductor attenuation with no surface roughness, ∆ is the RMS surface roughness,
δ is the skin depth, and αrc is the scaled attenuation constant of a rough conductor. This term was
used in [110, 106, 104] and others. The attenuation constant for the dielectric is found then by
subtracting the conduction and radiation contributions from the total, αd = α − αrc − αr.

This expression for the dielectric attenuation assumes there is no dispersion in the resonator
itself, that is, the mode propagating in the transmission line is a Transverse Electro-Magnetic
(TEM) mode and is not affected by the feed structures [105]. Since these transmission lines are
microstrip structures, they support quasi-TEM propagation. So, around a curve, additional sources
of error exist which are substantial for the first few resonances. At higher order resonances, the
curve of the ring is longer with respect to a wavelength and more closely approximates a straight
line microstrip which does not have this additional dispersion. This additional dispersion would
not exist for true-TEM wave-guiding structures, e.g., stripline.

In addition to dispersion along a curve, field interactions across the ring, and field perturbations
from the coupling gap also lower the measured quality factor of the resonator which would lead
to an artificially high calculated loss tangent. Since the resonators in this work use an annular
coupling structure, field perturbations are more pronounced.

[105] suggests using resonators that are at least five wavelengths long to measure dispersion.
A limitation of AJP is the quantity of material which can be deposited so fabricating large enough

43

structures to accommodate the size of resonators required is challenging. To account for these
additional dispersive effects, the attenuation constant is scaled based on simulations which do not
use a frequency dependent loss model for dielectrics. Using a known loss tangent that is enforced
in a simulation, these additional perturbations to the attenuation can be accounted for.

= αd (Cn

s ) where Cn

The numerically calculated attenuation constant at each resonance from a simulation is scaled by
s is the value that makes tan δ equal to the value enforced in the simulation
αn
ds
is the scaled dielectric attenuation constant for resonance n.
for a particular resonance n, and αn
ds
This scaling factor decreases for higher n as some of these effects become less prominent. To the
best of my knowledge, this strategy has not been previously reported to account for these extra
dispersive effects.

An expression for the dielectric loss tangent at each resonant frequency is written for the

in [105] and as described above as
attenuation constant from the dielectric, αn
ds
ds λg
 e f f (
r − 1)
αn
π
r (
 e f f − 1)

tan δ =

.

(3.8)

The ring resonators provide material characteristics at higher frequencies and the MIM capaci-
tors provide material characteristics at low frequencies. From measured S-parameters of a one port
network comprising a shunt capacitor 
r is calculated by,

C =

−1

2π f(cid:61){Zin},

(3.9)

where Zin for a one port network is the input impedance which, in Z parameters, is Z11 = z11Z0
normalized by the port reference impedance. The input impedance can therefore be written in
terms of S-parameters as

(cid:19)

(cid:18)1 + S11

1 − S11

Zin = Z11 =

Z0.

(3.10)

The expected value of capacitance for a single dielectric layer capacitor neglecting fringing fields is
C = 
r 
0 A/h, where A is the area of the plate and h is again the height of the dielectric. Similarly,
the loss tangent, defined as the ratio of the real and imaginary components of the impedance can

44

be written as

tan δ =

(cid:60){Zin}
(cid:61){Zin} =

(cid:60)(cid:8) 1+S11
1−S11 Z0(cid:9)
(cid:61)(cid:8) 1+S11
1−S11 Z0(cid:9) .

(3.11)

Equations (3.11) and (3.9) assume that the parasitic effects, that is, fringing fields, of the
conductive structures are negligible so this characterization is not valid at high frequencies. Using
equations (3.1), (3.2) and (3.8) with the measured resonance frequencies and bandwidths of the
two ring resonator structures and equations (3.9), (3.11) with the measured capacitor structure give
data points for 
 and tan δ of the material vs. a broad frequency range.

3.4 Fabrication

3.4.1

Initial Ink Formulations

This demonstration of MMAJP for microwave electronics uses PI and BaTiO3 nanocomposites.

The initial polyimide ink for MMAJP nanocomposites was identical to the PI ink used in Chapter
2, that is, 15% wt. polyamic acid solution in N-methyl-pyrrolidone (NMP) (Sigma Aldrich) diluted
further to 5% wt. polyamic acid in NMP. As described later, a second PI ink was prepared using
a low CTE PI, PI2611 (HD Microsystems) which was also diluted in NMP. The polyamic acid
content of PI2611 is unknown so the exact final concentration of polyamic acid in the mixed ink is
unknown.

For this project, the BaTiO3 ink was developed from scratch. At the start of this project a
ready-made BaTiO3 dispersion was not available with a compatible solvent so this ink development
focused on creating a stable dispersion of BaTiO3 nanoparticles in a solvent. The BaTiO3 ink
uses 99.9% cubic 50 nm BaTiO3 nanoparticles (US Research Nanomaterials, Inc.). The solvent
to disperse these nanoparticles in needed to be compatible with the PI ink, so the two solvents
investigated were NMP and dimethyl sulfoxide (DMSO). In the course of investigating dispersions,
DMSO was found to support more stable dispersions, possibly related to its slightly higher viscosity
than NMP, but the resulting inks were found to be too viscous to be efficiently atomized in the

45

Figure 3.5 Dispersant investigation for BaTiO3 nanoparticles.

ultrasonic atomizer of the AJP that is also partially pressurized by a pneumatic atomizer. As
discussed below, the final ink used a small amount of DMSO with NMP as the primary solvent.

A range of dispersants were investigated. Disperbyk-140 (BYK), BYK-W 9010 (BYK),
polyvinylpyrrolidone (PVP) (Fraunhofer), oleic acid (Sigma Aldrich), Polyacrylic acid (Sigma
Aldrich), SURFVS0010 (US Research Nanomaterials) were investigated in concentrations ranging
from approximately 0.1% wt. to 1% wt. for suspensions of 7% wt. BaTiO3 in a solvent or mixture
of solvents. 7% wt. nanomaterial was chosen to match the approximate concentration in the silver
ink used in Chapter 2 and described below since this was known to be a well performing ink. These
test dispersions were all mixed under ultrasonication for one hour and dispensed in test tubes to
observe how long it took for the suspension to separate. Ideally, the inks would first be centrifuged
to separate out larger agglomerations of nanoparticles but this equipment was not available at the

46

time of this work. Some of these samples are shown in Fig. 3.5.

This initial investigation yielded the best results using about 0.1% wt. BYK-W910 (BYK), a
commercial dispersant as in [84], or PVP–both suspensions remained stable for a long period of
time. The initial formulation of the BaTiO3 was therefore prepared by first preparing a solution of
DMSO and 2% wt. BYK-W910. Next, mixing a slurry of 67% wt. BaTiO3 nanoparticles and 33%
wt. of the DMSO and 2% wt. BYK-W910. This slurry is diluted with NMP to a mixture that is 7%
wt. BaTiO3 nanoparticles, 3.4% DMSO, 0.07% wt. BYK-W910 (BYK) in NMP. This suspension
is then mixed under ultrasonication for one hour.

As discussed later, during the course of this work a commercially available suspension of
BaTiO3 nanoparticles became available so the improved performance of these MMAJP materials
discussed at the end of this chapter used these commercially available suspensions mixed to match
the previously described BaTiO3 ink, sans a dispersant.

Initial samples using sub-optimal BaTiO3 inks were not as successful, however the results
obtained from these initial samples did correlate well with the final inks used whose data is
presented in this chapter. When mixing these sub-optimal inks, the primary failure mechanism
observed was related to the adhesion of the MMAJP films to the carrier. This poor adhesion was a
result of the aerosols not mixing thoroughly creating regular weak spots in the films. An example
of a film which has begun to de-laminate using inks with poor dispersion is shown in Fig. 3.6.
These failures were not observed with BaTiO3 inks in more stable suspensions.

As in Chapter 2, the silver ink used to fabricate the conductive features and circuits is made with
Clariant Prelect TPS 50 nanoparticle ink (Clariant). This ink is composed of 25% wt. Clariant
prelect TPS 50 in deionized water, diluted to improve atomization. This ink was mixed under
ultrasonication for one hour.

3.4.2 Y-Fitting

The concept of MMAJP is conceptually simple; aerosols are combined and mixed in place. The
mixing itself and how the aerosol flows are combined are important design considerations since

47

Figure 3.6 Third from left MMAJP film using sub-optimal BaTiO3 seen de-laminating from an Al
carrier in initial feasibility study of MMAJP.

aerosol flows must remain laminar as discussed in Section 3.2. This challenge is made more difficult
due to the need for this fixture to be simple enough to prevent the buildup of aerosols in the fitting
and the tubes so the printer does not clog.

Commercially available Y-fittings for gasses and liquids do not all have smooth transitions
between channels. Because of this and in an attempt to prevent turbulence in the aerosol I designed
a fitting to be 3D printed with a smooth transition between tubes. The design of this fitting is
shown in Fig. 3.7, and this fitting is shown in use in Fig. 3.8. The fitting was printed with an Objet
Connex350 polyjet printer This fitting was used to print initial proof of concept parts.

This design worked acceptably well, however, the walls of the channels for the aerosol are much
rougher in the 3D printed part than in the extruded tube. Because of this the part tended to clog.

48

a)

b)

c)

Figure 3.7 Design of 3D printed Y fitting for MMAJP a) side, b) side, c) top.

Better performance could be expected from machined Teflon where smoother walls can be achieved
than a 3D printed part. The parts reported in this chapter were printed with a simple commercially
available Y fitting (McMaster Carr). That part was found to perform comparably to the 3D printed
part but clog less.

Future investigations may produce better results in the mixing point of the aerosols by incor-
porating concepts from the study of microfluidics. One solution which would address these needs
is to use a focusing stage with a sheath gas, as in the print head itself, confining the aerosol flows
(mixed or un-mixed) and allowing mixing to take place while limiting contact of the aerosol with
the mixing apparatus. Adding another sheath gas would however impact the performance of the
atomizers by increasing the pressure.

With modifications to this process, more than two inks/aerosols can be combined and printed

49

a)

b)

Figure 3.8 MMAJP Fabrication setup a) 3D printed fitting, b) commercially available fitting.

50

at a time. Some concepts of mixing arrangements are shown in Fig. 3.9.

3.4.3 Printing

MMAJP ring resonators and MIM capacitors whose measurements are presented here were printed
on a molybdenum copper alloy of 85 % Mo and 15 % Cu (American Elements). 5 mm × 5 mm
films for microscopy and spectroscopy and for demonstrating material gradients were printed on
glass substrates. For each sample, the MoCu surface was mechanically polished, cleaned and an
adhesion promoter applied. The adhesion promoter used was VM652 (HD MicroSystems).

As in previously presented work in Chapter 2, parts were printed with an Optomec Aerosol
Jet 5x printer. The ambient temperature during printing ranged from 20 to 23 °C. The PI ink
was maintained at 25°C during printing so that it would not vary with the room temperature. The
BaTiO3 ink was maintained at 27.5°C. The BaTiO3 ink is heated to a higher temperature in order to
lower the ink viscosity to improve the atomization rate. The ultrasonic atomizer may be damaged
above 30 °C so the ink cannot be heated above that temperature. All samples were printed on a
printing platform heated to 100 °C. heating the print stage allows the polyamic acid to dry as it is
deposited. The goal is for the deposited ink to be wet enough when it is deposited to whet to ink
around it, creating a smooth surface, but to dry fast enough that by the time another layer is being
deposited over the same spot that ink is a dry, rigid film that will not move. The stage must not be
heated above 120 °C at which point imidization of the polyamic acid would begin. The polyamic
acid must be imidized in an environment free of water vapor and oxygen or else the films will
become brittle and break easily.

Table 3.1 shows the mixes of aerosol flow rates for each ink that was used. Mix numbers 1,
6 and 11 were used for fabricating material characterization circuits as well as the three films for
microscopy and spectroscopy. For all mixes, the sheath (focusing) gas flow rate used was 91 SCCM.
The film of a first dielectric gradient was printed using mix numbers 1 through 11 sequentially.
Another dielectric gradient was printed by programming even stepped changes in aerosol deposition
rates for each line printed on a pitch of approximately 50 µm. Changing and settling pressures in

51

a)

b)

b)

Figure 3.9 Concept of alternate embodiment of mixing channel for more than two aerosols
including computer controlled valves to completely shut off or open aerosol flows a) perspective,
b) side, c) bottom.

52

the AJP cause even the small changes in the aerosol flow rates between these steps to be naturally
evened out resulting in a smooth gradient. The printing setup used for these parts is shown in Fig.
3.8 b.

As discussed in Section 3.2, the length of the tube carrying the combined aerosol impacts
how well-mixed the composite aerosol is when it is deposited. The tubes used were made of
polyethylene, with lengths la = 355 mm from the ultrasonic atomizer, lb = 455 mm from the
pneumatic atomizer, and lc = 85 mm as in Fig. 3.1. The lengths from the atomizer to the Y-fitting
are not critical to printing a well-mixed composite but their lengths impact how quickly the tubes
clog and the pressures inside each of the atomizers which impacts the rate of atomization.
In
general, these tubes should be as short as possible while still allowing free movement of the print
head.

Dielectric and conductor patterns for the ring resonator circuits as shown in Fig. 3.2 and 5 mm×
5 mm films were printed from printing path patterns with a path overlap ranging from 15% to 30%.
The PI nanocomposites were printed with a 300 µm diameter nozzle. Both the larger and smaller
ring resonators for each aerosol mix were printed with 20 consecutive layers. Dimensions closer
to the design target can be achieved by checking the height of the printed part midway through
printing to account for process drift. In doing that, one must account for the expected loss in volume
of cured PI compared to uncured polyamic acid of about 50 %.

The MIM capacitors and the 5 mm× 5 mm films for scanning electron microscopy (SEM) and
energy dispersive X-ray spectroscopy (EDS) characterization were printed with two consecutive
layers. The MIM capacitors perform best with the thinnest layer which increases the capacitance
and decreases the effect of fringing fields but composites like this are prone to pinholes in the film
which is why two layers were printed instead of one. The printing time for the dielectrics was 2
hours 5 minutes total for the six ring resonators and three MIM capacitors and 35 minutes total for
four 5 mm× 5 mm films. Increasing the flow rate of the aerosols and also increasing the print speed
would allow for these print times to be reduced. Larger diameter nozzles could also be used which
would accommodate higher volumetric flows which are laminar.

53

Table 3.1 Aerosol Mix Ratios for Combined Aerosol

Mix No. PI (SCCM) BaTiO3 (SCCM)

1
2
3
4
5
6
7
8
9
10
11

15
14.7
14.3
14
13.7
13.3
13
12.7
12.3
12
11.7

10
11
12
13
14
15
16
17
18
19
20

©2020 IEEE.

Before the silver conductive features were printed, the dielectric was heat cured to imidize the
polyamic acid at 200°C. Silver ink features were printed three layers thick using a 100 µm diameter
nozzle with which we can achieve a 10 µm line width. The aerosol flow rate for silver printing was
20 SCCM with a sheath gas flow rate of 32 SCCM. Silver ink features were similarly heat cured to
make the ink conductive up to a temperature of 200°C.

The printed parts are shown in Fig. 3.14 and 3.15. Fig. 3.15 shows the printed gradient film
with mixes 1 through 11. That is, it shows a film with a gradual change in BaTiO3 loading from
one side to the other, and thus an increase in 
r from one side to the other. The printing time for
the silver was 27 minutes total for the six ring resonators and the three MIM capacitors.

3.4.4

Initial Fabricated Device Characterization

Cured sample thicknesses and roughness were measured with a NanoMap-500LS Surface pro-
filometer. A summary of these measurements is shown in Table 3.2. Variations in device thick-
nesses are due to variations in the printing process. That is, the atomization rate of inks can be

54

variable over time, that is, process drift.

On a representative sample of cured printed ink we achieved 30 % of the bulk conductivity
of silver. This representative sample was measured with a Lucas Labs Pro4 four-point resistivity
system. The effective conductivity of the transmission line cannot be considered to be only that
of the printed silver which is much higher than the conductivity of the MoCu substrate (which is
approximately 32% that of Cu), so an average of the measured silver conductivity and the MoCu
substrate were taken to account for the low conductivity of the ground reference in calculating the
loss as described in Section 3.3.2.

The surface roughness of the nanocomposites was measured to be less than or equal to 0.32µm
RMS in all cases. The printed silver is characteristically rougher than the printed composites,
with surface roughnesses reaching 0.42 µm RMS. Nanocomposite mixes 1, 6 and 11 respectively
averaged 1.39, 1.49 and 1.08 µm per printed layer. Adhesion of the nanocomposites was not
observed to be a problem even for high loading factors of BaTiO3 but as discussed in Section 3.4.1,
samples printed with poorly dispersed BaTiO3 were observed to de-laminate from the surface
with high temperature gradients. Additionally, we observed better adhesion of the silver ink to
composites with higher concentrations of BaTiO3 as compared to bare PI.

3.5 Initial Results

3.5.1 Electrical Characterization

S-Parameter measurements were taken using a MPI TS150-THZ Probe System, a Keysight N5227
PNA, and 100 µm pitch 67 GHz probes manufactured by Cascade Microtech. W-band measure-
ments were taken using Virginia Diodes’ Vector Network Analyzer Extender WR10 75-110 GHz
with Cascade Microtech 100 µm pitch W-band probes. In both cases, TRL calibration was per-
formed using a CS-5 substrate to bring the measurement reference plane to the probe tips which
corresponds to the start of the CPWG transmission line and CPWG to MS transition shown in Fig.
3.2. The measured S-Parameters are shown in Fig. 3.10 and Fig. 3.11 for both ring resonator

55

a)

b)

c)

Figure 3.10 S-Parameters for large ring resonators 0-67 GHz. Simulations use measured and
calculated material parameters a) Mix 1, b) Mix 6, c) Mix 11. ©2020 IEEE.

56

a)

b)

c)

Figure 3.11 S-Parameters for small ring resonators 75-110 GHz. Simulations use measured and
calculated material parameters. a) Mix 1, b) Mix 6, c) Mix 11. ©2020 IEEE.

57

a)

b)

c)

Figure 3.12 Calculations from measured MIM capacitors. Parasitic effects lead to error increasing
with frequency. a) capacitance, b) tan δ, c) 
r. ©2020 IEEE.

58

a)

b)

Figure 3.13 Calculations from ring resonator measurements a) 
r, b) tan δ. ©2020 IEEE.

59

Table 3.2 Measured Thickness and Roughness of Circuits

Sample

Composite Composite
Mix No.
Thickness

Large Resonator
Small Resonator
MIM Capacitor
Large Resonator
Small Resonator
MIM Capacitor
Large Resonator
Small Resonator
MIM Capacitor
Large Resonator
(Improved Loss)
Large Resonator
(Improved Loss)

1
1
1
6
6
6
11
11
11
1

11

©2020 IEEE.

(µm)
24.47
26.26
5.56
19.06
25.20
3.75
24.85
19.62
2.10
24.61

23.01

Composite
Roughness
(RMS) (µm)

0.26
0.32
0.27
0.21
0.17
0.22
0.23
0.24
0.09
0.27

0.29

Silver Ink
Thickness

(µm)
1.80
1.56
2.31
1.56
1.35
1.93
0.62
1.51
3.18
2.5

5.0

Silver Ink
Roughness
(RMS) (µm)

0.31
0.29
0.28
0.27
0.12
0.31
0.32
0.42
0.34
0.35

0.26

designs compared to simulated resonators with the calculated material properties and the physical
dimensions Table 3.2.

Due to variable overspray in the printed silver, changes in the magnitude of S21 can be seen at
the resonances of the measured ring resonators. The printed designed gap is close to the edge of the
dimensional tolerances of the printer. Because of this and because the substrate is thin relative to the
coupling gap, the resonators are sensitive to small changes in coupling caused by overspray. Because
my calculation takes into account such perturbations in the field in the resonator, these coupling
changes should not affect the results and increased coupling does not change the characteristic
resonance. As the annular coupling structure begins to radiate as its length approaches one-quarter
wavelength, this behavior obstructs higher order resonances in these samples.

It is clear from these measurements a shift down in frequency of the resonant peaks which is

60

due to the change in 
r from the introduction of the BaTiO3 nanoparticles into the polymer matrix.
A change in the quality factor of the resonance is also apparent from the increased loss due to the
nanoparticles. Table 3.4 shows the quality factors of the measured and simulated resonators.

Using the equations laid out in Section III. B, 
r and tan δ of the composites were calculated.
The results are shown in Fig. 3.12 and 3.13. To extend the useful range of data from the MIM
capacitors, the parasitic effects from the launch/probing structure must be accounted for since this
structure is not matched to 50Ω. To accomplish this, I used simulated short, open, and matched
calibration standards to perform a Short-Open-Load (SOL) calibration approximation. This brings
the reference plane of the measurement to the edge of the circular capacitor structure. There is likely
field interaction between the launch structure and the capacitor itself due to their size and proximity
which a calibration like this cannot take into account. This calibration also does not account for
fringing effects in the capacitor itself. All of these behaviors, as well as the self-resonance of the
capacitor itself, lead to increasing error with frequency in the calculation of material properties.
This behavior is most apparent in the tan δ calculation as seen in Fig. 3.12. The loss tangent of the
three composites is calculated with equation (3.11) to be 0.020, 0.030, and 0.028 for Mixes 1, 6,
and 11, respectively at 0.5 GHz.

The simulated resonators using the calculated and measured properties correlate well with
measured samples as shown in Fig. 3.10 and 3.11 and Table 3.4. Differences between simulations
and measurements are due to overspray in the coupling area as discussed earlier. Differences are
also due to the inability to account for the unusual frequency dependency of BaTiO3.

The 
r of these composites, calculated with equation (3.2) and the expected capacitances yielded
values ranging from 3.4 to 8.9 for these initial ink formulations. These results are shown in Fig.
3.12 and Fig. 3.13. The measured thickness of the capacitors is between 2 and 4 µm thick. The
difference in the thickness of these films and the distribution of particles in a thinner film is likely
a contributing factor to the difference between the capacitor measurements and the ring resonator
measurements and calculations however these results generally agree.

61

Figure 3.14 The fabricated ring resonators and MIM capacitors for mix numbers 1, 6, and 11.
©2020 IEEE.

3.5.2 Microscopy and Spectroscopy

Three 5 mm by 5 mm films of aerosol mix 1, 6, and 11 were fabricated for SEM and EDS analysis.
The two primary purposes of this exercise is to evaluate the quality of the dispersion and to obtain
a data point with EDS analysis of the elemental content of the films. These samples were prepared
by drying and coating with 5 nm of osmium (Os) in an NEOC-AT osmium coater manufactured
by Meiwafosis Co., Ltd. The samples were then mounted on an aluminum stub with carbon tape
manufactured by Ted Pella, Inc. We used a JEOL JSM-7500F cold field emission electron emitter
scanning electron microscope manufactured by JEOL Ltd. Elemental analysis by means of EDS
was done using an Oxford Instruments AZtec system with a 30 mm2 SiLi detector crystal and an
ultrathin window.

62

a)

b)

Figure 3.15 Fabricated 5-mm by 5-mm films, a) fabricated gradient film with mix 1-11 ©2020
IEEE, b) six films labeled with their 
r fabricated with improved ink formulation including a
nearly continuous gradient.

63

a)

b)

c)

Figure 3.16 SEM images for a) mix 1, b) mix 6, and c) mix 11. In each image, the magnification is
5000 at 5.0 kV beam energy. ©2020 IEEE.

64

a)

b)

c)

Figure 3.17 EDS analysis for a) mix 1, b) mix 6, and c) mix 11, showing an increase in the
constituent elements of BaTiO3. ©2020 IEEE.

65

Fig. 3.16 shows SEM images taken of printed films of composites of mix 1, 6, and 11. It is
clear the increase in BaTiO3 from mix 1 to 11. These images also show some agglomeration of
particles however most of the visible clusters are not tightly packed. These clusters are on the order
of magnitude of the expected aerosol droplet size. The aerosol droplets are expected to range from
0.5 µm to 6 µm [103]. These clusters are therefore likely a result of the aerosol mixing process
rather than poor dispersion of the nanoparticles in the ink. The agglomeration can be improved
with better dispersion of the BaTiO3. As discussed previously, the BaTiO3 ink can be improved by
centrifuging the ink to separate out agglomerations leaving well dispersed nanoparticles. Further
improvements to the ink formulation can likely be achieved with other dispersants or surfactants
to better disperse the particles. The dispersion of nanoparticles in the polymer appears to improve
with higher mixing ratios.

Another notable observation from these SEM images are small air bubbles, also approximately
the same size of aerosol droplets. These bubbles may be reduced by heating the uncured ink for
a longer period and holding the samples under vacuum before imidization. These bubbles may
also be reduced by printing on a higher temperature printing platform so that bubbles can escape
more quickly between printed layers.
It is likely that some bubble formation in this process is
unavoidable and may be a byproduct of AJP. To print thick films of PI, it is necessary to heat the
print stage which dries out the deposited films and likely traps many of these bubbles in the film.
These observations were used to improve the AJP and MMAJP processes and procedures in future
work.

The results of EDS analysis are shown in Fig. 3.17 for the three sample films. These plots
also show an increasing amount of BaTiO3 nanoparticles in the films. Using the lowest common
denominator of atomic content of elements present in BaTiO3, an estimate of the % vol. of BaTiO3
in the composite using for the density of BaTiO3, ρ = 6.02 g/cm3, and for PI, 1.42 g/cm3 [111]
can be calculated. For mix 11, from the EDS analysis, there is about 58.25 wt. % BaTiO3, which
corresponds to 24.8 vol. %. For mix 6, there is about 15.41 wt. % BaTiO3 corresponding to 4.12
vol. %.

66

To compare these results with the measured dielectric constant, an estimate of how the volume
of BaTiO3 affects 
r must be found. This estimate was calculated using the Modified Lichtenecker
(ML) equation as in [82, 90, 112] and the Maxwell-Garnett (MG) approximation as in [113, 114].
The Modified Lichtenecker equation is a logarithmic law equation given by

log 
r = log 
 p + fc(1 − k) log( 
c

 p

),

(3.12)

where 
r is the relative dielectric constant of the composite and 
 p is the relative dielectric constant
of the polymer matrix and fc is the volume fraction of the ceramic. For this analysis, the polymer
matrix is the PI and 
c is the relative dielectric constant of the ceramic filler material, BaTiO3. k
is a fitting constant. A value of k = 0.3 was used as in [82] and references therein who note that
this value corresponds to a well-dispersed composite. The Maxwell-Garnett approximation is

(cid:18)2
 p + 
c + 2 fc(
c − 
 p)

2
 p + 
c − fc(
c − 
 p)

(cid:19)


r = 
 p

,

(3.13)

using the same notation as in equation (3.12).

The results of this analysis are summarized in Table 3.3. This table gives both what we expect

r to be given a volume content of BaTiO3 and what we expect the volume content of BaTiO3 to
be given the measured 
r.

EDS analysis may not represent the bulk composite material since it is a characterization of
what is on or close to the surface of the material being analyzed. This is why in Table 3.3 mix no.
1 shows a Vol % of BaTiO3 of zero even while particles can clearly be seen in Fig. 3.16.

Another important source of error in these 
r calculations is that neither the Modified Lichte-
necker equation or the Maxwell-Garnett approximation take into account frequency dependence of
composite materials, and it is known that BaTiO3 has a strong frequency dependency as demon-
strated in composite materials by [115, 116, 113] all of whom also showed a decrease in 
r of
BaTiO3 and its composites with increasing frequency, all below 20 GHz.

67

Table 3.3 Comparison of Results with Modified Lichtenecker (ML) and Maxwell-Garnett (MG)
Equations

Measured

Mix
No.

1
6
11

Meas. 
r

Vol. %
(first resonance) BaTiO3
(EDS)
0.0
4.1
24.8

3.8
5.0
9.0

ML
Vol. %

r BaTiO3

MG
Vol. %

r BaTiO3

3.5
4.0
8.3

2.2
10.2
27.1

3.5
3.9
6.9

2.7
12.7
34.9

©2020 IEEE.

3.6 Process Improvement for Lower Loss

The loss of the composites is expected to be higher than PI alone due to the loss characteristics of
BaTiO3 nanoparticles. There is relatively little characterization information available for BaTiO3
itself, especially at frequencies as high as the W-band. The losses that we observed are in line
with available published data however. A complicating factor is that BaTiO3 shows material
characteristics that are strongly dependent on the preparation methods used, the size of the particles,
and the temperature of the material. [117] demonstrated that the loss of BaTiO3 behaves above the
curie temperature as tan δ = (0.004/D)(T − T0)−1 where D is the particle diameter and T0 is the
Curie temperature. That is, loss is inversely proportional to particle size.

If that behavior is accurate, since the nanoparticles used in this work were small (50 nm),
I expect loss to be higher potentially than other published work on nanocomposites with larger
particle sizes as in [117]. [118] showed PMMA + BaTiO3 composites with loss tangents as high
as 0.065 at 3 GHz, measured at room temperature.
[119] showed loss tangents for 30 vol. %
BaTiO3 + polyethylene glycol diacrylate composites as high as 0.2 at 10 GHz, measured at 22°C.
[120] reported the real and imaginary parts of the dielectric constant of BaTiO3 nanoparticles up
to 18 GHz (presumably measured at room temperature, though a measurement temperature was
not given explicitly). Using tan δ = 
(cid:48)(cid:48)/
(cid:48) their results correlate approximately to a loss tangent

68

of 0.5, 0.6, and 0.7 at 10, 14, and 18 GHz respectively. The same results were published in
[116].
[121] reported on fine grain BaTiO3 material properties. They demonstrate a loss for
BaTiO3 nanoparticles as high as 0.3 at 1.6 GHz measured at room temperature. They additionally
measured polypropylene composites with 50 vol. % loading loss from 0.02 to 0.05, though all the
particles used in their composites were larger than what was used in our work. They note that the
fine grain particles they studied differ dramatically from the coarse grain particles. Their results
seem to contradict [117] as they note that their fine grain samples exhibited the lowest loss, though
they also noted that a similar spike in the loss tangent occurred at a higher frequency relative to
their other samples. They did not present any data above 10 GHz.

While the loss characteristics of the films presented are not out of line with previous work, I
endeavored to improve the performance of the printed composites. This was done by investigating
changes in material processing and ink formulations due to observations from SEM and EDS
analysis. First, I switched the PI ink to HD Microsystems PI2611 polyamic acid polyimide
precursor, which has a published (imidized) 
r of 2.9 and a loss tangent of 0.002, which we expect
to be slightly lower than the PI formulation used above. Previously, the polyimide was cured
at 200°C which correlates to achieving approximately 80% imidization [122]. New dielectric
films were cured at 350°C to ensure 100% imidization. Polyimide films absorb moisture, and
films processed at lower temperatures tend to absorb more moisture so we expect the presence
of moisture in our samples also played a large role in the loss of the samples discussed above.
Finally, all dielectric films were held under vacuum for a minimum of 12 hours prior to curing.
The purpose of this is to ensure that no moisture is absorbed by the films and also to mitigate air
bubbles incorporated into the films during the aerosol jet printing process.

For new samples printed with modified processing, twice as many silver layers were printed.
The angle of the coupling arc in the annular coupling structure was reduced in these parts from 80°
to 60° as in Fig. 3.2. This shorter coupling arc allows measurement of two more resonances with
the X-band resonator before the annular coupling arc begins to radiate at a quarter wavelength. All
other printing and measurement procedures were performed as prescribed above for two X-band

69

fundamental resonators with mixing ratios 1 and 11.

The two resonators dimensions are shown in Table 3.2 and the calculated loss tangent and 
r
are shown in Fig. 3.13 compared with the initial ink formulations. The loss tangent ranged from
about 0.002 in mix 1, which is equal to the published loss tangent of the bulk PI2611 material
the ink was based on to 0.03 at 44.15 GHz in mix 2. Both results represent a large improvement
compared to the equivalent mixtures with the initial ink formulation and processing.

With these results, it is clear that the source of loss in the previously fabricated circuits stems
from the materials used and processing parameters. That is, loss is not a product inherent to the
MMAJP process. A decrease in 
r is also observed compared to the initial formulations. 
r of
these films ranged from 3.1 to 6.3. This change is due to the change in the polyamic acid from
which the PI inks are based. PI2611 has a slightly lower published 
r than most other generic
polyimide formulations available.

The measured S-parameters of these samples are shown in Fig. 3.18 and in Table 3.4 which
compares a simulated ring resonator with ideal properties to measured results from fabricated
x-band ring resonators fabricated with mix 1 and 11 as described. This ideal ring resonator uses
a known and not frequency dependent loss tangent and 
r using published values for PI2611 with
no surface roughness. Fig. 3.18 and in Table 3.4 include a second simulation using the calculated
material properties for either mix 1 or 11 employing a Djordjevic-Sarkar material model built
into our simulation tool. The calculated results correlate well with simulated materials of similar
material properties. The loss characteristics also compare favorably with the loss of other BaTiO3
composites and films mentioned above.

3.7 Conclusion

This chapter demonstrates a novel method for manufacturing composites using MMAJP.

This process, while similar processes have been demonstrated for solid oxide fuel cells and
variably conductive materials, has not been characterized before for use at microwave and mm-wave

70

a)

b)

Figure 3.18 S21 of lower loss composites compared to simulations a) large ring resonator: 0-67
GHz, mix 1 b) large ring resonator: 0-67 GHz, mix 11. ©2020 IEEE.

71

frequencies. This chapter also lays out a more thorough process for fabrication than previously
shown, with relevant design equations and limitations on the procedure. Using this process,
continuous gradients of materials can be realized in three dimensions, which using conventional
fabrication would be impractical, if not impossible.
In demonstrating this process, I showed
polyimide based polymer matrix BaTiO3 nanocomposite films of 
r constant ranging from 3.1 to
8.9, and I show how loss can be controlled in these printed films.

With very few papers published on similar procedures for multi-material or nanocomposite
manufacturing, this strategy is still very new. Many improvements in this procedure may be
found by looking to microfluidic mixing techniques to create more well-mixed aerosols. Because
agglomeration of particles may occur in the aerosol itself which is not a concern addressed in
other works on nanocomposite formulations, the field of microfluidics may also provide insight
into filtering out these larger agglomerations of particles prior to mixing.

With MMAJP, material properties can be tuned in place during fabrication. Some prior uses
of the term “tunable material” referring to nanocomposites, are only tunable in the sense that they
show strategies for engineering a mixture one at a time. The process demonstrated here shows the
manufacturing of a material which is tuned in real time during printing. This process is extendable
to other composites and is not limited to polymer matrix composites as are shown in later chapters.
There are many ways this process can be expanded upon. For example, mixing more than two
materials at a time for another dimension of material adjustment. More than two materials would
require an additional atomizer for each material and would likely benefit from a more elaborate
mixing strategy. MMAJP opens the possibility for functional electronic material printing akin to
RGB color printing where the values of 
, µ, and ρ can be chosen for any point in the space a part
is printed in. This flexibility in materials may facilitate many novel microwave circuits that would
otherwise not be practical to fabricate.

The focus in the demonstration in this chapter is adjusting electrical properties, but as is noted
by others, the mechanical properties of materials can also be tuned using this process. For example,
the CTE has been demonstrated to be adjustable by increasing the content of ceramics in a polymer

72

matrix. Thus, this process can be used to match the CTE of two materials gradually in order to
reduce stress in heterogeneous systems of materials. This process has wide ranging applications
from packaging to filters and other periodic structures.

73

Table 3.4 Unloaded Quality Factor of Measured and Simulated Resonances

Mix

Res. n = 1 Res. n = 2 Res. n = 3 Res. n = 4 Res. n = 5 Res. n = 1a

Mix 1

8.8

12.4

14.2

N/A

N/A

17.9

Simulated

Mix 1

Mix 6

Simulated

Mix 6

Mix 11

Simulated
Mix 11

Mix 1

(Imp. Loss)

Simulated

Mix 1

(Imp. Loss)

Mix 11

(Imp. Loss)

Simulated
Mix 11

(Imp. Loss)

11.4

14.1

15.3

N/A

N/A

20.8

6.8

8.7

3.5

4.8

15.1

16.9

11.2

9.8

11.4

N/A

N/A

11.1

10.5

5.6

4.2

22.7

19.7

14.9

8.2

5.7

1.9

28.7

25.3

15.6

N/A

N/A

12.9

N/A

N/A

11.6

N/A

31.6

27.3

15.9

N/A

34.3

22.5

15.5

13.0

N/A

N/A

N/A

11.9

14.9

16.4

18.3

17.1

N/A

aSmall (W-band) resonator ©2020 IEEE.

74

CHAPTER 4

APPLICATIONS OF MMAJP: OTHER COMPOSITES

As indicated in Chapter 3, it may be possible to tune material properties in place with MMAJP
by fabricating composites composed of materials with high values of 
, µ, and ρ. In Chapter 3, I
introduced the concept of MMAJP and demonstrated the ability to tune the dielectric constant of a
printed material. This chapter expands on that work and demonstrates composites that tune values
of µ and ρ. Tuning all three of these basic properties allows one to design a value of electrical
permittivity, permeability, and resistivity in a package at any point in 3-D space. This goes beyond
the capability to integrate basic electronic components like capacitors (as demonstrated in the last
chapter), inductors, and resistors. The amount of material flexibility MMAJP allows may enable
heretofore not yet demonstrated microwave structures which take advantage of the unique ability
of MMAJP to fabricate material gradients and patterns.

In the context of this dissertation, my goal is to demonstrate the ability to tune these material
properties, but demonstrating the breadth of tuning capability or the extent of microwave appli-
cations MMAJP has is beyond the scope of this work. To demonstrate magnetic and resistive
materials I use composites of polyimide and nickel zinc ferrite and of silver and carbon, respec-
tively. In this case, the carbon structures include carbon black, carbon nanotubes, and graphene.
Many other composite component materials would be possible to accomplish similar results; e.g.,
[91] demonstrated polyimide and carbon nanotube composites which demonstrate some tunability
of resistivity. Since polyimide is essentially non-conductive, it would be more desirable to use
composites of two conductive materials. In this way a more useful range of resistivities can be
realized from very low resistivity materials (e.g. silver) to materials with resistivity values useful
for fabricating resistors.

Magnetic materials have important applications in fundamental components like inductors as
well as in sensing [123]. Magnetic materials have also been demonstrated for applications in EMI

75

and crosstalk suppression as absorbing materials [124], as well as antenna miniaturization [125].
Magnetic materials have historically been more challenging to incorporate into packaging. This
is partially due to the relative bulk of magnetic materials required for many components which
use them. At mm-wave frequencies these structures are smaller, however. Magnetic materials
also tend to be more difficult to incorporate into electronics packaging and so are less frequently
used in conventional processes despite the potential benefits to using such materials, especially in
microwave structures. When magnetic materials are incorporated into packaging, often the package
is designed around the magnetics rather than the other way around such as in [126].

Many ferrite materials cannot easily be used because of the high sintering temperatures required.
This means that materials are limited to those that can be deposited by other means, such as physical
vapor deposition [127] or similar techniques. While magnetics may be deposited in composites such
as in [112], they may be challenging to pattern by lithography without leaving residual nanomaterial
on a part. Solution cast methods have also been used as in [128], but are limited as they do not
easily permit monolithic integration.

Given these limitations, AM is clearly a potentially useful method for integrating magnetic
materials. There are some prior examples of magnetic materials and composites deposited by
additive means such as by screen printing as in [129, 130], by paste extrusion [131, 12], or selective
laser melting (SLM)/sintering as in [132, 133, 134]. A drawback of these prior works is the inability
to pattern these magnetic materials with high resolution.

With the exception of [132], these previous works have used individually mixed formulations of
magnetic composites or pastes prior to depositing and patterning. Chaudhary et al. [132] showed
graded magnetic materials with SLM, varying the composition of the feedstock powder. That
process, however, requires materials compatible with SLM which is more limited than processes
such as AJP and the present MMAJP processes. To demonstrate MMAJP of magnetic materials, I
show an inductor fabricated with a MMAJP composite compared to inductors fabricated without
magnetic materials, to show the increase in inductance. I also show transmission lines fabricated
on the composite, subjected to a DC magnetic bias, and finally a ring resonator printed on the

76

composite, similar to those demonstrated in Chapter 3. Both the inductor and transmission lines
are compared to similarly fabricated circuits on polyimide (PI) alone. Using MMAJP, I show an
inductor with the highest inductance density presented to-date to the best of my knowledge of any
previously reported planar AM inductor. Energy dispersive spectroscopy (EDS) is also used to
further characterize these composites. MMAJP allows for a new method for integrating magnetic
composites into microwave packaging and circuits with high resolution patterning, expanding on
the work in the previous chapter.

Resistive materials, or metals with high resistivity compared to materials such as Cu, Au, or
Ag, are more commonly integrated into conventional packaging typically using thin film processes.
For microwave circuits, or for resistors requiring tight resistor tolerances, nichrome is frequently
sputtered and patterned by etching and sometimes tuned with laser etching. Resistors or resistive
materials can also be fabricated with semiconductor processes by selectively doping regions so they
become more conductive. Semiconductor processes are less useful, except when SoC strategies
are used. Thin film processes such as sputtering NiCr typically require extra processing steps so
may add significant cost to a packaging strategy. Again, AM as a flexible fabrication strategy may
allow for an alternative method for depositing such materials without as many pieces of additional
equipment or processing requirements.

To demonstrate MMAJP of resistive materials, I show composites of silver, a characteristically
high conductivity metal, with carbon nanotubes, graphene and carbon black, which have higher
resistivity. Besides basic circuit components like resistors, these composites may also have impor-
tant applications in sensing. Gradients of these materials may also allow for the basis of broadband
load structures as well.

4.1 Magnetic Composite Circuit Design

To demonstrate MMAJP of magnetic composites, I designed three types of circuits to fabricate.
First, a compact 4.5 turn spiral microstrip inductor fitting in a 1 mm by 1 mm square. This inductor

77

was fabricated on a printed nanocomposite as well as polyimide alone for comparison. Second,
microstrip transmission lines again intended to be printed both on polyimide and the composite.
These transmission lines will be subjected to a magnetic bias to analyze the composite’s behavior.
Finally, a ring resonator to be printed on the magnetic composite similar to in Chapter 3. All of
these circuits are designed to be fabricated on a MoCu carrier, acting as a ground reference. As in
prior chapters, these designs include a grounded coplanar waveguide to microstrip transition, with
coplanar grounds connected to the carrier.

The 4.5 turn microstrip inductor uses circular turns to achieve a higher quality factor. The
center of the inductor is bridged to the opposite microstrip feed line by way of a printed polyimide
bridge. In one sample the inductor is printed on the magnetic composite under the inductor area
only. In the other, this area is filled with polyimide. The transmission lines are printed on two
substrate thicknesses for comparison.

In all designs, a border of 70 µm of PI is included, as explained in the next section. The design

of these structures is shown in Fig. 4.1 with different material regions labeled.

4.2 Magnetic Composite Circuit Fabrication

In this example of MMAJP, three different inks are used and all circuits are fully additively manu-
factured as in previous chapters. The PI ink is composed of 40 vol. % PI2611 (HD Microsystems)
polyamic acid dilution plus N-Methyl-2-Pyrrolidone (NMP) (Sigma Aldrich). The magnetic ink
is Ni0.5Zn0.5Fe2O4 based, and composed of 20 wt. % Ni0.5Zn0.5Fe2O4 10 - 30 nm diameter
nanoparticles in NMP (US Research Nanomaterials). Ni0.5Zn0.5Fe2O4 was chosen because of its
high magnetic saturation. Since composites are printed with MMAJP, the magnetic ink does not
require any PI content. Finally, as in previous chapters, silver ink composed of 25 wt. % Clariant
Prelect TPS 50 in deionized water.

The MMAJP process is conducted as outlined in Chapter 3, mixing aerosols in place by
convective/advective mixing. For this chapter, the length of the combined aerosol tube (as indicated

78

(a)

(b)

(c)

Figure 4.1 Annotated design in millimeters with labeled material regions. Dimensions in
parenthesis are target material thicknesses. (a) Inductor design. (b) Transmission line design. (c)
Ring resonator design.

79

1.001.080.060.26PI + Ni0.5Zn0.5Fe2O4Ag (0.004)PI BridgeAg BridgePI (0.03)PI (0.001)MoCu CarrierPI + Ni0.5Zn0.5Fe2O4 (0.06)PI + Ni0.5Zn0.5Fe2O4 (0.03)PI (o.06)5.67PI (0.02)PI (0.001)MoCu CarrierPI (0.001)PI (0.03)PI + Ni0.5Zn0.5Fe2O4 (0.03)R2.58R2.52R2.62R2.6560°0.060.40in Fig. 3.1) of 85 mm and a focusing gas to composite (combined) aerosol flow rate ratio of 1.2.
Composites were printed with a polyimide aerosol flow rate of 75 sccm using a pneumatic atomizer
and a Ni0.5Zn0.5Fe2O4 aerosol flow rate of 50 sccm with an ultrasonic atomizer. The MMAJP
composite was printed using a 300 µm diameter nozzle.

Similar to previous chapters, these structures were fully printed on a mechanically polished
85% Mo 15% Cu carrier. VM651 (HD Microsystems) adhesion promoter was applied to this
carrier. A single layer of PI approximately 1 µm thick, was printed in all regions containing PI or
Ni0.5Zn0.5Fe2O4 composite in Fig 4.1 acting as an intermediate adhesion layer.

The magnetic ink has a low viscosity, so when a large amount of it is deposited with MMAJP,
the ink tends to flow away from the area it is printed. This is because it cannot dry fast enough,
even with a heated print stage. An example of this result is shown in Fig. 4.2 from an early sample.
I solved this issue and allowed for faster material deposition by printing a border of PI 70 µm wide
around all composite regions. This is the nominal printed line width for a 300 µm nozzle with the
print parameters I use. The PI ink alone has a much higher viscosity than the composite and dries
much more quickly. This allows for a walled in region to fill with the composite material. This
unique structure would not be practical to fabricate by conventional methods whereas MMAJP
allows both the composite and PI to be deposited simultaneously. The dielectric and magnetic
structures are cured by heating to 295°C in a nitrogen environment.

Silver features are printed on top of cured dielectrics and sintered at 180°C for four hours.
Finally, for inductors, a PI bridge was printed as in Fig. 4.1a. The PI bridge was cured as before
and the silver bridge was printed over the bridge and cured as before.

The two thickness regions targeted as indicated in Fig. 4.1 of 30 µm and 60 µm measured 37.4
µm thick and 59.5 µm thick, respectively for composites and 25.0 µm thick and 51.2 µm thick,
respectively for PI. The sintered silver measured 8.1 µm thick. The inductor PI bridge measured
8.0 µm thick. The fabricated parts are shown in Fig. 4.3.

80

Figure 4.2 Result of composite printing without PI border.

4.3 Magnetic Composite Measurement Results

All circuits were measured with a MPI TS150-THz Probe System, a Keysight N5227 PNA and
250 µm pitch non-magnetic (beryllium copper) GGB probes. LRRM calibration was performed
to bring the reference plane to the tip of the probes corresponding to the start of the CPW launch
structure in circuits. Measurements were taken up to 40 GHz, all referenced to a 50Ω characteristic
impedance.

4.3.1

Inductors

Using simultaneously fabricated through and open calibration standards, the printed inductor
measurements were de-embedded to isolate the inductor performance. The measurement reference
plane is shown in Fig. 4.3. The inductance and quality factor can be extracted from Y-parameters
by

(cid:21) 1

ω

(cid:20)

4

Y11 + Y22 − Y21 − Y12

81

L = (cid:61)

and

,

(4.1)

(a)

(b)

Figure 4.3 (a) Fabricated inductors. (b) Fabricated transmission lines and ring resonator on
composite. Measurement reference planes are indicated by dotted red lines.

82

Q = − (cid:61)[Y11 + Y22 − Y21 − Y12]
(cid:60)[Y11 + Y22 − Y21 − Y12] .

(4.2)

Fig. 4.4 shows extracted inductance and quality factor up to 8 GHz, covering the frequency
range below the self-resonant frequency (SRF) of the inductors. Simulated inductor models are
fitted to measurements, giving an estimate of the relative permeability of the composite. I estimate
with this method µr = 1.6. These simulations overestimate the quality factor of the inductors but
otherwise show excellent correlation with measured parts.

Without the magnetic composite, the inductors have an inductance of 3.0 nH. When the magnetic
composite is introduced, the inductance increases 40%, to 4.2 nH. The quality factor of these
inductors is 6.6 and 6.7, respectively. This is a small difference, but it is significant that an increase
in inductance density is achieved without sacrificing quality factor. The inductance density of the
inductors is 3 and 4.2 nH/mm2, respectively.

Using the MMAJP process, these results compare favorably to previously reported AM induc-
tors. A comparison with prior work is shown in Table 4.1. Both with and without the composite
material, these inductors are, to the best of my knowledge, the highest inductance density demon-
strated in an additively manufactured planar inductor–2.4 times higher than the next highest previ-
ously reported AM inductor at 1.75 nH/mm2 [135]. While these inductors compare favorably with
AM inductors, they are not competitive with inductors fabricated using semiconductor processes,
such as [136], which has far higher feature resolution capabilities. The AM inductors I present may
still be improved on by fabricating features closer to tolerance limits of AJP.

83

(a)

(b)

Figure 4.4 Inductor parameters extracted from de-embedded S-parameters. (a) Inductance, (b)
Quality factor.

84

Reference

Technology

Topology

Turns

L (nH)

Q

Table 4.1 Comparison with other planar inductors

This
This

McKerricher et al. [135]
McKerricher et al. [135]
McKerricher et al. [135]

Gu et al. [137]

Vaseem et al. [138]
Bidoki et al. [139]
Mariotti et al. [140]
Mariotti et al. [140]

Lee et al. [141]
Lee et al. [141]
Long et al. [136]

AJP
Inkjet
Inkjet
Inkjet
AJP
Inkjet
Inkjet
Inkjet
Inkjet
Inkjet
Inkjet
Si VLSI

MMAJP MS Spiral
MS Spiral
MS Spiral
MS Spiral
MS Spiral

Spiral
Spiral
Spiral
Spiral
Spiral
Meander
Meander
MS Spiral

4.5
4.5
1.5
2.5
3.5
31
1.5
9
1.5
2.5
1
1
4.5

4.2 (1 GHz)
3.0 (1 GHz)
9.7 (10 MHz)
23 (10 MHz)
76 (10 MHz)
2.7 (<25 MHz)

8 (1 GHz)
20 (N/A)

7 ( 300 MHz)
18 ( 300 MHz)
7.1 (1 GHz)
7.3 (1 GHz)
5.7 (2 GHz)

6.7 (3 GHz)
6.6 (3 GHz)
4.4 (500 MHz)
3.9 (300 MHz)
3.3 (100 MHz)
1 (25 MHz)
6 (1 GHz)

N/A

11 ( 300 MHz)
10 ( 300 MHz)
25 (1.7 GHz)
5 (1.7 GHz)
5.5 (2 GHz)

85

SRF
(GHz)
6.2
7.1
2
1.5
1

0.047

4
N/A
2.45
2
8
6.4
N/A

Area
mm2
1
1

13.99
13.14
43.56
78.5
12

10000
> 5.76
> 12.96
26.25
26.25
0.047

Density
nH/mm2

4.2
3
0.69
1.75
1.74
0.035
0.67
0.002
< 1.22
< 1.39
0.27
0.28
122.06

4.3.2 Transmission Lines

S-parameter measurements with a port power of 6 dBm of the transmission lines on the polyimide
and composite substrates were taken with and without a DC magnetic bias applied. The magnetic
bias was applied perpendicular to the substrate, that is, in the center of the cross section of a
microstrip transmission line, parallel to the direction of the electric field below the cutoff frequency
of TEM (or, quasi-TEM) propagation. The bias was applied with two different permanent magnets
and the magnitude of the field was confirmed with a Gauss meter.

These measurements can be seen in Fig. 4.5, 4.6, and 4.7. These figures correspond to the 2x
through, line 1 (L1), and line 2 (L2), respectively, as indicated in Fig. 4.1 and Fig. 4.3. For all
circuits, the return loss stays below 10 dB in the range of measurement, though it is clear from the
S-parameters that the microstrip on the magnetic composite have a greater impedance mismatch
due to the introduction of the magnetic material.

When a magnetic bias is applied, a resonance appears which shifts up in frequency when the
magnetic bias is increased. The resonance is independent of the thickness of the substrate as is
clear from Fig. 4.6 and 4.7, where the two resonances are approximately equal in magnitude.
The magnitude of the resonance reduces for the shorter transmission line, so is proportional to the
length of the transmission line as is clear from the much smaller resonances in 4.5 which are at
approximately the same frequency as the longer transmission lines.

These magnetic resonances are not ferromagnetic resonances which was confirmed by testing
up to an RF input power of 26 dBm using a Maury Microwave MT2000 Mixed Signal Active
Load-Pull System from 8 to 12 GHz with matched load and source. Increasing (or decreasing) the
RF power applied had no corresponding effect on the resonances. No resonance appears without a
magnetic bias, and no resonance appears in transmission lines of the same design on PI alone.

86

(a)

(b)

Figure 4.5 Transmission line S-Parameter measurements on composite with and without DC
magnetic bias. (a) 2x transmission (b) 2x reflection.

87

(a)

(b)

Figure 4.6 Transmission line S-Parameter measurements on composite and PI and with and
without DC magnetic bias. (a) Line 1 transmission (b) Line 1 reflection.

88

(a)

(b)

Figure 4.7 Transmission line S-Parameter measurements on composite and PI and with and
without DC magnetic bias. (a) Line 2 transmission (b) Line 2 reflection.

89

4.3.3 Resonator

Measured resonator S-parameters are shown in Fig. 4.8. Since the permittivity of PI is known, the
effective permeability of the composite material can be estimated based on the resonance locations
which are a function of √
µe f f 
 e f f as noted by [130]. The resonant frequencies supported by this
ring resonator can be expressed as:

fn =

2πr√

nc
µe f f 
 e f f

.

(4.3)

In this expression, n is the mode number corresponding to the resonance, 
 e f f is the effective
relative dielectric constant of the microstrip transmission line and µe f f is the effective relative
permeability. This equation is the same as the one used in Chapter 3, with the addition of the factor
of µe f f . Using equation (4.3), the product µe f f 
 e f f can be calculated for measured resonant
frequencies.

An equation for 
r µr for the substrate can be derived as from the equation for 
 e f f given by

[107] and used in Chapter 3 [73] again substituting the product µe f f 
 e f f , as

(cid:20)

(cid:21)−1/2 − 1

2µe f f 
 e f f +

(cid:20)


r µr =

1 + 12h
We f f

(cid:21)−1/2

+ 1

1 + 12h
We f f

,

(4.4)

where We f f
is the effective microstrip width as defined in [108] and in Chapter 3. Using the
expected value of 
r = 2.9 for PI2611 polyimide, and assuming the relative dielectric constant of
the ferrite does not significantly impact the dielectric constant of the composite, an estimate for the
relative permeability of the material can be calculated. For this measurement, the resonances are
centered at 10.5 GHz, 20.6 GHz, and 31.4 GHz. The product of µr 
r at these resonances is 4.14,
4.32, and 4.18, respectively, yielding an estimate for µr of 1.43, 1.49, and 1.44, respectively. These
values correlate fairly well with the fitted simulation model used to model the inductor printed on
the composite using µr = 1.6. The discrepancy may be related to the amount of field fringing
outside of the composite region in both cases since for both structures a different patterned dielectric
was used.

90

Figure 4.8 Composite ring resonator S-parameters.

4.3.4 EDS

To further characterize the magnetic composites, I used scanning electron microscopy, via backscat-
ter electron imaging, and energy dispersive spectroscopy. These were performed with a JEOL 7500F
scanning electron microscope with a cold field emission emitter using an Oxford EDS system. A
sample was coated with iridium for analysis. Fig. 4.9 shows a low magnification composition
image using a backscattered electron detector and EDS analysis of the highlighted region in the
image. In Fig. 4.9a, the PI border around the composite is visible as a slightly darker (less dense)
region. EDS analysis can be used to estimate the wt. % of Ni0.5Zn0.5Fe2O4 contained in the
composite. In the EDS analysis, the whole calculated content of Ni, Zn, and Fe contributes to the
composite content. Since the molar content of Fe should be half that of oxygen, the oxygen that
is bound in the composite can be found by dividing the EDS calculated Fe wt. % by the atomic
weight of Fe which yields a relative molar content. Doubling this figure yields the relative molar

91

content of oxygen bound in the composite. Multiplying that by the atomic weight of oxygen yields
an estimate of the wt. % of oxygen bound in the composite. Summing the oxygen content with the
EDS calculated Ni, Zn, and Fe content yields 18 wt. % Ni0.5Zn0.5Fe2O4 contained in the region
analyzed. As noted in the previous chapter, this estimate is representative of the material content
on the surface, rather than inside the material which may be slightly different since the material is
in a liquid state when deposited.

4.4 Resistive Composite Formulations

For resistive composites, rather than using a polymer-matrix nanocomposite, it makes more sense
to use composites of two conductive materials–one with high conductivity and another with low
relative to the first. This enables fabrication of a range of conductivities more conducive to resistor
fabrication, which is the primary end goal of this exercise. Polymer matrix nanocomposites would
also yield useful properties as essentially non-conductive, but lossy materials for EMI suppression.
For this reason, resistive composites were investigated in the form of silver and carbon, taking
the form of carbon black nanoparticles, graphene, and carbon nanotubes. The first carbon ink
investigated was a mixture of all three of these forms mixed 20 wt. % in water. The second, a
20 wt. % suspension of graphene and carbon black, and finally, a 13.5 % by vol. graphene and
carbon black mixture + 1.9 % silver from Clariant TPS 50 ink as has been used previously in this
dissertation in water. Silver was added in an attempt to stabilize the suspension.

A 6 wt. % suspension of multi-walled carbon nanotubes in water was also investigated, however
these films did not result in resistivities high enough to be useful for components such as resistors,
due to the low content of carbon. Composites with this formulation may be useful in flexible circuits
by increasing the strength of Ag films without negatively effecting the conductivity [142, 143].

These composites were printed in a similar fashion to the other MMAJP composites in this
dissertation. The carbon-based inks were printed with both a pneumatic atomizer as well as an
ultrasonic atomizer. Ultimately, it was found that the pneumatic atomizer yielded the best results.

92

(a)

(b)

Figure 4.9 (a) Composition backscatter electron image of composite with region highlighted in
red where EDS analysis was performed. (b) EDS analysis.

93

Silver was printed with a pneumatic atomizer when the carbon-based inks were printed with an
ultrasonic atomizer, and when the carbon-based ink was printed with an pneumatic atomizer the
silver was printed with an ultrasonic atomizer as in previous chapters.

4.5 Resistive Composite Measurement Results

First samples, printed with the graphene and carbon black ink, yielded promising results but due
to process drift in the printing process and rapid formation of clogs, other ink formulations were
investigated as well. A mixture of graphene, carbon black, and carbon nanotubes was next printed
with results correlating well with the established trend line of results from the first ink. This second
ink formulation was again found to form clogs in the printer too quickly. Finally, the most stable
ink formulation used was a mixture of 13.5 % by vol. graphene and carbon black mixture + 1.9 %
silver in the form of Clariant TPS 50 ink as described above.

All samples were 7 mm by 2 mm printed rectangles. These samples were cured under identical
conditions with a 2°C per minute temperature gradient from 20°C to 180°C for a four hour soak time
followed by an approximately 1°C per minute negative temperature gradient down to 20°C. Samples
were measured with a Lucas Labs Pro-4 four point probe system. To calculate the resistivity from
the measurement V/I,

ρ =

V
I

d
s

Ct,

(4.5)

as in [144], where d/s is the ratio of the shorter dimension of the printed rectangle to the space
between the probes, t is the thickness of the film and C is a correction factor given in [144] for
specified values of d/s. In this case, d/s = 1.2598, and C, interpolated from values given by [144],
is 0.9972. These calculations are shown in Fig. 4.10. and some of the fabricated films are shown
in Fig. 4.11. The dents where the four point probes landed on the films are visible in Fig. 4.11 b
through e. Fig. 4.11a additionally shows a printed gradient from 100% carbon based ink to 100%
silver ink.

94

Figure 4.10 Calculated resistivity from four-point-probe measurements of MMAJP resistive
composites.

Even while using three types of carbon based inks, these results show good correlation between
each other and together form a clear trend allowing for resistivity to be adjusted from 3.7−6
Ω− cm,
approximately 2.5 time that of bulk silver (using the present curing profile–approximately 40% of
the bulk conductivity), to 5.8 Ω − cm. On a logarithmic scale, the trend of increasing resistivity
with the percentage of carbon-based aerosol is approximately linear between 30% and 80% carbon
based aerosol content (the remainder being silver). Changing sintering times or temperatures
would shift these results somewhat. The highest conductivity I have achieved in a printed silver
film is approximately 80% of the bulk conductivity of silver. Lower temperatures and sintering
times would not, however, be a good method of tuning higher or lower conductivities in a printed
film. Such films would be less stable mechanically and would be prone to additional sintering, for
example, when high power or temperature was applied locally when in use in a circuit.

Using the MMAJP process and an approximately 50% carbon ink aerosol of 13.5 % by vol.

95

(a)

(b)

(d)

(c)

(e)

Figure 4.11 (a) Samples printed with carbon ink composed 13.5 % by vol. graphene and carbon
black mixture + 1.9 % silver, (b) detail of 40% carbon ink MMAJP mix, (c) detail of 50% carbon
ink MMAJP mix, (d) detail of 60% carbon ink MMAJP mix, (e) detail of 70% carbon ink
MMAJP mix.

96

graphene and carbon black mixture + 1.9 % silver mixed with pure silver ink, the dimensions of
resistor films can be calculated by

l = R wt
ρ

,

(4.6)

where l is the length, w is the width, t is the thickness, ρ is the resistivity, and R is the target
resistance of the film. For an approximately 53% carbon ink aerosol, I measured the film to have a
resistivity of 0.001 Ω− cm. For this value, a 4 µm thick film, 80 µm wide, will yield a resistance of
10 Ω at 0.28 mm long, and 50 Ω at 1.4 mm long. These dimensions are well within the capabilities
of AJP. Adjusting the aerosol content, the dimensions of these resistors can be tuned to whatever
purpose required–for example, in a microwave circuit the dimensions of a microstrip transmission
line can easily be maintained.

The carbon based ink can possibly be improved on by exploring other solvents which may allow
for a more stable dispersion of graphene. New ink formulations may also address process drift
issues in printing with these inks and the MMAJP process.

4.6 Conclusions

In this chapter, I expand on my work in Chapter 3 to magnetic and resistive composites for
microwave packaging applications. To the best of my knowledge, this is the first demonstration
of MMAJP magnetic materials. Using this process I increased the inductance density of a fully
printed microstrip spiral inductor by 40 %, reaching 4.2 nH/mm2. This result is, to the best of
my knowledge the highest inductance density presented to date for a planar AM inductor. This
magnetic composite exhibits a magnetic resonance behavior when subjected to a magnetic bias.
Using simulation models and ring resonator analysis, I estimate the permeability of this material to
be between µr = 1.4 and µr = 1.6. Using EDS analysis, I show that the material has approximately
18 wt. % ferrite loading. The ability to incorporate magnetic materials in printed circuits expands
on the capability of AM packaging processes, and allows the monolithic integration of inductors and
magnetic materials with the chip-first AJP process I present in this dissertation. This strategy may

97

allow for magnetic materials to be used as a microwave absorbing material, antenna miniaturization,
as well as for structures for sensing and filtering. With higher ferrite loading, these composites may
also be used to directly print structures such as circulators and isolators.

The magnetic composite fabrication strategy I present may be improved upon by improving the
quality of the ferrite nanoparticle suspension. Larger nanoparticles may also improve the range
of permeabilities that can be realized. Larger nanoparticles, closer to the atomization limit of the
ultrasonic atomizer of 50 nm, would have a larger magnetic domain, and would therefore potentially
display a higher effective permeability.

The ability to print directly resistive materials with a range of values allows direct printing of
resistors obviously, but it also may allow for novel structures for sensing using carbon nanotubes
or for other structures such as microwave loads or absorbing materials. Others have shown that
introducing carbon nanotubes into silver nanoparticle films may additionally increase the strength
of these conductive films which would have important application in flexible electronics.

98

CHAPTER 5

INTEGRATING CHIP-FIRST PACKAGING AND MMAJP

Chapters 2 [61, 68], 3 [73], and 4 outline the necessary basic tools of the chip-first packaging process
this dissertation presents–namely, the packaging procedure with the formation of interconnects, and
the printing of composites, facilitating the ability to directly print capacitors, inductors, and resistors.
What remains is combining these basic components into a functional microwave package with active
microwave components. This chapter shows two fully additively manufactured microwave packages
with integrated active and passive components using the procedures outlined in previous chapters.
That is, the package substrates (dielectrics) and interconnects (conductors) are built up around
power amplifier bare die attached to carriers. Multi-material aerosol jet printing as in Chapter 3
and 4 is used to mix and deposit thin film high dielectric constant polymer matrix nanocomposite
film for bypass capacitors integrated into the packages. The material properties of these films are
characterized with three separately printed MIM capacitors to show the consistency of the MMAJP
process with the results in Chapter 3.

This chapter shows two packaged GaAs pHEMT commercial-off-the-shelf (COTS) medium
power amplifiers (MPAs), HMC451 (Analog Devices) [1] as well as the fabrication process used
and the characterization of the printed bypass capacitors composed of a polyimide (PI) and barium
titanate (BaTiO3) nanocomposite fabricated with MMAJP. The process used in this demonstration
improves upon the process outlined in Chapter 2 based largely upon the improvements learned from
Chapter 3. These differences are highlighted below.

Prior to the work presented here and in [145], the only other report I obtained of a fully AM
[67] which
packaged active microwave component with an AM substrate is in Tehrani et al.
used inkjet printing and surface mount chip bypass capacitors. That work did not use the printed
substrate as an RF reference as the transmission line structures were coplanar waveguides, that
is, the thickness of the printed dielectric was not a critical dimension. Ramirez et al. [146] also

99

reported a hybrid packaging strategy using AM and laser micro-machining to create a microwave
package.

The fully additively manufactured chip-first package demonstration in this chapter achieves a
package loss < 2.3 dB across the entire 5 - 20 GHz pass band with an average pass band loss <
1.3 dB for both fabricated packages, which improves on the losses reported in [67]. The maximum
packaged gain is 21.7 dB, compared to the nominal gain of the bare die part of 22 dB. This
chapter also demonstrates large-signal measurements of these packages, achieving a maximum
Pout = 21.9 dBm compared to the manufacturer specified Psat ∼ 22 dBm.

While this work showed improved package performance, it also is, to the best of my knowl-
edge, the first integration of AM capacitors in such a package, as well as the highest RF power
demonstration to date. In the context of this dissertation, this chapter shows the most complete
packaging strategy, integrating the work in previous chapters. Finally, stable performance is mea-
sured over 1,000 power cycles and 7 temperature cycles up to 200 °C with a temperature gradient of
10 °C/minute. This demonstrates negligible performance degradation with moderate use and offers
evidence of the robustness of this packaging approach. These packaged circuits show improve-
ments in gain, output power, and bandwidth relative to the COTS packaged version of the MPA,
HMC451LP3 (Analog Devices) [2] in a QFN chip-scale lead frame package requiring external
capacitors. We also report the highest demonstrated AJP capacitance values to date as compared
with literature values.

5.1 Design

The packages are designed to be manufactured on a molybdenum copper (MoCu) alloy carrier to
match the coefficient of thermal expansion (CTE) of the COTS GaAs die and the low CTE polyimide
used for the dielectric. The 85% Mo and 15% Cu alloy used has a CTE of 6.2 ppm/°C. The package
design is similar to that in Chapter 2 [68], consisting of a printed dielectric substrate with printed
Ag microstrip (MS) transmission lines and power traces which form the die interconnects as well as

100

(a)

(b)
Figure 5.1 (a) Exploded view of package illustrating fabrication procedure and (b) cross section
showing material layers. ©2020 IEEE.

a conductor backed coplanar waveguide (CPW) to MS transition. The CPW is meant to facilitate
measurement with ground-signal-ground (GSG) probes.

The CPW to MS transition, as used in parts presented in Chapter 2, 4, and 4, is designed
to perform from DC to above the operating frequency of the die. This structure is similar to
Zheng et al.
[77] but includes additional grounding straps to the MoCu carrier as in [68, 73],
allowing performance to DC. The printed Ag additionally composes the top metal layer of the
MIM capacitors. The bottom layer of the MIM capacitor is the MoCu carrier as well as the ground
reference for the CPW, MS and the die.

101

(a)

(b)

Figure 5.2 Package design for both fabricated packages (package 1 and package 2) with
dimensions in millimeters. Dimensions in parentheses are target material thicknesses of PI (a) and
Ag (b). ©2020 IEEE.

102

(a)

(b)

Figure 5.3 Capacitor layout for nanocomposite characterization with dimensions in millimeters of
PI (a) and Ag (b). ©2020 IEEE.

A unique feature of this package design is that the dielectric surfaces are not coplanar. Such a
multi-leveled structure would not be practical to fabricate by conventional lithography and etching
processes. These levels consist of a thin high dielectric film composing the capacitor dielectric, a
thin PI layer bordering the capacitors where the power connections do not need to form impedance
matched transmission lines, a thicker layer on which the microstrip lines are printed, and dielectric
ramps or fillets up to the height of the die surface. The package and transmission lines are designed
for a 50 Ω environment. The design is shown in Fig. 5.2 and Fig. 5.1 shows an exploded diagram
of the different material layers and an illustration of the cross section of the package. Besides the
packages, to characterize the loss the MS, a 2x through structure was designed and fabricated. This
is shown in Fig. 5.4. The packaged die are passivated as received, but a modified version of this
process could include a final printed passivation layer of PI or some other material if required.

To characterize the material properties of the MMAJP high dielectric films, additional capacitors
were fabricated as shown in Fig. 5.3. These capacitors are smaller in area than the bypass capacitors
in the packages but otherwise of identical thickness, consisting of a dielectric film with three MIM

103

(a)

(b)

Figure 5.4 2x through design for AJP packages with dimensions in mm for PI (a) and Ag (b).
©2020 IEEE.

capacitors printed on top. Unlike the bypass capacitor in the packages, these capacitors are circular
structures with diameters of 0.4, 0.6, and 0.8 mm for characterization and analysis. From the
S-parameter measurements of these capacitors, the relative dielectric constant (
r) and loss tangent
(tan δ) can be extracted. The materials characterization method, along with the derivations and
assumptions related to obtaining these quantities can be found in Chapter 3 [73].

As previously explained in Chapter 3, simulated Short-Open-Load (SOL) calibration standards
were used to de-embed the launch of the capacitors to account for some of the parasitic effects of
the launch. This brings the reference plane to the edge of the MIM capacitor. This calibration is
imperfect, as it does not account for fringing fields and other parasitics in the capacitor structure
itself, including the interaction between the launch and the capacitor. Error in the calculation of

104

material properties increases with frequency but is adequate for the purposes of characterizing the
capacitor dielectric. All simulations were performed using ANSYS HFSS©.

In Chapter 2, close to the device, the microstrip was narrowed to compensate for the capacitive
discontinuity of the bond pad. In this chapter however, this compensation is not necessary because
the capacitive discontinuity of the bond pad does not affect the performance within the operating
range of the device. As previously noted, the larger bond pads typically used in COTS die to
accommodate multiple wire bonds are not necessary in this approach. However, devices without
bond pads are not typically available as off the shelf components.

This package can be integrated into a larger system in several ways. A microwave system-in-
package could be designed which includes down mixing to baseband frequencies where wire bonds
will not present a bottleneck to performance. Kaestle et al. [147] demonstrated wire bonding to
printed silver. Packages could also be printed directly onto a printed circuit board material (PCB)
rather than using a MoCu carrier as is demonstrated in this paper, or placed in a cavity of a PCB.
Other carriers such as alumina or silicon would also be suitable for this packaging strategy.

5.2 Fabrication

The fabrication process for these packages largely mirrors the process presented in Chapter 2, but
with some important modifications which improve yield and package robustness. As before, the
first step of this procedure is the die attach so that the packages can be fabricated with respect to the
die placement. I first prepared a 0.5 mm thick 85% Mo 15% Cu plate by mechanical polishing. Die
attach was performed with Epo-Tek H20E Ag epoxy and the manufacturers recommended curing
profile in a nitrogen environment to prevent oxidization of the MoCu. Immediately prior to printing
in order to improve the adhesion of the PI and nanocomposites, I applied diluted VM651 adhesion
promoter (HD Microsystems) by dropping on and drying off. Other adhesion promoters, such as
hexamethyldisilazane (HMDS) based adhesion promoters would likely be equally effective.

As in Chapter 3, the PI ink was composed of PI2611 polyamic acid (HD Microsystems) diluted

105

(a)

(b)

(c)

(d)

(e)

Figure 5.5 One of the two fabricated packages and material characterization structures: (a) bypass
capacitors, (b) packaged die, (c) perspectives of the package in (a) and (b), (d) 2x through, and (e)
capacitors for material characterization. ©2020 IEEE.

106

to 33 vol. % PI2611 + N-Methyl-2-Pyrrolidone (NMP) (Sigma Aldrich). As PI2611 is a proprietary
formulation, it is unknown the exact polyamic acid concentration. The polyamic acid was diluted
in order to improve atomization during printing. A concentration of 33 to 45 vol. % PI2611 was
found to be acceptable for printing of the PI ink. Cured PI2611 has a published 
r = 2.9 and
tan δ = 0.002 at 1 kHz. PI2611 was chosen for its low CTE to match to the die and the MoCu
carrier. The published CTE of cured PI2611 is 3 ppm / °C. The significance of these changes are
discussed further below.

BaTiO3 ink was composed of 20 wt. % 50 nm cubic phase BaTiO3 dispersed in NMP (US
Research Nanomaterials). A nanocomposite of BaTiO3 and PI was fabricated with MMAJP as
in Chapter 3 and [73]. It is not necessary to mix a separate ink of a BaTiO3 and polyamic acid
composite prior to deposition. As in previous chapters, the Ag ink was composed of 25 wt. %
Clariant Prelect TPS 50 + deionized water. All nanomaterial based inks–the Ag and BaTiO3 inks
were mixed under ultrasonication for at least 1 hour prior to printing.

As discussed in more detail below, all printing of the packages was performed with the MoCu
carrier grounded in order to prevent any accidental static charge build up during the aerosol
deposition process. During printing, the ambient temperature varied between 22 and 24 °C. The
printing was performed with an Optomec Aerosol Jet 5X printer as before, and for all inks at a print
speed of 1 mm/s.

The PI ink was deposited with a pneumatic atomizer and an aerosol flow rate of 55 sccm with
a focusing gas (sheath, N2) flow rate of 90 sccm. The PI ink was heated to a temperature of 25 °C
during printing so that it would not vary with the ambient temperature of the room and to reduce the
viscosity of the ink, increasing atomization. MMAJP printing of PI and BaTiO3 was accomplished
by mixing aerosols as in Chapter 3. The BaTiO3 was printed with an ultrasonic atomizer. The
liquid BaTiO3 ink was maintained at a temperature of 27.5°C. The BaTiO3 ink was heated to reduce
its viscosity and therefore improve atomization. The temperature of the BaTiO3 ink is limited by
the safe operating temperature of the ultrasonic transducer which is 30°C to prevent damage.

Since in this chapter the aim of MMAJP is to integrate printed bypass capacitors into a microwave

107

package, multiple aerosol mixing ratios are not needed, and only a PI composite with a high
concentration of BaTiO3 is required. So, for this chapter, for the MMAJP films, the PI aerosol flow
rate was maintained at 55 sccm and the BaTiO3 aerosol flow rate was 25 sccm. For the dielectrics,
the print stage was heated to approximately 100 °C. This allows the ink to dry as it is printed which
improves the surface quality and allows for thicker films to be deposited. For all dielectric inks, a
300 µm diameter nozzle was used.

After applying the adhesion promoter to the MoCu carrier, the first four layers of PI and to
layers of the nanocomposite were deposited. Four layers of PI ink were deposited that coincide
with those shown in Figs. 5.2, 5.3, and 5.4 in the areas marked with a 5 µm design target thickness.
The PI fillets were printed at this time by printing at an angle of 30° to normal along the edges
of the die, which is 0.1 mm thick. This fillet protects the sides of the die and allows a smooth
transition from the PI to the die. The area that the fillet occupies is indicated in Fig. 5.2 as “PI
Ramp.” These printed layers cover all the area that dielectric will exist in the package design so
additional application of adhesion promoter is not required.

A soft bake of these initial layers was performed at a temperature of 200 °C for a 2 minute hold-
time, with a 2 °C/minute maximum temperature gradient in a nitrogen environment, a rise-time
of approximately 90 minutes. Curing these layers before depositing additional layers reduces the
tension in the films so they are less likely to pull away from the surfaces they are printed on. Of
particular concern are the edges and surface of the die which is much smoother and because of the
sharp edges covered by the PI, holds more tension in the film.

Following the initial layers and the PI soft bake, 20 layers of PI were printed in the 30 µm
design target thickness area followed by another soft bake. Finally, 10 layers of PI were printed
followed by a hard bake to achieve 100% imidization by heating the package to 295 °C for a 1 hour
hold-time with a 2 °C/minute temperature gradient, a rise-time of approximately 138 minutes, also
in a nitrogen environment. An additional change compared to Chapter 2 due to what was learned
in Chapter 3, is prior to each curing step, the packages were held under vacuum for several hours to
prevent the films from absorbing moisture and to mitigate air bubbles that can become trapped in

108

the films during printing. Previous AM packages presented by [67] have used thin layers of printed
dielectric and coplanar waveguide transmission lines which are not as sensitive to dielectric layer
thicknesses as microstrip transmission lines. This work in this chapter as well as the packages in
Chapter 2 [61, 68] all use microstrip transmission lines. This is possible because we can more
accurately deposit and control thicker dielectric layers.

Finally, with all the dielectric layers printed to the design height, the Ag ink was printed to
form electrical connections to the die. Ag ink was printed using a 150 µm diameter nozzle with
an ultrasonic atomizer and an aerosol flow rate of 22 sccm with a sheath gas flow rate of 52 sccm.
Three layers of Ag ink were printed with an additional three layers on the PI fillets at an angle
of 15° to normal. Additional layers of Ag were printed to ensure adequate Ag coverage in these
areas which are up a dielectric ramp. The Ag ink was cured in air at 180 °C with a 2 °C/minute
temperature gradient, a rise-time of approximately 80 minutes, for a 4 hour hold time. On a
representative sample of printed Ag, this curing profile achieved a conductivity of 39% of bulk Ag.
The total active printing time for a single package is approximately 100 minutes. Pictures of one
of the two fabricated packages of the same design and the capacitors for material characterization,
as well as the 2x through transmission line are shown in Fig. 5.5.

5.2.1 Packaging Considerations for Active Components

This process resolved some failures that were previously observed. As discussed in Chapter 3,
significantly improved microwave loss performance was observed with PI that was fully imidized.
Furthermore, PI that is not entirely imidized is more likely to pull away from bare die parts during
processing. While processing temperatures in semiconductor front end of line fabrication are
very high for diffusing dopants into semiconductors, some back end of line processes such as the
formation of air bridges are more sensitive to high temperatures.
Ideally, the PI ink would be
cured at 350°C, but this is beyond the processing temperature limits recommended for many COTS
MMIC parts. At temperatures close to 300°C, these sensitive structures should not be damaged
and it is possible to achieve 100% imidization of the polyamic acid [122]. This is the reason why

109

the packages presented in this chapter were only processed to 295°C.

Another potential source of failure in packages that are otherwise mechanically robust is
electrostatic discharge (ESD) from the AJP process itself. Aerosol droplets may build up charge
on a device high enough to cause damage. While the print stage in the Optomec AJP 5x printer is
grounded, it is made of anodized aluminum so no electrical connection is made to parts placed on
the print stage. For this reason, the MoCu carrier was grounded by placing it on an intermediate
copper coated surface that was grounded, so that an electrical connection was made to dissipate
any charge that might build up from the aerosol. Passive parts, such as those packaged in Chapter
2, would not suffer failures from ESD or high temperatures, but for active components such as
the power amplifiers packaged in this chapter, these are important considerations. Without these
considerations, package failures were observed in packages which did not have obvious mechanical
failures such as the package in Fig. 5.6.

Finally, an important consideration for any packaged semiconductors is matching the coefficient
of thermal expansion (CTE) of materials. Many polymers, including more conventional polyimide
formulations typically available, have a high CTE relative to semiconductor materials which fall
between 2-6 ppm/°C. Conventional polyimide such as Kapton (DuPont) has a CTE of 20 ppm/°C,
but low CTE formulations of PI such as PI2611 (HD Microsystems) has a CTE of 3 ppm/°C which
is well matched to silicon. This is the primary reason for the switch to PI2611 rather than the
polyamic acid used in Chapter 2. Other materials used for AM packaging include SU-8 epoxy,
as used in [67], has a CTE as high as 102 ppm/°C. BCB, another commonly used material in
packaging has a CTE of 42 ppm/°C. These other materials, with a CTE an order of magnitude or
higher than semiconductor materials, may not produce as robust a package.

5.3 Results

The package dielectric and conductor thicknesses were measured with a NanoMap-500LS surface
profilometer. The package profile measurements are shown in Table 5.1. Variations from the

110

Figure 5.6 A mechanically good package which failed prior to high temperature and ESD process
changes.

design target thicknesses are due to process drift during printing such as variations in atomization.
The profile measurements are consistent with the target design thicknesses.
In considering the
effect of the surface roughness on RF circuit performance, we expect the roughness of the surface
of the dielectric to be more relevant figure than the surface roughness measured on the top of the
conductors. The surface roughness of the metal structures at the interface with the PI will be closer
to that of the dielectric surface roughness. In microstrip transmission lines, most of the current
flows on the bottom face of the conductor as dictated by the skin depth. The skin depth of the
printed silver features is ∼ 0.7 µm at 20 GHz and ∼ 1.4 µm at 5 GHz.

111

Figure 5.7 Measurement setup.

Table 5.1 Measured Printed Material Thicknesses

Measurement
4 layers PI
35 layers PI

2 layers BaTiO3

3 layers Ag

Thickness (µm) RMS Roughness (µm)

4.649
27.260
2.754
1.977

0.061
0.050
0.210
0.393

©2020 IEEE.

5.3.1 Material Characterization

S-parameter measurements were taken with a MPI TS150-THZ Probe System, a Keysight N5227
PNA, 100 µm pitch 67 GHz probes, and 100 µm pitch power-ground-power probes, both manufac-
tured by Cascade Microtech. TRL calibration was performed to bring the measurement reference
plane to the probe tips, corresponding to the start of the CPW transmission line and transition to
MS as shown in Fig. 5.5. Fig. 5.7 shows a part being probed for measurement. As explained
above and in Chapter 3, I simulated SOL standards for capacitor structures using measured material

112

thicknesses and expected material properties extrapolated from [68] of 
r = 7 and tan δ = 0.02.
That is, this composite mixture is expected to have a slightly higher concentration of the BaTiO3
than Mix 11 of the improved ink formulations in Chapter 3. As discussed in Chapter 3, the SOL
calibration helps account for the parasitic effects of the launch structure. This simulated calibration
does not account for fringing effects of the capacitor structure itself so error still increases with
frequency after applying this second calibration. Characterization of the dielectric films is therefore
most accurate at the lowest frequencies.

The results of the measurements, including the extracted 
r and tan δ for the three MIM
capacitors is shown in Fig. 5.8. Averaging the calculated 
r of the three capacitor characterizations
yields an 
r of 7.1, which correlates well to the expected value of 7 used for the simulated calibration.
Based on Chapter 3 and [73], the material properties evaluated at lower frequencies should remain
fairly flat with increasing frequency, though for these packages, since the capacitors are power
bypass capacitors, the material characterization is most relevant at lower frequencies. The increase
in tan δ seen in Fig. 5.8c is an artifact of parasitic effects not accounted for by the calibration
rather than the actual physical behavior of the material. The tan δ calculation is clearly particularly
sensitive to these parasitic effects. The effect of parasitics is also apparent when comparing the
three calculations. Since the capacitors are different sizes, they are impacted unequally by parasitic
effects. That is, larger capacitors have more fringing fields and are closer to the launch allowing for
more coupling to the capacitor. These results are consistent with the results obtained in Chapter 3
[73], demonstrating the repeatability of the MMAJP process.

Based on these capacitors, the printed bypass capacitors in the packages have a value of 28.4
pF. To the best of my knowledge, these printed capacitors are the highest demonstrated capacitance
value of any previously reported AJP capacitor. The capacitance density is also higher than all
other previous reports with the exception of [135, 58, 148] who all reported a lower quality factor
than the capacitors presented here. The MIM capacitors here are also competitive with the IDC
capacitors I presented in [65]. A summary of these comparisons is shown in Table 5.2.

113

(a)

(b)

(c)

Figure 5.8 Measured capacitors and extracted dielectric properties. Parasitic effects lead to error
increasing with frequency: (a) capacitance, (b) 
r, and (c) tan δ. ©2020 IEEE.

114

Table 5.2 Comparison with current state-of-the-art AM capacitors

Ref. Geom.

Process

C (pF)

C

Q

(pF/mm2)

This MIM
This MIM
This MIM
Thisa MIM
IDC
[65]
[65]
IDC
IDC
[149]
[150]
IDC
[135] MIM
[140] MIM
[151] MIM
[151] MIM
[152] MIM

AJP
AJP
AJP
AJP
AJP
AJP
AJP

LEDPAMc

Inkjet
Inkjet
Inkjet
Inkjet
Inkjet

2.38 (0.5 GHz)
5.46 (0.5 GHz)
9.35 (0.5 GHz)
28.43 (0.5 GHz)
1.10 (0.5 GHz)
1.28 (0.5 GHz)
6.24 (1 MHz)
0.14 (1 GHz)
16.6 (14 MHz)
5.10 (0.5 GHz)
12 (200 MHz)
25 (100 MHz)
50 (0.5 GHz)

18.95
19.31
18.59
18.87
1.70
1.98
0.325
0.533
48.82
6.5
15.8
33
22.2

45.38
52.43
70.20
N/A
126b
30b
15
750
12.9
15
22
25
3.9

aExtrapolated from measured capacitors, bIncludes line loss, measured at 0.5 GHz, cLaser
enhanced direct printing additive manufacturing. ©2020 IEEE.

5.3.2 Small-Signal Package Measurement

The package S-parameters were measured with the same experimental setup described in IV.A. A
TRL calibration was performed using a CS-5 substrate bringing the measurement reference plane
to the probe tips. On the package, this corresponds to the start of the CPW transmission line and
transition to MS. Small signal S-parameters were measured with an input power set at -15 dBm.
The measurements of the printed package compared to the manufacturer’s published bare die data
as well as the manufacturer’s published packaged die data are shown in Fig. 5.9.

My packages achieved a maximum gain of 21.2 dB during these initial measurements. After
operating the package at Psat, the maximum gain improved to 21.7 dB as described below. The
advertised gain for the bare die part is 22 dB. The commercially available packaged version of the
die is advertised to have a gain of 18 dB. Furthermore, it is clear from Fig. 5.9 that the bandwidth

115

of my printed packages is wider than the COTS package. This increased bandwidth is possibly
due to the bandwidth of the interconnect methodology. The area the printed package occupies
is 20.63 mm2. This area could be reduced by reducing the microstrip length and removing the
CPW to MS launch, which in an implementation which integrates this package into a larger system
would not be necessary to include. The area could also be decreased by using multilayer capacitors.
The area of the COTS package is approximately 9 mm2, but the COTS package does not include
required bypass capacitors.

Fig. 5.10 shows the 2x through measurement for the printed packages and an image of the
fabricated structure is shown in Fig. 5.5d. To calculate the performance of the die and interconnects
alone, The loss of the 2x through is subtracted from the measured package. This result is shown
in Fig. 5.11. The calculation is shown using S21 from Fig. 5.9, that is, before large-signal
measurements and power cycling as described later in this paper, as well as after. As described
later, the package performance improved slightly after large signal measurements. The difference
between the bare die and the de-embedded measurements in Fig. 5.11 is an estimation of the
interconnect loss.

The loss of the printed package components, that is, including the MS line constituting the
loss of the 2x through and the interconnects, is the difference between the measured packages and
the bare die performance. This loss is better than 2.3 dB across the entire passband of the die
for both package 1 and 2. The average package loss is for package 1, package 2, and package 2
after large signal measurements, respectively 1.3 dB, 0.8 dB, 0.6 dB. These averages are taken over
the passband of the amplifier, 5-20 GHz. Some of the differences between the two packages can
be attributed to the differences between individual die. The performance of the printed packages
compares favorably to the performance of the advertised performance of the bare die amplifier.

Table 5.3 shows a comparison between this work and similar AM/primarily AM packaging
strategies. For comparison points from the packages in this chapter, points were taken at 15 GHz
rather than 20 GHz. 20 GHz is a comparison point more readily available from other works but
20 GHz is the very edge of the passband and at 20 GHz the insertion loss of the package is lower

116

(a)

(c)

(b)

(d)

Figure 5.9 Measured small signal S-parameters of the AJP packages compared to the published
bare die [1] and COTS packaged die [2]: (a) S11, (b) S12, (c) S22, and (d) S21. ©2020 IEEE.

than the bare die measurement. This is likely due to factors such as the biasing point of the die and
possible physical changes to the die during processing (e.g. exposure to high temperatures) which
increased the bandwidth and/or shifted the passband of the die up in frequency slightly.

The line loss of the package I calculated uses the 2x through measurement. The line loss of the
package is 0.2 dB/mm at 30 GHz, and between 0.1 and 0.2 dB/mm in the range 5 - 20 GHz which
is the operating range of the amplifier.

117

Figure 5.10 Measured 2x through. ©2020 IEEE.

Figure 5.11 Packages with loss of 2x through removed compared to bare die performance [1].
©2020 IEEE.

118

Table 5.3 Comparison with other AM Packages

Ref.

Die

Process

Estimated

This (package 1)

GaAs PA

This (package 2)

GaAs PA

AJP

AJP

Chapter 2 [68]

GaAs 0 dB

AJP

Chapter 2 [68]

GaAs 0 dB

AJP

Chapter 2 [61]

Si 0 dB

AJP

Tehrani et al. [67]

GaAs 0 dB

Inkjet

Tehrani et al. [67]

GaAs LNA

Inkjet

Ramirez et al. [146] GaAs LNA Mixed b

Package Loss

1.7 dB
(15 GHz)
1.0 dB
(15 GHz)
0.58 dB
(20 GHz)
0.42 dB
(20 GHz)
0.44 dB
(20 GHz)
1.08 dB
(24.5 GHz)
2.9 dB a
(24.5 GHz)

2.4 dB
(20 GHz)

Estimated Line and Capacitors
Interconnect Loss

Pout

0.55 dB/mm
(15 GHz)
0.33 dB/mm
(15 GHz)
0.2 dB/mm
(20 GHz)
0.15 dB/mm
(20 GHz)
0.12 dB/mm
(20 GHz)
0.45 dB/mm
(24.5 GHz)
0.57 dB/mm a
(24.5 GHz)
0.2 dB/mm
(20 GHz)

Integrated

Integrated

20.9 dBm
(9 GHz)
21.9 dBm
(9 GHz)

N/A

N/A

N/A

N/A

Chip

N/A

N/A

N/A

N/A

N/A

4.2 dBm
(24.5 GHz)

N/A

aMeasurement taken at a low-loss point.
bFDM, micro-dispensing, laser micro-machining ©2020 IEEE.

119

5.3.3 Large-Signal Measurements

The work in this chapter also includes large-signal measurements, which to the best of my knowledge
is the first such demonstration for any similar packaging strategy for microwave electronics. Large
signal measurements were taken with a Maury Microwave MT2000 Mixed Signal Active Load-Pull
System from 8 - 11 GHz within external amplifier limitations. These measurements are shown in
Fig. 5.12, 5.13, 5.14, and 5.15 for frequencies of 8, 9, 10, and 11 GHz, respectively. These figures
show the Pout load-pull measurements. For each frequency, the input power deviation was less
than 0.05 dBm and the input power was nominally set to 2 dBm. For each measurement, the source
impedance was set to 50 Ω.

The maximum measured package 1 output power was 20.9 dBm. The maximum measured
package 2 output power was 21.9 dBm. Both measurements were taken at 9 GHz. The manufacturer
specifies a saturated output power of 22 dBm for the COTS bare die part in a 50 Ω environment. The
packaged version of this COTS die, HMC451LP3, specifies a saturated output power of 21 dBm in
a 50 Ω environment, that is, approximately the output power that we achieved with printed package
1 and less than the maximum output power we achieved with printed package 2. The maximum
power point of our package was not at a matched load, and was shifted to ΓL = −0.2 + 0.1j which
is equivalent to ZL = 31.3 + 5.5j Ω for package 1 and to ΓL = −0.4 + 0.1j which is equivalent to
ZL = 19.7+3.9j Ω for package 2. This maximum power point is marked in Fig. 5.13. At a matched
load, we measured the maximum output power at 8 GHz as 21.0 dBm. In these measurements, I
did not attempt to exceed the advertised power or go beyond Psat.

5.3.4 Reliability

A rigorous reliability investigation of these packages was not possible with available equipment,
but to provide some characterization of the reliability of these packages and observe any change in
RF performance over time, I power cycled one of the two packages and temperature cycled both
packages. Power cycling was done at a duty cycle of approximately 2 seconds, with approximately

120

(a)

(b)

Figure 5.12 Large signal load pull measurements of the AJP package at 8 GHz: (a) package 1 and
(b) package 2. ©2020 IEEE.

121

(a)

(b)

Figure 5.13 Large signal load pull measurements of the AJP package at 9 GHz: (a) package 1 and
(b) package 2. ©2020 IEEE.

122

(a)

(b)

Figure 5.14 Large signal load pull measurements of the AJP package at 10 GHz: (a) package 1
and (b) package 2. ©2020 IEEE.

123

(a)

(b)

Figure 5.15 Large signal load pull measurements of the AJP package at 11 GHz: (a) package 1
and (b) package 2. ©2020 IEEE.

124

1.5 seconds on time and 0.5 seconds off time. During power cycling, the maximum gain deviated
by 0.32 dB, and increased slightly after the first 10 power cycles. This deviation after the first few
power cycles may be due to additional microwave sintering of the Ag nanoparticle ink during load
pull measurements. leading to lower loss of the package.

This deviation is not significant relative to the package performance but is too large to be
accounted for by a change in the measurement probe landing position. Microwave sintering of
silver nanoparticle inks has been demonstrated by Perelaer et al. [153] and others. After 1,000
power cycles, the package performance was basically unchanged. The result of these measurements
is shown in Fig. 5.16 a.

Following this, temperature cycling was done on both parts from 20 °C to 200 °C with a
+10 °C/minute temperature gradient increasing to 200 °C for a rise-time of approximately 18
minutes and a hold-time of 1 minute. The negative temperature gradient at the end of this cycle
is approximately -1 °C/minute due to equipment limitations. These cycles were additionally
completed in an oxidizing environment. Small signal measurements were repeated after 1, 3, and 7
temperature cycles. During these cycles, no physical changes in the packages were observed such
as cracking of the dielectric or metallic structures. The only changes observed were additional
oxidization around the printed package of the MoCu. The results of these measurements are shown
in Fig. 5.16 b.

After temperature cycling, the performance of package 2 reduced back to close to the initial
measurements but essentially no change in performance was observed after 7 temperature cycles.
After 7 temperature cycles, package 1 began to show some decrease in performance at the higher
end of the amplifiers pass band, but the change in performance was less than 0.5 dB. To our
knowledge, this is the first demonstration of temperature cycling of an AM microwave package.

125

(a)

(b)

Figure 5.16 Deviation in performance with power and temperature cycles: (a) maximum gain and
(b) comparison between initial gain measurement and after 1,000 power cycles and after 1, 3 and 7
temperature cycles. ©2020 IEEE.

126

5.4 Conclusions

This chapter demonstrates two fully AM packaged MPAs, fabricated using a chip-first process
as in Chapter 2 and including for the first time MMAJP capacitors using the process developed
in Chapter 3. Both packages demonstrated improved performance over previous reports of AM
microwave packages, as well as improved performance over the commercial packaged version of
the same. The maximum packaged gain is within 0.3 dB of the bare die performance and the output
power within 0.2 dB of the bare die performance. The printed bypass capacitors also exhibit the
largest capacitances of AJP capacitors to date. Finally, I demonstrate that with moderate use the
package performance does not degrade by stressing the package performance with power cycling
and temperature cycling.

The significance of this work in the context of this dissertation is extending and combining the
work in chapters 2 and 3. This shows a more complete and mature packaging process using AJP.

127

CHAPTER 6

EXTENDING THE CHIP-FIRST PACKAGING STRATEGY TO W-BAND

FREQUENCIES

While previous chapters have discussed and demonstrated the usefulness of AJP and MMAJP
for heterogeneous-compatible microwave circuit packaging, what remains is to demonstrate the
extent to which these strategies can be extended into higher frequencies. At mm-wave frequencies,
performance bottlenecks and manufacturing challenges to overcome them are far more significant,
as discussed in Chapter 1. At mm-wave frequencies, heterogeneous integration techniques are
critical to building systems that are cost-effective. With the rise in demand for technologies like
automotive radar and mm-wave imaging and sensing, there is a growing demand for broadly useful
packaging strategies for mm-wave electronics in particular.

The flexibility in design and fabrication offered by AM strategies is particularly desirable at
mm-wave frequencies where small features have a large impact on performance. Since other AM
strategies that have been used for packaging like inkjet printing have a characteristically much
lower feature resolution, AJP is one of the only AM strategies that can reasonably be implemented
for mm-wave circuits while still taking advantage of the unique flexibility of AM. Incorporating
MMAJP extends the flexibility of materials available even further beyond standard AJP processes.
Together these strategies enable application-specific packaging at mm-wave frequencies.

To the best of my knowledge, there have been no reports of an AM packaged commercial off the
shelf device operating in the W-band. [154] showed a printed CPW on a dielectric ramp operating
to 110 GHz. [155] showed a similar interconnect to a transmission line on a GaAs die operating to
210 GHz.

In previous chapters I showed a heterogeneous-compatible fully additive chip-first packaging
approach operating up the K and V bands. In this chapter, I extend on the strategy shown in Chapter
5 and demonstrate a similar strategy tailored for use in the W-band. To this end, this chapter

128

demonstrates a fully additively manufactured W-band package with integrated bypass capacitors.
This package is fabricated using a chip first approach and aerosol jet printing using a process similar
to that laid out in Chapter 5. The packaged die is a commercial-off-the-shelf W-band GaAs HEMT
amplifier. Notably, this chapter expands on the present strategy by demonstrating shaped ramp
interconnects from microstrip transmission lines which allow for frequency agnostic interconnect
performance with no impedance discontinuity within the physical resolution limits of the printer.
High dielectric thin films for bypass capacitors are demonstrated in an identical fashion to the
process laid out in Chapter 5 using MMAJP. This package achieves a maximum measured gain of
14.9 dB in the W-band, compared to the maximum bare die gain of 15.5 dB at the same frequency.
The average package loss is 1.1 dB from 75 to 92 GHz, including a 3 mm long transmission line,
interconnects and the GSG launch CPW to MS transition used in previous chapters. This average
total loss correlates to an average loss of 0.37 dB/mm.

6.1 Design

The package design in this chapter is similar to that in Chapter 5 with some notable differences.
The general structure parallels that in Chapter 5, where the package is printed around a die placed
first on a MoCu alloy carrier. The amplifier chosen was HMC-ALH509 (Analog Devices), a GaAs
HEMT MMIC amplifier with a passband from 71-86 GHz and a nominal gain of 14 dB. This die
is 100 µm thick. As before, the package is multi-leveled, allowing transmission lines on thicker
substrates 30 µm thick, and low frequency power interconnects on thin substrates 5 µm thick without
the use of vias to interconnect levels. The high dielectric composite film is designed to be 2 µm
thick. PI is used for the bulk of the dielectric with a PI and BaTiO3 composite for bypass capacitor
films as previously demonstrated. As before, the package uses printed dielectric ramps around the
die to support the interconnect. On top of the dielectric, the conductors are printed composing
the transmission lines, interconnects, CPW to MS transition, capacitor top metal layer, and power
connections conformally on this multi-leveled structure. As in previous chapters, the CPW to MS

129

transition is included to facilitate measurement with ground-signal-ground (GSG) probes. The
MoCu carrier acts as the ground reference for the microstrip and CPW transmission lines as well
as the lower metal layer of the MIM capacitor and the electrical connection to the backside of the
die. This multi-leveled structure with conformal electrical connections would not be practical to
fabricate by conventional means.

In Chapter 2, interconnects were narrowed on ramps to compensate for the capacitance of the
bond pad. That is, this feature was treated as a lumped inductive element in the circuit. For the
W-band die in this chapter, the small bond pads create a negligible capacitive discontinuity that does
not need to be compensated for in this manner. However, the length of the ramp itself is significant
with respect to a wavelength in the W-band, so leaving the interconnect geometry constant as in
Chapter 5 will not allow for acceptable performance. Unlike in previous chapters, the interconnects
are shaped, so that with the increasingly thick substrate up the ramp to the die surface a 50 Ω
characteristic impedance is maintained. The transmission line widens to roughly the width of a 50
Ω transmission line on 100 µm of PI at its widest before narrowing to 10 µm narrower than the
width of the bond pad. This interconnect was optimized in simulation with ANSYS HFSS. With
this method, an almost frequency agnostic interconnect with no impedance discontinuity within
physical limitations of the printer is achieved.

The design of this package is shown in Fig. 6.1. Fig. 6.2 shows an illustration of the multi-
leveled package structure and an image of the shaped ramp interconnect design. A simulated
tolerance sensitivity analysis was performed of the printed microwave structures on either side of
the die including the CPW launch, microstrip, and shaped interconnect. This analysis was done
to determine how sensitive the design is to process drift during fabrication by varying substrate
thickness, and conductor features between +/- 4 µm. For all simulations, the return loss of the
structure stays below 13 dB in the frequency range of interest, and below 19 dB for the nominal
case. The entire structure, including the amplifier was analyzed by concatenating the S-parameters
for the nominal case of the package design with the manufacturer’s provided S-parameters of the
bare die performance at the input and the output. The results of this link simulation are shown in

130

Figure 6.1 Annotated package design in millimeters. Dimensions in parenthesis are target
material thicknesses.

Fig. 6.5.

6.2 Fabrication

Following the chip-first packaging process, the die attach occurs first on a 0.5 mm thick 85%
Mo 15% Cu polished carrier using Epo-Tek H20E Ag epoxy. The epoxy is cured in a nitrogen
environment to prevent oxidization of the carrier. Prior to printing the first layers of dielectric,
VM651 (HD Microsystems) adhesion promoter is applied. All printing is accomplished with an
Optomec Aerosol Jet 5x printer as in previous chapters.

131

HMC-ALH509Ag (0.004)PI (0.03)PI RampPI + BaTiO3PI (0.005)1.5000.1001.4251.500(a)

(b)

(c)

Figure 6.2 (a) Exploded view of package illustrating fabrication procedure. (b) Illustration of the
multi-leveled structure of the package design. (c) Illustration of interconnect detail.

132

The PI ink was composed of 40 vol. % low CTE PI, PI2611 (HD Microsystems) in N-Methyl-
2-Pyrrolidone (NMP) (Sigma Aldrich). This is a slightly higher concentration than used in Chapter
5, allowing a better balance of material deposition and atomization rate of ink during printing. The
BaTiO3 ink is again composed of 20 wt. % 50-nm cubic phase BaTiO3 dispersed in NMP (U.S.
Research Nanomaterials) as in Chapter 5.

Fig. 6.2 indicates the first layers of dielectric to be printed are the 5 µm layers of PI and the
ramps around the die. Following these initial layers of PI, the high dielectric films for capacitors
are printed using the MMAJP process I develop in Chapter 3. This covers all areas where PI or the
high dielectric composite will be deposited in the completed package so that no additional adhesion
promoter needs to be applied. These first layers are then partially cured/imidized in a soft bake at
200°C in a nitrogen environment. Following the soft bake, an additional 25 µm of PI is printed in
areas targeted for a 30 µm dielectric thickness as indicated in Fig. 6.1. The complete dielectric
structure is fully imidized at 295°C, again in a non-oxidizing (nitrogen) environment.

Conductive features are conformally printed on the resulting cured multi-leveled dielectric
structure. The same silver ink used in previous chapters was again used here, composed of 25
wt. % Clariant Prelect TPS 50 in deionized water. As before, the Ag ink is cured in air at 180°C
for four hours. The active printing time for this package was 1 hour 11 seconds. This printing time
could be improved using wider diameter deposition nozzles or further optimizing ink composition
and flow rates. Fig. 6.3 shows an image of the fabricated package as well as a perspective view of
the fabricated package.

6.3 Results

The profile of the printed package features were measured with a NanoMap-500LS surface pro-
filometer. The area with a target thickness of 5 µm measured 6.7 µm thick, the area with a target
thickness of 30 µm measured 29.5 µm thick, and the BaTiO3 measured 3.7 µm thick. The silver
was measured to be µm thick. These regions are indicated in Fig. 6.1.

133

(a)

(b)

Figure 6.3 Annotated fabricated package composite image.

134

S-parameter measurements were taken from 75 to 110 GHz within equipment limitations with a
MPI TS150-THz Probe System, a Keysight N5227 PNA, 100 µm pitch 110-GHz Cascade Microtech
Waveguide Infinity Probes, and Virginia Diodes 75 GHz to 110 GHz Waveguide Extension Modules.
The test setup is shown in Fig. 6.4. LRRM calibration was performed to bring the reference plane
to the tip of the probes, illustrated in Fig. 6.3. The output power at the probes was attenuated to
approximately -7 dBm to avoid saturating the amplifier.

Fig. 6.5 shows the result of this measurement compared to the manufacturer’s data for the
amplifier and the link simulation of the package and amplifier as described in the design section.
The simulated performance of the package correlates very well with the measured package.

To estimate the package loss due to printed components, that is, not from the die itself, the die
transmission is subtracted from the package transmission S-parameters. This is only an estimate
of package loss because of die to die variations in performance. The transmission lines connected
to the input and output of the amplifier are 1.5 mm long each for a total of 3 mm long. Calculating
the loss per mm gives another perspective of the loss that can be expected using this packaging
strategy. The total package loss estimate as well as the line loss per mm is shown in Fig. 6.6. The
average loss of the package from 75 to 92 GHz, the range of data provided for the amplifier that
falls in the range of measurement, is 1.1 dB correlating to an average loss of 0.37 dB/mm including
interconnects.

Table 6.1 shows a comparison of this package with other AM packages, including [146] and
[155] who used hybrid processes incorporating some conventional techniques as well as AM
techniques. This result compares favorably to other examples of packages and interconnects using
AM techniques, achieving a maximum gain in the measurement range of 14.9 dB compared to a
gain of 15.5 dB of the amplifier alone at the same frequency. The work in this chapter demonstrates
a low loss, broad-band packaging strategy suitable for mm-wave/W-band operation and is to the
best of my knowledge, the only example to date of a fully additively manufactured package of an
active component in the W-band.

135

Figure 6.4 Test setup.

136

(a)

(b)

Figure 6.5 Measured S-parameters (a) Transmission S21 (b) Reflection S11 and S22.

137

Figure 6.6 Package loss estimate.

138

Table 6.1 Comparison with other AM Packaged Semiconductor Devices

Die

Process

GaAs LNA

AJP

GaAs CPW AJP, micro-dispensing

Estimated

Package Loss

Estimated Line and Capacitors
Interconnect Loss

Pout

1.1 dB
(85 GHz)

2 dB

(140 GHz)
1.0 dB
(15 GHz)
0.42 dB
(20 GHz)
0.44 dB
(20 GHz)
1.08 dB

(24.5 GHz)

2.9 dB

(24.5 GHz)a

2.4 dB
(20 GHz)

0.37 dB/mm
(85 GHz)
0.8 dB/mm
(140 GHz)
0.33 dB/mm
(15 GHz)
0.15 dB/mm
(20 GHz)
0.12 dB/mm
(20 GHz)
0.45 dB/mm
(24.5 GHz)
0.57 dB/mm
(24.5 GHz)a
0.2 dB/mm
(20 GHz)

Integrated

N/A

Integrated

N/A

N/A

N/A

Chip

N/A

6.2 dBm
(85 GHz)

N/A

21.9 dBm
(9 GHz)

N/A

N/A

N/A

4.2 dBm
(24.5 GHz)

N/A

AJP

AJP

AJP

Inkjet

Inkjet

Ref.

This

Ihle et al.

[155]

Craton et al.

[145]

Craton et al.

[68]

Craton et al.

[61]

GaAs PA

GaAs 0 dB

Si 0 dB

Tehrani et al. [67] GaAs 0 dB

Tehrani et al.

GaAs LNA

[67]

Ramirez et al.

[146]

GaAs LNA FDM, micro-dispensing,
laser micro-machining

aMeasurement taken at a low-loss point.

139

6.4 Conclusion

This chapter extends the capability of the chip-first packaging process to W-band frequencies, and
demonstrates a shaped interconnect which allows for frequency agnostic interconnect performance
within the resolution limits of AJP technology. Apart from the packages presented in Chapter 5,
this is the only other example of a package with integrated capacitors using MMAJP.

140

CHAPTER 7

CONCLUSIONS AND FUTURE WORK

This dissertation first presents the current state of the art in microwave packaging and system
integration strategies and introduces the concept of using AJP as a supplement or replacement
to these conventional strategies in order to address the performance and design bottlenecks they
present. The basic strategies for using AJP as a packaging and integration tool for microwave
electronics are laid out in Chapter 2, introducing the chip-first packaging and interconnect process.
In Chapter 3 and Chapter 4 the use of multi-material aerosol jet printing for microwave electronics
is introduced as a tool, demonstrating how MMAJP can be used to monolithically fabricate and
integrate basic circuit components like capacitors resistors and inductors in these packages as well as
fabricate novel material patterns and gradients which are also of use in microwave circuits. Finally,
in Chapter 5 and Chapter 6, these ideas are used together to demonstrate functional packages using
active components and the required passive elements in the package such as bypass capacitors and
extending interconnect performance into the W-band. Here, I will elaborate on what performance
and design flexibility gains one may obtain using the present AJP packaging processes, as well as
a discussion of what the future of this research is.

7.1 Potential System Performance Gains from AM Integrated Microwave

Circuits

It is challenging to ascertain the performance gains possible from the chip-first AJP packaging
process this dissertation presents since it is not easy to compare one packaged device to another
since most of the overall performance of a particular system-in-package is dictated primarily by the
devices themselves as well as how carefully designed the microwave circuits contained in a package
are.
It is possible, however, to compare individual package components such as interconnects,
and the dielectric loss of materials used. The packaging process I present in this dissertation

141

is essentially a novel strategy facilitating heterogeneous integration, so much of the performance
benefits possible using this strategy boil down to the benefits gained when heterogeneous integration
is made possible.

7.1.1 Heterogeneous Integration Benefits

While there are many strategies for heterogeneous integration in packages, this is still an active
area of research and the strategies that exist still have many shortcomings as explained in Chapter
1. While it is possible to fabricate a system on a single chip, that is System-on-Chip (SoC), in
particular at mm-wave frequencies it is a very expensive solution. System-on-Chip solutions are
frequently used at mm-wave frequencies because of the limitations of conventional interconnect
methodologies which make it difficult to go off chip without sacrificing performance. This is
due to the materials that must be used to fabricate mm-wave circuits. High speed circuits require
semiconductor material with high electron mobility such as III-V semiconductors such as GaAs
which has an electron mobility of 8500 cm2/Vs or InP which has an electron mobility of 4000
cm2/Vs [9], both of which are far more expensive than silicon technology. SiGe technology has
been demonstrated up to THz frequencies [156, 157], but yields circuits with a high noise figure
and low dynamic range. III-V semiconductors also tend to have much wider band gaps so tend to
have lower noise figures than SiGe which has a band gap between 0.66 eV and 1.12 eV. GaAs and
InP have band gaps of 1.42 and 1.35, respectively. Therefore there is a clear trade-off between cost
and performance without access to a suitable heterogeneous integration strategy. With an effective
strategy for heterogeneous integration and interconnects however, expensive III-V technologies can
be used only where they are necessary, that is, in the RF circuitry or in low noise amplifiers etc.,
and elsewhere less expensive Si and SiGe technologies can be used.

To further illustrate this trade-off, consider a wide band, W-band receiver. To the best of my
knowledge, the lowest reported noise figure (NF) for CMOS/SiGe technology amplifiers covering
the W-band is 4 dB [158], corresponding to a noise temperature of 438 K. For III-V/HEMT devices,
2 dB [159], corresponding to a noise temperature of 167 K. If such a receiver were used in a passive

142

imaging system front end, the difference in the minimum resolvable temperature is [160]

(cid:115)

(cid:18) ∆G

(cid:19)2

,

G

∆T = Ts

1
Bτ

+

(7.1)

where Ts is the system noise temperature, B is the bandwidth, τ is the integration time, ∆G is
the RMS variation in gain of the receiver, and G is the nominal gain of the receiver. Using the
previously cited amplifiers as examples of the state of the art, a 1 s integration time, a 35 GHz
bandwidth (covering the W-band), an RMS gain from the data provide in those publications of 10.6
dB for SiGe and 1.4 dB for GaAs, the minimum resolvable temperature for each of these theoretical
receivers is 500 mK for SiGe, and 23 mK for III-V semiconductors/GaAs. Using Dicke switching,
a common strategy for increasing the resolution of such passive imaging systems, these minimum
resolvable temperatures reduce to 2.3 and 0.89 mK, respectively.

As another example, consider a transmitter working in the W-band. Comparing the state-of-the-
art in power amplifiers in SiGe and III-V semiconductors, the highest output powers achieved are 91
mW [161] for SiGe and 2.5 W [162] for III-V technology (in this case, GaN). If these transmitters
were used in a communications system, the maximum range can be calculated with

(7.2)

(cid:18)

Rmax =

PtGtGr λ

2

(4π)3(SN R)kbT0F∆ f

(cid:19)1/4

,

where Pt is the transmitted power, Gt and Gr are the gain of the transmitter and receiver, respectively,
T0 is the temperature, F is the noise figure of the receiver, kb is Boltzmann’s constant, and ∆ f is
the bandwidth. Assuming no path loss or atmospheric attenuation, i.e., outer space, a temperature
of 100 K, a bandwidth of 10 GHz centered at 100 GHz, a transmitter and receiver antenna gain of
55 dBi (corresponding roughly to a 0.5 m by 0.5 m array), a required SNR of 6 dB, and the noise
figures of the receivers in the previous examples, the maximum range is 109 km and 3362 km for
the SiGe and the III-V systems respectively.

7.1.2

Interconnect Performance Gains

The easiest point of comparison is the interconnect performance when comparing packaging
technologies because it is the most easily isolated. It is also an important bottleneck to system

143

performance. The main purpose of Chapter 2 was to characterize the interconnect performance,
and I showed that interconnect loss as low as 0.1 dB at 50 GHz. That result correlates with the
package and interconnect loss found in Chapter 6 in the W-band.

Using flip chip strategies, it is possible to achieve very low loss interconnects however these
parts are not easy to rework and add many additional mechanical constraints on a system such as
surface planarity, as well as making thermal sinking much more challenging. Microbumps are also
required to form flip chip interconnects. There are other interconnect strategies as well, and this
continues to be an active area of research, but wire bonds and ribbon bonds remain the dominant
interconnect methodology, even at W-band frequencies.

Conventional strategies for compensating for the inductive discontinuity of a bond wire or
ribbon usually involve adding capacitive stubs and it is common practice to use multiple bonds to
reduce this discontinuity. [74] showed a “low-loss, low-cost IC to board bond wire interconnect”
strategy with an insertion loss of approximately 4 dB at 45 GHz and 2 dB at 60 GHz. These losses
are only achieved over narrow bandwidths.

The significance of the low loss interconnects demonstrated in this dissertation is they have a
200 % fractional bandwidth, that is, they are essentially frequency agnostic. Some compensation
may be necessary as in Chapter 2 where traces were narrowed to compensate for the capacitive
discontinuity of a large bond pad, but for this strategy a large bond pad is not required, or shaped
as in Chapter 6 where at high frequencies the geometry of the microstrip needs to be widened on
the ramp up to the die, but these compensations do not create large discontinuities like bond wires
and the stubs used to compensate for them.

7.1.3 Processing Gains

Microwave circuits in particular often require fine features to be patterned, so an important factor
in a processing strategy is feature resolution. Conventional heterogeneous integration strategies for
SiP/SoP typically use a printed circuit board (PCB) as a carrier, usually of a material that has low
material loss characteristics. Two materials that are active areas of research for this purpose are

144

Liquid Crystal Polymer (LCP) and Low Temperature Co-fired Ceramic (LTCC). Conventional PCB
processing limits are typically 50-90 µm line widths and spaces for etched features, though LTCC
limits are not as good because of material shrinkage during firing. Layer alignment for multilayer
circuits is often done mechanically with pins. Furthermore, layers are of a fixed thickness, usually
>50 µm. Integration of thin films for structures such as capacitors or resistors must be completed
as a separate post-processing step. Layer to layer interconnects are typically formed as an extra
processing step for blind/buried and through vias whose spacing and aspect ratio are limited.

AJP processing allows for finer feature sizes, though not to the scale of semiconductor processes.
With AJP, line widths less than 10 µm and gaps as small as 15 µm can be reliably printed.
Layer repeatability, aligned optically is between 2-5 µm. An advantage of AJP processing over
conventional PCB carriers to carry out heterogeneous integration is the ability to print layer
thicknesses in a very wide range, from < 0.5 µm and >100 µm. This allows seamless integration
of thin films for capacitors or resistors which can both be printed in place using the strategies laid
out in Chapter 3 and Chapter 4. Multilayered circuits with AJP can be more seamlessly connected
with ramps rather than vias (though vias have also been demonstrated). Conventional processes
often use rows of vias to improve crosstalk and isolation, but AJP can fabricate continuous metal
walls which would outperform vias for that application.

A major difference between AJP and conventional processing strategies is that AJP is a serial
rather than parallel process. This means that fabricating large numbers of a design may be less
cost effective with AJP, and is in many ways the biggest drawback to the technology. It also means,
however, that AJP allows for designs to be easily tailored to a specific application for application
specific packaging. Design changes for conventional processes are often very expensive and would
only be done if absolutely necessary. AJP is ideally suited to bespoke packaging solutions where
the performance benefits and design flexibility can be fully exercised. Hybrid processes using a
mixture of AJP and conventional processes may also allow for the benefits of both technologies
without sacrificing throughput or performance, allowing more flexibility in a package design.

Finally, AJP as an AM technology has very low material waste, and eliminates etching processes

145

which often use dangerous chemicals and gasses. This makes the AJP processes generally much
safer and more environmentally friendly compared to conventional ones. The strategies presented
in this dissertation all are performed with a single piece of equipment, an aerosol jet printer, and
an oven for curing/sintering, rather than a fully equipped clean room as would be necessary to
accomplish the same using conventional processes. This significantly lowers the barrier to entry
into this research field, and may make the technology possible to produce with these processes
more widely available.

7.1.4 Materials Performance Gains

Materials used in this dissertation have competitive loss characteristics with those used in conven-
tional SiP/SoP packaging for microwave circuits. PI used throughout this dissertation has a tan δ of
0.002, which was measured in Chapter 3 and a 
r of 2.9. PI is already a commonly used material in
conventional packaging and semiconductor processing. The material characteristics of PI are close
to LCP which has 
r = 2.9 and tan δ = 0.0025. LTCC has a higher 
r of 7.1, which is desirable
in some cases, but often requires finer feature resolution since the wavelength in the material is
shorter.

A major advantage of AJP processes is the wide range of materials that can be used. While
Ag nanoparticle based ink was primarily used to fabricate conductive parts of circuits in this
dissertation, other metal nanoparticle, nano-flake, or metal-organic decomposition inks without
nanomaterials could all be printed with AJP with little modification to the processes presented.
Likewise, while PI was the primary material used, many other dielectric materials are possible to
print, and have been used by myself and others. The high viscosity limit of the pneumatic atomizer
of 1000 cps allows viscous materials to be printed, allowing for high solid content aerosols to be
deposited without adding excessive solvent to inks. With MMAJP, the material flexibility is even
greater, allowing for composites and material gradients to be printed in place, as well as allowing
for inks to be seamlessly switched without pauses in printing. AJP can also possibly integrate
absorbing materials as in Chapter 4 to achieve better crosstalk and isolation performance, and thus

146

more compact circuits and packages. This material flexibility is not available with conventional
processes.

Because the dielectric loss of materials is similar, transmission lines can achieve similar in-
sertion loss performance. Microwave structures such as filters will also be able to achieve similar
performance, though at higher frequencies into the mm-wave regime, AJP circuits may outperform
similar conventionally fabricated circuits owing to the higher feature resolution possible.

7.2 Extension of this Research

This dissertation presents a toolbox of strategies to package and integrate systems using AJP but
there are several areas where additional research would complement the work in this dissertation.
The expansion of this research can extend in several directions by using these processes to demon-
strate systems which use conventional and AJP processes, expanding on the work in this dissertation
to show multilayer packages and stacked packages, expanding on Chapter 3 and Chapter 4 to show
more uses of the MMAJP process, and finally showing more rigorous reliability testing of the
chip-first packaging strategy as well as expanding it to higher power and even higher frequencies.
Incorporating some conventional processes may also yield the best of both worlds in terms of
performance and flexibility of the design of microwave and mm-wave packaging. Supplementing
AJP/MMAJP processes with laser micro-machining as in [146] would also likely offer significant
performance gains without sacrificing much of the flexibility afforded to AJP processes.

7.2.1 Multilayer Packaging

Following in the trend of conventional packaging, a way to reduce the physical footprint of AJP
packages and increase functional density is to stack active components as with conventional strate-
gies like PoP, CoC, etc. Because AJP layers are not fixed and are not deposited and etched in a
serial manner, novel stacking structures not possible with conventional packaging could be possible
with AJP. This dissertation has shown multi-leveled structures in several places though not multi-

147

Figure 7.1 Initial samples of AJP microbumps.

layered circuits. Multi-layered circuits have been shown by others with AJP however. Because both
multi-leveled and multi-layered circuits are possible, novel coupling structures could be developed
to be fabricated with AJP to potentially develop more efficient and better performing couplers or
filter structures.

Multilayer packaging could also include flip-chip implementations enabled with AJP. Some
preliminary work has been done to fabricate AJP microbumps as shown in Fig. 7.1 where printed
microbumps of target diameter 20 µm, 15 µm, and 10 µm are shown. All the printed bumps have a
thickness of approximately 10 µm. These mechanical proofs of concept are not perfect and push the
limits of the AJP’s resolution, but are promising first samples for dense microbump arrays printed
with AJP. 10 µm bumps with 30 µm pitch are likely possible based on these initial samples.

7.2.2 Conformal Packaging

With conventional packaging strategies, conformal packaging on non-planar surfaces would not
be possible however there are no significant challenges to accomplishing such a task with AJP.
Conformal antennas have been an active area of research for some time, including using AJP, but to
expand on that work by integrating active devices on a conformal surface would allow for improved

148

system performance by lowering transmission line loss from those radiating structures to a receiver
or transmitter.

7.2.3 High Power & High Reliability Packaging

This dissertation does not show particularly high RF power systems. Using printed Ag for conduc-
tors as was done for the majority of the work presented here may not be feasible for higher power
applications due to electromigration of the Ag. It is possible however to use printed silver as a
catalyst for an electroless plating reaction to plate printed Ag structures with copper, mitigating
the electromigration problem. It may also be possible to use other kinds of metallic inks for high
power applications. Besides high RF power, it would also make sense to integrate switching power
supplies into packaging solutions, whose packaging may also benefit from the use of AJP. The
pairing of packages for microwave electronics and for power electronics would provide a significant
increase in the functional density of a system fabricated with that strategy. Higher power systems
will require more thermal control to prevent devices from overheating. It may be possible to print
microfluidic cooling channels with AJP, or otherwise integrate AM cooling solutions as I showed
in [163] and in Fig. 7.2.

Related to concerns for high power applications, a more rigorous investigation of AJP packaging
reliability is necessary to move the present packaging strategy from the realm of research to practical
use. While a preliminary but somewhat cursory reliability study is shown in Chapter 5, a more
thorough investigation is called for in future work in this area.

7.2.4 Novel Microwave Structures with MMAJP

MMAJP and the processes laid out in Chapter 3 and 4 have a particularly wide range of areas
future research can explore. Because of the difficulty involved in fabricating them, exploiting
material gradients or patterned/periodic materials in microwave circuits have not been thoroughly
investigated. So-called metamaterials and metamaterial based lenses and filters may benefit in

149

(a)

(b)

(c)

Figure 7.2 AM microfluidic cooling: (a) cavity design for constant flow, (b) serpentine design, (c)
device under test. ©2017 IEEE.

particular from the capabilities of MMAJP since they rely on periodic structures to create material
properties not found in nature. By replacing the BaTiO3 used in Chapter 3 with a material such
as barium strontium titanate (BST), electrically tunable materials could be deposited. Dielectric
lenses, which may benefit from the ability to print dielectric gradients for thin film optics, antennas
and arrays could be fabricated with the MMAJP process as well. It may be possible to use MMAJP
to perform active chemistry in the aerosol or immediately as the aerosol is deposited. While
the examples of MMAJP in this dissertation all use only two aerosols to create composites, as
discussed in Chapter 3, it would be possible to extend this concept to more than two aerosols. More

150

elaborate mixing mechanisms or manifolds for controlling the flow and mixing of these aerosols
could be conceived to fabricate more elaborate composites or switch between ink sources so that
multiple base materials with useful properties, i.e., low permittivity (e.g. PI), high permittivity (e.g.
BaTiO3), high permeability (e.g. Ni0.5Zn0.5Fe2O4), low resistivity (e.g. Ag), and high resistivity
(e.g. C or NiCr) materials could all be present and available to mix so that without changing inks,
these base inks can be mixed with MMAJP to select values of 
, µ, and ρ at any point in space for
a design.

7.3 Final Thoughts

These strategies for AJP packaging and system integration show great promise as useful tools for
enhancing microwave and mm-wave electronics and systems. These strategies are not a silver bullet
to solving all of the challenges involved in microwave package and system design and do come with
some shortcomings and further challenges as previously discussed. The proper application of these
processes alongside cleaning and maintenance procedures for the equipment used are necessary
to achieve the results presented in this dissertation. The appendices detail my recommended
procedures to carry out this process based on their most mature form in my development of them.

151

APPENDICES

152

APPENDIX A

AEROSOL JET PRINTING NOTES AND PROCEDURES

This appendix details the cleaning, preparation, maintenance and printing procedures required to
build the AM packages described in this dissertation. These procedures are not well documented
anywhere in literature that I am aware of, and some details of these procedures are required for the
successful implementation of the strategies presented in this dissertation.

All of the AJP in this dissertation was performed with an Optomec Aerosol Jet 5x Printer
equipped with a pneumatic atomizer (PA), an ultrasonic atomizer (UA), five axes of movement and
Kewa control software. Additionally, all the circuit demonstrations used the fine feature deposition
head and nozzles. Large features (such as polyimide/dielectric features) could be deposited orders
of magnitude faster use the large feature deposition head included with the printer. Aerosolized
materials are focused through a printing nozzle with a sheath of nitrogen gas which prevents the
printing material from coming into contact with the printing nozzle. This printer is capable of
printing features below 10 µm. This manufacturing process allows for printing of material at
standoff heights as great as 10 mm without significant loss in resolution. The PA generates an
aerosol by partially pressurizing the ink and, by the venturi effect, jetting nitrogen over a column
of ink. The quantity of aerosol and the flow rate of the PA jet is too high to be printed directly so
the aerosol flow is divided between an exhaust which is filtered of aerosol and vented through a
vacuum pump and the aerosol flow to the print head. The UA generates an aerosol by ultrasonic
energy alone so all of the UA flow rate is directed to the print head.

Temperature of the ink is controlled in the PA with a heat/stir controller and in the UA with a
chiller circulating water around the ink vial which also serves to carry heat away from the ultrasonic
transducer. The UA may not be used without the chiller. Non-standard equipment I installed in the
printer for this work includes a heated print stage, heated aerosol tubes, a thermocouple for the print
stage, a relative humidity sensor for the print chamber and a MMAJP combiner. The temperature of

153

inks should be kept stable during printing. Increasing the temperature of PA may lower the viscosity
of the ink and increase the rate of atomization. The UA ink can be affected similarly but because the
temperature of the UA ink is controlled by the chiller for the ultrasonic transducer, the temperature
should not exceed 30°C to prevent damage to the transducer. The temperature of the print stage
depends entirely on the ink in use. Some inks (the silver ink used in this dissertation for example)
do not need to be printed on a heated stage, but still may benefit from a heated stage. Heating the
print stage helps to dry the ink as it is deposited. Aerosol tube heaters composed of heating wire
wrapped around tubes from the atomizers to the print head and monitored by thermocouples may
be helpful for some applications, but should be carefully controlled so that they do not melt the
tubes.

The user has control over the sheath gas flow rate, the pneumatic atomizer flow rate, the exhaust
flow rate, and the ultrasonic atomizer flow rate. These flow rates are controlled by programmable
mass flow controllers. Additionally, the pressure of each of these flow rates is monitored at the
location of the flow controller. Nitrogen gas is used as a carrier gas for aerosols as well as for the
focusing sheath. The printing speed, also defined by the user, should be optimized for the ink in use
and should in general not exceed 8 mm/s. At high speeds, vibrational dampening in the printer is
not sufficient to prevent printer deviation from the specified path, and the momentum of the printer
can spray ink over the print surface when the print head changes direction.

Inks with viscosity as high as 1000 cps may be printed, but care should be taken to ensure that
inks are compatible with the materials they will come in contact with. O-ring gaskets of Viton (syn-
thetic rubber and fluoropolymer elastomer) are suitable for most applications. Perfluoroelastomer,
fluorinated ethylene propylene (FEP) coated Viton, or polytetrafluoroethylene (PTFE) O-rings may
be necessary for some materials. Some common solvents such as dimethyl sulfoxide (DMSO)
and N-methyl-pyrrolidone will dissolve Viton so inks using such solvents (polyimide for example)
should never be used with Viton gaskets.

Sealing off the print chamber entirely so that the print area can be purged with nitrogen may offer
significant benefits and increased print times but would add significant complexity to the system.

154

This would allow for a wider range of materials to be used which may otherwise interact with oxygen
in the air in their aerosol form. Take care in making such a modification so that oxygen outside
the print chamber is not displaced, asphyxiating the user. Another potentially useful modification
would be including UV filters or masks on light sources in and around the printer so that UV
curing of materials can be done in-place. Additionally, filtering the air in the room to clean room
specifications would avoid many potential pitfalls. Many materials are sensitive to moisture so the
relative humidity in the print area should be monitored and kept as low as possible.

A.1 Cleaning

The cleanliness of the printing equipment is essential for good printing results. Lab procedures to
leave the equipment clean must be completed after printing. Droplets of ink may partially cure if
not quickly removed and will be more challenging to clean if left for a long period of time.

i. Tubes carrying ink to the print head (aerosol tubes) should be discarded as hazardous waste

and replaced each time the printer is used, and whenever ink is changed.

ii. During printing, aerosol tubes may be either cleaned out (taking care to not use incompatible
chemicals) or replaced periodically as ink tends to build up on the inner walls of these tubes
over time. The rate at which material builds up depends on the ink and material in use.

iii. The exhaust tube must also be inspected and periodically cleared of blockage.

If enough
ink builds up to block the exhaust, inconsistent flow rates in the PA aerosol will result and
potentially lead to a clog in the aerosol tube.

iv. The shutter which controls the flow of ink to the print stage must be periodically cleaned so
that ink deposited on it does not splash onto the print surface and so that it does not build up
and clog the nozzle.

v. O-ring gaskets must be regularly inspected and replaced when damaged.

155

vi. PA exhaust filters should be inspected regularly. The final filter before the mass flow controller,
a HEPA filter, should be removed when not in use. Solvent can build up in the filter and prevent
flow control of the exhaust.

vii. Cleaning printer parts properly depends on the inks in use and may dictate the chemicals used.

A general process is given here:

a. Disassemble components and wipe gaskets clean of residual grease. If gaskets are greased

properly, there should be no residual grease.

b. Clean parts first in the solvent used with the ink currently in use (e.g. NMP for PI) in an

ultrasonic bath for 30 min.

c. Next clean parts in a cleaning solution of Neutrad/Branson IS or similar in an ultrasonic

bath for 30 min.

d. Clean parts in IPA or DI water in an ultrasonic bath for 30 min.

e. Blow parts dry, or dry in an oven before using in the printer.

f. Sensitive components such as nozzles or the PA atomizer jet may require additional
cleaning with harsher chemicals. E.g., silver may need to be etched off using ammonium
hydroxide or similar. Conventional cleaning strategies such as RCA clean may be modified
for thorough cleaning of AJP parts so long as compatibility with the AJP component
materials (notably the stainless steel of most parts) is insured. It should be obvious that
many metal etchants should never be used, such as sodium persulfate.

g.

Inspect sensitive components under a microscope for debris.

h. Some nanomaterials will be difficult or impossible to clean entirely. For such materials,

use dedicated nozzles and atomizers to prevent contamination of other inks.

i. disassembled parts should be stored in a dust and moisture-free location such as in a

desiccator.

156

(a)

(b)

(c)

Figure A.1 (a) Disassembled PA, (b) Assembled PA, (c) assembled print head.

A.2 Assembly

If the printer is not assembled properly, it will not operate as intended. Gaskets must be properly
greased with a thin sheen of grease (e.g. Apiezon L grease) covering the gasket only. Grease on
other surfaces may contaminate the ink and disrupt aerosol flows.

Aerosol tubes must be cut as flat as possible. The face of the cut should be flush with the
face of the adapter. If there is any mushrooming effect on the edges of the cut, the tube should be
re-cut. Ensure that the tube itself is clean and completely free of dust inside and out. If the cut
of the tube is not straight, the resolution of the printer will decrease significantly. The material
of the tube should be chosen based on the ink used and the hydrophilic or hydrophobic properties
of the material. Polyethylene or PTFE are suitable for most inks. Tubes with an inner diameter
greater than the nominal 1.595 mm can be used and may increase the length of continuous print
time possible but may also disrupt the aerosol flow enough to lead to overspray of the ink. Tubes
must also be fully seated in the fittings on the atomizer as well as the print head.

The movement of the printer is calibrated for the weight of the moving printer parts, so the printer
should not be used disassembled. The disassembled PA, the assembled PA, and the assembled print
head are shown in Fig. A.1.

157

A.3 Development of Printing Parameters

A laminar flow of aerosol must be maintained. For pure nitrogen in the range of fine feature nozzle
sizes, 100 µm to 300 µm, yields a Reynolds number of approximately 2000 for flow rates 425 sccm,
and 1275 sccm, respectively. For an increase in density of 25%, these flow rates decrease to 340
sccm, and 1020 sccm. These should be thought of as absolute limits rather than recommendations
for practical flow rates to use with the printer. This absolute limit is for the sum of the sheath gas
flow rate and the aerosol flow rate(s). The flow rates I have had the most success with range from
10 to 30 % of these values.

The UA atomizer flow rate is a straightforward setting since there is no exhaust as there is for
the PA. The rate of atomization can be controlled somewhat by reducing power to the ultrasonic
transducer, but in general you would want the best possible atomization rate. Controlling the PA is
somewhat more complicated. Depending on how stable a particular ink is, or how sensitive it is to
factors like moisture, or what the atmospheric pressure is at the time of printing, or how tight the
seals on gaskets are the print parameters can change significantly. In general I recommend keeping
the pressure of the PA flow rate constant as well as the difference in the flow rate of the PA and
the exhaust, and adjusting the flow rate of the PA and exhaust to keep pressure constant. There is
significantly more process drift in the PA than the UA, but the PA may be used with a wider range
of inks.

When developing print parameters for a new ink, start with a sheath gas flow rate of 50 to 100
sccm depending on the nozzle diameter. Start with an aerosol flow rate of 20 to 60 sccm for the
UA and 25 to 70 sccm for the PA. For the PA, start at a low pressure and flow rate and adjust the
PA and exhaust flow rates up, keeping their difference constant, until sufficient atomization begins.
Finally, adjust the sheath until the printed lines are smooth and consistent without overspray.

The height of the print head from the print stage, that is the distance from the face of the nozzle
to the print surface, can be between approximately 3 mm and 10 mm. Optimally, it will be between
3 mm and 5 mm.

158

A.4 Inspection and Validation of Procedures and Printed Parts

line width, line height (start and end of printing period before tubes are changed)

When developing printing parameters, the critical physical characteristics of the printed line
are the line width, the line thickness, and amount of overspray, and the line consistency. Due to
process drift, the line thickness tends to increase over time. The rate of increase is highly dependent
on the ink. Choosing a tube material that the ink does not tend to whet to will also help prevent
this process drift. If the line width is inconsistent it may indicate the formation of a clog or that the
sheath gas flow rate is either too high or too low. Line consistency can also depend on the ink and
the surface it is being printed on. For example, a water-based ink (such as the silver ink formulation
used in this dissertation) will not tend to print well on highly hydrophobic surfaces. This effect can
be mitigated on rougher surfaces or with a heated print stage.

The layer thickness depends on the overlap between adjacent lines as well. In this work, an
overlap between 20 and 40 % was used. More of an overlap will create thicker layers. Test structures
should include low aspect ratio structures of several layers so that an average layer thickness can
be determined. The characteristic surface roughness can also be determined from such a structure.
Thin films will not yield a characteristic roughness because the roughness may be influenced by
the surface the material is printed on. A structure 4 to 5 µm thick is sufficient. Similar to the
resistive films in Chapter 4, a 2 by 7 mm film was printed and also used to characterize the film
resistivity/conductivity.

Dielectric properties are more challenging to determine but can be measured using techniques
as used in Chapter 3 and Chapter 4. Material processing will have the largest impact on dielectric
performance. As noted in Chapter 3, some microbubbles may form in dielectric films which may
impact the resulting films electrical characteristics. Nanoparticle based dielectrics will behave
similarly to metal nanoparticle based inks and may tend to be more porous than the bulk material.

159

APPENDIX B

MATERIAL PROCESSING AND PREPARATION

This appendix focuses on the material processing required for the materials used in this dissertation
as well as some associated materials and commonly used materials in conventional packaging
strategies which may also be suitable for AJP. These are primarily post processing procedures after
material deposition as well as preparing inks suitable for printing.

B.1 Silver

The silver ink used throughout this dissertation was 25 wt. % Clariant Prelect TPS 50 in deionized
water printed primarily with the UA. With the PA, this ink can be used diluted or undiluted. This
is a proprietary ink so the content is not entirely known. Based on EDS analysis of this ink and the
presence of Cl, I assume that silver chloride is used as a surfactant to disperse the nanoparticles.
After sintering, these silver films turn white in color which can also be explained by the presence
of silver chloride.

With a sintering profile of 20°C to 180°C with a 2°C per minute with a 4 hour hold time and
then 180 °C to 20°C with a temperature gradient of approximately 1°C per minute, the resulting
films have a conductivity of 39% of the bulk conductivity of silver. This was confirmed over several
test samples printed at different times. The material may be dried before sintering, but it is not
necessary. An SEM image of sintered silver using this profile and EDS analysis is shown in Fig.
B.1.

This nanoparticle suspension is very stable and does not need extra mixing prior to printing,
but should be mixed under ultrasonic energy for 60 minutes when diluted. This ink requires oxygen
to sinter properly. As in Chapter 2, it is possible to sinter this ink with the oxygen released as the
byproduct of PI imidization, but the results will not be optimal. Conductivity as high as 80% of

160

(a)

(b)

Figure B.1 (a) SEM image of silver after sintering at 180°C, (b) EDS analysis of sintered silver.

the bulk conductivity of silver can be achieved with long sintering times and higher temperatures.
Sintering can be performed on a hot plate or in an oven with similar results. This ink adheres well
to most materials. It does not adhere as well to PI, but adhesion was acceptable for the purposes of
the work in this dissertation. Adhesion improved on PI based polymer matrix nanocomposites.

B.2 Polyamic Acid/Polyimide

PI as well as other materials used in conventional electronics and packaging fabrication can in
general be processed as they would be in those circumstances.

PI is expected to lose approximately 50% of its volume when cured. [122] gives an estimate
as to the % imidization for curing times between 100 and 250 °C. A soft bake at 200°C therefore

161

yields films approximately 80 to 90% imidized. Full imidization requires higher temperatures.
Manufacturers typically recommend 350°C, but at 295°C, films should be 100% imidized. Because
PI films are derived from polyamic acid, the surface tension of the liquid polyamic acid allows the
films to be smooth. If thick films are deposited without drying the material as it is printed the
surface may become wrinkled and uneven. If the material is dried as it is printed, films as thick as
30 to 50 µm (once cured) can be deposited in a single pass. On sharp corners (such as the edge of
a die) thinner layers on the order of 3 to 5 µm thick, should be printed and soft baked before adding
additional material. Prior to curing, insure that films are dry. Drying can be done on a hot plate, or
on the print stage during printing. Parts should be held under vacuum for 12 hours prior to curing
to allow bubbles trapped in the film to escape. Some bubble formation is likely unavoidable with
AJP.

PI is very sensitive to moisture so all printing parts must be completely dry before printing. PI
ink must also be stored carefully, in an oxygen free environment if possible. Bottles containing
polyamic acid should be left with very little head room and stored below 0°C. Before opening
bottles, they must be first allowed to come to room temperature. Curing must be done in an oxygen
free environment. Oxygen breaks up imide chains and makes the resulting films lossy and brittle
with poor adhesion. In an inert gas oven, ultra-high purity grade inert gas should be used to purge
the oven. An adhesion promoter should be used on surfaces before printing.

B.3 Other Polymers and Epoxies

Other polymers and epoxies derived from polymer precursors should be processed and prepared
similar to the template laid out for polyimide. AJP inks are necessarily lower viscosity than
for conventional application via spin coating so typically need to be diluted with a compatible
solvent. Not all of these materials may be well suited for packaging applications as many have
a characteristically high CTE relative to semiconductor material. Most materials with a liquid
precursor should be possible to print with AJP when diluted. Depending on the material, there

162

may be additional processing or storage considerations. Since these materials tend to have high
viscosity, they will in general need to be printed with a PA.

B.4 Other Nanoparticle Based Materials

Other nanoparticle based materials including those used in this dissertation, BaTiO3 and Ni0.5Zn0.5Fe2O4,
may be printable with AJP. The primary consideration for these materials is creating a stable disper-
sion in a solvent. The better quality dispersion achieved, the better the print quality will be. With
the UA, if nanoparticle suspensions are not stable and material precipitates out of the suspension
it will settle on the bottom of the ink vial and stop atomization from occurring. Suspensions that
are less stable may still be printed with a PA. Using MMAJP, the amount of agglomeration in the
suspension will also affect the quality of the mixing.

163

APPENDIX C

AJP CHIP-FIRST PROCESS

Following is a generic process for the AJP chip-first packaging approach:

1. Die attach (chip-first): single or multiple die on low CTE substrate; MoCu as used in this

dissertation, LTCC, other ceramics, semiconductors, or glass.

2. Surface treatment: Apply adhesion promoter if necessary, suitable for the carrier and
printed dielectric. For the work in this dissertation using low CTE polyamic acid PI2611,
VM651 or VM652 (HD Microsystems) are suitable. Other surface treatments such as plasma
cleaning may be done at this time to prepare surfaces for printing.

3. Initial printed dielectric layers: print the first dielectric layers, including MMAJP thin films
for capacitors. These layers act as an intermediate adhesion layer so additional adhesion
promoters are not necessary. The thickness of these layers is not important, but all the areas
where dielectric will exist in the design should be covered enough to prevent pinholes in the
film.

4. Dielectric ramp printing: print dielectric so that the edges of the die are all covered. This
printing is necessary to perform at an angle (e.g. 30°as used in this dissertation). Some
dielectric should be printed on the surface of the die to anchor the material to the die. These
layers are necessary to complete first to reduce the stress on the films in contact with the die.
If too much material exists around the die, the printed dielectric may crack and pull away
from the die when cured.

5. Soft bake: For PI and many other dielectrics, soft bake the first printed material layers. For
PI, heat in a non-oxidizing nitrogen atmosphere to 200°C for one minute with a maximum
temperature gradient of 2°C per minute.

164

6. Surface treatment: This step may not be necessary for all designs, but some surface
treatment may be necessary especially if printing in step 3 did not cover all areas where
dielectric may exist. Some materials may require cleaning steps prior to additional printing
for optimal surface quality.

7. Dielectric printing: Print remainder of dielectric to design thickness, typically 5 to 100 µm
thick, though as thin as 0.5µm thick is possible (t. Intermediate soft bakes during this process
may be necessary depending on the thickness and geometry of the films required. As much
as 50µm of PI can be deposited and cured (in a soft-bake), reliably producing high quality
structures. The limits of film thickness acceptable prior to cure depend on if the film is dried
as it is deposited on the print stage, and the material being used. Materials may include
dielectric, or MMAJP magnetic, or MMAJP high dielectric films.

8. Soft/hard bake: If only a single dielectric layer is needed or if this is the final dielectric layer,
this step is a full cure of the dielectric material. For PI, 295°C for 30 min with a maximum
temperature gradient of 2°C per minute. If multiple layers of dielectric are needed, this step
should be a soft bake as in step 5.

9. Metal printing: Print conductive structures forming interconnects, transmission lines, an-
tennas filters, and other circuits on the current metal layer. These structures may be conformal
to a multi-leveled structure. This may include conductive inks like silver as well as MMAJP
resistive composites. These are not necessarily nanoparticle based inks as there have been
examples of printed metal organic decomposition inks that can be processed at low temper-
atures.

10. Sinter metallic structures: Metal features should be fully cured before adding additional
dielectric layers. The silver curing process I recommend is as in the previous appendix;
180°C for 4 hours.
If the dielectric is fully cured a higher temperature gradient may be

165

acceptable. If the dielectric has not gone through a hard bake process, a low temperature
gradient should be used to reach the soak temperature.

11. Repeat steps 6 through 10: as needed for multiple layers of circuitry.

166

APPENDIX D

LIST OF PUBLICATIONS

• Michael Thomas Craton, John D. Albrecht, Premjeet Chahal, John Papapolymerou, “Multi-
material Aerosol Jet Printed Magnetic Nanocomposites for Microwave Circuits,” IEEE Trans-
actions on Components, Packaging and Manufacturing Technology, submitted, 2020.

• Michael Thomas Craton, John D. Albrecht, Premjeet Chahal, John Papapolymerou, “Addi-
tive Manufacturing of Frequency-Agile W-Band Packaging,” IEEE International Microwave
Symposium (IMS), submitted, 2020.

• Michael Thomas Craton, Ioannis Papapolymerou, Premjeet Chahal, John D. Albrecht,
“Material Mixing for Additive Manufacturing Apparatus,” United States Patent Application
Pub. No. Pending, Provisional Patent Application No. 63/067001 Aug. 2020.

• Michael Thomas Craton, “In Situ Nanocomposite Fabrication for RF Electronics Appli-
cations With Additive Manufacturing,” in International Symposium on Microelectronics
(iMAPS) (invited talk), 2020 (presented virtually due to COVID-19).

• Michael Thomas Craton, “Strategies for Microwave System Integration Using Additive
Manufacturing,” in IEEE Electronic Components and Technology Conference (ECTC) (in-
vited talk), 2020 (presented virtually due to COVID-19).

• Michael Thomas Craton, Xenofon Konstantinou, John D. Albrecht, Premjeet Chahal,
John Papapolymerou, “A Chip-First Microwave Package Using Multimaterial Aerosol Jet
Printing,” IEEE Transactions on Microwave Theory and Techniques, vol. 68, no. 8, pp.
3418-3427, August 2020, doi: 10.1109/TMTT.2020.2992074.

• Xenofon Konstantinou, Cristian J. Herrera-Rodriguez, Aaron Hardy, Michael Thomas Cra-
ton, John D. Albrecht, Qi Hua Fan, Timothy Grotjohn, John Papapolymerou, "A Monolithic

167

Wilkinson Power Divider on Diamond via a Combination of Additive Manufacturing and
Thin-Film Process," in Radio & Wireless Week, 2020,
doi: 10.1109/RWS45077.2020.9050128.

• Michael Thomas Craton, John D. Albrecht, Premjeet Chahal, John Papapolymerou, "In Situ
Nanocomposite Fabrication for RF Electronics Applications With Additive Manufacturing,"
IEEE Transactions on Microwave Theory and Techniques, vol. 68, no. 5, pp. 1646-1659,
May 2020, doi: 10.1109/TMTT.2020.2977030.

• Michael T. Craton, Ioannis Papapolymerou, Premjeet J. Chahal, John D. Albrecht, "Hetero-
geneous microwave and millimeter wave interconnect and packaging by additive manufac-
turing," United States Patent Application Pub. No. Pending, Provisional Patent Application
62/673403, PCT Patent Publication No. WO 2019/222410 A1, May 2018.

• Michael Thomas Craton, Yining He, Aljoscha Roch, Premjeet Chahal, and John Pa-
papolymerou, "Additively Manufactured Interdigitated Capacitors Using Barium Titanate
Nanocomposite Inks", in European Microwave Conference (EuMC), 2019,
doi: 10.23919/EuMC.2019.8910820. Michael T. Craton, "Paddle Card Having Shortened
Signal Contact Pads," United States Patent No. US 10249988B2, Apr. 2019.

• Michael Scholeno, Michael T. Craton, "Electrical Connector Assembly," United States

Patent No. US 10218108B2, Feb. 2019.

• Michael Thomas Craton, John D. Albrecht, Premjeet Chahal, and John Papapolymerou, "A
Chip-First Approach to Millimeter-Wave Circuit Packaging", IEEE Microwave and Wireless
Components Letters, vol. 29, no. 2, Feb. 2019, pp. 116-118,
doi: 10.1109/LMWC.2018.2886737.

• Michael Thomas Craton, Christopher Oakley, John D. Albrecht, Premjeet Chahal, and
John Papapolymerou, "Fully Additively Manufactured Broadband Low Loss High Frequency

168

Interconnects", in Asia Pacific Microwave Conference (APMC), 2018,
doi: 10.23919/APMC.2018.8617327.

• Michael Thomas Craton, Jakub Sorocki, Ilona Piekarz, Slawomir Gruszczynski, Krzysztof
Wincza, and John Papapolymerou, "Realization of Fully 3D Printed W-band Bandpass Filters
Using Aerosol Jet Printing Technology", in IEEE European Microwave Conference (EuMC),
2018, doi: 10.23919/EuMC.2018.8541416.

• Ilona Piekarz, Jakub Sorocki, Michael Thomas Craton, Krzysztof Wincza, Slawomir
Gruszczynski, and John Papapolymerou, "Application of Aerosol Jet 3D Printing with Con-
ductive and Non-Conductive Inks for Manufacturing mm-Wave Circuits", IEEE Transactions
on Components Packaging and Manufacturing Technology (CPMT), vol. 9, no. 3, pp.
586-595, Mar. 2019, doi: 10.1109/TCPMT.2018.2889698.

• Michael Thomas Craton, Vincens Gjokaj, Christopher Oakley, Brian Wright, John D. Al-
brecht, Premjeet Chahal, and John Papapolymerou "A Broadband (10-20 GHz) Lightweight
Receive Module on Multilayer LCP Technology for Radar Applications", in IEEE Radar
Conference (RadarConf), 2018, doi: 10.1109/RADAR.2018.8378545.

• Yuxiao He, Michael Thomas Craton, John Papapolymerou, and Premjeet Chahal, "A Bi-
material Fully Aerosol Jet printed W-band Quasi-Yagi-Uda Antenna", in Global Symposium
on Millimeter Waves (GSMM), 2018, doi: 10.1109/GSMM.2018.8439233.

• Michael Craton, Jennifer A. Byford, John Papapolymerou, Premjeet Chahal, and John D.
Albrecht, "A Polyjet 3D Printed Alternative For Package to RFIC Interconnects", in European
Microwave Conference (EuMC), 2017, doi: 10.23919/EuMC.2017.8230899.

• Michael Craton, Mohd Ifwat Mohd Ghazali, Brian Wright, Kyoung Youl Park, Premjeet
Chahal, and John Papapolymerou, "3D Printed Integrated Microfluidic Cooling for High
Power RF Applications" in International Symposium on Microelectronics (IMAPS), 2017,
doi: 10.4071/isom-2017-Poster6\_098.

169

• Michael Craton, Jennifer A. Byford, Vincens Gjokaj, John Papapolymerou, and Premjeet
Chahal, "3D Printed High Frequency Coaxial Transmission Line Based Circuits" in IEEE
Electronic Components and Technology Conference (ECTC), 2017,
doi: 10.1109/ECTC.2017.180.

• Stephen B. Smith, Khushboo Patel, Michael Craton, Madhumitha Rengarajan, Steven A.
Blasko, and Yifan Huang, "To Couple or Not to Couple, That is the Question", DesignCon,
2017.

• Contributing author to Serial Attached SCSI-4 (SAS-4), "Annex K (Informative) Recom-
mended electrical performance limits for mated connector pairs and cable assemblies sup-
porting rates of 22.5 Gbit/s," Project T10/BSR INCITS 534, 2017.

170

BIBLIOGRAPHY

171

BIBLIOGRAPHY

[1]

[2]

[3]

[4]

[5]

Analog Devices. HMC451 GaAs PHEMT MMIC Medium Power Amplifier, 5-20 GHz,
2020.

Analog Devices. HMC451LP3 GaAs PHEMT MMIC Medium Power Amplifier, 5-18 GHz,
2020.

Jun Fan, Xiaoning Ye, Jingook Kim, Bruce Archambeault, and Antonio Orlandi. Signal
integrity design for high-speed digital circuits: Progress and directions. IEEE Transactions
on Electromagnetic Compatibility, 52(2):392–400, 2010.
Daniel Chow. The basics of digital signal spectra. EDN Network, page 25, 2013.
Eric Beyne. The rise of the 3rd dimension for system integration. International Interconnect
Technology Conference, pages 1–5, 2006.

[7]

[6] M. J. Wolf, P. Ramm, A. Klumpp, and H. Reichl. Technologies for 3D wafer level hetero-
geneous integration. Symposium on Design, Test, Integration and Packaging of MEMS/-
MOEMS, pages 123–126, 2008.
Konstantin Statnikov, Janusz Grzyb, Bernd Heinemann, and Ullrich R. Pfeiffer. 160-GHz
to 1-THz Multi-Color Active Imaging With a Lens-Coupled SiGe HBT Chip-Set.
IEEE
Transactions on Microwave Theory and Techniques, 63(2):520–532, 2015.
Johnna Powell, Helen Kim, and Charles G. Sodini.
Millimeter-Wave Passive Imaging.
niques, 56(11):2416–2425, 2008.
B. G. Streetman. Solid State Electronic Devices. 2008.

SiGe Receiver Front Ends for
IEEE Transactions on Microwave Theory and Tech-

[8]

[9]

[10] Lorene A. Samoska. An Overview of Solid-State Integrated Circuit Amplifiers in the
Submillimeter-Wave and THz Regime. IEEE Transactions on Terahertz Science and Tech-
nology, 1(1):9–24, 2011.

[11] Telesphor Kamgaing, Adel A. Elsherbini, Sasha N. Oster, Brandon M. Rawlings, and Kyu Oh
Lee. Ultra-thin dual polarized millimeter-wave phased array system-in-package with em-
bedded transceiver chip. International Microwave Symposium, 2:1–4, 2015.

[12] Young Chul Lee and Chul Soon Park. LTCC-based monolithic system-in-package (SiP)
International Journal of RF and Microwave

module for millimeter-wave applications.
Computer-Aided Engineering, 26:803–811, 2016.

[13] Peng Wu, Fengman Liu, Jun Li, Cheng Chen, Fengze Hou, Liqiang Cao, and Lixi Wan.
Design and implementation of a rigid-flex RF front-end system-in-package. Microsystem
Technologies, 23(10):4579–4589, 2017.

172

[15]

[14] Gilles Poupon, Nicolas Sillon, David Henry, Charlotte Gillot, Alan Mathewson, Lea Di
Cioccio, Barbara Charlet, Patrick Leduc, Maud Vinet, and Perrine Batude. System on wafer:
A new silicon concept in SiP. Proceedings of the IEEE, 97(1):60–69, 2009.
J. U. Knickerbocker, P. S. Andry, L. P. Buchwalter, A. Deutsch, R. R. Horton, K. A. Jenkins,
Y. H. Kwark, G. McVicker, C. S. Patel, R. J. Polastre, C. D. Schuster, A. Sharma, S. M.
Sri-Jayantha, C. W. Surovic, C. K. Tsang, B. C. Webb, S. L. Wright, S. R. McKnight,
E. J. Sprogis, and B. Dang. Development of next-generation system-on-package (SOP)
technology based on silicon carriers with fine-pitch chip interconnection. IBM Journal of
Research and Development, 49(4.5):725–753, 2010.

[16] Vaidyanathan Kripesh, Seung Wook Yoon, V. P. Ganesh, Navas Khan, Mihai D. Rotaru,
Wang Fang, and Mahadevan K. Iyer. Three-dimensional system-in-package using stacked
silicon platform technology. IEEE Transactions on Advanced Packaging, 28(3):377–386,
2005.

[17] Atabak Rashidian, Saman Jafarlou, Alexander Tomkins, Kim Law, Mihai Tazlauanu, and
Kenji Hayashi. Compact 60 GHz phased-array antennas with enhanced radiation properties
in flip-chip BGA packages. IEEE Transactions on Antennas and Propagation, 67(3):1605–
1619, 2019.

[18] Dane C. Thompson, Olivier Tantot, Hubert Jallageas, George E. Ponchak, Manos M.
Tentzeris, and John Papapolymerou. Characterization of liquid crystal polymer (LCP)
material and transmission lines on LCP substrates from 30 to 110 GHz. IEEE Transactions
on Microwave Theory and Techniques, 52(4):1343–1352, 2004.

[19] Dane Thompson. Characterization and Design of Liquid Crystal Polymer (LCP) Based
Multilayer Rf Components and Packages. PhD thesis, Georgia Institute of Technology,
2006.

[20] Peter Ramm, Armin Klumpp, Josef Weber, Nicolas Lietaer, Maaike Taklo, Walter De Raedt,
Thomas Fritzsch, and Pascal Couderc. 3D integration technology: Status and application
development. European Solid State Circuits Conference, pages 9–16, 2010.

[21] Debabani Choudhury. 3D integration technologies for emerging microsystems. In IEEE

International Microwave Symposium Digest, pages 1–4. IEEE, 2010.

[22] Rao R. Tummala. Packaging: Past, present and future. International Conference on Elec-

tronics Packaging Technology, 2005, 2005.

[23] Rao R. Tummala. Moore’s Law for Packaging to Replace Moore’s Law for ICS. Pan Pacific

Microelectronics Symposium, pages 1–6, 2019.

[24] Tailong Shi, Chintan Buch, Vanessa Smet, Yoichiro Sato, Lutz Parthier, Frank Wei, Cody
Lee, Venky Sundaram, and Rao Tummala. First Demonstration of Panel Glass Fan-Out
(GFO) Packages for High I/O Density and High Frequency Multi-chip Integration.
In
Electronic Components and Technology Conference, pages 41–46. IEEE, 2017.

173

[25] Scott R. McCann, Venkatesh Sundaram, Rao R. Tummala, and Suresh K. Sitaraman. Flip-
chip on glass (FCOG) package for low warpage. In Electronic Components and Technology
Conference, pages 2189–2193. IEEE, 2014.

[26] Vivek Sridharan, Sunghwan Min, Venky Sundaram, Vijay Sukumaran, Seunghyun Hwang,
Hunter Chan, Fuhan Liu, Christian Nopper, and Rao Tummala. Design and fabrication
of bandpass filters in glass interposer with through-package-vias (TPV).
In Electronic
Components and Technology Conference, pages 530–535. IEEE, 2010.

[27] Venky Sundaram, Yoichiro Sato, Toshitake Seki, Yutaka Takagi, Vanessa Smet, Makoto
Kobayashi, and Rao Tummala. First demonstration of a surface mountable, ultra-thin glass
BGA package for smart mobile logic devices. In Electronic Components and Technology
Conference, pages 365–370. IEEE, 2014.
Jie-Ying Zhong, Wen-Jie Lin, Jen-Hao Cheng, Yi-Hsiang Kung, Jiun-Peng Chen, Jeng-Han
Tsai, Powen Hsu, and Tian-Wei Huang. A High Spectral Efficiency Receiver at 57-66 GHz
Using 65-nm CMOS in LTCC Package With Polarization MIMO. IEEE Access, 7:129466–
129479, 2019.

[28]

[29] Guang Chen, Yuan Gu, Harvey Tsang, Daniel R. Hines, and Siddhartha Das. The Effect
of Droplet Sizes on Overspray in Aerosol-Jet Printing. Advanced Engineering Materials,
1701084:1–13, 2018.

[30]

Jameel Showail, Markku Lahti, Kautio Kari, Eyad Arabi, Pekka Rantakari, Ismo Huhtinen,
Tauno Vaha-Heikkila, and Atif Shamim. SIW Cavity Filters with Embedded Planar Res-
onators in LTCC Package for 5G Applications. European Microwave Conference, pages
757–760, 2018.

[31] Dingding Kuang, Gang Dong, Hui Nie, Wei Xiong, and Yintang Yang. Novel double fractal
patches structure Antenna-in-Package based on LTCC technology for 2.4 GHz applications.
International Journal of RF and Microwave Computer-Aided Engineering, 28(5), 2018.
Jianfeng Zhu, Student Member, Yang Yang, Senior Member, Chenhao Chu, Shufang Li,
and Senior Member. Low-Profile Wideband and High Gain LTCC Patch Antenna Array for
60-GHz Applications. IEEE Transactions on Antennas and Propagation, PP(c):1, 2019.

[32]

[33] Alessandro Di Carlofelice, Francesco De Paulis, Antonio Fina, Ulisse Di Marcantonio,
Antonio Orlandi, and Piero Tognolatti. Compact and Reliable T/R Module Prototype for
Advanced Space Active Electronically Steerable Antenna in 3-D LTCC Technology. IEEE
Transactions on Microwave Theory and Techniques, 66(6):2746–2756, 2018.

[34] Thomas P Budka. Wide-Bandwidth Millimeter-Wave Bond-Wire Interconnects. IEEE Trans-

actions on Microwave Theory and Techniques, 49(4):715–718, 2001.

[35] Albert Sutono, N. Gio Cafaro, Joy Laskar, and Manos M. Tentzeris. Experimental modeling,
repeatability investigation and optimization of microwave bond wire interconnects. IEEE
Transactions on Advanced Packaging, 24(4):595–603, 2001.

174

[36] T. Krems, W. Haydl, H. Massler, and J. Rudiger. Millimeter-wave performance of chip
interconnections using wire bonding and flip chip. In International Microwave Symposium,
pages 247–250. IEEE, 1996.

[37] Andrea Jentzsch and Wolfgang Heinrich. Theory and measurements of flip-chip inter-
IEEE Transactions on Microwave Theory and

connects for frequencies up to 100 GHz.
Techniques, 49(5):871–878, 2001.

[39]

[38] Bosui Liu, Xun Gong, and William J. Chappell. Applications of layer-by-layer polymer
stereolithography for three-dimensional high-frequency components. IEEE Transactions on
Microwave Theory and Techniques, 52(11):2567–2575, 2004.
Jimmy G. Hester, Sangkil Kim, Jo Bito, Taoran Le, John Kimionis, Daniel Revier, Christy
Saintsing, Wenjing Su, Bijan Tehrani, Anya Traille, Benjamin S. Cook, and Manos M.
Tentzeris. Additively manufactured nanotechnology and origami-enabled flexible microwave
electronics. Proceedings of the IEEE, 103(4):583–606, 2015.
Jennifer A Byford, Mohd Ifwat Mohd Ghazali, Saranraj Karuppuswami, Brian L Wright,
and Premjeet Chahal. Demonstration of RF and Microwave Passive Circuits Through 3-D
Printing and Selective Metalization.
IEEE Transactions on Components, Packaging and
Manufacturing Technology, 7(3):463–471, 2017.

[40]

[41] Ryan Bahr, Bijan Tehrani, and Manos M. Tentzeris. Exploring 3-D printing for new appli-

cations. IEEE Microwave Magazine, 19(1):57–66, 2018.

[42] Thomas P Ketterl, Yaniel Vega, Nicholas C Arnal, John W I Stratton, Eduardo A Rojas-
Nastrucci, María F Córdoba-Erazo, Mohamed M Abdin, Casey W Perkowski, Paul I Deffen-
baugh, Kenneth H Church, and Thomas M Weller. A 2.45 GHz Phased Array Antenna Unit
Cell Fabricated Using 3-D Muliti-Layer Direct Digital Manufacturing. IEEE Transactions
on Microwave Theory and Techniques, 63(12):4382–4394, 2015.

[43] Mohd Ifwat Mohd Ghazali, Jennifer A. Byford, Saranraj Karuppuswami, Amanpreet Kaur,
James Lennon, and Premjeet Chahal. 3D Printed Out-of-Plane Antennas for Use on High
Density Boards. In IEEE Electronics Components and Technology Conference, pages 1835–
1842. IEEE, 2017.

[44] Yuxiao He, Michael Thomas Craton, Premjeet Chahal, and John Papapolymerou. A Bi-
material Fully Aerosol Jet printed W-band Quasi-Yagi-Uda Antenna. Global Symposium on
Millimeter Waves (GSMM), pages 1–3, 2018.
Ibrahim T. Nassar and Thomas M. Weller. An electrically-small, 3-D cube antenna fabricated
with additive manufacturing. In IEEE Conference on Power Amplifiers for Wireless and Radio
Applications, pages 85–87. IEEE, 2013.

[45]

[46] Min Liang, Corey Shemelya, Eric MacDonald, Ryan Wicker, and Hao Xin. 3-D printed
microwave patch antenna via fused deposition method and ultrasonic wire mesh embedding
technique. IEEE Antennas and Wireless Propagation Letters, 14:1346–1349, 2015.

175

[47] K Lomakin, M Ankenbrand, M Sippel, J Franke, K Helmreich, and G Gold. Nanojet Printed
Coplanar Waveguides on Flexible Polyimide Substrate up to 24 GHz. In IEEE International
Conference on Flexible and Printable Sensors and Systems, 2019.

[48] Eduardo A. Rojas-Nastrucci, T. Weller, Vera Lopez Aida, Fan Cai, and John Papapolymerou.
A study on 3D-printed coplanar waveguide with meshed and finite ground planes. In IEEE
Wireless and Microwave Technology Conference, pages 3–5. IEEE, 2014.

[49] Bing Zhang and Herbert Zirath. Metallic 3-D Printed Rectangular Waveguides for
Millimeter-Wave Applications. IEEE Transactions on Components, Packaging and Manu-
facturing Technology, 6(5):796–804, 2016.

[50] Zhenzhen Shen and Aleksey Reiderman. Additive Manufacturing for Multi-chip Modules.

In International Symposium on Microeletronics, 2018.

[51] Mario D’Auria, William J. Otter, Jonathan Hazell, Brendan T.W. Gillatt, Callum Long-
Collins, Nick M. Ridler, and Stepan Lucyszyn. 3-D Printed Metal-Pipe Rectangular Waveg-
uides.
IEEE Transactions on Components, Packaging and Manufacturing Technology,
5(9):1339–1349, 2015.

[52] Michael Craton, Jennifer A. Byford, Vincens Gjokaj, John Papapolymerou, and Premjeet
Chahal. 3D Printed High Frequency Coaxial Transmission Line Based Circuits. In IEEE
Electronics Components and Technology Conference, pages 1080–1087. IEEE, 2017.

[53] Fan Cai, Yung-hang Chang, Kan Wang, Chuck Zhang, Ben Wang, and John Papapolymerou.
Low-Loss 3-D Multilayer Transmission Lines and Interconnects Fabricated by Additive
Manufacturing Technologies.
IEEE Transactions on Microwave Theory and Techniques,
64(10):3208–3216, 2016.

[54] Adrian Tamayo-Dominguez, Jose Manuel Fernandez-Gonzalez, and Manuel Sierra-Perez.
Groove Gap Waveguide in 3-D Printed Technology for Low Loss, Weight, and Cost Distribu-
tion Networks. IEEE Transactions on Microwave Theory and Techniques, 65(11):4138–4147,
2017.

[55] Fan Cai, Tanveer Khan, and John Papapolymerou. A Low Loss X-band Filter Using 3-D

Poly jet Technology. In IEEE International Microwave Symposium. IEEE, 2015.

[56] Cheng Guo, Xiaobang Shang, Michael J. Lancaster, and Jun Xu. A 3-D Printed Lightweight
X-Band Waveguide Filter Based on Spherical Resonators. IEEE Microwave and Wireless
Components Letters, 25(7):442–444, 2015.

[57] Amanpreet Kaur, Joshua C Myers, Mohd Ifwat Mohd Ghazali, Jennifer Byford, and Premjeet
Chahal. Affordable Terahertz Components using 3D Printing. In Electronic Components &
Technology Conference, pages 2071–2076. IEEE, 2015.

[58] Benjamin S Cook, James R Cooper, and Manos M Tentzeris. An Inkjet-Printed Micro
fluidic RFID-Enabled Platform for Wireless Lab-on-Chip Applications. IEEE Transactions
on Microwave Theory and Techniques, 61(12):4714–4723, 2013.

176

[59] Cristiano Tomassoni, Ryan Bahr, Manos Tentzeris, Maurizio Bozzi, and Luca Perregrini.
3D Printed Substrate Integrated Waveguide Filters with Locally Controlled Dielectric Per-
mittivity. In European Microwave Conference, pages 253–256. IEEE, 2016.
Ilona Piekarz, Jakub Sorocki, Michael Thomas Craton, Krzysztof Wincza, Slawomir
Gruszczynski, and John Papapolymerou. Application of Aerosol Jet 3D Printing with Con-
ductive and Non-Conductive Inks for Manufacturing mm-Wave Circuits. IEEE Transactions
on Components, Packaging and Manufacturing Technology, PP(c):1–1, 2018.

[60]

[61] Michael Thomas Craton, Christopher Oakley, John D Albrecht, Premjeet Chahal, and John
Papapolymerou. Fully Additively Manufactured Broadband Low Loss High Frequency
Interconnects. In Asia Pacific Microwave Conference, 2018.

[62] Aimeric Bisognin, Diane Titz, Fabien Ferrero, Romain Pilard, Carlos A. Fernandes, Jorge R.
Costa, Christian Corre, Pierino Calascibetta, Jean Michel Riviere, Alexis Poulain, Christian
Badard, Frederic Gianesello, Cyril Luxey, Pierre Busson, Daniel Gloria, and Didier Belot.
3D printed plastic 60 GHz lens: Enabling innovative millimeter wave antenna solution and
system. In IEEE International Microwave Symposium Digest. IEEE, 2014.

[63] Mehmet A. Belen and Peyman Mahouti. Design of nonuniform substrate dielectric lens anten-
nas using 3D printing technology. Microwave and Optical Technology Letters, (February):1–
7, 2019.

[64] Shiyu Zhang, Yiannis Vardaxoglou, Will Whittow, and Raj Mittra. 3D-printed flat lens for
microwave applications. In Loughborough Antennas and Propagation Conference, pages
31–33. IEEE, 2015.

[65] Michael Thomas Craton, Yining He, Aljoscha Roch, Premjeet Chahal, and John Papapoly-
merou. Additively Manufactured Interdigitated Capacitors Using Barium Titanate Nanocom-
posite Inks. In European Microwave Conference, 2019.

[66] Garret Mckerricher, Mohammad Vaseem, and Atif Shamim. Fully inkjet-printed microwave

passive electronics. Microsystems & Nanoengineering, 3:1–7, 2017.

[67] Bijan K Tehrani and Manos M Tentzeris. Fully Inkjet-Printed Ramp Interconnects for
Wireless Ka-Band MMIC Devices and Multi-Chip Module Packaging. European Microwave
Conference (EuMC), pages 1037–1040, 2018.

[68] Michael Thomas Craton, John D. Albrecht, Premjeet Chahal, and John Papapolymerou. A
Chip-First Approach to Millimeter-Wave Circuit Packaging. IEEE Microwave and Wireless
Components Letters, 29(2):116–118, 2019.

[69] Ahmed Elmogi, Hannes Ramon, Joris Lambrecht, Peter Ossieur, Guy Torfs, Jeroen Missinne,
Peter De Heyn, Yoojin Ban, Marianna Pantouvaki, Joris Van Campenhout, and Geert Van
Steenberge. Aerosol-Jet Printed Interconnects for 60-Gb / s CMOS Driver and Microring
Modulator Transmitter Assembly. IEEE Photonics Technology Letters, 30(22):1944–1947,
2018.

177

[70] Ville Pekkanen, Matti Mäntysalo, Kimmo Kaija, Pauliina Mansikkamäki, Esa Kunnari, Katja
Laine, Juha Niittynen, Santtu Koskinen, Eerik Halonen, and Umur Caglar. Utilizing inkjet
printing to fabricate electrical interconnections in a system-in-package. Microelectronic
Engineering, 87(11):2382–2390, 2010.

[71] Franz Xaver Röhrl, Johannes Jakob, Werner Bogner, Robert Weigel, Stefan Zorn, T H D
Technische, and Hochschule Deggendorf. Bare Die Connections via Aerosol Jet Technology
for Millimeter Wave Applications.
In European Microwave Conference (EuMC), pages
1033–1036, 2018.

[72] Stefano Moscato, Ryan Bahr, Taoran Le, Marco Pasian, Maurizio Bozzi, Luca Perregrini, and
Manos M. Tentzeris. Infill-Dependent 3-D-Printed Material Based on NinjaFlex Filament
for Antenna Applications. IEEE Antennas and Wireless Propagation Letters, 15:1506–1509,
2016.

[73] Michael Thomas Craton, John D. Albrecht, Premjeet Chahal, and John Papapolymerou.
In Situ Nanocomposite Fabrication for RF Electronics Applications With Additive Man-
ufacturing. IEEE Transactions on Microwave Theory and Techniques, 68(5):1646–1659,
2020.

[74] Gang Liu, Andreas Trasser, A. ÇaÇğri Ulusoy, and Hermann Schumacher. Low-loss,
In IEEE

low-cost, IC-to-board bondwire interconnects for millimeter-wave applications.
International Microwave Symposium Digest, volume 1, pages 3–6, 2011.

[75] Manfred Mengel and Ivan Nikitin. Inkjet printed dielectrics for electronic packaging of chip

embedding modules. Microelectronic Engineering, 87(4):593–596, 2010.

[76] Bijan K. Tehrani, Benjamin S. Cook, and Manos M. Tentzeris. Inkjet-printed 3D intercon-
nects for millimeter-wave system-on-package solutions. In IEEE International Microwave
Symposium Digest. IEEE, 2016.

[77] Guizhen Zheng, John Papapolymerou, and Manos M. Tentzeris. Wideband Coplanar Waveg-
IEEE Microwave and

uide RF Probe Pad to Microstrip Transitions without via Holes.
Wireless Components Letters, 13(12):544–546, 2003.

[78] K Lomakin, M Sippel, G Gold, J Ringel, D Weiß, K Helmreich, M Ankenbrand, and
J Franke. Substituting Bond Wires by Additively Manufactured Interconnections. In German
Microwave Conference, pages 367–370, 2018.

[79] G. Arlt, D. Hennings, and G. de With. Dielectric properties of fine-grained barium titanate

ceramics. Journal of Applied Physics, 58(4):1619–1625, 1985.

[80] Vladimir Petrovsky, Tatiana Petrovsky, Swetha Kamlapurkar, and Fatih Dogan. Dielectric
constant of barium titanate powders near curie temperature. Journal of the American Ceramic
Society, 91(11):3590–3592, 2008.

[81] Sewoong Oh, Jae Hyuk Park, and Jun Akedo. Dielectric characteristics of barium strontium
IEEE Transactions on

titanate films prepared by aerosol deposition on a Cu substrate.
Ultrasonics, Ferroelectrics, and Frequency Control, 56(3):421–424, 2009.

178

[82] Tao Hu, Jari Juuti, Heli Jantunen, and Taisto Vilkman. Dielectric properties of BST/polymer

composite. Journal of the European Ceramic Society, 27(13-15):3997–4001, 2007.

[83] Nicholas A D Burke, Harald D H Stöver, and Francis P. Dawson. Magnetic nanocompos-
ites: Preparation and characterization of polymer-coated iron nanoparticles. Chemistry of
Materials, 14(11):4752–4761, 2002.

[84] Naga Gopi Devaraju, Eung Soo Kim, and Burtrand I. Lee. The synthesis and dielectric study
of BaTiO3/polyimide nanocomposite films. Microelectronic Engineering, 82(1):71–83,
2005.

[86]

[85] Sea Fue Wang, Yuh Ruey Wang, Kuo Chung Cheng, and Yu Ping Hsaio. Characteristics of
polyimide/barium titanate composite films. Ceramics International, 35(1):265–268, 2009.
Jong Jin Choi, Byung Dong Hahn, Jungho Ryu, Woon Ha Yoon, Byoung Kuk Lee, and
Dong Soo Park. Preparation and characterization of piezoelectric ceramic-polymer com-
posite thick films by aerosol deposition for sensor application. Sensors and Actuators,
153:89–95, 2009.

[87] Motoyuki Iijima, Nobuhiro Sato, I. Wuled Lenggoro, and Hidehiro Kamiya. Surface modifi-
cation of BaTiO3 particles by silane coupling agents in different solvents and their effect on
dielectric properties of BaTiO3/epoxy composites. Colloids and Surfaces A: Physicochem-
ical and Engineering Aspects, 352(1-3):88–93, 2009.

[88] Xianbo Yang and Premjeet Prem Chahal. On-wafer terahertz ribbon waveguides using
polymer-ceramic nanocomposites. IEEE Transactions on Components, Packaging and Man-
ufacturing Technology, 5(2):245–255, 2015.

[89] Premjeet Chahal, Rao R. Tummala, Mark G. Allen, and Madhavan Swaminathan. A novel
integrated decoupling capacitor for MCM-L technology. IEEE Transactions on Components
Packaging and Manufacturing Technology, 21(2):184–192, 1998.

[90] R. K. Goyal, S. S. Katkade, and D. M. Mule. Dielectric, mechanical and thermal properties
of polymer/BaTiO 3 composites for embedded capacitor. Composites Part B: Engineering,
44(1):128–132, 2013.

[91] Kan Wang, Yung-Hang Chang, Chuck Zhang, and Ben Wang. Conductive-on-demand:
Tailorable polyimide/carbon nanotube nanocomposite thin film by dual-material aerosol jet
printing. Carbon, 98:397–403, 2016.

[92] A. M. Sukeshini, P. Gardner, F. Meisenkothen, T. Jenkins, R. Miller, M. Rottmayer, and T. L.
Reitz. Aerosol Jet Printing and Microstructure of SOFC Electrolyte and Cathode Layers.
Electrochemical Society Transactions, 35(1):2151–2160, 2011.

[93] H. Y. Yang and N. G. Alexopoulos. Gain Enhancement Methods for Printed Circuit An-
tennas Through Multiple Superstrates. IEEE Transactions on Antennas and Propagation,
35(7):860–863, 1987.

179

[94] D. R. Jackson and N. G. Alexopoulos. Gain Enhancement Methods for Printed Circuit

Antennas. IEEE Transactions on Antennas and Propagation, 33(9):976–987, 1985.

[95] L. Shafai. Dielectric Loaded Antennas, 2005.

[96] L. D’Auria, J. P. Huignard, A. M. Roy, and E. Spitz. Photolithographic fabrication of thin

film lenses. Optics Communications, 5(4):232–235, 1972.

[97] Nasser Ghassemi and Ke Wu. Planar high-gain dielectric-loaded antipodal linearly tapered
slot antenna for E-and W-band gigabyte point-to-point wireless services. IEEE Transactions
on Antennas and Propagation, 61(4):1747–1755, 2013.

[98] C. T. Chan, K. M. Ho, and C. M. Soukoulis. Photonic band gaps in experimentally realizable

periodic dielectric structures. Europhysics Letters, 16(6):563–568, 1991.

[99] K. M. Ho, C. T. Chan, C. M. Soukoulis, R. Biswas, and M. Sigalas. Photonic band gaps
in three dimensions: New layer-by-layer periodic structures. Solid State Communications,
89(5):413–416, 1994.

[100] Hung Yu David Yang and Jianpei Wang. Surface waves of printed antennas on planar
artificial periodic dielectric structures. IEEE Transactions on Antennas and Propagation,
49(3):444–450, 2001.

[101] Zhiyi Zhang, Ping Zhao, Gaozhi Xiao, Min Lin, and Xudong Cao. Focusing-enhanced

mixing in microfluidic channels. Biomicrofluidics, 2(1):1–9, 2008.

[102] Hye Yoon Park, Xiangyun Qiu, Elizabeth Rhoades, Jonas Korlach, Lisa W. Kwok, Warren R.
Zipfel, Watt W. Webb, and Lois Pollack. Achieving uniform mixing in a microfluidic device:
Hydrodynamic focusing prior to mixing. Analytical Chemistry, 78(13):4465–4473, 2006.
[103] James Q. Feng and Michael J. Renn. Aerosol Jet ® Direct-Write for Microscale Additive

Manufacturing. Journal of Micro and Nano-Manufacturing, 7(1), 2019.

[104] Lung-Hwa Hsieh and Kai Chang. Equivalent lumped elements G, L, C, and unloaded Q’s
of closed- and open-loop ring resonators. IEEE Transactions on Microwave Theory and
Techniques, 50(2):453–460, 2002.

[105] Kai Chang. Microwave Ring Circuits and Antennas. Wiley, New York, 1996.
[106] Gang Zou, Hans Grönqvist, J. Piotr Starski, and Johan Liu. Characterization of liquid
crystal polymer for high frequency system-in-a-package applications. IEEE Transactions on
Advanced Packaging, 25(4):503–508, 2002.

[107] I. J. Bahl and D. K. Trivedi. A designer’s guide to microstrip line. Microwaves, 16(5):174–

176, 1977.

[108] K. C. Gupta, Ramesh Garg, Inder Bahl, and Prakash Bhartia. Microstrip Lines and Slot

Lines. Artech House, 2 edition, 1996.

180

[109] M. D. Abouzahra and L. Lewin. Radiation from Microstrip Discontinuities. IEEE Transac-

tions on Microwave Theory and Techniques, 27(8):722–723, 1979.

[110] Brian C. Wadell. Transmission Line Design Handbook, 1991.
[111] DuPont. DuPont âĎć Kapton ® Summary of Properties. DuPont Kap:1–7, 2017.
[112] P. Chahal, R. R. Tummala, and M. G. Allen. Integrated Capacitors using Polymer-Ceramic
Composites for MCM-L. In International Symposium on Microeletronics, pages 126–131,
1996.

[113] H. C. Pant, M. K. Patra, Aditya Verma, S. R. Vadera, and N. Kumar. Study of the dielectric
properties of barium titanate-polymer composites. Acta Materialia, 54(12):3163–3169,
2006.

[114] Ari Sihvola. Electromagnetic Mixing Formulas and Applications. Institution of Elcctrical

Engineers, London, 1999.

[115] S. M. Abbas, A. K. Dixit, R. Chatterjee, and T. C. Goel. Complex permittivity and microwave
absorption properties of BaTiO 3-polyaniline composite. Materials Science and Engineering
B: Solid-State Materials for Advanced Technology, 123(2):167–171, 2005.

[116] Xiaodong Chen, Guiqin Wang, Yuping Duan, and Shunhua Liu. Electromagnetic character-
istics of barium titanate/epoxide resin composites in X and Ku bands. Journal of Alloys and
Compounds, 453(1-2):433–436, 2008.

[117] A. J. Pointon and R. M. Henson. Dielectric Loss Mechanisms in the Paraelectric Region for

Ceramic Barium Titanate. Ferroelectrics, 7(1):359–360, 1974.

[118] Omodara Gbotemi, Sami Myllymaki, Jani Kallioinen, Jari Juuti, Merja Teirikangas, Heli
Jantunen, Marjeta Mac Ek Krzmanc, Danilo Suvorov, Marcin Sloma, and Malgorzata
Jakubowska. Characterization of PMMA/BaTiO3 Composite Layers Through Printed Ca-
pacitor Structures for Microwave Frequency Applications. IEEE Transactions on Microwave
Theory and Techniques, 66(4):1736–1743, 2018.

[119] R. Popielarz, C. K. Chiang, R. Nozaki, and J. Obrzut. Dielectric properties of polymer/ferro-
electric ceramic composites from 100 Hz to 10 GHz. Macromolecules, 34(17):5910–5915,
2001.

[120] Guiqin Wang, Xiaodong Chen, Yuping Duan, and Shunhua Liu. Electromagnetic properties
of carbon black and barium titanate composite materials. Journal of Alloys and Compounds,
454(1-2):340–346, 2008.

[121] Mark P. McNeal, Sei Joo Jang, and Robert E. Newnham. The effect of grain and particle
size on the microwave properties of barium titanate (BaTiO3). Journal of Applied Physics,
83(6):3288–3297, 1998.

[122] V. S. Kishanprasad and P. H. Gedam. Polyamic acids: Thermal and microwave imidization

and film properties. Journal of Applied Polymer Science, 50(3):419–429, 1993.

181

[123] Min Sang, Sheng Wang, Mei Liu, Linfeng Bai, Wanquan Jiang, Shouhu Xuan, and Xing-
long Gong. Fabrication of a piezoelectric polyvinylidene fluoride/carbonyl iron (PVDF/CI)
magnetic composite film towards the magnetic field and deformation bi-sensor. Composites
Science and Technology, 165:31–38, 2018.

[124] Baoshan Zhang, Yong Feng, Jie Xiong, Yi Yang, and Huaixian Lu. Microwave-Absorbing
Properties of De-Aggregated Flake-Shaped Carbonyl-Iron Particle Composites at 2-18 GHz.
IEEE Transactions on Magnetics, 42(7):1778–1781, 2006.

[125] Pekka M.T. Ikonen, Konstantin N. Rozanov, Alexey V. Osipov, Pekka Alitalo, and Sergei A.
Tretyakov. Magnetodielectric substrates in antenna miniaturization: Potential and limita-
tions. IEEE Transactions on Antennas and Propagation, 54(11):3391–3399, 2006.

[126] P. S. Keatley, E. R. Glover, B. Tremain, I. R. Hooper, A. P. Hibbins, and R. J. Hicken. A
Ferrite-Filled Cavity Resonator for Electronic Article Surveillance on Metallic Packaging.
IEEE Transactions on Magnetics, 55(12), 2019.

[127] H. Meng, R. Sbiaa, S. Y.H. Lua, C. C. Wang, M. A.K. Akhtar, S. K. Wong, P. Luo, C. J.P.
Carlberg, and K. S.A. Ang. Low current density induced spin-transfer torque switching in
CoFeB-MgO magnetic tunnel junctions with perpendicular anisotropy. Journal of Physics
D: Applied Physics, 44, 2011.

[128] Yuxiao He, Eric Drew, Z. John Zhang, and John Papapolymerou. Compact Microstrip Patch
Antenna Utilizing Low Cost Solution Cast Nanomagnetic Thin Film. EEE International
Symposium on Antennas and Propagation, pages 1118–1121, 2019.

[129] Jae Y. Park and Mark G. Allen. Packaging-compatible microinductors and microtrans-
formers with screen-printed ferrite using low temperature processes. IEEE Transactions on
Magnetics, 34(4):1366–1368, 1998.

[130] Farhan A. Ghaffar, Mohammad Vaseem, Langis Roy, and Atif Shamim. Design and Fab-
rication of a Frequency and Polarization Reconfigurable Microwave Antenna on a Printed
Partially Magnetized Ferrite Substrate. IEEE Transactions on Antennas and Propagation,
66(9):4866–4871, 2018.

[131] Lanbing Liu, Chao Ding, Shengchang Lu, Ting Ge, Yi Yan, Yunhui Mei, Khai D.T. Ngo, and
Guo Quan Lu. Design and Additive Manufacturing of Multipermeability Magnetic Cores.
IEEE Transactions on Industry Applications, 54(4):3541–3547, 2018.

[132] Varun Chaudhary, Nartu Mohan Sai Kiran Kumar Yadav, Srinivas Aditya Mantri, Sriswaroop
Dasari, Abhinav Jagetia, R. V. Ramanujan, and R. Banerjee. Additive manufacturing of
functionally graded CoâĂŞFe and NiâĂŞFe magnetic materials. Journal of Alloys and
Compounds, 823:1–8, 2020.

[133] C. V. Mikler, V. Chaudhary, T. Borkar, V. Soni, D. Jaeger, X. Chen, R. Contieri, R. V.
Ramanujan, and R. Banerjee. Laser Additive Manufacturing of Magnetic Materials. The
Journal of The Minerals, Metals & Materials Society, 69(3):532–543, 2017.

182

[134] D. Goll, D. Schuller, G. Martinek, T. Kunert, J. Schurr, C. Sinz, T. Schubert, T. Bern-
thaler, H. Riegel, and G. Schneider. Additive manufacturing of soft magnetic materials and
components. Additive Manufacturing, 27:428–439, 2019.

[135] Garret McKerricher, Jose Gonzalez Perez, and Atif Shamim. Fully inkjet printed RF in-
ductors and capacitors using polymer dielectric and silver conductive ink with through vias.
IEEE Transactions on Electron Devices, 62(3):1002–1009, 2015.

[136] John R. Long and Miles A. Copeland. The modeling, characterization, and design of
monolithic inductors for silicon RF IC’s. IEEE Journal of Solid-State Circuits, 32(3):357–
369, 1997.

[137] Yuan Gu, Donghun Park, David Bowen, Siddhartha Das, and Daniel R. Hines. Direct-Write
Printed, Solid-Core Solenoid Inductors with Commercially Relevant Inductances. Advanced
Materials Technologies, 1800312:1–7, 2018.

[138] Mohammad Vaseem, Garret McKerricher, and Atif Shamim. 3D inkjet printed radio fre-
quency inductors and capacitors. European Microwave Integrated Circuits Conference,
pages 544–547, 2016.

[139] S. M. Bidoki, J. Nouri, and A. A. Heidari. Inkjet deposited circuit components. Journal of

Micromechanics and Microengineering, 20(055023), 2010.

[140] Chiara Mariotti, Federico Alimenti, Luca Roselli, and Manos M. Tentzeris. High-
Performance RF Devices and Components on Flexible Cellulose Substrate by Vertically
Integrated Additive Manufacturing Technologies. IEEE Transactions on Microwave Theory
and Techniques, 65(1):62–71, 2017.

[141] Hoseon Lee, Benjamin S. Cook, K. P. Murali, Markondeya Raj, and Manos M. Tentzeris.
Inkjet Printed High-Q RF Inductors on Paper Substrate with Ferromagnetic Nanomaterial.
IEEE Microwave and Wireless Components Letters, 26(6):419–421, 2016.

[142] Hyun Jun Hwang, Sung Jun Joo, and Hak Sung Kim. Copper Nanoparticle/Multiwalled
Carbon Nanotube Composite Films with High Electrical Conductivity and Fatigue Resistance
Fabricated via Flash Light Sintering. ACS Applied Materials and Interfaces, 7(45):25413–
25423, 2015.

[143] Ji Hoon Lee, Na Rae Kim, Byoung Joon Kim, and Young Chang Joo. Improved mechanical
performance of solution-processed MWCNT/Ag nanoparticle composite films with oxygen-
pressure-controlled annealing. Carbon, 50(1):98–106, 2012.

[144] F. M. Smits. Measurement of Sheet Resistivities with the FourâĂŘPoint Probe. Bell System

Technical Journal, 37(3):711–718, 1958.

[145] Michael Thomas Craton, Student Member, Xenofon Konstantinou, Student Member, John D
Albrecht, Premjeet Chahal, Senior Member, and John Papapolymerou. A Chip-First Mi-
crowave Package Using Multimaterial Aerosol Jet Printing. IEEE Trans. Microw. Theory
Techn., 68(8):3418–3427, 2020.

183

[146] Ramiro A. Ramirez, Di Lan, Jing Wang, and Thomas M. Weller. MMIC packaging and
on-chip low-loss lateral interconnection using additive manufacturing and laser machining.
In IEEE International Microwave Symposium Digest, pages 38–40, 2017.

[147] C Kaestle, J Hoerber, F Oechsner, and J Franke. Prospects of wire bonding as an approach
for contacting additive manufactured Aerosol Jet printed structures. In European Microelec-
tronics Packaging Conference (EMPC), number September, pages 1–6. IEEE, 2015.

[148] Chiara Mariotti, Student Member, Wenjing Su, Benjamin S Cook, Luca Roselli, Senior
Member, and Manos M Tentzeris. Development of Low Cost , Wireless , Inkjet Printed
Microfluidic RF Systems and Devices for Sensing or Tunable Electronics. IEEE Sensors
Journal, 15(6):3156–3163, 2015.

[149] Eun-Seong Kim, Jun-Ge Liang, Cong Wang, Myung-Yeon Cho, Jong-Min Oh, and Nam-
Young Kim. Inter-digital capacitors with aerosol-deposited high-K dielectric layer for highest
capacitance value in capacitive super-sensing applications. Scientific Reports, 9(1):680,
2019.

[150] Ramiro A Ramirez, Di Lan, Eduardo A Rojas-nastrucci, and Thomas M Weller. Laser
Assisted Additive Manufacturing of CPW mm-Wave Interdigital Capacitors. International
Microwave Symposium, 1:1553–1556, 2018.

[151] Chiara Mariotti, Benjamin S. Cook, Luca Roselli, and Manos M. Tentzeris. State-of-the-art
inkjet-printed metal-insulator-metal (MIM) capacitors on silicon substrate. IEEE Microwave
and Wireless Components Letters, 25(1):13–15, 2015.

[152] Benjamin S. Cook, James R. Cooper, and Manos M. Tentzeris. Multi-layer RF capacitors on
flexible substrates utilizing inkjet printed dielectric polymers. IEEE Microwave and Wireless
Components Letters, 23(7):353–355, 2013.

[153] Jolke Perelaer, Berend Jan De Gans, and Ulrich S. Schubert. Ink-jet printing and microwave

sintering of conductive silver tracks. Advanced Materials, 18(16):2101–2104, 2006.

[154] Jubaid Abdul Qayyum, Marvin Abt, Aljoscha Roch, Ahmet Cagri Ulusoy, and John Pa-
papolymerou. Ultra wideband 3D interconnects using aerosol jet printing up to 110 GHz.
European Microwave Conference (EuMC), pages 1112–1115, 2017.

[155] Martin Ihle, Steffen Ziesche, Christian Zech, and Benjamin Baumann. Functional Printing
of MMIC-Interconnects on LTCC Packages for sub-THz applications. European Microelec-
tronics and Packaging Conference (EMPC), pages 1–4, 2019.

[156] Neelanjan Sarmah, Janusz Grzyb, Konstantin Statnikov, Stefan Malz, Pedro Rodriguez
Vazquez, Wolfgang Föerster, Bernd Heinemann, and Ullrich R. Pfeiffer. A Fully Integrated
240-GHz Direct-Conversion Quadrature Transmitter and Receiver Chipset in SiGe Technol-
ogy. IEEE Transactions on Microwave Theory and Techniques, 64(2):562–574, 2016.

[157] J Grzyb, K Statnikov, and U Pfeiffer. A Lens-Coupled All-Silicon Integrated 2x2 Array
of Harmonic Receivers for THz Multi-Color Active Imaging. European Conference on
Antennas and Propagation (EuCAP), pages 1–5, 2015.

184

[158] A. ÇaÇğri Ulusoy, Mehmet Kaynak, Václav Valenta, Bernd Tillack, and Hermann Schu-
macher. A 110 GHz LNA with 20dB gain and 4dB noise figure in an 0.13 µm SiGe BiCMOS
technology. In IEEE International Microwave Symposium Digest, pages 23–25, 2013.

[159] A. Leuther, A. Tessmann, I. Kallfass, R. Lösch, M. Seelmann-Eggebert, N. Wadefalk,
F. Schafer, J. D.Gallego Puyol, M. Schlechtweg, M. Mikulla, and O. Ambacher. Metamorphic
HEMT technology for low-noise applications. nternational Conference on Indium Phosphide
and Related Materials, (V):188–191, 2009.

[160] M. E. Tiuri. Radio Astronomy Receivers. IEEE Transactions on Antennas and Propagation,

AP-12(7):930–938, 1964.

[161] Michael Chang and Gabriel M. Rebeiz. A wideband high-efficiency 79-97 GHz SiGe linear
power amplifier with > 90 mW output. In IEEE Bipolar/BiCMOS Circuits and Technology
Meeting, pages 69–72, 2008.

[162] James M. Schellenberg. A 2-W W-Band GaN Traveling-Wave Amplifier With 25-GHz
Bandwidth. IEEE Transactions on Microwave Theory and Techniques, 63(9):2833–2840,
2015.

[163] Michael Craton, Mohd Ifwat, Mohd Ghazali, Brian Wright, Kyoung Youl Park, Premjeet
Chahal, and John Papapolymerou. 3D Printed Integrated Microfluidic Cooling for High
Power RF Applications. In International Microelectronics Assembly and Packaging Society,
2017.

185