A CMOCS-Comnatihle
C omnct T)Dignlav
....
"
.............
by
MAssACHUSS
OF TECh'Nt )OGY
Andrew R. Chen
OCT 21
Bachelor of Science in Electrical Science and Engineering
Massachusetts Institute ot 'lechnology, June 1990
~
,
. .
~
.
..
r
ran
,r
t
T~
r
AN~
2005
LIBRARIES
/-1
Master of Engineering in Electrical Engineering and Computer Science
Massachusetts Institute of Technology, June 1997
Submitted to the Department of Electrical Engineering and Computer Science in Partial
Fulfillment of the Requirements for the Degree of
Doctor of Philosophy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June, 2005
MASSACHUSETTS INST1ITUTE
OF TECHNOLOGY
© 2005 Massachusetts Institute of Technology
OCTLIBRARIES2005
All Rights Reserved
LIBRARIES
Signature of Author ...................
, ..... Y.
Department of Electrical Engineering and Computer Science
April 27, 2005
Certified
by..............
..............
.. ... , .....................
Hae-Seung Lee
Professor of Electrical Engineering
/
Certified by ..............................
Accepted
Acceptedby.(y,-~C.i'
by ...........
..
.. ....
[ /if
ThesisSupervisor
.........
..............
~~/ ~
Akintunde I. Akinwande
Professor of Electrical Engineering
Thesis Supervisor
.L
....................
-:
..--.
Arthur C. Smith
Chairman, Committee on Graduate Students
Department of Electrical Engineering and Computer Science
A CMOS-Compatible Compact Display
by
Andrew R. Chen
Submitted to the Department of Electrical Engineering and Computer Science
on April 27, 2005, in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Electrical Engineering and Computer Science
Abstract
Portable information devices demand displays with high resolution and high image
quality that are increasingly compact and energy-efficient. Microdisplays consisting of a
silicon CMOS backplane integrated with light generating or modifying devices, are being
developed for direct-view and projection applications.
A microdisplay architecture using silicon light emitters and image intensification suitable for a micro-projector application is developed. A standard low-voltage CMOS IC
incorporating display drivers and an array of avalanche diodes produces a faint optical
image, and an image intensifier efficiently amplifies the image to useful brightness. This
architecture has high efficiency and the potential to achieve adequate luminance for pro-
jection applications. A proof-of-concept system with 16x32 arrays is implemented and
evaluated.
A high-performance silicon backplane for the above system is designed, implemented,
and evaluated. The backplane is a standard CMOS die including a 360x200 pixel array
with silicon light emitters, and 10b precision current-mode driver circuits. The driver cir-
cuits can support a number of emissive display technologies including silicon light emitters
and organic light emitting diode (OLED). They employ a self-calibration technique based
on the current copier circuit to minimize variation and fixed-pattern noise while reducing
circuit area by a factor of five to seven compared to a conventional solution. A circuit
technique to improve the retention time of dynamic analog memories is also presented.
This technique allows a dynamic analog memory to retain 10b precision for 500ms at room
temperature.
Thesis Supervisor: Hae-Seung Lee
Title: Professor of Electrical Engineering
Thesis Supervisor: Akintunde I. (Tayo) Akinwande
Title: Professor of Electrical Engineering
4
Acknowledgements
FIRST
I would like to thank my co-advisors Harry Lee and Tayo Akinwande for their
support and thoughtful guidance throughout my doctoral program. They have helped
me with matters large and small, from technical answers to publishing and professional
planning. Their straightforward style and pursuit of quality results are a source of inspiration for many students. I would also like to thank Tom Knight for reading my thesis and
providing insights into the design of microdisplays.
It has been a privilege to be part of the Lee-Sodini lab, and I am thankful to my friends
and labmates for the discussions, help, and fun times we have shared. Iliana Fujimori
showed great patience and teaching skills not only as a co-teaching assistant, but also by
introducing me to imagers, and sharing test chips with me for making initial measurements.
I've learned a lot from Todd Sepke, Anh Pham, and Albert Jerng with whom I've shared
a cubicle for the last two years. Mark Spaeth provided a much wisdom in printed circuit
board design, and arcade games. Pablo Acosta, Ayman Shabra, Mark Peng, Don Hitko,
and Dan McMahill all had answers to my persistent questions as I started my research.
I am thankful to Duane Boning for introducing me to the intricacies of process variation
in semiconductor manufacturing.
and system design.
This basic knowledge is valuable at all levels of circuit
Thanks to Professor Vladimir Bulovic and Yakov Tischler for help in making optical
measurements, and for advice relating to organic LED technology. John Kmyssis and
Annie Wang provided help with electro-optical measurements and data analysis.
I am grateful for the support of family and
encouraged me and helped me to grow. She has
hours. Thanks to my friends for sharing these
scattered around the country, we are still close.
years, and I will always love them.
friends. My fiancee Lucy has consistently
also put up with my graduate student work
years together. Although many of us are
My parents have cared for me for all these
Integrated circuit fabrication was provided by National Semiconductor. Peter Holloway,
Matt Courcy, Mike Guidry, Sangamesh Buddhiraju, and the physical design staff all contributed valuable advice and assistance.
This research was funded by the MARCO Focus Center for Circuit & System Solutions
(C2S2, www.c2s2.org), under contract 2003-CT-888.
6
Biography
ANDREW
R. Chen received the Bachelor of Science and Master of Engineering degrees
in electrical engineering from the Massachusetts Institute of Technology, Cambridge,
in 1996 and 1997, respectively.
From 1997 to 1999 he was a design engineer at Intel
Corporation, Santa Clara CA, designing circuits for Intel Architecture 32b and 64b microprocessors. He is currently a Ph.D. candidate at the Massachusetts Institute of Technology
studying analog and mixed-signal integrated circuit design.
His research interests include analog and mixed-signal integrated circuit design, imaging systems, and electronic display systems.
8
Contents
1
Introduction to Displays
1.1 Overview.
1.2 Electronic Display Applications and Technologies
1.3 Brightness and Efficiency .............
. . .
1.4 Microdisplays ..............
.
1.5 Micro-projector .............
.
2
3
Background
2.1 Silicon Light Emission ..........
2.1.1 Previous Work and Applications .
2.1.2 Avalanche Breakdown ......
2.2 Device Measurements ...........
2.2.1 0.8 /tm Technology ........
2.2.2 0.35/im Technology .......
2.2.3 0.18tm Technology.
2.2.4 Measurement Summary .....
2.3 Image Intensification ...........
2.4 Circuit Design for Displays ........
2.4.1 MOS Transistors .........
2.4.2 DAC Fundamentals ........
2.4.3 Calibration Techniques ......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
17
. . 17
. . 18
. . 20
. . 22
. . 23
25
... . 25
. . . . 25
.
.
.
.
26
28
28
30
..................
..................
..................
..................
..................
33
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
...
...
...
31
....
Micro-projector Design
3.1
3.4
3.3.6
3.3.7
Results
37
37
40
41
45
Design Overview .............
3.1.1 Background: Two-stage displays .
3.2 Micro-projector Efficacy .........
3.3 Test Chip Implementation .........
3.3.1 IC overview ............
3.3.2 Pixel ...............
3.3.3
3.3.4
3.3.5
...
...
...
.
.
.
.
.
.
Pixel Arrays .............
.
Top-level IC Design and Packaging
.
System Design
..
. . . . . . .
.
Image Intensifier and Optics .....
.
Proof-of-Concept System Performance
.
. . . . . . .
. . . . . . . . . . . . . . . . .
9
.
·
.
.
.
.
.
.
.
.
.
. .
...
...
...
...
...
...
...
...
...
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
45
46
47
50
51
51
52
53
54
55
56
. 57
...
CONTENTS
10
3.4.1
3.4.2
3.4.3
3.4.4
4
Single Pixel ..............................
Arrays.
................................
Power Measurements .........................
Discussic)n ...................
...........
Backplane IC Design
4.1 Pixel Array.
4.2 Row Drivers ...........
4.3
Column Drivers and Calibration
4.3.1 Overview........
5
6
4.4
4.3.2 Area Comparison ....
4.3.3 Current Copier .....
Biasing .............
4.5
4.6
RefDAC .............
ArrayDAC ............
4.7
Design for Testability ......
4.8
Top-level IC ...........
4.9
System Board and Optics . . . .
Measured results
5.1 Electrical Measurements
5.1.1
RefDAC Linearity
5.1.2
ArrayDAC ....
5.2
5.1.3 Column Variation .
Optical Measurements
..
5.3
5.4
Power Measurements . . .
Sample Images ......
Summary
Topics for Future Investigation
A Precision Oscilloscope Measurements
58
59
60
63
.
.
.
.
.
.
.
.
.
.
.
.
.................
.................
.................
.................
.................
.................
.................
.................
.................
.................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
63
67
68
68
69
70
75
79
83
86
88
89
95
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . .
. . .
95
95
......
101
......
......
......
......
103
105
108
108
Discussion
6.1
6.2
57
111
. . . . . . . . . . . . . . . . . .
.. 111
............ ........ 112
115
List of Figures
.
1.1 Applications of electronic displays .............
1.2
Flat-panel
1.3
Micro-projector
. . . . . . . . . . . . . . .
display technologies
I-V characteristic
......
20
. . . . . . . . . ....
concept (Fraunhofer-Gesellschaft)
breakdown
19
24
2.1
Avalanche
2.2
Cross-section of twin-well CMOS process
2.3
Diode I-V characteristic,
2.4
2.5
2.6
PMOS test device ...............................
I-V curve for p+/nwell junction in 0.35/um CMOS technology .......
I-V curve for n+/psub junction in 0.35tim CMOS technology . . . . . ...
2.7
Light emission
2.8
I-V curves for two p+/nwell junctions in 0.18um CMOS technology . . . .
0 . 8 /m
. . . . . . . . . . . . . . .....
process
27
..................
.
from six parallel test devices
27
.................
.
29
29
30
30
. . . . . . . . . . . . .....
31
32
2.9 Spectraof light emissionfromjunctionson CMOSdie . . . . . . . . . . . 32
.
2.10 Light output vs. current for junction in 0.18,um CMOS process .....
2.11
. .
.
. .
Intensifier gain vs. number of MCPs . . . . . . . . . . . . . . .
Intensifier resolution vs. number of MCPs . . . . . . . . . . . . . .....
.
Simple image intensifier
. . . . . . . . . . . . . . .
application
.......................
2.12 Microchannel plate (Proxitronic)
2.13
2.14
32
2.16 Gate current for an NMOS device .
34
.
2.15 GaAs Photocathode Sensitivity (Hamamatsu)
34
35
35
...............
36
.....................
38
2.17 Sample and Hold Circuit models. Left: ideal, Right: realistic
........
39
2.18
Current mirror measurements
. . . . . . . . . . . . . . . .
.
......
40
3.1
CMOS IC and image intensifier
. . . . . . . . . . . . . .
.
. .
3.2
Two-stage display architecture for projection application (JVC) .......
46
3.3
Circuit schematic of a single pixel
52
3.4
Block diagram
3.5
IC photomicrograph
3.6
3.7
..
Proof-of-concept system board
Proof-of-concept system with intensifier
3.8
Pixel I-V characteristic, VHI sweep
3.9
Test image of IC under microscope
of pixel array
.
..
.....................
. . . . . . . . . . . . . . .
53
. ......
54
..............................
.
54
56
......................
..................
58
......................
.
45
. . . . . . . . . . . . .
.
58
3.10 32-level grayscale gradient. Bottom two rows are at full intensity ......
59
3.11 Output window of intensifier (20mm) with sample image . . . . . . . ...
60
11
LIST OFFIGURES
12
4.1
4.2
4.3
Overview of CMOS backplane .....
.
.
.
Pixel schematic .............
.
4.4 OLED material layers ..........
.
4.5 OLED pixel schematic .........
.
4.6 Row driver circuit . . . . . . . .. . .
.
4.7 Rnw drive.rtiming
.
4.8 Two-stage calibration technique ........
. . .
.
4.9 Column driver calibration overview ..
4.10 Simple MOS current source ...........
.
4.11 Simple current copier ..............
.
4.12 Reduced transconductance current copier . . .
.
4.13 Improved current copier ..
.
. . . . . .. . . .
4.14 Passgate leakage scenarios: (a) conventional, (b) voltage
Array floorplan
a.e,'.,
4.15
4.16
4.17
4.18
.a·'v ~ s
.A
.............
e,
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Conventional (left) and improved (right) passgate
RefDAC reference current generator and switch
Bleeder bias voltage generator.
VBIAS circuit ................
4.19 Testability control circuit ..........
4.20 Testability scan chain ............
4.21 RefDAC block diagram ...........
4.22 RefDAC current copier circuit .. . . . . .
4.23 RefDAC row driver .............
4.24 RefDAC column driver ...........
4.25 RefDAC cell pattern .............
4.26 RefDAC copier cell settling time simulation
4.27 Retention time improvement circuit . . .
4.28 ArrayDAC copier schematic ........
4.29 ArrayDAC block diagram ..........
4.30 ArrayDAC registers .............
.
4.31 ArrayDAC data load control.
4.32
4.33
4.34
4.35
4.36
4.37
4.38
4.39
4.40
ArrayDAC calibration control .......
ArrayDAC calibration control, improved . .
5.1
Synchronization circuit ........................
5.2
5.3
RefDAC bit currents, first attempt ..................
RefDAC bit currents, second attempt .................
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
control
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
64
64
65
66
66
67
67
68
69
70
71
72
73
74
...
...
...
...
...
...
...
...
...
...
...
...
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
76
77
78
78
78
79
80
80
80
81
82
82
83
84
85
85
85
86
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
:s ..............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
74
Micrograph showing Si-LED pixels, OLED pixel s, and ArrayDACs . . . . 87
Column transimpedance amplifier .....
Die micrograph ...............
High-level program flowchart ........
Clock oscillator ...............
Timing diagram - simple program .....
Timing diagram - optimized ........
4.41 Assembled system board ..........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
88
89
92
93
93
94
94
95
96
97
...
...
...
...
...
...
...
LIST OFFIGURES
5.4
5.5
5.6
5.7
5.8
5.9
5.10
5.11
5.12
5.13
5.14
5.15
5.16
5.17
5.18
13
RefDAC INL, second attempt ..........
RefDAC bit currents, final ...........
RefDAC INL, final ...............
RefDAC retention test ..............
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . . . . .
RefDAC retention test, zoomed in . . . . . . .
Transimpedance
amplifier circuit . . . . . . . .
...............
100
ArrayDAC INL .................
ArrayDAC retention test ............
ArrayDAC output current distribution .....
ArrayDAC outputs before and after calibration .
Sample image for pixel luminance analysis . .
Pixel luminance variation by column ......
Column luminance ...............
...............
102
...............
...............
104
105
...............
...............
98
98
99
100
101
.........
......104
103
...............
......... ...... 106
......... ...... 107
Silicon backplane IC emitting light, showing a se,ries of ramps .......
Proof-of-concept system with calibrated backplai ne .............
109
109
5.19 Proof-of-concept system with calibrated backplane and alternative optics . .110
6.1
RefDAC, ArrayDAC, and pixel arrangement.
6.2
Alternative
RefDAC, ArrayDAC,
and pixel arrangement
.116
.116
.117
time . . . . . . . . . . . . . . . . . . . . . . . . . . .117
.118
step response . . . . . . . . . . . . . . . . . . .
A.1 Step responses of four oscilloscope probes.
A.2 Probe settling behavior ...........................
A.3 Four oscilloscope channels, same input ...................
A.4
A.5
Longer observation
Newer oscilloscope
.112
. . . . . . . . . .113
...
...
14
LIST OF FIGURES
List of Tables
1.1
Typical illumination
levels
1.2
1.3
Projector illuminance by application .
Efficiency of light sources .......
Intensifier Performance with two MCPs
3.2 0.18/um CMOS technology ......
3.3 Microcontroller interface definition . . .
3.1
4.1
4.2
4.3
4.4
4.5
PreRef switch truth table ........
ArrayDAC current mirror biasing .
IC power supplies ............
Input pins ................
Output pins
. 20
. . . . . . .
. . . . . . . . . . . . . . .
15
. . . . .
. 21
. 22
...................
...................
...................
...................
...................
...................
...................
...................
49
50
55
76
84
90
91
92
16
LIST OF TABLES
Chapter 1
Introduction to Displays
1.1
Overview
This thesis presents work in electronic display design and integrated circuit design. Electronic systems are becoming increasingly ubiquitous, and displays for a wide spectrum of
applications are being developed. An area of special interest is that of personal portable
electronics, where all system components must be designed for high levels of integration,
compact size, and high energy efficiency. The design of a "micro-projector" for portable
electronic systems is presented. The micro-projector produces an image with adequate
luminance for projection with high efficiency for battery-powered systems.
We propose a micro-projector based on a combination of silicon light emission and
image intensification. A standard low-voltage CMOS IC incorporating display drivers and
an array of avalanche diodes produces a faint optical image. The image is coupled into an
image intensifier that amplifies the image to useful luminance levels. This architecture has
high efficiency and the potential to produce adequate luminance for projection applications.
The advantages of this architecture are compatibility with mainstream CMOS processes,
high efficiency,and potential for projection. Junction diodes which are available in all standard CMOS processes are used to emit light. Process modifications which are often used
in imagers [26] and microdisplay ICs are not necessary. These junctions can be integrated
with high-performance circuits such as calibrated display drivers. The integrated display
can also be integrated with other circuit blocks to achieve a high level of system integration
on a single IC.
The image intensifier is a vacuum device capable of high gain. It uses the mechanism
of cathodoluminescence
to efficiently generate light, and can produce adequate luminance
for projection applications by using the same principles as projection CRT tubes.
A high-performance silicon backplane for the above system is designed and evaluated.
17
CHAPTER 1. INTRODUCTION TO DISPLAYS
18
The backplane is an IC produced in a standard CMOS technology. It includes a 360x200
pixel array with silicon light emitters and 10b precision current-mode driver circuits. The
driver circuits can support a number of emissive display technologies including silicon light
emitters and organic light emitting diode (OLED). They employ a self-calibration technique
to minimize variation and fixed-pattern noise while reducing circuit area by a factor of five
to seven compared to a conventional solution. A circuit technique to improve the retention
time of dynamic analog memories is also presented.
The organization of this document is as follows. The remainder of this chapter is an
overview of electronic display technologies and applications. Chapter 2 provides background on silicon light emission, operation of the image intensifier, and circuit design
concepts relating to display drivers. Chapter 3 presents the design, implementation,
and
results of a proof-of-concept display system using two-stage image generation. Chapter 4
presents the design of a high-performance CMOS backplane for the previously described
microdisplay architecture. Chapter 5 describes characterization and measurements of the
system with calibrated backplane. Finally, chapter 6 summarizes and suggests topics for
future investigation.
1.2 Electronic Display Applications and Technologies
An electronic display is a device that converts time-sequential electrical signals into spatially and temporally configured light signals (images) useful to the viewer [1]. Electronic
displays are used in a broad variety of applications, some of which are shown in Figure 1.1.
The variety of applications results in a large number of technologies currently in use
[2, 3]. The cathode-ray tube (CRT) has been a workhorse for many years. Newer displays
are based on flat-panel technologies which have the advantages of smaller form factor,
lighter weight, and the ability to achieve larger sizes than CRTs.
Flat panel display technologies can be divided into categories as shown in Figure 1.2.
Displays that generate light in response to electrical signals are called emissive displays,
and ones that modulate light are called light valve displays. Examples of mechanisms
used in emissive displays are electron-hole recombination in semiconductors, ultraviolet
radiation exciting a phosphor, or an electron beam exciting a phosphor. Light valves are
examples of a broader class of devices known as spatial light modulators (SLM). These
create spatial modulation on a plane wave of light by various means, for example by reflection or change in polarization. SLMs require an external source of light which may be a
lamp or ambient illumination. The most widely used SLM is the liquid-crystal display. It
is versatile and inexpensive, but transmission is typically low and the operation at extreme
1.2. ELECTRONICDISPLAY APPLICATIONSAND TECHNOLOGIES
19
Pixels
10M
IM
lOOK
10K
1K
0.3"
lin
3in
10in
30in
l00in diagonal
Figure 1.1: Applications of electronic displays
temperatures is difficult.
Emissive displays potentially have higher efficiency than SLMs because they generate light only when and were it is needed.
In contrast, SLM's require a light source to
constantly produce the maximum display luminance, and dissipate unwanted light as heat.
Emissive and transmissive displays work well under moderate illumination, but have difficulty producing very high luminance for applications in sunlight. Reflective SLMs work
well under high illuminance but have reduced energy efficiency under low and moderate
illumination levels where they require an external light source.
Projection systems have the advantage of producing a larger image than than the sys-
tem form factor. Two applications of projectors are digital cinema and portable information
systems. Digital cinema is an application where very large video-rate images are desirable.
Flat-panel displays are available in sizes up to 40 inches, but a projector is lighter, poten-
tially cheaper, and easier to scale to larger sizes. Portable electronic information systems
that are becoming smaller and handling increasing quantities of still and video images.
In this application the goal is to produce page-sized images for personal viewing while
keeping a pocket-sized form factor.
The focus of this work is a micro-projector intended personal information systems.
These are compact electronic systems that will include communication, computing, multimedia manipulation, and data storage. They have been described in literature with various
CHAPTER 1. INTRODUCTION TO DISPLAYS
20
Flat panel displays
Emissive
semiconductor: OLED
UV+phosphor: Plasma
electron beam + phosphor:
Cathodoluminescence
Modulator
polarization: Liquid crystal
reflection: micromirror
transmission: electrowetting
Figure 1.2: Flat-panel display technologies
Conditions
Direct sunlight
Open shade
Overcast day
Twilight
Full moon
Indoors, exacting visual tasks
Indoors, fine assembly work
Indoors, office
Indoors, home
Illuminance, lux
100,000
10,000
1,000
1 to 10
0.1
15,000
1,000
500
100-200
Table 1.1: Typical illumination levels
terms, such as "information bank system" [4]. The constraints of these systems are compact size, high energy efficiency to maximize battery life, high resolution, high bit depth,
and video speed.
1.3 Brightness and Efficiency
This section describes typical illumination levels of various environments, and the required
luminance or illuminance levels for displays. Typical levels of illumination are listed in
Table 1.1. The ratio of illumination varies over a 1000:1 ratio from full sunlight to in-
doors at night. This results in a variety of displays optimized for different ambient lighting
conditions.
The luminance of an emissive display for indoor use such as a television or computer
1.3. BRIGHTNESS AND EFFICIENCY
Application
Movie theater
21
Illuminance (lux) Comments
full control of ambient light
50-80
Office
120-150
Ballroom (dimmed)
Training room (dimmed)
Training room (full light)
150-200
200
400
high background illumination
Table 1.2: Projector illuminance by application
monitor is 25-100 footlamberts. Emissive microdisplays used in head-mounted display systems (HMDs) require much higher luminances because optical transmission (throughput)
is around 15%. Typical luminance is 1000fL at the display output, which appears as 150fL
to the eye.
The required luminance of a projector depends on a number of factors. The contrast ra-
tio determines the ratio between maximum projector illuminance and ambient illuminance.
The output of the projector is spread over the image area, leading to a tradeoff between
illuminance and image size. The properties of the surface where the image is projected
also affect illuminance. An ordinary white surface has a gain of unity. Projection screens
with non-uniform reflection patterns can increase gain in a preferred direction, at the cost
of limited viewing angle and contrast. The illuminance of a projector is:
luminance(lumens)
2
ae(m
area(m2)
* gain = illuminance(lux)
(1.1)
where the luminous output of the projector is in lumens, the area of the projected image is
in square meters, the gain is unitless, and illuminance is in lux. Required illuminance by
application is shown in Table 1.2.
A variety of light-emitting technologies are used today. Table 1.3 compares some com-
monly used ones. Metal halide bulbs are attractive for projectors because of their high
efficiency, high luminous output, and near point-source emission. Organic and inorganic
LEDs are making rapid progress in efficiency, brightness, and the variety of output spectra
available. Phosphor-based light emitters include fluorescent lamps and cathodoluminescent
devices. In these devices, a phosphor converts energy from ultraviolet light or an electron
flux into visible light. Fluorescent lamps are efficient, but because of their large sizes they
can only be used as backlights for modulator-based displays which have low efficiency due
to transmission losses. The cathode ray tube is an example of a cathodoluminescent dis-
play. These can produce extremely high luminance levels, as illustrated by liquid-cooled
tubes used in projection televisions.
CHAPTER 1. INTRODUCTION TO DISPLAYS
22
Efficiency (lumens/watt) Notes
Technology
Metal halide (HID)
50-100
10-400W
T8 Fluorescent
Compact fluorescent
80
30-60
4-foot, 32W
Cathodoluminescence
OLED
48
10-30+
[5]
White inorganic LED
32+
Halogen
12-20
10-100W
Incandescent
6-17
7-100W
Table 1.3: Efficiency of light sources
1.4
Microdisplays
Microdisplays produce high-resolution images typically less than one inch on a side. They
are commonly used in both the smallest and the largest display applications. Typical nearto-eye devices are head-mounted displays [6] and video camera viewfinders. In these devices, optics provide magnification for comfortable viewing. Projection applications include high-definition televisions and computer projectors. In these systems, the microdisplay modulates very high luminance levels from an external light source. Microdisplay
technologies are also being applied to novel applications such as three-dimensional
dis-
plays [7], switching of optical communication signals [8], high-end photo printers [9], and
optical tweezers [10].
An emerging opportunity for microdisplays is the micro-projector. The micro-projector
is part of a pocket-sized personal information device. It projects a high-resolution image
approximately the size of a piece of paper (8in = 20cm) for personal viewing. Its main
advantage is the ability to produce images of reasonable size with a very small form factor.
The micro-projector must produce high luminance, high resolution images while being
compact and highly energy efficient. Emissive microdisplays can be designed to have these
characteristics.
Presently the leading technologies for microdisplays are liquid crystal on silicon (LCOS)
[11], micromirrors, and organic LED. These technologies can be constructed on top of a
silicon integrated circuit with additional processing. Using a silicon backplane provides
access to high-performance circuits at the pixel or array level, and alleviates the problem
of large numbers of interconnect wires between the pixel array and driver circuits.
Liquid crystal displays use an organic material to modify the polarization of light. Pixels are modelled as capacitive elements and driven with AC voltage waveforms. A variety
of circuit techniques for driver circuits have been investigated [12, 13]. Color is usually
1.5. MICRO-PROJECTOR
23
achieved by using patterned color filters and sub-pixels for red, green, and blue, at each
pixel location.
Micromirrors are an array of mirrors on top of a silicon chip that are controlled electrically. At each pixel location, a mirror reflects light to either the output or a heat sink [14].
Color output is achieved in frame-sequential mode using a rotating color filter wheel.
Organic LED is an emissive technology that promises high image quality and reasonable manufacturing cost. OLED microdisplays with silicon backplanes have been demon-
strated [15, 16]. They promise excellent image quality and potential power savings because
of their emissive nature. OLED performance has been steadily improving, and packaging
and sealing [17] have also improved in recent years. Color can be produced either by filter-
ing of a white image (similar to the case of liquid crystal), or by stacking layers of different
light-emitting materials to produce different colors.
1.5
Micro-projector
A micro-projector would allow a pocket-sized electronic system to display high information content images and video at a reasonable size for comfortable viewing. Because a
projector produces images larger than its own form factor, the system can be pocket-sized
while producing page-sized images. The cost of computation has decreased to a point
where the display, user interface, and battery are primary bottlenecks in portable system
design. A compact display capable of presenting high resolution images and video enables
new uses for portable electronic systems. Figure 1.3 is an illustration of this concept [18].
This contrasts with the flat-panel display approach where the system must be larger than
the image size, or the display must be designed to fold or roll-up when not in use.
Compact projector systems have been demonstrated and are described in the literature.
Keuper [19] describes a projector consisting of a white LED light source, liquid-crystal
modulator, and lens. Projectors using lasers, either a raster-scanned single beam or multiple
beams in parallel, have also been demonstrated [20, 21, 22]. For near-to-eye applications,
a laser beam can be modulated and swept onto the retina to produce an image [23, 24, 25].
Laser-based projectors do not require refractive optics for focusing, which may permit
them to be more compact than ones using lenses. They also have good energy efficiency.
The disadvantages of these systems are (1) they require precision mechanical systems to
position the laser beam, (2) they must address the problem of image flicker because they
operate at low duty cycles, and (3) safety concerns if the beam is pointed at the eye, or if
higher power is used to produce high luminance images.
24
CHAPTER 1. INTRODUCTION TO DISPLAYS
Figure 1.3: Micro-projectorconcept (Fraunhofer-Gesellschaft)
Chapter 2
Background
This chapter presents previous work and preliminary studies on components of our microprojector design: silicon light emitters, image intensifiers, and calibrated driver circuits.
An introduction to light emission from silicon is presented along with measurements of
light emitters in standard CMOS processes. The image intensifier and its basic concepts
are introduced.
Display driver circuits are digital-to-analog converters (DACs). An overview of DACs
is presented along with circuit and device considerations that influence their design. A
major concern in data converter design is the management of variation that arises from the
manufacturing process. A circuit called the current copier can produce precise currents
while occupying less area than a conventional solution. The current copier is introduced
here, and its implementation is discussed in depth in Chapter 4.
2.1
Silicon Light Emission
2.1.1
Previous Work and Applications
The emission of light from silicon p-n junctions was first reported in 1955 [27, 28]. Since
then, it has been used for a variety of purposes. Silicon light emitters were used in a data
storage system to produce dots on photographic film [29]. Light emission from the channels
of MOSFET transistors has been used for debugging and fault analysis of microprocessors
[30]. The idea of using integrating light emitters and detectors on the same chip dates back
to at least 1965 [31], and the application of on-chip optical interconnect is receiving atten-
tion as metal wires have become a bottleneck in the design of high performance integrated
circuits [32].
Silicon is an indirect bandgap material, therefore when electrons and holes recombine
25
CHAPTER 2. BACKGROUND
26
the probability of a photon being emitted is small. This is in contrast to the III-V materials
used commercially in LEDs that are efficient photon sources.
In forward bias, p-n junctions emit photons with energy equal to the bandgap energy. In
silicon this is 1.12eV, corresponding to photons with a wavelength of 1100nm. In avalanche
breakdown, hot electrons can cause photons with higher energies to be emitted. Lightemitting devices in standard CMOS technologies have been studied [33, 34], and the best
reported quantum efficiency is 5 *
10
- 7
phot/elec [35].
Efficiently generating light from a silicon substrate is an area of research interest [36,
37]. Efficiency can be improved by various dopants or altering the shape or material properties of silicon devices [38, 39]. These advances may eventually be integrated into mainstream manufacturing processes. Another solution is the integration of efficient optoelectronic devices on a standard silicon substrate [40]. Novel device and circuit designs for
light emitters are being explored, for example a structure to lower the breakdown voltage
of silicon light emitters [41].
2.1.2
Avalanche Breakdown
This section explores the properties of silicon junctions in avalanche breakdown used as
light emitters. Light emission from avalanche breakdown produces photons with larger
energies than the bandgap energy. Emission can be in the visible or near-infrared regions
of the spectrum.
When a junction diode is biased with a large negative voltage, the electric field across
the junction is very large. When a conduction band electron enters the field, or is created
by thermal processes, collisions with valence electrons can move them into the conduction
band. A cascading or multiplying effect occurs, and the current grows dramatically over
a small voltage range. The I-V characteristic of a silicon p+/nwell junction in avalanche
breakdown has the shape shown in Figure 2.1. Current through the junction grows until limited by external components such as series resistance or a current source circuit. Operating
a junction diode in reverse breakdown is nondestructive if the current through the junction
is limited. Spinelli and Lacaita [43] provide a more detailed description of avalanche diode
physics.
This abrupt characteristic allows the use of low-voltage control circuits. A large DC
bias is needed to make the junction enter avalanche breakdown, but once avalanche breakdown occurs the change in voltage required to modulate the current is very small. Consider
a junction with the characteristic in Figure 2.1. The cathode is connected to a +12 volt supply, and the anode to a "leakage "current sink of lnA in parallel with an adjustable current
2.1. SILICON LIGHT EMISSION
nited
le-3
le-4
avalanche
le-5
breakdown
~le-6
•
le-7
le-8
le-9
leakage
le-10
-11
-10
volts
-9
-8
Figure 2.1: Avalanche breakdown I-V characteristic
NMOS
PMOS
~
~
~///////////l/~n///~//1.
~/////////N/////r/////N///
a~dri
~
'/////I//NN//N///~////,
~~
Figure 2.2: Cross-section of twin-well CMOS process
sink. The voltage at the cathode will be 12-10.5=1.5V for currents up to 100PA, producing
a 105 dynamic range which is more than adequate for display applications.
In a twin-well CMOS process, as shown in Figure 2.2, the p+/nwell junction is the best
suited for integration. It is electrically isolated from other devices on chip. When the nwell is connected to a high voltage and the p+ region connected to a low voltage, avalanche
breakdown occurs. No substrate currents are created by this structure, and the n-well voltage can be safely controlled by MOS devices. The p+/nwell junction breakdown voltage
is designed to be much higher than the nominal power supply voltage, to ensure proper
operation of CMOS circuits. This reduces the energy efficiency of light emitting junctions,
which should be designed for avalanche breakdown at the smallest possible voltage.
The p+/nwell junction is available in all standard CMOS processes, and low-voltage
CHAPTER 2. BACKGROUND
28
devices can be used to drive them. This eliminates the need for more costly high-voltage
processes. In addition, other circuit functions that require high-performance devices such
as processors or memory can be integrated together with light emitters.
In unmodified CMOS processes, avalanche breakdown occurs first around the perimeter
of the junction area. This is due to a combination of surface states and the high electric
fields at the edges of the junction area. For a microdisplay using front-emission from a
silicon IC, lateral breakdown is desirable over vertical breakdown because it occurs closer
to the surface and allows more photons to escape. Lightly doped guard rings can be used
to modify the doping profile of avalanche diodes if bottom-surface breakdown is preferred,
for example for back-emission from a thin substrate [44].
Avalanche diodes are more commonly used as in sensitive imaging devices for photon
counting and operation at low light levels, and many articles have been published on their
design in that context [42]. An important design consideration is to avoid covering light-
emitting junctions with silicide. Most CMOS processes include a mask to selectively block
deposition of silicide, and it should be utilized for light-emitting junction areas. Silicides
are deposited over diffusion regions to reduce their sheet resistance. They are nearly opaque
to light and reduce optical transmission by 80-90% [45].
2.2
Device Measurements
A preliminary study was performed to characterize the light-emitting properties and reverse-
breakdown characteristics of p-n junctions manufactured in commercial CMOS processes.
Limited information and data on the use of these junctions as light emitters is available,
therefore we characterized devices manufactured in commercial 0.8um , 0.35um , and
0. 18pm CMOS processes. In all three processes, the p+/nwell junction showed avalanche
breakdown behavior when a large reverse bias was applied.
2.2.1
0.8pum Technology
A test device in the Hewlett Packard 0.8[tm (CMOS26G) process is measured. The process
has n-wells, one poly layer, and three metal layers. Nominal operating voltage is 5V. The
test device is the junction between the a source/drain diffusion of a PMOS transistor and
its n-well.
Junction area is 50 x 1.5um , and the device is covered with silicide. The
electrical characteristics of this device in reverse bias are shown in Figure 2.3. Reverse
leakage current rises gradually between 0 and -18V. At -18V, avalanche breakdown occurs
and current increases by over six orders of magnitude over a very small voltage range until
2.2. DEVICE MEASUREMENTS
Id vs Vd N76VAF
-20
-15
-10
-5
0
Diodevoltage (V)
Figure 2.3: Diode I-V characteristic,0.8upm process
Figure 2.4: PMOS test device
it is limited by series resistance.
The pictures in Figure 2.4 were created with a CCD camera mounted on a microscope. On the left, the microscope light is on and the four-terminal test device is visible.
The device is connected to four metal pads for use with a probe station. The n-well and
source/drain pads were bonded to package pins for easier testing, and solder from the bonding process is visible on the two pads on the right side. On the right, the microscope light
and room illumination were turned off and a long-exposure image shows light emission
from the p-n junction. Current through the junction is 130PA, and light emission is concentrated at the sides and corners of the p+ diffusion area. Light emission was visible in
a darkened room, in agreement with published reports of broadband emission from silicon
junctions in avalanche breakdown.
CHAPTER 2. BACKGROUND
30
Chipl1 - d2626Rverse
03111102achen
1.OE-02-
- -----------
1.0E-03
1.0E-04
i
1.0E-05
X
1.0E-07
.0E6oE
.
,
1.OE-08
1.OE-091.OE-10
-11
-10
-9
-7
-8
-6
Vd (volts)
Figure 2.5: I-V curve for p+/nwell junction in 0.35p1mCMOS technology
Chip #1 d2726
03/11/02 achen
0.010.0075 0.005
0.0025
o
-
0
-0.0025
-1.5
-0.5
0
0.5
-0.005
-0.0075
-0.01 .....
Vd (volts)
Figure 2.6: I-V curve for n+/psub junction in 0 . 35 p1mCMOS technology
2.2.2
0.35pm Technology
The characteristics of the p+/nwell junction in the TSMC 0.35/um CMOS technology were
measured. The process has n-wells, one poly layer, four metal layers, and silicide block.
Nominal operating voltage is 3.3V. Avalanche breakdown of the p+/nwell junction occurs
at -10.OV as shown in Figure 2.5.
An n+/psub junction in this process was also measured, as shown in Figure 2.6. Reverse
breakdown occurs due to tunnelling instead of avalanche breakdown, and the I-V characteristic is much less abrupt. Compared to avalanche breakdown, a much larger voltage swing
is needed to produce the same change in current. This makes control with low-voltage
devices difficult or impossible. The mechanism of zener tunnelling is not expected to emit
photons, therefore this junction is not useful for light emission purposes.
2.2. DEVICE MEASUREMENTS
Figure 2.7: Light emissionfrom six paralleltest devices
2.2.3
0.18pm Technology
Devices in the TSMC 0.18prm CMOS technology were measured. This process has nwells, thin and thick oxide transistors, one poly layer, six metal layers, and silicide block.
Nominal operating voltages are 1.8V for thin oxide devices, and 3.3V for thick oxide devices. The test structure consists of six 12x14[pm junctions connected in parallel, as shown
in Figure 2.7. When this image was taken, ID = 700pLA and VD = -10.OV. Reverse
breakdown current-voltage characteristics are shown in Figure 2.8.
Emission occurs at discrete points mostly around the junction perimeter, and luminance
varies between the junction areas due to random manufacturing variation. This variation is
exacerbated by the parallel connection of pixels which allow unevenly located "hotspots"
to carry most of the total current. To improve uniformity, pixels should be small in size
and driven individually by current sources. This technique allows well-controlled transistor electrical characteristics to control light emission instead of poorly controlled junction
characteristics.
The emission spectrum for a p+/nwell junction on the 0.18p/m CMOS chip is shown
in Figure 2.9. In forward bias, the junction emitted photons at its bandgap energy (1.1eV
= 1100nm). In reverse bias, a broadband emission spanning the visible and near-infrared
wavelengths was observed. This is consistent with published data.
By comparing images of the display with a reference target, we estimated the quantum
efficiency of silicon light emitters to be - 10- 7 photon/electron. This value are comparable
to published results for light emission from silicon devices. The silicide layer was not
blocked, so the efficiency of the p-n junction without silicide should be higher.
32
CHAPTER 2. BACKGROUND
Chip3 -d3331reverseI-V
0824101 achen -100uA a -10.1V
Chip3 -d3334
08124101
achen-100uA @-10.1v
1.OE-03 4 A
. t..............................................................
._
_'
..
1.0E-04
1.0E-05
1.0OE-06
I 1.OE-07
"n4
__
o!
_Seris_|
-
1.OE-08
r
1.0E-09
.u
t
***146-
1.OE-10
I
-- I
1.OE-11
-11
-10.5
-10
Vd (volts)
i
i
1.0E-12j
-11
-9.5
-10.5
-10
-9.5
-9
Vd(volts)
Figure 2.8: I-V curves for two p+/nwell junctions in 0.18pm CMOS technology
Chip #3 d3334
forward20s -08.27.01-achen
Chip #3 d3334
reverse20s -08.27.01
35
600
30.0
10
500
, _
-,
r-4,
- f:'"
-
25
400
- -
20
-Series1
15
300
200
1010
0
100
5
800
850
900
950
1000 1050
04
1100 1150
1200
450
550
wavelength(nm)
i3s
.
.
.
650
750
850
wavelength
(nm)
Forward bias
Reverse bias
Figure 2.9: Spectra of light emission from junctions on CMOS die
PMTVoutvs. d - Chip 3 logilog
10102102
achen,Johnkym
10000
1000
0
100
10
10
01
100
-Id(mA)
Figure 2.10: Light output vs. current for junction in 0.18,um CMOS process
950
-Snes1|
2.3. IMAGE INTENSIFICATION
33
The relationship between junction current and luminance was measured in a darkened
room using a photomultiplier tube. A linear dynamic range of over 100:1 was measured, as
shown in Figure 2.10. This measurement was limited by the test apparatus, and the linear
dynamic range of the junction is expected to be larger. The linear relationship between
current and luminance suggests the use of linear current-mode driver circuits for silicon
light emitting junctions.
2.2.4
Measurement Summary
The reverse breakdown characteristics of junctions three commercial CMOS technologies
were observed. Avalanche breakdown with light emission was observed from the p+/nwell
junction in all three technologies, and breakdown voltages ranged between -18 and -10OV.
A
rough measurement of quantum efficiency agrees with published data, and the emission of
a broadband spectrum covering visible and near-infrared also agrees with published data.
The linear relationship between current and luminance suggests the use of current-mode
driver circuits.
2.3
Image Intensification
The image intensifier is a vacuum device that accepts a two-dimensional optical image and
produces an amplified version of its input [46]. It is commonly used in night vision and
scientific applications. In its simplest form it has two parts: photocathode and phosphor
screen, as shown in Figure 2.11. In this example, a lens focuses an image onto the pho-
tocathode. The photocathode is a metallic material which emits electrons in response to
light by the photoelectric effect. A two-dimensional flux of electrons is accelerated toward
the phosphor screen by a large electric field. Electrons strike the phosphor screen, causing
photons to be emitted in the same two-dimensional pattern. These typically have gains up
to 100 photon/photon. Dark emission from the photocathode is typically very low, resulting in a large dynamic range. Power consumed is proportional to light output, an important
trait for portable systems where energy is limited.
The device is very efficient because light is produced by cathodoluminescence, using
electrons to excite phosphor. A type P-22 phosphor produces 40 lumens/watt with electrons accelerated with 6kV [5, 47]. In addition, phosphor screens are capable of handling
large amounts of power and producing high luminance levels as shown by their application
to projection tubes.
Adding one or more microchannel plates (MCP) increases the gain of the image in-
CHAPTER 2. BACKGROUND
34
Objective lens
Eyepiece
Scene
Eye
Image
Image
Figure 2.11: Simple image intensifier application
to the
Screen
Figure 2.12: Microchannel plate (Proxitronic)
tensifier. An MCP is a plate of glass with many microscopic holes running through it. A
large DC voltage is applied across the plate, creating an electric field through the channels.
An electron striking one end of a channel causes secondary electrons to be released, as
shown in Figure 2.12. An avalanche effect occurs, and gain of a single MCP can be as
high as 5000 electron/electron.
104
The overall gain of an image intensifier with one MCP is
105 . With two MCPs, gain of 106 is achieved. Gain of intensifiers with varying
numbers of MCPs is shown in Figure 2.13. While gain increases with more MCPs, spatial
resolution decreases because the gaps between MCPs allow electrons to cross-over into
adjacent channels. Resolution versus number of MCPs is shown in Figure 2.14.
A DC bias current called the "strip current" flows through the MCP. Typical MCP
resistances range from 2 * 106 to 1082Q. For best linearity, the strip current needs to be
larger than the current of the electron flux being amplified. This additional current reduces
the energy efficiency of the intensifier.
Intensifier gain and dynamic range can be improved by using a semiconductor material
as the photocathode instead of a metal. Quantum efficiency improves by a factor of four
over a broad range of wavelengths, resulting in higher gain. The sensitivity of a galliumarsenide based photocathode is shown in Figure 2.15. This cathode has peak efficiency
between 650nm and 850nm, and is well matched to the avalanche mode silicon light emitter
with spectrum shown in Figure 2.9. Matching photocathode sensitivity to the light source
2.3. IMAGE INTENSIFICATION
35
IntensifierGain vs # of MCPs
1.UtI-+Ud
1.OE+07
= 1.OE+06
0
:1 1.OE+05
0
r- 1.OE+04
-
9 1.OE+03
1.OE+02
1.OE+01
0
1
2
3
MCPs
Figure 2.13: Intensifiergain vs. numberof MCPs
Resolutionvs # of MCPs
E 20
E
E
i 15
5
10
o
5
0
0
1
2
3
MCPs
Figure 2.14: Intensifier resolution vs. number of MCPs
CHAPTER 2. BACKGROUND
36
103
102
H-
-z
-0UJ
Z
I U
On LL
i--LL
Z LU
101
z
100
0T1
-
RADIANT SENSITIVITY
- - -'
10-1f
300
ANTIMFFFIINCY.
I
F . " I..' .. 'I----'-- ' I I
400
500
600
700
800
I
900
1000
WAVELENGTH (nm)
Figure 2.15: GaAs Photocathode Sensitivity (Hamamatsu)
also improves rejection of spurious input signals and reduces background noise.
In the future, the image intensifier could be replaced by smaller or more efficient structures. MEMS micromachining might help to miniaturize the intensifier itself [48], and
aid in integrating the intensifier with a silicon integrated circuit. The large glass vacuum
envelope could be replaced with a smaller vacuum package.
An up-converter might replace the intensifier if the efficiency of silicon light emitters
were increased, or if efficient infrared emitters were integrated onto an IC. The up-converter
converts photons at one wavelength to shorter wavelength, for example an infrared image
can be converted to a visible one. A pixelless up-converter for mid-infrared imaging is
described by Luo et al. [49]. It could eventually compete with image intensifiers, with the
advantages of being smaller and requiring lower operating voltages.
2.4. CIRCUIT DESIGN FOR DISPLAYS
2.4
37
Circuit Design for Displays
The driver circuits for an electronic display are typically digital to analog converters which
convert stored digital data into into analog waveforms used to produce an image. Management of leakage currents and device variation are key issues in data converter design.
Introductions to these topics are presented, followed by a survey of digital to analog converter topologies and techniques, with emphasis on circuits for display applications. The
current copier is a circuit that uses a self-calibration technique to reduce sensitivity to device variation. Its basic concept is presented, along with potential advantages in the design
of display driver circuits.
2.4.1
MOS Transistors
Leakage Currents
Circuit techniques to reduce leakage currents have been proposed, for example a 12b accurate analog storage cell [54]. Our design, as described in Chapter 4, uses circuit techniques
to minimize the effects of overlap capacitance and leakage currents on the current copier
circuit.
In the analog context, the main effect of leakage currents is to reduce circuit precision.
An important property of CMOS circuits is their capability to store charge on floating
nodes. The sample-and-hold
circuit shown in Figure 2.17 is a widely used example of a
charge-based circuit. An important figure of merit is how long it can retain a value within
a specified level of precision. Non-ideal effects in the sample and hold circuit include off-
current through the switch, junction leakage through parasitic source/drain diffusions, and
charge injection due to overlap capacitance (CGD) of the switch device.
Gate and subthreshold leakage are serious problems for charge-based circuits such as
the sample-and-hold.
In charge-based circuits, the retention time is how long a circuit can
preserve a stored value within a specified level of precision. Long retention times allow the
circuits to function longer between calibration cycles, and reduce the amount of resources
spent on periodic calibration.
A transistor is often used as a MOS capacitor because gate oxide has the highest ca-
pacitance per area (fF/,um2 ) of structures available in common CMOS processes. Charge
leakage through the gate limits the time period that a sampled value can be retained, and
changing device area does not remedy the problem because both capacitance and leakage
current are proportional to area. Poly-poly or metal-metal capacitors are available in some
processes, although their capacitance per unit area is much lower than the MOS capacitor.
38
CHAPTER2. BACKGROUND
Leakagetest
NIIOS Vs--Vb=VdO - Die 0
-n
41,UU-I
Z
8/02/0
achen
. . _...._ ....
I
1.OOE-13
Z011",
i 1.00E-14
i
i
i
i
i
Efa1
1,00E-15
i
i
41
1 OOE-16
0
0.5
1
1.5
2
Vg (volts)
Figure 2.16: Gate current for an NMOS device
Gate leakage of a large 1.8 volt NMOS device was measured with a semiconductor
parameter analyzer at room temperature as shown in Figure 2.16. For this measurement the
source, drain, and bulk are at zero volts.
The resulting data are consistent with published results [50, 51], and the primary mechanism for gate leakage is direct tunnelling through the thin (32A) gate oxide. To assess the
impact of gate leakage, consider a sample-and-hold implemented in this technology with a
voltage tolerance of lmV and a stored voltage around 1.OV.
I
area
Current density at 1.OV is lOfA/,m
2
C AV
area
(2.1)
(2.1)
and capacitance per unit area is 8fF/um2 . The
retention time is 800us. Display applications require retention times that are larger by
orders of magnitude. For example, a display operating at 100 frames per second with one
refresh cycle per frame requires a minimum retention time of 1/100 second, or 10ms.
In a 0.18tim technology, one solution is to use thick-oxide transistors (70A vs. 32A)
at circuit nodes where charge is stored. Thick-oxide transistors are available in most commercial CMOS processes. These devices have thicker gate dielectrics resulting in orders
of magnitude less gate leakage. They also have larger minimum feature sizes and worse
high-frequency performance than thin oxide devices. Capacitance per unit area is lower
than for thin oxide devices, but still higher than other capacitor design options.
As CMOS processes scale to smaller dimensions and power supply voltages are reduced, there is pressure to reduce the threshold voltage. Reducing the threshold voltage
increases on-current due to a larger (VGS - VT) factor. However it also increases subthreshold conduction by a factor of ten for every 90-100mv decrease in VT. This can be
2.4. CIRCUITDESIGN FOR DISPLAYS
39
I
Figure 2.17: Sample and Hold Circuit models. Left: ideal, Right: realistic
seen by setting VGS to zero in the equation for subthreshold conduction:
ID,OFF= Iseq(- V T)/nkT
(2.2)
In digital CMOS circuits, leakage currents cause static current to flow from from power
to ground. It increases power dissipation and causes reduced noise margins due to signal
degradation. In designs with millions of transistors per chip, leakage currents can exceed
the currents used in active operation.
Negative gate drive (V
< 0) and the "stacking
effect" have been used to reduce subthreshold conduction currents [52, 53].
Device Variation
The design of high-performance data converters depends on an understanding of device
variation and how to manage it. Variation is caused by random fluctuations which oc-
cur during the manufacturing process, and affects both material properties (doping levels,
threshold voltage) and geometries (transistor width and length). The most commonly used
variation model for MOSFET threshold voltage and currents is of the form aID k
[55]. Other models have also been proposed, i.e. [56].
Maintaining consistent image qualities across a display is a challenging problem, and
becomes even more difficult as integrated circuit feature sizes decrease. Traditional analog
designs use layout techniques such as symmetry and dummy devices to improve matching.
In a microdisplay, devices must be spread over a wide area and such techniques are not
practical. To measure variation across large length scales, we designed an NMOS current
mirror with 22 legs spanning 2.0 millimeters in the TSMC 0.18pm process. This structure represents a realistic design for a microdisplay where drivers are spaced across a die.
All devices were 2*2ipm in size. In a microdisplay, the currents through individual driver
circuits are small because there are many operating in parallel. The input current was set
CHAPTER 2. BACKGROUND
40
Die 0. ceumrrrt
rlrrr
tLuA
VdmOi, M
1.2DE
1.10-
.
1~SH7
1
a
5
7
pIillen
9
11
Figure 2.18: Current mirror measurements
at 1LA, and the drain voltage of mirror devices was 0.9V. Threshold voltage for NMOS
devices is 380mV from simulation models, and the VGS corresponding to 1,uA is 495mV.
The variation between output currents is aID = 2.5%. When there are 1,000 drivers on a
chip, circuits should be designed at the 4a or 5a certainty level. This represents 3.5b precision, far short of current application demands. Modem commercial display systems have
8b resolution, and 10-12b displays are being developed. To achieve reasonable resolution,
circuit techniques to improve matching between display drivers will be necessary.
2.4.2
DAC Fundamentals
Digital to analog converters (DACs) are used in display systems to convert digital data into
analog waveforms that control light emitters or light modulators. This section is survey of
DAC circuit design techniques with emphasis on converters useful for displays.
An ideal unipolar voltage-mode DAC has an output range from zero volts to (2 N - 1) *
VLSB volts. It accepts a digital data word d(N: 1) and performs the function:
N-1
VOUT = VLSB * i,
d(i) * 2
(2.3)
i=O
The output voltage is discrete, and quantization noise is: VQ,rms= VLSB/xV
natively, this can be expressed as the signal-to-noise
. Alter-
ratio: SNR = 6.02N + 1.76 dB, where
N is the number of bits. Johns and Martin [57] provide a discussion of non-ideal behavior
in DACs.
DACs can be divided into two groups: Nyquist rate DACs that change their outputs
once per cycle, and oversampling DACs that change their outputs multiple times per cy-
2.4. CIRCUIT DESIGNFOR DISPLAYS
41
cle. Nyquist DACs can be implemented using voltage-mode or current-mode techniques.
Voltage-mode DACs produce an analog voltage using techniques such as a resistor lad-
der or switched capacitor circuits. The output voltage typically requires buffering with an
operational amplifier to reduce output impedance. As an example, Bult and Geelen [58]
describe a circuit topology based on the R-2R ladder.
Current-mode DACs are attractive in CMOS technology because MOS transistors are
voltage-controlled current sources (VCCS) with medium transconductances, allowing precise control of current. A current-mode DAC can be simpler to implement than a voltagemode one because currents can be summed by connecting current sources in parallel. A
straightforward example is DAC made by switching 1024 identical current sources, achieving 10b precision at 40MHz in 0.8pum CMOS [59]. Note that a voltage output can be
produced by attaching a transimpedance amplifier to the output of a current-mode DAC.
Segmented DAC architectures use different techniques to generate portions of the output current, for example [60] describes a DAC with a unary decoded architecture for its
MSBs and a binary architecture for its LSBs. The MSBs are optimized for high precision,
while the LSBs use a structure that minimizes circuit area.
Oversampling converters are attractive for moderate speed applications where digital
circuits run much faster than the sampling frequency. For more information, Norwsowthy,
Schrier, and Temes have a book on delta-sigma converters [61]. Display drivers typically
require large numbers of converters that operate in parallel. Low-frequency operation, low
power dissipation, and small circuit area are important, therefore Nyquist rate converters
are preferable.
Dynamic element matching [66, 67] converts errors due to component mismatch to
uniformly distributed white noise. For example in a fully segmented DAC, if all cells
have exactly the same duty cycle then the mismatch is averaged out and much greater
dynamic precision achieved. Two more examples are swapping current sources at half the
time period to build a 14b accurate divide-by-two network [68], and a 14b DAC using
a "random walk" cycling scheme [69]. This technique can be used in conjunction with
calibration techniques to further improve the performance of data converters. The cost is
the complexity of choosing and switching arbitrary elements instead of using fixed patterns.
2.4.3
Calibration Techniques
In Chapter 4, a calibration technique based on the current copier circuit is applied to display
driver circuits which are current-mode digital to analog converters. Calibration techniques
optimized for large numbers of data converters on a single chip are applicable to a variety
CHAPTER 2. BACKGROUND
42
of applications. Display drivers are one such application, where a single chip incorporates
hundreds of DACs. Other applications currently under development are MEMS actuators,
neural network computation systems, and bioelectrical stimulators. In these types of chips,
the cost or "overhead" of a calibration technique is divided by the number of converters,
and the per-converter cost of adding calibration is reduced.
Digital algorithms have been applied to data converters to improve performance, for example calibration and storage of digital correction coefficients in memory [62, 63]. Pipeline
converters use redundancy to allow digital calibration and error correction [64]. Digital
calibration techniques have been applied to large arrays of moderate resolution DACs for
neural type computation cells [65]. The limitation of digital calibration is the cost of ad-
ditional memory and computation. In a DSP system that already includes a processor and
large memory this is not an issue. In applications with resource constraints such as embedded systems or ICs with large numbers of converters on-chip, the digital overhead can be
prohibitive. Digital calibration of a data converter may require one or more lookup tables,
and the amount of memory needed becomes large if there are many converters on a single
chip.
Analog calibration techniques have the advantages of full-speed operation, with no digital computation or memory overhead. In applications that require the highest operating
frequencies, or where large numbers of converters are included on a single chip, analog calibration is a practical way to improve precision. The current copier uses a self-calibration
technique to create a precise replica of a reference current. Its key characteristic is immunity to process variation, allowing it to make a replica current with greater precision than
a conventional current mirror occupying the same area. An overview of its history and
applications is given here, and it is described in detail in Chapter 4.
The current copier circuit was first published in 1989 by a number of groups. Wegmann
and Vittoz presented techniques to produce a continuous output current and to produce integer current ratios [70]. Wouter et al. and Groeneveld et al. described a 16b DAC with
self-calibration technique [71, 72]. Nairn and Salama proposed an algorithmic ADC based
on current copiers [73]. Other DACs using current copiers include a 5b 128x oversampled 44kHz DAC for audio applications made by combining a 4b PMOS and a 4b NMOS
converter [74]; a 16b segmented DAC using 32 current copier cells to produce 5 MSB's,
and 1 lb divider [75]; and a 14b segmented DAC using with 5b thermometer MSB, 5b
thermometer upper LSB, and 4b binary lower LSB [76].
Current copiers can be used to implement switched-current circuits which are analogous to switched-capacitor circuits. Fiez et al. [77] describe a variety of current-mode
circuit techniques such as track-and-holds, V-I/I-V converters, and ways to cancel charge
2.4. CIRCUITDESIGN FOR DISPLAYS
43
injection. A good collection of information on switched-current circuits is compiled by
Toumazou [78]. Noise analysis of current copiers is complicated by time-dependent sampling behavior. Daubert provides a detailed analysis in [79].
Innovative circuit techniques have been applied to the current copier circuit, for example a regulated-cascode current copier [80]. This boosts output resistance, at the cost of
increased power dissipation and complexity. Related calibration techniques for currentmode circuits have been proposed, including a 14b DAC using MOS devices in parallel to
create "calibrated" devices [81].
44
CHAPTER 2. BACKGROUND
Chapter 3
Micro-projector Design
3.1
Design Overview
A micro-projector utilizing silicon light emission and image intensification is described in
this chapter. The goal of this display design is to produce an image with adequate brightness
for micro-projection applications while being highly energy efficient to maximize battery
life of a portable system. The performance of a proof-of-concept system is calculated, and
confirmed on the lab bench.
A schematic of the ideal system is shown in Figure 3.1. An integrated circuit produces
a faint image which is proximity coupled to the photocathode of the image intensifier. The
image intensifier efficiently amplifies the image to useful brightness levels. The output of
the image intensifier is a high luminance image that can be viewed directly, or projected
with a lens onto a viewing surface.
light
Figure 3.1: CMOS IC and image intensifier
CHAPTER 3. MICRO-PROJECTOR DESIGN
46
high intensity
light source
A
to
screen
writing
light
CRT
Modulator
beamsplitter
V
lens
Figure 3.2: Two-stage display architecturefor projection application (JVC)
3.1.1
Background: Two-stage displays
The operation of a display can be divided into two parts. Addressing is the creation of timevarying electrical signals and routing them into an array of display elements. Emission or
modulation of light occurs in response to the electrical signals at each display element.
Dividing the tasks of addressing and light modulation into two stages allows allows a
display to combine the best characteristics of multiple technologies. For example, high
speed addressing and high precision driver circuits can operate at low signal levels. Meanwhile the SLM consists of a two-dimensional array of elements operating in parallel at
moderate speeds to modulate high optical power.
One such system uses a cathode-ray tube and a photoaddressed liquid crystal SLM [82].
First, electrical signals drive a CRT display working at relatively small signal levels. The
output of the CRT is a two-dimensional image. A photoconductor receives this image and
converts it to a two-dimensional
pattern of voltages which control a liquid-crystal light
valve. The liquid-crystal light valve modulates light from a high intensity lamp for projection. This architecture, shown in Figure 3.2 was used in the first-generation digital cinema
projectors [83, 84]. To produce color images three separate projectors are used to generate
the primary colors, and their outputs superimposed.
A similar architecture has been used to produce an infrared display for military apA visible light display controls a photoaddressed liquid crystal panel which
modulates infrared radiation [85]. A liquid crystal SLM can be used to control a laser [86].
An SLM has been used to produce [87]. It uses a conventional microdisplay to generate a
plications.
write-light image, and an optically addressed SLM (liquid crystal) to produce the final image. The above examples use liquid crystal SLM's, but it should be noted that other SLM
technologies exist, for example one that uses a microchannel plate and dielectric mirror
[88].
3.2. MICRO-PROJECTOREFFICACY
3.2
47
Micro-projector Efficacy
In this section the efficacy of the two-stage micro-projector is calculated. Efficacy is defined
as the lumens available on the projection screen per watt of input power, and has units of
lumens per watt. We will assume that all conversions in the system are linear, and that no
components are saturated.
The best reported external quantum efficiency of light emitters in a standard CMOS
process is 5 * 10 - 7 photons/electron [35]. Our measurements also yielded
2 * 10 - 7
photons/electron for a 0.18 pm CMOS process without silicide block, which is consistent
with the published result. This figure is the external quantum efficiency for photons emitted
from the front face of the IC. The use of microlenses or light-reflecting metal structures
could improve quantum efficiency.
External Q.E. = 5 * 10 - 7 photon/electron
Photons travel between the silicon integrated circuit and the photocathode of the image
intensifier. We will assume the ideal case where the two components are placed in contact
with each other. Then, the ratio of emitted photons to photons striking the photocathode is
unity.
Transmission from IC to photocathode = 100% (ideal)
The image intensifier used in the proof-of-concept system is a Hamamatsu V7090U-61N130. It is a Generation III intensifier, with high efficiency gallium arsenide photocathode,
one microchannel plate, and P-43 phosphor screen. The photocathode quantum efficiency
is 0.25 for 600-850nm photons [89]. This range of wavelengths includes the majority of
the silicon emission spectrum.
Photocathode Q.E. = 25%
The maximum voltage across the microchannel plate (MCP) is 900 volts [89]. At this
voltage, MCP gain is 1500 electron/electron.
MCP gain = 1500 electron/electron
The image intensifier output screen converts electrical energy to optical energy at the
output window. It has a sharp emission peak at 540nm, and its quantum efficiency is 95
photons/electron for electrons at 6,000 eV [89, 46].
Phosphor Q.E. = 95 photons/electron
48
CHAPTER 3. MICRO-PROJECTOR DESIGN
Multiplying these efficiencies together, the overall intensifier gain is 3 * 104 phot/phot.
To calculate luminous efficacy, diode breakdown voltage is - 10OV.
The display driver circuit
requires some headroom, so the total voltage drop through the silicon light emitting pixels
and driver circuits is about 11V. The wavelength of emitted light is 540nm, so the energy
per photon can be calculated:
A= 5.4 * 10- 7 m
v= c
A = 5.56 * 1014hertz
E = h*v
E = 6.626 * 10- 3 4 * 5.56 * 1014= 3.68* 10- 19joules/photon
The eye is most sensitive to radiation at 550nm, where one watt of energy equals 680
lumens. The standard relative luminosity factor for 540nm is 0.954 [90].
Luminosity = 680 * 0.954 = 649 lumens/watt
The strip current of the MCP must be be considered. For low signal levels, the minimum
required strip current is simply the MCP voltage divided by the MCP resistance. Typical
values are VMCP= 900V and R = 107Q. For high current applications, the signal current
must be less than 7% of total strip current to maintain linear operation [91].
Suppose the total average current through the pixel array is 90mA. The silicon IC dissipates 90mA * 1V = 1W. The image intensifier dissipates negligible energy at the photocathode, 220mW in the MCP (signal=7% of total current), and 100mW at the output
screen. Output luminance is 2.4 lumens with total power dissipation of 1.3 watts, yielding
1.8 lumens/watt. Assume the output screen is a lambertian surface, and the projector lens
aperture is f/1.4 (NA = 0.36, 12.7% transmission). Photographic objectives with aperture
of f/1.4 are common, and projection lenses with apertures from f/1.0 to f/3 are available.
Output luminance = 1.8 lumens
Transmission of lens = NA
2
= 12.7%
Let the final image size be a half-page (5.5x8in, or 14x20cm), and assume the projection
surface has unity gain.
Area = .028m2
The illuminance from the micro-projector is:
1.81umens+ .028m2 = lllux
3.2. MICRO-PROJECTOR EFFICACY
Total pixel array current
Electron flux
Photon flux from Si
Photocathode electron flux
MCP1 electron output
MCP2 electron output
Phosphor screen output
Luminance output
Efficacy
49
600 ,uA
3.75 * 1015 e/sec
1.9 * 10 9 phot/sec
4.7 * 108
4.7 * 1011
4.7 * 1014
4.5 * 1016
10.6
7.1
e/sec
e/sec
e/sec
phot/sec
lumens
lumens/watt
Table 3.1: Intensifier Performance with two MCPs
With one MCP, illuminance is below the level necessary for a micro-projector application (see Table 1.2). The bottleneck is image intensifier gain. Efficacy improves signif-
icantly if another MCP is added to the image intensifier. The combination of two MCPs
in series boosts image intensifier gain to 106. This decreases power dissipated by the silicon IC and at the same time increases output luminance. With two MCPs, performance is
summarized in Table 3.1.
Power dissipated by the silicon IC is reduced to only 7mW. The MCPs dissipate 1.1W,
and the output screen dissipates .45W. Total power dissipation is 1.5W, and efficacy is 7.1
lumens/watt. This efficacy is high compared to other display technologies. For comparison,
a liquid crystal SLM with metal halide light source produces only 0.5-3 lumens/watt due
to the low transmission of the SLM [92]. This figure is for large-size projectors using
metal halide lamps that produce 100 lumens/watt. These lamps cannot be used in compact
projectors, and if they are replaced with LEDs efficacy will be reduced by at least a factor
of three, to 0.1-1 lumen/watt. A recently published organic-LED-on-silicon display [93]
produces approximately 2 lumens/watt.
With two MCPs, the MCP strip current and output screen dominate power dissipation.
Assuming the same f/1.4 lens as before, 12.7% of 10.6 lumens are spread over an area of
.028m 2 , yielding 49 lux which is adequate for a projector application. A micro-projector
with high image intensifier gain can create half-page sized images with worst-case 1.5W
power dissipation. This is for the entire image at maximum intensity. If the display is
emissive, then realistic estimates of relative brightness are 13% for a screen of text, 68%
for a personal computer desktop, and 40% for photographic or video images.
realistic estimate is 50% of full intensity, resulting in 0.75W power dissipation.
A more
Considering that lithium ion batteries in personal portable systems typically have capacities of 2-7 watt-hours [94], a portable system with emissive micro-projector and other
functions could have a battery life of 2-7 hours.
50
CHAPTER 3. MICRO-PROJECTOR DESIGN
Operating voltage VDD
Physical gate length
Gate oxide thickness (physical)
3.3 and 1.8
0.35 and 0.16
70 and 32
Silicide (self-aligned)
Metal layers
volts
,rm
A
CoSi2
6
Al
Table 3.2: TSMC 0.18pum CMOS technology characteristics
The dynamic range of the image intensifier is calculated below. Assume all components are operating in a linear manner. The maximum photon flux into the photocathode
is 2.8
*
1011 phot/sec. We approximate all photons to be the peak wavelength of the sil-
icon emitters, 700nm. The photocathode diameter is 18mm, so its area is 2.5 * 10- 4 m 2 .
The radiant flux is 3.1 * 10- 8 W/cm 2 . The effective background illuminance (EBL) of the
V7090U-61-N130 image intensifier is specified as 1.0 * 10- 13 W/cm 2 . The ratio of these
two levels is the dynamic range: 3 * 105, which is significantly larger than the 103 ratio
required for 10b resolution. In reality, the image intensifier gain rolls off or "saturates"
at high input luminance levels. The linear dynamic range is limited at the low end by the
EBL, and at the high end by saturation effects. From Hamamatsu datasheets, the linear
dynamic range with one MCP is 105. When two or three MCPs are included, the increased
gain causes a proportional increase in EBL while saturation occurs at the same level, thus
linear dynamic range is decreased.
Linear dynamic range is about 200 and 30, respec-
tively. The V7090 series image intensifiers produce low output luminance because they
are intended for night vision and scientific applications.
An image intensifier optimized
for micro-projection could be designed to accomodate higher current levels and provide a
larger dynamic range.
3.3
Test Chip Implementation
Process Technology
A simple microdisplay backplane was designed to test the idea of a display using silicon
light emission and image intensification. The TSMC 0. 18,m CMOS technology was used.
Some key process features are listed in Table 3.2. This process was chosen to emphasize
the ability to integrate a microdisplay in a modern high-performance CMOS process. The
display's pixels include lb static memory, therefore area is shared between circuits and
light emitters. A small feature size reduces circuit area resulting in higher fill factor.
3.3. TEST CHIPIMPLEMENTATION
3.3.1
51
IC overview
The integrated circuit includes light-emitting pixel arrays and other device and circuit experiments. The arrays are described first, followed by the other experiments.
The integrated circuit includes two display arrays. Each array measures 16x32 pixels,
with a pixel pitch of 25m . Each pixel contains a light emitter, a current source, and a lb
SRAM cell. Integrating static memory into each pixel eliminates the need for refreshing.
Each array has sixteen digital inputs driving the bitlines. A full decoder permits random
row access. Grayscaling can be achieved using pulse-width modulation.
The fill factor
of the IC array is 3%, and photon and electron scattering increase the apparent fill factor.
Simulation indicates that the write time for an optimized 1000x1000 pixel array is 30ns,
adequate for real-time video display.
3.3.2
Pixel
A pixel element was designed to match the resolution of a commercially available image
intensifier. Its size is 25*25m
. A schematic diagram of the pixel circuit is shown in Fig-
ure 3.3. At its core, cross-coupled inverters form a lb SRAM cell. A complementary passgate is controlled by the WRITEEN
signal, which is driven by the rowline. The NMOS
transistor in the bottom inverter combined with the VBIAS transistor control the current
through the silicon light emitter. VBIAS is a single voltage reference distributed across the
chip. For precise control of pixel luminance, current-mode control of diodes is preferable
to voltage-mode control because of their sharp avalanche breakdown characteristic.
The high voltage supply, Vhi, is 11.5V. Reverse breakdown of the p+/nwell junction
occurs at -10V. Because the avalanche breakdown characteristic is extremely abrupt, the
MOS transistors see only 11.0-10.0 = 1.0 volt, and this value changes very little regardless of whether the pixel is on or off. For added robustness, a protection diode limits the
maximum voltage seen by MOS devices.
The light-emitting diode is a p+/nwell junction measuring 10*2um . Luminance is
proportional to current, and to the first order independent of junction area. The properties
of junctions in TSMC 0.18[m technology were described in section 2.2.3, and Figures 2.8,
2.9, and 2.10. There is additional space available in the pixel array, and the fill factor could
be increased by increasing junction area. Micro-lenses can also be added to spread light
and improve fill factor.
The array is fully static when used in black-and-white (lb) mode. It can also display
grayscale images when driven with pulse-width modulation (PWM) input signals.
The digital portion of the pixel was designed for a supply voltage VDD of 1.80V. A high
CHAPTER 3. MICRO-PROJECTOR DESIGN
52
-0
0
:3
Figure 3.3: Circuit schematic of a single pixel
voltage (Vhi = 11V) is used to drive light-emitting diodes in avalanche breakdown.
Current vs. Voltage Drive
Light-emitting diodes produce light in proportion to diode current. Linear current-mode
drivers can provide good control of light output. However, limiting the current leads to
long switching transients on the capacitive column lines. Rearranging the fundamental
capacitor equation:
AT = CCOLUMNAV
(3.1)
IDRIVE
Voltage drive can switch column lines faster because current is not limited, however
precise control of light emitters is difficult. The proof-of-concept system uses a hybrid
approach. Hybrid approaches have been described in the literature [95], and can provide
both fast addressing and good linearity. The per-pixel memory is written by a voltage, so
bitline transitions and writes to memory are fast. Within a pixel, the silicon light emitter
is controlled by a MOSFET transistor acting as a current source. Switching transitions are
fast and provide good timing precision. The luminance of the light emitter is precisely
controlled at a fixed level by the current source MOSFET, and can be modulated using
pulse width modulation.
3.3.3
Pixel Arrays
Two pixel arrays were implemented on the test chip. Each consisted of a 16x32 pixel matrix, wordline decoder, and input buffer circuits as shown in Figure 3.4. The only difference
3.3. TEST CHIPIMPLEMENTATION
53
Pixel
Array
o
0
16 x 32
U0
a0
111
1111
addr<4:0>
I
l l l l
l l
l
data<15:0>
Figure 3.4: Block diagramof pixel array
was resizing of the bias transistor in each pixel. One array was designed for a maximum
current of 100uAlpixel, the other was designed for 40[A/pixel.
Because of the lb SRAM per pixel architecture, a full decoder was used to control rowlines. The decoder generates a 1-of-32 output when ENABLE is true, or forces all outputs
to zero when ENABLE is false. It is built from static CMOS standard cells. Combined with
the memory-per-pixel arrray, random access to rowlines allows writing to occur in arbitrary
order, which can reduce energy used in addressing.
3.3.4
Top-level IC Design and Packaging
The die measures 2.7x2.7mm (7mm 2 ), and was designed for wire bonding into an open
cavity DIP40 package. The DIP40 package is used for ease of testing; simple measurements are made with the IC on a solderless breadboard, and a test system is built on a
printed circuit board. A chip photomicrograph is shown in Figure 3.5. Bonding pads are
placed along all four sides of the die. Large metal busses around the perimeter of the chip
distribute VDDand ground. Electrostatic discharge protection circuits are included on all
CMOS inputs.
CHAPTER 3. MICRO-PROJECTOR DESIGN
Figure 3.5: ICphotomicrograph
3.3.5
System Design
A printed circuit board to drive the display arrays and to demonstrate its capabilities is
shown in Figure 3.6. It includes regulated power supplies, the test chip, and a microcontroller.
The Microchip PIC16F874 microcontroller is used for its ease of programming, ability to function with a 1.8V power supply, and large number of I/0 pins. The microcontroller operates at a frequency of 1MHz, and is programmed in assembly and PICBasic.
The pseudo-code fragment below is representative of the code used to drive the proof-of-
Figure 3.6: Proof-of-concept system board
3.3. TEST CHIPIMPLEMENTATION
PIC16F874
PORTA
PORTB
PORTC
PORTD
7
6
d15
spare
d7
d14
EN2
d6
5
LED1
d13
EN1
d5
55
4
LEDO
d12
a4
d4
3
spare
dl 1
a3
d3
2
1
spare spare
d10O
d9
a2
al
d2
dl
0
spare
d8
aO
dO
Table 3.3: Microcontroller interface definition
concept system. It blanks the display by writing zeroes to all pixel locations.
FOR
i = 0 to
ADDRESS
31
= i
DATA_HIGH = xOO
DATA_LOW = xOO
ENABLE
=
1
ENABLE
=
0
NEXT
i
From the point of view of the microcontroller, the display behaves like an SRAM. The
interface consists of a 5b address, 16b data, and an enable signal. The 16F874 has three
8b I/O ports and one 6b I/O port. The signal mapping for the microprocesor is shown in
Table 3.3.
3.3.6
Image Intensifier and Optics
The intensifier used in this system is the Hamamatsu V7090U-61-N130. The photocathode
is of a GaAs(Cs) material, sensitive from 400 to 900nm. It has one microchannel plate
(MCP) to provide high luminous gain, up to 3.5 * 104 phot/phot. The output window phosphor is type P-43 which emits light at 540nm, close to the eye's peak response wavelength
[89]. The unit is 23mm thick and 45mm in diameter. Since this unit is optimized for scientific and night vision applications, output luminance is low and the unit operates at low
power levels.
The intensifier requires DC voltages at 800V and 6kV. A matching power supply,
Hamamatsu C6706-30, is used. This power supply includes automatic power limiting and
shutoff circuitry.
In the proof-of-concept system, packaging and mechanical constraints prevent the photocathode from being in contact with the silicon die. A microscope objective lens was used
to couple light from the pixel arrays to the intensifier's photocathode, A 20x/0.4NA lens
CHAPTER 3. MICRO-PROJECTORDESIGN
Figure 3.7: Proof-of-concept system with intensifier
was mounted at one end of a hollow tube, and the intensifier attached to the other end.
This allowed us to evaluate the proof-of-concept system, although a real system should
have intensifier and silicon IC in contact ("proximity focusing") to minimize system volume. Assuming image source is a lambertian surface, the lens captures 16% of the energy
radiated by the source and focuses it into an image. In addition, reflections at air-glass
interfaces without anti-reflection coating reduce transmission by about 4% per interface.
The addition of micro-lenses on the IC could increase coupling by allowing more photons
to escape the front surface of the IC. A better solution is the ideal structure where the photocathode is deposited directly on top of the IC dielectric layers. This would eliminate
losses due to total internal reflection, increasing the effective quantum efficiency by 80%
to 9 * 10- 7 phot/elec. It would also reduce the number of interfaces in the optical path, and
associated reflections.
A picture of the final test setup is shown in Figure 3.7. The image intensifier is on top
of an optical tube attached to a 20x 0.4 N.A. microscope objective lens. The printed circuit
board is visible under the optical tube. Black cardboard (visible in the photo) was used to
reduce stray light and reflections in the test system. The letters "MIT" are visible in the
intensifier's output window.
3.3.7
Proof-of-Concept System Performance
This section calculates the performance of the proof-of-concept system built on the lab
bench, and compares it to measured results.
To compare calculations with measured efficiency, the display was configured with
all pixels on.The current through the IC high voltage supply was measured at 1.1mA, or
3.4. RESULTS
57
6.9 * 1015electron/sec. This current level is much smaller than full-scale to ensure linear
operation of the image intensifier. The quantum efficiency is the same as before, 5 * 10 - 7
photons/electron, however, in the proof-of-concept system the silicon junctions were covered with a salicide layer. This significantly reduces optical transmission to about 20%, so
the photon flux from the surface of the IC is 6.9 * 108 phot/sec.
The silicon surface acts as a lambertian source. The lens is a microscope objective
with 20x magnification and a numerical aperture (NA) of 0.4, and the system is in air. A
lens with this numerical aperture captures 16% of photons emitted from the IC. Assuming
reflections and other losses are small, 1.1 * 108 phot/sec strike the photocathode of the
image intensifier.
The gain of the image intensifier is 3.5 * 104 for Vcathode = -800, Vmcp = 900, and Va
= 6000. The flux leaving the output window is 3.9 * 1012 phot/sec., and the P-43 phosphor
screen has a sharp emission peak at a wavelength of 540nm. At this wavelength, each
photon carries 3.7e - 19 joule, so the radiant flux is 1.4 * 10-6 watt. The luminous output
is 1.4 * 10-6 watt * 649 lumens/watt = 9.2 * 10- 4 lumens.
The pixel array measures 400*800pLm , and the lens has 20x magnification,
so the
photon flux is spread over an area 400 larger than the source. The image area is 1.28 *
10- 4
m
3.4
3.4.1
2
.
The expected luminance of the display is 0.67 footlamberts.
Results
Single Pixel
The pixel with lb SRAM memory functioned as designed. DC measurements of pixel I-V
characteristics are presented below. Figure 3.8 shows current through the light-emitting
junction as a function of the high voltage supply VHI. When the pixel is storing a low value
(off-state), current is negligible for VHI less than 12.5V. Above this voltage, the protection
diode turns on, and a large current flows. For the on-state, diode current is constant at about
200[A for VHI between 11.5 and 12V. In this region there is adequate voltage to bias the
junction in avalanche breakdown, and current through the junction is regulated by the MOS
transistors acting as a current sink. Below 11V,the current source has inadequate headroom
and current is reduced. Simulated results are also included on this graph. Diode models
for this process were probably intended for capacitance estimates, and do not accurately
predict reverse breakdown I-V characteristics.
For light emission characteristics of the p+/nwell junctions in this process, refer to
Section 2.2.3.
CHAPTER 3. MICRO-PROJECTOR DESIGN
Pixel current, Vblas=Vdd=1.80v,Chip #3
3 OE-04
2.OE-04
----6-
OE-04
nnr.M
•r
.
(pixel) ON
I(pixel) OFF
Sim
~11
I M
_ _
10LT~
10
Vhi (volts)
Figure 3.8: Pixel I-V characteristic,VHI sweep
Figure 3.9: Test image of IlC under microscope
3.4.2
Arrays
The lower-current array was fully functional.
Test images were produced for lb and
grayscale modes of operation. This photo was taken in a darkened room with a CCD
camera mounted on a microscope. Exposure time was several seconds using a 20x 0.45
N.A. microscope objective lens. Figure 3.9 shows a static lb test pattern. The small white
dots in the picture are caused by noise in the CCD camera.
Grayscale images can be created using pulse-width modulation, as shown in Figure 3.10.
This is a photo of the image intensifier output window, demonstrating operation of the twostage display architecture. The bottom two rows of pixels are at maximum intensity to
provide a constant luminance reference. The rest of the pixels are switched to produce a 5b
3.4. RESULTS
Figure 3.10: 32-level grayscale gradient. Bottom two rows are at full intensity
linear gradient (32 time slots per frame, 60 frames/sec). Bit depth is limited in this demonstration by the speed of the microcontroller. Figure 3.11 shows another lb test pattern,
observed on the image intensifier output window.
3.4.3
Power Measurements
The proof-of-concept system was assembled in our laboratory. For a static test image with
all pixels at maximum brightness, total IC current from the 11V supply is 1.ImA, so the
IC dissipates 12mW. Current from the low voltage (1.8V) supply is negligible. The image
intensifier is set for maximum gain with Vk-mcp=800, Vmcp=900, Va=6000. Intensifier
power is dominated by MCP strip current which is approximately 9002V/107Q = 81mW.
Total display power is 12+81=93mW. Because the proof-of-concept system operates at low
signal levels, power is dominated by the constant MCP strip current.
Output luminance was 1.3 footlambers, measured with a Photo Research PR-880 photometer. The difference between this and the predicted value of 0.67 footlamberts in Section 3.3.7 may be caused by a number of sources. Image intensifier gain varies exponentially with applied voltage, so a small voltage error may cause a large difference in optical
gain. The quantum efficiency may be higher because of different device of different junction characteristics, and transmission out of the silicon may be higher because the junctions
in the 0.18/pm process are shallower than those reported by Snyman [35].
The image intensifier's high voltage power supply limits its output current to prevent
damage to the intensifier. The current limit is small, on the order of microamperes. Leakage currents and surface conduction currents were enough to trip the automatic shutoff
circuitry, and this prevented us from measuring the intensifier current directly. The im-
CHAPTER 3. MICRO-PROJECTORDESIGN
Figure 3.11: Output window of intensifier(20mm) with sample image
age intensifier and high-voltage power supply were tested in isolation, and the input power
to the power supply was measured. Ideally, efficiency is constant and total power is proportional to power delivered to the intensifier. Lab measurements show the power supply
dissipates 200mW while standing idle, and 360mW when showing the test pattern shown
in Figure 3.7. Design of an energy-efficient high voltage power supply for the image intensifier will have a significant impact on overall energy efficiency.
3.4.4
Discussion
We have demonstrated two-stage image generation using silicon light emitters and an image
intensifier. This architecture can produce high-resolution output with high energy efficiency
and adequate luminance for a micro-projector. A proof-of-concept system is functional,
and performance agrees with design calculations.
Two ways to increase system efficiency are increasing the gain of the image intensifier,
and improving the efficiency of silicon light emitters. Increasing image intensifier gain
allows lower power operation of silicon light emitters, and higher output luminance. As
gain increases, less light is required from the silicon light emitters. Power dissipation of
the silicon IC decreases, and energy efficiency is dominated by the performance of the
image intensifier. The image intensifier uses cathodoluminescence to produce light, and is
both energy-efficient and capable of producing high luminances comparable to projection
CRT's [96]. Image intensifiers with multiple MCP's can have gains of up to 107 , however
3.4. RESULTS
61
resolution decreases as more MCP's are added [89].
Improving the efficiency of light emission from silicon is an area of active research
interest [39, 38, 36], and silicide should be blocked from light-emitting areas to maximize
luminous output.
62
CHAPTER 3. MICRO-PROJECTOR DESIGN
Chapter 4
Backplane IC Design
This chapter describes the design of an integrated backplane for the two-stage microdisplay
architecture. The backplane accepts digital input signals and DC reference currents, and
produces an image which is coupled to an image intensifier.
Its major components are
a 360x200 pixel array, row drivers, calibrated column drivers, and testability circuits as
shown in Figure 4.1. The following sections describe each of these blocks in detail.
4.1
Pixel Array
The 360x200 pixel array occupies most of the IC area. Pixels measure 25/um square
because it results in reasonable demands for optical performance, and also allows pitchmatching of driver circuits to the array. The array operates in column-parallel mode, with
one row active at a time. The dimensions were chosen to match the 16:9 aspect ratio of the
HDTV format.
The pixel array includes features to facilitate characterization
and measurement, as
shown in Figure 4.2. The rightmost column of pixels (column 359) is removed, and the
output of its column driver is connected to a package pin for ease of measurement. The
three rightmost columns (356-358) are implemented with top-metal electrodes for future
testing with OLED technology. The top row of pixels (row 0) is removed, and the output of
its row driver is connected to a package pin. The next row (row 1) is replaced with a row of
transimpedance amplifiers to permit observation of column driver outputs without loading
of the column lines. The remaining 356x198 pixels include silicon light emitters.
A schematic of a pixel with silicon light emitter is shown in Figure 4.3. Device D1 is a
p+/nwell junction used to emit light. VLOWis set at -10OV.Current flows from the column
line through M2 causing D1 to break down and emit light at low levels. Avalanche breakdown in the p+/nwell junction occurs abruptly, therefore the voltage swing at nl necessary
63
CHAPTER 4. BACKPLANE IC DESIGN
Test
Pixel Array
(360x200)
Figure 4.1: Overview of CMOS backplane
t to pin
_1
rowO
rowl
row2
Amplifier row
0
O
r-
m
CD
C,,
Si-Pixel Array
356x198
u,
co
rowl99
00
Column Drivers (ArrayDACs)
Figure 4.2: Arrayfloorplan
B
-- pin
4.1. PIXEL ARRAY
65
rowline
D2
M1
V PCH
D
I
I"
\/
Figure 4.3: Pixel schematic
to turn the pixel on and off is less than 100mV. D2 is a junction diode that prevents ni from
falling below ground.
When column-driver DACs change their output values, large switching transients can
occur. In addition, charge stored on the column-line due to distributed resistance and capacitance can cause nonlinearity. An additional rowline and pass transistor are implemented to
suppress these effects. Ml sinks current from ni to VPCH at the beginning of each row pe-
riod, preventing transient currents from flowing through D1. The tradeoffs of this technique
are reduced duty cycle, and a 4% reduction in fill factor due to the additional rowline.
OLED pixel
Three columns of the pixel array were implemented with top-level metal pads instead of
silicon light emitters, for future testing with OLED material. The pixels are on the same
25,um pitch as the rest of the array, and the passivation opening measures 12.4x10.4tPm 2
resulting in a fill factor of 21%.
OLED material layers can be deposited on top of the IC. The top level metal pads are the
anodes of the OLED devices, and and a common cathode is formed by a layer of transparent
conductor such as ITO applied over the OLED layers, as illustrated in Figure 4.4.
A schematic of the OLED pixel is shown in Figure 4.5. The dual rowline structure, and
the protection diode are the same as in the pixel with silicon light emitter.
CHAPTER 4. BACKPLANE IC DESIGN
passivation
ITO cathode
e- injection layer
e- transport layer
hole transport layer
hole injection layer
}
OLED
layers
Figure 4.4: OLED materiallayers
rowline
..
I
,,,,
VPIHU I
-HZ 11"
rowlinepch
---'
'i
VDUMP
column
line
Figure 4.5: OLED pixel schematic
pad
4.2. ROW DRIVERS
67
chift in
"
"
~~
Figure 4.6: Row driver circuit
I~ /--\~--
rowclk
rowclkdn
shiftin
2'
enable
shiftout
I
rowline
rowlinepch
Figure 4.7: Row driver timing
4.2
Row Drivers
The row drivers are digital buffers controlled by a shift register. The shift register design
was chosen for its simplicity and compact area. It requires that rows be accessed in sequential order, which is acceptable given the row-at-a-time operation of the display. There are
two sets of rowlines. The first set controls current from the column to the light emitters. An
enable signal gates the shift register outputs to allow clearing of the shift register without
disturbing the rowlines. The second set of rowlines is used to reduce switching transients,
as described in Section 4.1. The logical functions for each of the i rows are:
rowline(i) =
rowlinepch(i) =
shiftout(i) AND enable
shiftout(i) AND rowclk AND rowclkdn
A rowline driver schematic is shown in Figure 4.6, and a timing diagram in Figure 4.7.
The signal rowclkdn is a delayed and inverted copy of rowclk, generated by the FPGA and
off-chip delay circuitry.
68
CHAPTER 4. BACKPLANE IC DESIGN
IIN
ITR<9:0>
-
calibration
calibration
-;·--i
r;..ii--i--!'i-i·;t
i.;.i~-
i ili
·--'-)
external
reference
r-i-i-i-i-ij
ii-ii
i-
:;....
·
i:
ifi
i:
!·i-··
i-t-iti')i-ii-i(
RefDAC
1023 copiers
Binary-weighted
ArrayDACs
IREF<i>=nI.*2i
Figure 4.8: Two-stage calibrationtechnique
4.3
Column Drivers and Calibration
4.3.1
Overview
The pixel array is driven in column-parallel manner with one row illuminated at a time.
Current-mode driver circuits permit precise control of light emitters, however transferring
small currents through long column lines with large parasitics is challenging. Long settling
times reduce precision at the pixel level, and degrade the quality of the image. One solution
is transferring large currents through column lines, and scaling them down at the pixel level
[97]. Precharge techniques to reduce settling time have also been presented [98]. In this
display, column driver current levels are increased by operating pixels at a low duty cycle.
The full-scale pixel current was set by considering the maximum allowable power dissipation for an IC. The maximum dissipation without design for thermal management is
about one watt. The high-voltage supply for avalanche junctions is about 10V. Consider
a one-megapixel display with one thousand rows. The average full-scale pixel current is
100nA. The row duty cycle is 1/1000, therefore the maximum column driver output current
100pA.
Mismatches between column-parallel driver circuits result in fixed pattern noise (FPN)
that the human eye detects readily. To achieve a high degree of device matching in column
drivers, conventional designs use large geometry devices. In this work, a self-calibration
technique which is based on the current copier circuit [99] is used to achieve high precision
while reducing circuit area compared to a conventional design.
Calibration is performed periodically in two stages as shown in Figure 4.8. The starting
4.3. COLUMN DRIVERS AND CALIBRATION
69
Figure 4.9: Column driver calibration overview
point is a DC reference current. This current can be generated on-chip with the use of a
bandgap or other precision reference source. For demonstration purposes, in our system it
was supplied by an off-chip source.
The reference current is copied 1,023 times by an array of identical current-copier cells.
This array is called the Reference DAC, or RefDAC. There is only one RefDAC per chip
and all column drivers are calibrated against it. It must be highly accurate while operating
in subthreshold, therefore a fully segmented architecture is used. The outputs of the cells
are grouped to produce ten very precise binary-weighted bit-currents which are used to
calibrate the column driver DACs (ArrayDACs).
Each ArrayDAC consists of ten current-copier cells that replicate the bit-currents from
RefDAC. A binary-weighted architecture is used to minimize area at the cost of some
accuracy because there are hundreds of ArrayDACs per chip.
The current copier is a dynamic analog memory, therefore the RefDAC and ArrayDACs
must be calibrated periodically. To reduce refresh rate and power dissipated during calibration, maximizing the retention time of the current-copier is of prime importance. A circuit
technique to improve retention time will be described in Section 4.5.
Figure 4.9 is a block diagram showing the major circuit blocks in the calibrated column
drivers. These blocks will be described in following sections.
4.3.2
Area Comparison
Consider a 10b current-mode DAC implemented as a binary-weighted design with MOS
transistors as current sources. Matching between drivers should also be at the 10b level.
The MSB is 50[LA, and 3
variation must be less than 0.025%. The MSBs are in strong
inversion, and LSBs in subthreshold. Array driver circuits are spatially distributed, so lay-
CHAPTER 4. BACKPLANE IC DESIGN
70
lOUT
1
Figure 4.10: Simple MOS current source
out techniques to improve matching such as common-centroid are not applicable. The total
area of a single DAC implemented this way is 50,000/m
2 , found
by extrapolating available
process variation data. In contrast, an entire ArrayDAC occupies 7,000,m
2.
In the limit
of a large display, the area used by the RefDAC and calibration overhead is insignificant
compared to the area of the ArrayDACs. On the implemented chip, one ArrayDAC plus
1/360 of the RefDAC occupy 10,600bum2. This represents a 5x to 7x reduction in circuit
area.
4.3.3
Current Copier
This section describes the motivation for the current copier circuit, its basic operation,
and improvements made in the implemented circuit. The circuits are explained as NMOS
circuits, although both NMOS and PMOS versions of the current copier are used.
In a conventional design, shown in Figure 4.10, MOSFET transistors can be used as
current sources. Common models for drain current and threshold voltage variation assume
that they are inversely proportional to the square root of device area [100].
kId
kct
(4.1)
kvt
aVt= x/-L
(4.2)
Uld =
where kId and kvt are empirically determined constants with units of A/lm
and V/um
respectively, and W and L are device dimensions. Large devices are required in order to
obtain precise currents. The problem is worse in subthreshold (weak inversion) because
4.3. COLUMNDRIVERS AND CALIBRATION
A
I
71
B
(%.jUUT
Figure 4.11: Simple current copier
threshold voltage variation has an exponential effect on drain current. Furthermore, process
variation is increasing in newer CMOS technologies [101].
Instead of using large devices to achieve high precision, a circuit technique can be used
to produce precise replicas of a reference current. The circuit is called the current copier
[70], and Figure 4.11 is a simplified schematic diagram showing its basic operation. The
circuit has two states. In the first state the A switches are closed, the B switch open, and
a DC reference current driven into node IIN. Device M1 is diode-connected, and the VGS
corresponding to the reference current is stored on capacitor C. Then the A switches are
opened and the B switch closed. Transistor Ml acts as a current source producing a replica
of the reference current. Device variation in M1 does not affect circuit precision, thus M1
can be small in size. However the circuit is a sample and hold, and its precision is limited
by charge injection and leakage currents.
Charge injection can be partially cancelled by a properly sized "dummy" device con-
nected to the storage node. Leakage currents are caused by subthreshold conduction in the
switch device, and reverse-bias leakage of source/drain diffusions.
To find the relation-
ship between leakage currents and change in copier output current, consider the storage
capacitor and the small-signal model of M1:
AV
ILEAK = CSTORE AT
iD = gmAvGS
Combining these, it can be seen:
(4.3)
(4.4)
CHAPTER4. BACKPLANEIC DESIGN
72
vbias
Figure 4.12: Reduced transconductance current copier
AiD = 9mILEAK
CSTORE
(4.5)
Output current fluctuation can be reduced by reducing transconductance and leakage
current. These will be explained shortly. The remaining two options, reducing
\T and
increasing CSTORE,translate into increased time spent in calibration and increased circuit
area, which are not desirable.
A current copier circuit with reduced copier transconductance is shown in Figure 4.12
[71]. M4, called the "bleeder device," is the output branch of a current mirror. There is one
reference (diode-connected) device per chip biasing many bleeder devices. The purpose
of M4 is to reduce the current through M3 in order to reduce its transconductance,
gm,M3.
Based on variation analysis, M4 carries the largest amount of current possible while being
less than IIN. This constraint is necessary because the NMOS-only current copier can only
sink current, not source it. If the nominal value of ID,M4 is chosen 80% of IIN, the current
through M3 is reduced by a factor of five. If M3 is in subthreshold operation, gm,M3 is also
reduced by a factor of five.
An improved current-copier circuit that we developed is shown in Figure 4.13. Transistor M3 is replaced by the stack M8-M10 which has a lower transconductance due to source
degeneration. M1 1-M12 are a cascoded "bleeder" device. M14 is the sampling switch and
M13 is the charge injection cancellation device. M8 and M9 also raise the voltage stored
on capacitor C2 to improve retention time by allowing more negative gate drive on M14
4.3. COLUMN DRIVERS AND CALIBRATION
73
lout
-I
lii
A-
n6
M13
I
I
C2
Figure 4.13: Improved current copier
vbias2
-
CHAPTER4. BACKPLANEIC DESIGN
74
0
0
I
0
Vstore
--
Vhold
F---
Vstore
Figure 4.14: Passgate leakage scenarios: (a) conventional, (b) voltage control
!,
0
I
Fh.
HOLD
Figure 4.15: Conventional (left) and improved (right) passgates
when turned off. The target voltage at n5 is 1.OV.
To further improve retention time, the voltage applied to the source of M14 is controlled. Consider a conventional NMOS passgate as shown in Figure 4.14. In the worst
case, the gate is at OV, the drain at the stored voltage (VSTORE),and the source is at OV.
The gate-to-source voltage is zero while VDS is greater than a few kT/q, and subthreshold
conduction can be calculated:
ID = Ise q(-VT)/kT
(4.6)
This subthreshold conduction will quickly discharge capacitor C2. A solution is to drive
the source voltage to a positive voltage VHOLD,ensuring that VGS,14< 0. Subthreshold
conduction is greatly reduced:
- VT)/kT
ID = Ise q(- VHOLD
(4.7)
Figure 4.15 shows the leakage reduction concept in its simplest form. On the left is
an ordinary NMOS transistor used as a passgate. On the right is the improved passgate.
When the clock is high, the two sides are connected. When the clock is low, the two sides
are disconnected and the middle node is set to voltage VHOLD- Subthreshold conduction
4.4. BIASING
75
is reduced, and junction leakage becomes the main cause of leakage currents. A similar
structure known as the "T-switch" has been used to reduce feedthrough in high frequency
switches [102].
In the conventional design (Figure 4.12), the voltage at node n3 may to drop to zero to
reduce power consumption by turning off the copier's output current. When this happens,
VGS
of M5 is zero and subthreshold leakage from n2 to n3 is significant. To avoid this, the
input and output current paths (n7 and n6) are separated in our design (Figure 4.13). The
voltage at n6 does not affect M14, and n6 can fall to ground while the bit is deselected, for
lower power.
The voltage at n7 is controlled by connecting it to a voltage source (VHOLD)when the
current copier is holding its value. Subthreshold leakage current through M14 is reduced
because a large negative gate drive is maintained, and VDS of M14 is also reduced.
Circuit operating points and DC operation were verified with simulation. Process variation was analyzed using models provided by the foundry. Predicting the dynamic behav-
ior of sampling circuits is difficult. Modelling charge injection requires two-dimensional
analysis of the MOSFET to produce correct results, and a complex task. Models have
been developed [103, 52], but accessibility of design tools is limited. For design purposes,
we estimated the dummy device M13 would cancel 80% of the charge injection from the
sampling switch M14.
4.4
Biasing
PreRef
The PreRef section supplies the RefDAC with the reference current IIN and the reference
voltages for "bleeder" devices. A simplified schematic of the IIN circuit is shown in Figure 4.16. The off-chip reference current is replicated by a ratioed current mirror. The ratio
is set to make off-chip currents larger than 1,uA for easier manipulation and noise immunity. This mirror also isolates on-chip circuits from the large capacitances associated with
off-chip circuits. A 1-to-4 analog demultiplexer connects the reference current to one of
four RefDAC quadrants, which will be explained in the RefDAC section. This multiplexer
also controls the voltage of the de-selected outputs. Its logical function is shown in Table 4.1, and it will be used to improve retention time of the current copiers as discussed in
a later section.
The circuit to generate reference voltages for "bleeder" devices is shown in Figure 4.17.
It consists of a ratioed PMOS current mirror at its input. Current from the PMOS mirror
CHAPTER4. BACKPLANE IC DESIGN
76
Vdd
Vdd
RefDAC
iin<3:0>
ost
rol
Figure 4.16: RefDAC reference current generator and switch
prerefsel3
0
0
0
1
2
1 0 VHOLDEN
o 0 1
0
0 1 0
0
1 0 0
0
0 0 0
0
O
0
0
O
O
O
1
1
0
0
RefDAC iin(3)
float
float
float
ii.
(2)
float
float
iin
float
(1)
float
iin
float
float
VHOLD
(0)
ii.
float
float
float
1
1
VHOLD
VHOLD
1
0
1
VHOLD
VHOLD
iin
VHOLD
0
0
1
VHOLD
iin
VHOLD
VHOLD
0
1
iin
VHOLD
VHOLD
VHOLD
Table 4.1: PreRef switch truth table
iin
4.4. BIASING
77
ib
Figure 4.17: Bleeder bias voltage generator
flows into an NMOS mirror. The devices in the NMOS mirror are layout-matched
to the
"bleeder" devices located in the RefDAC current copier cells, and the two output voltages
are distributed to the RefDAC.
VBIASGEN
This block accepts reference currents from off-chip sources and produces the reference
voltages for the ArrayDAC current mirror circuits. Figure 4.18 shows a single circuit.
There are four circuits total, one for each group of current copiers. This circuit is a mirrorimage of the one used in the PreRef block. Current from an off-chip reference flows into a
ratioed NMOS current mirror. The output feeds the input of a PMOS current mirror which
is layout-matched
to the current mirror output legs located in the ArrayDACs.
The two voltage outputs are supplied to the ArrayDACs. For testability, the VBIAS2
node is accessible on a pin through a passgate controlled by the testmode shift register. The
testmode shift register circuit is shown in Figure 4.19. When TESTEN is low, the passgate
is disabled. When TESTEN is high and the control register output is high, the passgate
connects the test node to a package pin. The SHIFTIN and SHIFTOUT pins are connected
to form a shift register as shown in Figure 4.20, with the clock, first input, and last output
connected to package pins.
CHAPTER4. BACKPLANE IC DESIGN
78
asl
as2
cu
n
Figure 4.18: VBIAS circuit
It
IS
To
Figure 4.19: Testability control circuit
vbiasd2
test
control I
to pins
Figure 4.20: Testability scan chain
shift out
79
4.5. REFDAC
RefDAC
INPI ITS
OUTPUTS
iout
iin
vbypl ,vbyp2
current
copier
rowsi
CD
rowclk
CD
o-
array
32x32
0)
colsi
colclk
r.
It 1 nnhlo
VUL
IC:L/Il
column select
--
Figure 4.21: RefDAC block diagram
4.5
RefDAC
The RefDAC is an array of 1023 identical NMOS current copier cells, based on the design
presented in section 4.3.3. Figure 4.21 shows the current copier array and row and column
drivers. A detailed circuit schematic is shown in Figure 4.22. The copier is implemented
with a mix of thick and thin oxide NMOS devices. Devices M1 and M12 are 1.8V devices
because they have lower variation (k in Equation 4.1). M2, M3, M7, and M9 are 3.3V
devices to minimize gate leakage. M4 and M5 are designed conservatively as 3.3V devices
in case of large voltage swings at the DAC outputs. The gate voltages of devices M3, M4,
and M5 are limited between 0 and 1.7V. When switches are turned on, they also act as
cascode devices to improve current copier output resistance.
During calibration, the OUT signal is low to disable outputs. Row and column control
signals are connected to an AND gate in each cell, so that only one copier is enabled at a
time.
The RefDAC rowline driver is shown in Figure 4.23, and the column driver in Fig-
ure 4.24. The rowline driver includes an edge-shaping circuit to guarantee non-overlap
when rowlines are switching [104], and the column driver is a simple shift register with
buffers. Row and column driver circuits are implemented with standard cells, and operate
between DVDDL=1.7V, and DGND=OV.
The RefDAC outputs are ten binary-weighted currents. Current copier cell outputs
are connected in parallel to produce the desired current levels. To compensate for process
gradients, cells were placed symmetrically around the center of the array The cell layout
pattern for the center of the RefDAC array is shown in Figure 4.25. The numbers corre-
CHAPTER4. BACKPLANE IC DESIGN
80
r
out
I
us
*_
I
iout
M5
F-
iin
"ll I
DVDDL
I
DVDDL
M4
inrow
incol
M2
M3
M3
I1IL
byp2
bypl
Uo
M7
l~~~~~~~~
a
X
n
v
I
I
IL M12
M10
I
XL
I
I
A J
FM',
_
M11
i
I
Figure 4.22: RefDAC current copier circuit
Figure 4.23: RefDAC row driver
power=DVDDL
si
col line
clk
so
ground=DGND
Figure 4.24: RefDAC column driver
M1l
4.5. REFDAC
81
89899898
9
6
9
7
7
9
6
9
8
9
9
8
9
8
9
7
7
9
6
9
8
9
9
8
9
8
9
0
1
9
7
9
9
1
x
9
7
9
8
9
9
8
9
8
9
7
7
9
6
9
8
9
9
8
9
8
Figure 4.25: ReJDACcell pattern
spond to the i t h output current where i = 0 is the LSB and i = 9 is the MSB. The it h output
current uses 2i current copier cells in parallel. This is a common-centroid pattern. The cell
marked "0" produces the LSB, and the cell marked "x" is used for test purposes.
The RefDAC array includes a row of dummy cells on each side of the array to improve
uniformity between active cells. The overall array size is 34x34 cells. Each cell measures
22.4* 28.8[tm , and the complete RefDAC occupies 1.lmm 2 .
The initial RefDAC design had all current copier inputs (IIN) connected together to a
single node. Simulations and calculated estimates of settling time indicated the settling
time of RefDAC current copiers was too long with this arrangement.
The cause was the
large capacitive load on the input node. This capacitance consists of gate and interconnect
capacitances within the RefDAC array; note that the PreRef block has a current mirror
that isolates this node from external capacitances.
The input node was divided into four
to reduce its capacitance. Each quadrant included eight columns of the RefDAC array.
During calibration, a demultiplexer in the PreRef block to steers the reference current to
the appropriate quadrant of the RefDAC array. Figure 4.26 is a simulation of RefDAC
cell settling behavior. The curves are currents through the copier (M7) and bleeder (M12)
devices during a calibration cycle. The reference current is 100nA, "bleeder" or mirror
current is constant at 85nA, and the copier settles to 5nA with 10b precision in 70ps.
Retention time improvement is implemented with a multiplexer in the PreRef block, and
the current copier cells in the RefDAC block. Figure 4.27 shows these elements together
Current copier cells are arranged in groups of 256 that share a common input node n7.
During calibration, n7 connects between IIN and one current copier cell. Its voltage is three
diode drops above ground, so deselected passgates have negative gate-to-source voltages
and leakage currents are small. When copier cells are holding their values, n7 is connected
to a voltage source and the voltage is set to VHOLD. The voltage at n7 is now controlled
at all times. Subthreshold conduction through M14 is reduced, and retention time of the
CHAPTER4. BACKPLANE IC DESIGN
82
test i estreTcell
snemaric:
AUg
I/:33:zL3
3
ZL4-
Transient Response
I
,: /I0/M12/D
-: /I0/M7/D
130n
100n
m n
·
n
·
I
70.0n
40.0n
10.0n
-20.n
_
. . . .. .. . . . . . . . . . . ..
0.00
30.0u
60.0u
time ( s )
11
III
I
.
90.0u
.
..
~
120u
Figure 4.26: RefDAC copier cell settling time simulation
Vh
Figure 4.27: Retention time improvementcircuit
83
4.6. ARRAYDAC
vDlasZ
o
vbias1
AVDDH
inn
in
iin
out
ni it
Figure 4.28: ArrayDAC copier schematic
current copier cells is greatly increased.
4.6
ArrayDAC
There are 360 column drivers (ArrayDACs) on the chip. Each ArrayDAC consists of ten
PMOS current copiers which replicate the ten binary-weighted currents from the RefDAC.
A circuit schematic is shown in Figure 4.28. It is a PMOS implementation of the RefDAC
current copier. ArrayDAC I/O pins are shown in Figure 4.29. The 10b data word controls
the OUT switches for the ten current copiers. The calibration input for each copier (IIN)
connects to the respective binary-weighted reference current. Signal IN is common to all
ten copiers within an ArrayDAC. The output IOUT is connected to the column line.
To reduce settling time during calibration, a multiplexer connects the RefDAC outputs
to either the left 180, or right 180 ArrayDACs. The multiplexer also implements the retention time improvement technique described in the RefDAC section. The ArrayDACs
occupy a total of 2.5mm2 .
The current copiers are individually optimized for their respective current levels. There
are four sets of bias voltages available for producing bleeder currents. In addition, a current
copier cell may contain 1, 2, or 4 legs of the current mirror to produce multiples of the
bleeder current. By varying both the number of legs and the bias voltages, a design with
very compact area can be implemented. Excluding the transistor used as a storage capacitor,
CHAPTER4. BACKPLANEIC DESIGN
84
ArrayDAC bit
nominal current bleeder voltage group
legs
0
100n
D
1
1
200n
C
1
2
3
4
5
6
7
8
400n
800n
1.6u
3.2u
6.4u
12.8u
25.6u
C
C
B
B
B
A
A
2
4
1
2
4
1
2
9
51.2u
A
4
Table 4.2: ArrayDAC current mirror biasing
lout
data<9:0>o----
0
iin<9:0> o> .
ino
inn o
ArrayDAC
1,
vbiasal a2,b1
b2,cl,c2,d1 ,d
Figure 4.29: ArrayDAC block diagram
the largest transistors are 4*45am , and the maximum ratio for current mirror devices is
4:1. Nominal currents and biasing arrangements are listed in Table 4.2.
Two sets of 10b digital registers are located below each ArrayDAC. One is used to store
the present value of the output, the second is used to load the next value in the background.
This allows all ArrayDACs to update their output values synchronously. The circuit in Figure 4.30 is used. The signal EN 1 is generated by a shift register that selects one ArrayDAC
register to be loaded. This is repeated 360 times to load new values into all 360 first-level
register banks. Then the signal EN2 is asserted for one clock cycle by the FPGA, causing
the new values to be loaded into the second-level register banks.
The EN1 shift register was implemented with the circuit shown in Figure 4.31. The
signal clear forces all registers to zero on the rising edge of clk. This eases the process
of resetting the display control logic, and also saves time during data loading when only a
portion of the display is being utilized.
Calibration of the ArrayDACs is controlled by a shift register. Before calibrating an
ArrayDAC, its output switches are turned off by setting the data word to zero (Ox3FF). To
4.6. ARRAYDAC
85
DVDDH
data<!
'<9:0>
VE
standard cells use DVDDH, DGND
Figure 4.30: ArrayDAC registers
si
enl
clk
SO
clear
Figure 4.31: ArrayDAC data load control
sout
enm
Figure 4.32: ArrayDAC calibration control
CHAPTER4. BACKPLANE IC DESIGN
86
sout
Figure 4.33: ArrayDAC calibration control, improved
reduce capacitance on the input node, the ArrayDACs are divided into two groups: columns
0-179, and 180-359 the RefDAC output demultiplexer must select the correct group, and
the RefDAC output control must be enabled.
A shift register composed of the cell shown in Figure 4.32 selects one ArrayDAC at
a time to use the RefDAC outputs as reference currents. A circuit improvement was to
change the voltage levels of the in and inn signals. These signals are switched between
AVDDH and VBIAS=1.OV,causing the switch devices inside the ArrayDAC current copier
to act as cascode devices. A more detailed schematic is shown in Figure 4.33
4.7 Design for Testability
Because of the widening gap between the sizes of on-chip circuits and board-level components, testing is an important design consideration. On-chip circuits to support testing and
observation are essential. Circuits such as an on-chip sample-and-hold to observe signals
without loading critical nodes has been described [105]. The idea has been taken farther,
for example an on-chip oscilloscope [106].
This chip included a number of testability features. First, some nodes were routed to
pins through passgates controlled by a testmode shift register. This allowed direct observation and measurement of electrical signals. One row driver and one ArrayDAC were
configured for direct measurement. The first row driver is identical to all others except that
it drives a pad instead of a rowline. The last ArrayDAC (column 359) drives a pad instead
of a pixel column.
To observe ArrayDAC outputs and switching behavior of the bitlines, the top row of
4.7. DESIGN FOR TESTABILITY
.F,
, 1ON
FO
Figure 4.34: Micrographshowing Si-LED pixels, OLED pixels, and ArrayDACs
pixels was removed and replaced with transimpedance amplifiers which are described below.
Both silicon light emitters and OLEDs can be supported by current-mode driver circuits, although OLEDs typically require larger voltage swings. For future testing, three
columns of pixels were built with top-level metal bondpads, as shown in Figure 4.34. This
picture shows the bottom-right corner of the pixel array. At top left, pixels with silicon light
emitters can be seen. The three rightmost columns of pixels are OLED-style bondpads. The
ArrayDACs are located directly below the pixel array.
Amplifier
Transimpedance amplifiers allow realtime observation of column driver outputs by isolating the column lines from large external capacitances. The circuit for the transimpedance
amplifier is shown in Figure 4.35. It is a single transistor common source amplifier with
resistor biasing. Nominal bias current is 2001pA, and the input and output operating point
CHAPTER4. BACKPLANEIC DESIGN
88
ampso
ampout
iND.
Figure 4.35: Column transimpedance amplifier
voltages are 650mV. The ideal transimpedance is 3.7KQ, so the output voltage swing for a
full-scale 100pA input is 370mV.
When the rowline for row 1 is enabled, the column lines are connected to the transimpedance amplifier inputs. A shift register selects a single amplifier, connects it to the power
supply, and enables it to drive a shared output pin. For characterization purposes, the input
of the amplifier can be connected to an input pin so that a known current can be applied
and the output observed.
4.8
Top-level IC
The IC is fabricated in a standard 0.18pm CMOS logic process with one poly and five metal
layers, and measures 8mm by 10mm. A micrograph is shown in Figure 4.36. The center of
the chip is occupied by the 360x200 pixel array. The RefDAC is located at bottom center.
The ArrayDACs are located along the bottom of the array, and row drivers are located along
the left edge of the array. The chip is packaged in an open-cavity PQFP package.
The RefDAC occupies 1.3mm2 , and the ArrayDACs occupy a total of 2.5mm 2 . Pixel
size is 25,um square, the pixel array measures 360x200 corresponding to a 16:9 aspect ratio,
and the fill factor of silicon light emitters is 22%. Three columns of pixels were built with
4.9. SYSTEM BOARD AND OPTICS
Figure 4.36: Die micrograph
top-level metal electrodes instead of silicon diodes to interface with OLED technology.
The electrical interface to the backplane IC is described in Tables 4.3, 4.4, and 4.5.
4.9
System Board and Optics
The microdisplay backplane IC is packaged in an open cavity PQFP package and mounted
on a printed circuit board. The board includes an FPGA to generate digital data and control
signals, and DC voltage and current sources. Programs are compiled on a computer and
downloaded to the FPGA through a programming cable.
Due to packaging and mechanical constraints it was not possible to put the silicon IC
in contact with the image intensifier's photocathode. For demonstration purposes, a microscope objective lens was used to couple the image from the silicon IC to the photocathode.
The lens and the image intensifier are mounted on opposite ends of an optical tube which
is positioned above the silicon IC.
The lens has a numerical aperture of 0.25, so it transmits 6% of the light emitted by the
silicon to the intensifier. The intensifier is a Hamamatsu V7090U-61-N130, a Generation
III intensifier with gallium arsenide photocathode and one microchannel plate.
CHAPTER4. BACKPLANE IC DESIGN
90
Name
AVDDH
VHINWELL
DVDDH
DVDDL
AGND
DGND
VBULK
VDUMP
Direction
3VDC
Power Supply
3VDC
3VDC
1.7VDC
Analog GND
Digital GND
Substrate GND
0.6VDC
-10VDC
VLOW
VHOLD
AMPVDDH
AMPGND
VPROT
Type
Notes
Power Supply Equal to AVDDH
Power Supply
Power Supply
Power Supply
Power Supply
Power Supply
Power Supply
Power Supply
Avalanche diode supply
0.3VDC
Power Supply
3VDC
Power Supply
Amplifier GND Power supply For transimpedance amps
0.3VDC
Power supply For protection circuits
Table 4.3: IC power supplies
FPGA and control
Control algorithms are implemented on an FPGA for maximum flexibility in testing. A
control board based on the Xilinx Spartan-II XC2S50 FPGA is used. This FPGA includes
50,000 gates, and is packaged in a 144-pin QFP package. The I/O circuits operate at up
to 3.3V, and the core operates at 2.5V. The register toggle frequency FTOG is stated as
263MHz which is ample for our application.
Finite state machines were used to control the various parts of the display backplane.
The controller initialized state registers, performed calibration of the RefDAC and ArrayDACs, and drove patterns using the ArrayDACs, as shown in Figure 4.37.
A 20MHz reference clock was generated by the crystal-controlled oscillator shown in
Figure 4.38. Frequency division and duty cycle control were implemented in the FPGA. DC
voltages were generated using linear voltage regulators. Current references were generated
using discrete bipolar transistors with large emitter degeneration resistances.
A simple control pattern used for electrical measurements is shown in Figure 4.39. First
the RefDAC outputs are used to calibrate the ArrayDACs, which requires 25ms. Then, the
RefDAC is calibrated against the reference current while the ArrayDACs drive the pixel
array. Calibration of the RefDAC takes 70ms, which is longer than the 10b retention time
of the ArrayDACs. Therefore, the ArrayDACs only produce output for 35ms (25ms+35ms
= 60ms), and are idle for the remaining 35ms. The pixel array is driven for 33% of total
time, so this pattern is only useful for testing.
An improved control pattern is shown in Figure 4.40. This pattern utilizes the long
4.9. SYSTEM BOARD AND OPTICS
Name
rowenable
rowsi
ampsi
vbtesten
vbsin
calsin
calen
shiftclr
shiftsi
data(9:0)
rowclkdn
rowclk
enable2
vbscanclk
prvholden
prselect(3:0)
vbiasa
vbiasb
vbiasc
vbiasd
priin
pribypin
prtestclk
prsin
prtesten
refrowsi
refrowclk
refcolclk
refvholden
refselect(2:0)
latchclk
shiftclk
calclk
outenable
ampclk
amptestiin
amptest
Type
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
analog current
analog current
analog current
analog current
analog current
analog current
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
digital
Direction
in
in
in
in
91
Notes
enable rowlines to turn ON
data input for row driver shift register
data in for test-amp shift register
connect VBIAS nodes to vbtest(3:0) pins
in
in
data in for vbias shift register. Negative true
data in for calibration shift reg. Negative true
in
in
in
enable for calibration shift register
clears ArrayDAC data register enable signals
data in for ArrayDAC enable register
in
10b image data
in
in
in
in
in
in
in
in
in
in
in
in
in
controls second set of rowlines
clock for rowline shift register
load second level of ArrayDAC registers
clock for vbias shift register
clock for pre-refdac shift register
drive preref(3:0) test outputs
bleeder current for bits 9,8,7
bleeder current for bits 6,5,4
bleeder current for bits 3,2,1
bleeder current for bit 0
sets refdac reference current
sets refdac bleeders
clock for preref shift register
in
data in for preref shift register, Negative true
in
enables preref shift register
in
input for refdac row shift register
in
in
in
clock for refdac row shift register
clock for refdac column shift register
enables VHOLD muxes in refdac
in
RefDAC outputs to left, right, or refiouttest
in
clock for array registers
in
clock for first-level ArrayDAC shift register
in
in
in
clock for arraydac calibration shift register
enables some digital outputs
clock for amplifier shift register
analog current
in
test current input for amplifiers
digital
in
select amplifier normal or test mode
Table 4.4: Input pins
CHAPTER4. BACKPLANE IC DESIGN
92
Name
Direction Notes
Type
vbso
arrrowso
digital
digital
out
out
output of vbias shift register
output of row driver shift register
prerefso
digital
out
output of preref shift register
refrowso
refcolso
digital
digital
out
out
output of refdac row shift register
output of refdac column shift register
prerefout(3:0)
a. voltage or current
out
test output from PreRef
vbtest(3:0)
refioutx
refiouttest(9:0)
arrshiftso
arrco1359
arrcalso
arrampso
arrampout
analog voltage
analog current
analog current
digital
analog current
digital
digital
analog voltage
out
out
out
out
out
out
out
out
VGS of VBIAS mirrors
output of extra copier cell in RefDAC
RefDAC output currents, for testing
arrayshift shift register output
column359 ArrayDAC current out
arraydac calibration shift register out
amplifier shift register out
amplifier current output
arrrowlinetest
arrrowpchtest
digital
digital
out
out
output of dummy rowline driver
output of dummy rowline driver
Table 4.5: Output pins
reset
set testmode signals
calibrate ArrayDACs
V
calibrate RefDAC
drive ArrayDAC pattern
NI--I
Figure 4.37: High-level program flowchart
4.9. SYSTEM BOARD AND OPTICS
1/6 74HC04
+o% COrDA
II
Figure 4.38: Clock oscillator
RefDAC out
I
ArrayDAC
calibration
g
b
25ms
35
a
35
60ms
Figure 4.39: Timing diagram - simple program
CHAPTER 4. BACKPLANE IC DESIGN
Ro
Ro
Ac Ao
Ac
25+ 35
Ro
Ao
60ms
Ro
Ro
Ro Rc Ro Re
Acc Ao lAc Ao JAc_ Ao Ac Ao
7 * 70 = 490ms
Figure 4.40: Timing diagram - optimized
Figure 4.41: Assembled system board
retention time of the RefDAC (500ms). The ArrayDACs alternate between 25ms calibration periods and 35ms array-driving periods. RefDAC calibration is divided between two
periods to minimize ArrayDAC idle time. The pixel array is driven for 58% of total time,
limited by the retention time of the ArrayDACs and the time required to calibrate them.
ArrayDAC calibration and output cycles can be subdivided to improve image quality.
For example a fraction of the ArrayDACs could be calibrated after each frame. This would
eliminate flicker caused by the idle time during ArrayDAC calibration (25ms). In future
implementations, the addition of one or more extra ArrayDACs could permit calibration in
the background. This would increase pixel duty cycle at a slight increase in control circuit
complexity.
The assembled system board is shown in Figure 4.41. The board has four signal layers
and measures 4x10 inches. The FPGA is mounted on a daughterboard located on the left
side of the main board. The backplane IC is at center, in an open-cavity package covered
with a lid. Many test terminals located around the perimeter of the IC. Adjustable DC
voltage and current source circuits are located on the right side of the board.
Chapter 5
Measured results
5.1
5.1.1
Electrical Measurements
RefDAC Linearity
To measure the performance of the RefDAC, the ten binary-weighted current outputs are
connected to ten pins by enabling a test mode. The RefDAC is calibrated, and the currents
are measured individually using an HP4140B picoammeter.
During calibration, the RefDAC output currents are disabled. The picoammeter must
be synchronized to measure only when outputs are valid. The FPGA generates a synchronization signal at the end of each calibration cycle, and the circuit shown in Figure 5.1 is
used to delay the sync signal and invert its polarity for connection to the 4140B trigger
input.
The following graphs show the operation of the RefDAC in response to varied reference
and bleeder current levels (IINand IBYP).The ten binary-weighted output currents are on
the X-axis, from LSB to MSB. The measured current is plotted on the Y-axis, expressed
as the ratio between measured and nominal values. First, the ordinary current mirrors are
measured alone by disabling the current copiers ("uncalibrated").
Calibration is enabled,
O
3.3K
_m
FPGA
'/VV~
.
.. N3904
I1OOuF 1
_"_2_
7
47uF
Figure 5.1: Synchronization circuit
95
to 4140B
sync input
96
CHAPTER 5. MEASURED RESULTS
mcr3yref4.m refdac nocal current error 25 aug 04 achen
o
C o
o
o
X
X
0O
0o
0
-5
C)
0-
ca
-a -10
>15
x
'o
X
: -20
0
--
X
X
X
I
x~~~~~~~~~~~~~~~~~~~~~~.
x uncalibrated
_
0
-2_:N
-,4n
0
2
I
calibrated
i
I
I
4
6
8
10
Bit (Isb-to-msb)
Figure 5.2: RefDAC bit currents,first attempt
the total "calibrated" output current is measured.
In the first measurement, the reference and bleeder current sources are set with IREF =
100nA and IBYP = 85nA. The outputs from the RefDAC are shown in Figure 5.2. The
uncalibrated currents are approximately 83% of nominal value (shown as 83%-100% = 17%). There is an offset between the BYP value of -15% and the output value of -17%.
This may be caused by mismatch in the current mirror input circuit. When calibration is
enabled the output currents are 105% of their nominal values, and 5% higher than the value
of IREF. Variation in the uncalibrated currents is mainly caused by mismatch between the
bleeder current source devices. Variation in the calibrated currents is caused mainly by
charge injection in current copier circuits, and will be explained in more detail later.
RefDAC precision can be improved by adjusting the two current references.
IREF is
reduced by 5%, and IBYP is increased by 3% to make the calibrated and uncalibrated output
currents 100% and 85% of nominal value, respectively. The results are shown in Figure 5.3.
Based on the measured currents, the response to a ramp input was calculated and corrected for gain and offet errors. These data were used to compare RefDAC performance to
ideal data converter performance, shown in Figure 5.4. Integral nonlinearity (INL) is less
than ±0.43 LSB, indicating 10.lb precision after calibration.
The main source of error in the RefDAC is variation in charge injection. A constant
amount of charge injection in all cells does not affect linearity because of the fully segmented architecture. It is variation between the sizes of the passgate and charge injection
97
5.1. ELECTRICAL MEASUREMENTS
mcr3yref5.m refdac nocal current error 25 aug 04 achen
Jl
U
)
O
O
O
X
X
O
O
a)
a)
-10
U)
', -15
X X
X
00
X
X
1-20
O -25_--_
qn
0
2
1.1 _
_ *
x
uncalloratea
0
calibrated
!
.
I
4
6
8
_
10
Bit (Isb-to-msb)
Figure 5.3: RefDAC bit currents, second attempt
cancellation devices that causes nonlinearity. In a sample-and-hold
circuit, the dummy de-
vice should exactly cancel the charge injection from the switch device. If the cancellation
is not exact, then the change in charge causes a change in voltage: \Q = CAV. If this
change in voltage is small, a small-signal
model can be used. The effect of AV on the
output current is Al = g, * AV. For a given copier, AQ and C are fixed. The change in
output current can be minimized by reducing the transconductance g, of the current copier.
Transconductance is monotonically increasing with drain current, so reducing the current
level through the current copier will reduce g, and the effect of charge injection and charge
injection variation on the output current.
To illustrate this, note the current copier carries 7nA (95-88=7%) in the second case.
Reducing this current should reduce transconductance of current copier circuits, and improve RefDAC precision.
In the next experiment, IREF = 96nA and IBYP = 90nA. Current copiers carry 6nA,
or 6% of nominal current. Performance of the RefDAC is shown in Figures 5.5 and 5.6.
Without calibration the RefDAC precision is 8b, about the same as in the previous measurement. With calibration, the RefDAC achieves 11.5b accuracy. In principle, reducing
the difference between IByp and IREF, will minimize charge injection error. However,
IBYP must be set to always be less than IREF with four or five sigma certainty because the
NMOS current copier can only sink current.
RefDAC retention time test results are shown in Figure 5.7 and 5.8. Node IIN (see Fig-
CHAPTER 5. MEASURED RESULTS
mcr3yref5 pseudo-inl 25 aug 04 achen
0
200
400
600
code
800
1000
Figure 5.4: RefDAC INL, second attempt
RefDAC bit-current values
100
o
O
95
90
85
-*
-÷
80
*
0
75
I
0
I
uncalibrated
calibrated
I
4
6
Bit (Isb-to-msb)
I
8
Figure 5.5: RefDAC bit currents,final
1200
99
5.1. ELECTRICAL MEASUREMENTS
RefDAC INL
-
1.
0.
.0
I
-J
z
-0.
1
iL
200
400
600
code
800
1000
Figure 5.6: RefDAC INL, final
ure 4.22 is shared by 256 current-copier inputs and if left floating, its voltage drops rapidly
due to parasitic junction leakage currents. In the worst case scenario with VHOLD set to
zero, retention time for 10b accuracy is less than
ms. When n7 was left floating, results
were similar to the line for VHOLD=SOmvand 10b retention time is a few milliseconds. As
higher VHOLDis applied, leakage current is reduced. With VHOLD> 300mV, the output
current is maintained with 10b accuracy for 5OOms at room temperature. In this case, sub-
threshold conduction is reduced to negligible levels and junction leakage from source/drain
diffusions dominates leakage from the storage node.
The RefDAC achieves its design goal of producing highly accurate output currents. The
fully segmented architecture improves linearity by causing systematic errors to appear as
gain error instead of nonlinearity. Imprecise cancellation of charge injection is the primary
source of error, and could be improved in a future manufacturing run by resizing the switch
and dummy devices. The retention time improvement is effective and the RefDAC can
maintain 10b precision for 500ms.
100
CHAPTER 5. MEASURED RESULTS
RefDAC Iref9 VHOLD sweep
0.98
- 0.96
Q 0.94
N
E
Z 0.92
0.9
0.88
0
100
200
300
Time milliseconds (ms)
400
Figure 5.7: RefDAC retention test
---A~
RefDAC Iref9 VHOLD=300mV Zoom
I.UUUV
1.0004
1.0003
1.0002
i
i 1.0001
1
i
. 0.9999
0.9998
0.9997
0.9996
n Qar
0
100
200
300
Time milliseconds (ms)
400
Figure 5.8: RefDAC retention test, zoomed in
5.1. ELECTRICAL MEASUREMENTS
101
C
from
cope
Figure 5.9: Transimpedance amplifier circuit
5.1.2
ArrayDAC
To measure the performance of the ArrayDACs, column 359 which drives an output pin
instead of a bitline is used. The output current is connected into an op-amp based transimpedance amplifier, and the voltage output connected to an oscilloscope. Averaging is
used to improve measurement precision. The transimpedance amplifier circuit is shown
in Figure 5.9. Input bias current must be minimized because of the small currents being
measured (100nA LSB). The OP-42EJ was chosen for its combination of low input bias
current (200pA max), high gain, and fast settling time.
The RefDAC is calibrated, then RefDAC outputs are used to calibrate the ArrayDAC.
The ArrayDAC is driven with a digital ramp pattern, and its outputs recorded by the oscilloscope.
In order to reduce transient effects, the period of each step is 100OOs.Data
from the oscilloscope are transferred to a computer and analyzed in MATLAB. Results are
normalized to remove gain and offset errors.
The ArrayDAC has 7b precision without calibration, and calibration improves its precision to 8.5b. The lower precision is due to the binary-weighted structure of the ArrayDAC's
as an area saving measure. In the layout, a single layout block containing the passgate and
charge injection cancellation transistors was designed for the ArrayDAC. The same cell
is used in all ArrayDAC copiers. A fixed amount of imbalance between the passgate and
charge injection cancellation devices results in the same amount of charge injection in all
102
CHAPTER 5. MEASURED RESULTS
ArrayDAC INL
200
400
600
800
1000
code
Figure 5.10: ArrayDAC INL
copiers. Because the ten current copier cells carry different currents, the transconductances
of the copiers are all different, and the same amount of charge error has a different effect
on each copier, and causes nonlinearity. This is in contrast to the RefDAC, where all cells
are designed to be identical and a fixed offset does not cause nonlinearity.
Retention time tests for the ArrayDAC show similar trends to the RefDAC. The ArrayDAC maintains 10b precision for 60ms at room temperature as shown in Figure 5.11. The
data for normalized current in Figure 5.11 start at a value greater than unity because of overshoot in the oscilloscope step response. The lower retention time is due to the use of PMOS
devices in the ArrayDAC in contrast to NMOS devices in the RefDAC. The substrate and
n-well doping levels are different, resulting in different leakage current densities.
To summarize, at room temperature the RefDAC requires calibration once every 500ms
and the ArrayDACs must be calibrated once every 60ms. These times are much longer than
the conventional case without the retention time improvement technique.
5.1. ELECTRICAL MEASUREMENTS
103
mcr6yb.m arraydac leakage 30 aug 04 achen
I .UUU3
0.9995
0.999
0
20
40
60
time (ms)
80
100
Figure 5.11: ArrayDAC retention test
5.1.3
Column Variation
Variation between the ArrayDAC column drivers results in fixed pattern noise, and must be
minimized. The amount of inter-column variation was quantified before and after calibration.
In this experiment, 357 ArrayDACs were set to drive the same output value (1/4 scale =
256LSB). The output current of each ArrayDAC was measured using the transimpedance
amplifier located at the top of each column and an oscilloscope. Without calibration, the
mean current is 217LSB and the standard deviation is 1.2%. The standard deviation is a
measure of fixed pattern noise. With calibration the mean current is 256LSB (by definition),
and the standard deviation is reduced to 0.55%. Fixed pattern noise has been reduced by a
factor of two. The results are plotted in Figure 5.12.
Another way of viewing the same data is to compare the output of each ArrayDAC
before and after calibration. This is plotted in Figure 5.13. Ideally, the calibrated output
currents should be constant and independent of uncalibrated current. This would appear
as a horizontal line on the graph. The measured results show a weak correlation between
uncalibrated and calibrated currents (m=0.15, b=17.8/uA, and r=-0.33). This weak correlation suggests that the calibration technique is functioning, and improving the uniformity
between columns in the display.
CHAPTER 5. MEASURED RESULTS
104
ArrayDAC currents
45
40
35
30
25
20
15
10
5
0
210
220
230
240
250
260 LSB
Figure 5.12: ArrayDAC output current distribution
x 10 - 5 mamp3a2 scatterplot 12 nov 04 achen
2.1
2.08
03
2.06
E
1 2.04
a(b
, 2.02
-o
2
1.98
1.98
1.96
1 QA
1.6
1.65
1.7
1.75
1.8
uncalibratedcurrent - amps
1.85
x 10-
Figure 5.13: ArrayDAC outputs before and after calibration
5.2. OPTICAL MEASUREMENTS
105
Figure 5.14: Sample image for pixel luminance analysis
5.2
Optical Measurements
Optical measurements were performed to measure the uniformity between pixel luminances. Measurements were taken using a sensitive CCD camera mounted on a microscope,
and with a photometer.
Microscope and CCD
For the microscope setup, the PCB with silicon IC backplane was placed on the microscope
platen. Drivers were set to maximum current, and the output current time-multiplexed
between sixteen rows. Only sixteen rows were evaluated at a time in order to increase the
duty cycle, and reduce the long integration times required by the CCD camera.
The microscope is a Nikon Eclipse L200, used with a 20x/0.45 NA objective lens. A
Qimaging Retiga 1300 camera is attached to the camera port, and controlled by a computer with QCapture software. For each measurement, two images are taken to implement
correlated double sampling (CDS). First, an image is captured with the silicon backplane
emitting light. Then, the VLOW supply is turned off, and an image of background illumination captured. The second image is subtracted from the first to remove background
illumination and fixed noise from the imager (hot pixels). A sample image showing the top
left comer of the pixel array is in Figure 5.14.
This image is divided into colums (16 pixels each), and statistics are computed for variation within the column and variation between columns. Ideally, pixel luminance within
a column should be constant because all pixels share the same driver circuit. Measurements show there is signficant pixel-to-pixel variation due to manufacturing variation. In
Figure 5.15, the standard deviation of pixel luminance is plotted for each of 24 measured
columns, and averages 6.8%.
106
CHAPTER 5. MEASURED RESULTS
Pixel Variation vs. Column - i8889u
'u
-- ·
9
8
75
'a
4
cl
0
5
10
15
20
25
column
Figure 5.15: Pixel luminance variation by column
The measured variation is the sum of multiple independent sources. These include pixel
luminance variation which we want to measure, variation from the imaging process (optical
transmission, camera artifacts), and variation from image recognition and processing.
The microscope and optics were tested with a uniform test target. Uniformity was
within a 5% range across the entire image. The 5% figure includes light falloff, dust, and
in-camera variation. This suggests
UOPTICS
is about 1%. Variation in pixel recognition
occurs because the boundary of a pixel, and thus its area, may be interpreted differently by
different identification methods. For example, a digitized image was loaded into MATLAB
and the boundary of a single pixel was manually identified a number of times. The result
of ten trials was 9SOFTWARE
=
1.7%. Assuming these effects are independent, OcPIXEL is
6.5%.
This result is useful for characterizing the display, unfortunately the amount of pixel
luminance variation makes it difficult to quantify the difference between calibrated and
uncalibrated driver circuits. The measurement precision needed to measure 8.5b driver
circuits is less than 0.5%, which is much smaller than the variation due to other sources. In
addition, a gradient was observed in the mean column luminances as shown in Figure 5.16,
which may have been caused by focus error.
Photometer
For photometer measurements, the PCB with silicon backplane IC was rigidly attached to
an X-Y micrometer stage. The X-Y stage was placed on top of a Z-stage. This combination
allowed precise control of X-Y position and focusing.
107
5.2. OPTICAL MEASUREMENTS
Mean Pixel Luminance vs. Column - i8889u
._d
a0
xCa
a)
0
5
10
15
20
25
column
Figure 5.16: Column luminance
The PhotoResearch PR-880 photometer was used with MS-lOx lens. When used with
the 1/4 degree aperture, the measurement area is a 25,um diameter.
The photometer is
capable of measuring the luminance of a single pixel on the IC. The photometer with lens
are mounted on a sturdy tripod with lens pointed downward at the IC. Light shields made
of dark colored paper were used to reduce background illumination. To reduce vibration,
the photometer was operated via RS-232 serial link to a computer.
In a room with dim light, the light emission from silicon was visible through the
viewfinder. This was encouraging to observe, and also aided in aiming the photometer.
With the pixels off, background luminance was 1.04 * 10 -
3
(arbitrary units) with a standard
deviation of 1.4 * 10 - 4 . With the full-intensity test pattern, measurements were around
1.3 * 10-2, so background uncertainty CrBACKGND= 1%. Measuring a single pixel repeatedly indicates same-pixel measurement variation is UMEAS = 2.4%. The variation
between six illuminated pixels in a single column was aCOL = 6.9%. This suggests pixel
luminance variation alone is 6.4%, which agrees with the result in the previous section.
Summary
Two sets of measurements using different equipment indicate that the standard deviation of
pixel luminance variation for a constant drive current is 6%. The equipment used consisted
of a CCD camera mounted on a precision microscope, and a PR-880 photometer. Measurement of ArrayDAC performance via optical measurements is difficult because the pixel
variation is much larger than the 8-10b precision levels achieved by the electrical circuits.
Although this variation level seems large, it is uncorrelated with position. Randomly
CHAPTER 5. MEASURED RESULTS
108
distributed variation is smoothed by the eye as long as there are no low-frequency artifacts such as fixed pattern noise. We have shown that calibration improves the matching
between column drivers to reduce fixed-pattern noise. Electrical measurements show this
result clearly; optical measurements with larger sample sizes should also support this con-
clusion. Pixel variation could be reduced with the addition of sensors and feedback circuitry. Photodetectors in each pixel would enable the use of analog feedback techniques
[107], or digital correction techniques to reduce pixel variation.
5.3
Power Measurements
Power is dominated by the current supplied to the pixels, where the maximum current with
all pixels on is 360 * 100,uA = 36mA. This current flows to the VLOW supply at -11 volts,
so total power is 400mW. Calibration of the RefDAC and ArrayDAC is much smaller;
the RefDAC uses two reference currents: 85nA and 100A,
and the ArrayDAC uses the
RefDAC outputs totalling 102gA.
5.4 Sample Images
A sample image consisting of a series of 7b grayscale ramps is shown in Figure 5.17. This
is a long-exposure image of the silicon IC, taken with a sensitive CCD camera. It illustrates
light emission from silicon. The banding and irregularity in this image are caused by postprocessing; the image from the camera was very smooth. For this test pattern, the row time
was 100ps and frame rate was controller-limited at 22fps.
The proof-of-concept system with the calibrated backplane IC is shown in Figure 5.18.
The IC is mounted on a printed circuit board at bottom. A microscope objective is used to
couple light from the IC to the input window of the image intensifier. The image intensifier
is located on top of an optical tube. A zoomed-in portion of the ramp test image is visible
in the image intensifier's output window.
An alternative setup of the proof-of-concept system is shown in Figure 5.19. The mi-
croscope objective lens is replaced with a lens intended for video that provides lower magnification. This makes the entire display is visible in the intensifier output window. Hardware was machined from aluminum stock to position the lens and intensifier. The lens is a
1:1.4/25mm Schneider-Kreuznach Xenon in CS-mount. It is mounted on the bottom side
of the aluminum bar with extension tubes. The intensifier is mounted on top of the bar, and
the "ramps" test image can be seen in the output window.
5.4. SAMPLE IMAGES
Figure 5.17: Silicon backplane IC emitting light, showing a series of ramps
Figure 5.18: Proof-of-concept system with calibratedbackplane
109
110
CHAPTER 5. MEASURED RESULTS
Figure 5.19: Proof-of-conceptsystem with calibratedbackplane and alternative optics
Chapter 6
Discussion
6.1
Summary
A microdisplay architecture using silicon light emitters and image intensification is designed and evaluated. A standard low-voltage CMOS IC incorporating display drivers and
an array of avalanche diodes produces a faint optical image, and an image intensifier efficiently amplifies the image to useful brightness.
This architecture has high efficiency
and the potential to achieve high brightness suitable for micro-projection. Calculations
indicate that a high-resolution monochrome micro-projector can achieve an efficiency of
7 lumens/watt.
A proof-of-concept system with 16x32 pixel arrays is implemented and
evaluated.
A CMOS-based microdisplay backplane for use in conjunction with the above system
is designed, implemented, and tested. The backplane is a standard CMOS die including
pixels with avalanche diodes, and current-mode driver circuits. Current-mode driver circuits support a number of emissive display technologies including silicon light emitters and
organic LED (OLED). The integrated display drivers employ a self-calibration technique
to minimize variation while reducing circuit area.
Area occupied by display driver circuits has been reduced a factor of five to seven
compared to a conventional design. Calibration improves RefDAC precision from 8b to
11.5b, and ArrayDAC precision from 7b to 8.5b. ArrayDAC precision could be further
improved by adjusting the sizes of charge injection cancellation devices.
Circuit techniques to improve retention time in current copiers have been developed.
Retention time of the RefDAC and ArrayDAC have been improved from a few milliseconds
to 500ms and 60ms at room temperature, respectively.
Variation in column driver (ArrayDAC) outputs, which produces fixed pattern noise,
has been reduced by more than a factor of two. Pixel luminance variation is measured
111
CHAPTER6. DISCUSSION
112
ArrayDACs
IN .
7
:9 0
REF<
>
Ref DAC
Si
i- i
11
V LOW
OLED
01
v/K
Figure 6.1: RefDAC, ArrayDAC, and pixel arrangement
at 6.4%, which is comparable to other display technologies currently in use and could be
reduced by the use of additional calibration or feedback techniques.
6.2
Topics for Future Investigation
Improving Calibrated Precision
The reason for the reduced precision of the ArrayDACs is an offset in the charge injection
cancellation scheme as explained in Section 5.1.2. In a textbook example, the dummy device is exactly half the size of the passgate. When the passgate switches, exactly half of the
channel charge and overlap capacitance couples to the storage node, and the dummy device
cancels this coupling. In practice, size mismatches and differences in node impedances affect the distribution of charge. In both the RefDAC and ArrayDAC circuits, there is an
imbalance between the passgate and dummy device sizes, resulting in uncancelled charge
injection. This problem could be fixed by resizing the passgate and dummy device transistors for better cancellation. If re-sizing is not feasible, the voltage used to control charge
cancellation devices could be adjusted. This would allow fine adjustment of the amount of
charge injected by cancellation devices.
The retention time measurements show that the NMOS current copiers used in the RefDAC have a much longer retention time than the PMOS current copiers in the ArrayDACs.
The present arrangement of RefDAC, ArrayDAC, and pixels and the polarity of currents is
shown in Figure 6.1. To improve calibration performance, the RefDAC should be implemented with PMOS devices and ArrayDACs with NMOS devices, as shown in Figure 6.2.
The RefDAC only needs to maintain high precision for long enough to calibrate the ArrayDACs, or about 25ms. In contrast, the ArrayDAC retention time must be maximized
6.2. TOPICSFOR FUTUREINVESTIGATION
113
VHI
VA
RefDAC
Si
IIN i
I
OLED
N
~REF<9 : 0 >
ArrayDACs
Figure 6.2: Alternative RefDAC, ArrayDAC, and pixel arrangement
in order to minimize the amount of time spent in calibration which forces the array to be
dark. Changing the circuits to reverse current polarities of the RefDAC and ArrayDACs is
straightforward.
An effect of this change is the common terminal of the light emitters is
now the anode, and compatibility with OLED materials may be affected.
Technology Choices
Because of the dynamic nature of the MOS sample-and-hold structure, calibration must be
done at frequent intervals. The use of non-volatile analog memory could greatly reduce the
frequency of calibration [108]. The backplane could be calibrated infrequently, or only in
response to predetermined events such as power-on reset or temperature changes. Dualgate CMOS technologies that accommodate floating-gate memory structures are available,
and commercial applications are appearing [109].
The 360x200 CMOS backplane IC supports both silicon light emitters and organic LED
technologies. This document has focused on the silicon light emitters. Manufacturing
processes to deposit OLED materials on silicon ICs are becoming stable and available,
and an OLED-on-silicon demonstration system can be assembled. This would demonstrate
the versatility of the circuit techniques implemented on the backplane IC. The circuits
described in this work may eventually be implemented with organic transistors, leading to
a fully integrated display based on organic semiconductors [110].
The circuit techniques in this document can also be used in conjunction with field emitters integrated on a silicon substrate. A silicon IC with field emitters [111] could replace the
photocathode in an image intensifier structure. It would directly emit electrons instead of
converting junction currents to light on the IC, and light to electrons in the image intensifier.
114
CHAPTER 6. DISCUSSION
Appendix A
Precision Oscilloscope Measurements
This section describes some of the issues involved in obtaining precise voltage measurements using an oscilloscope. Traditional oscilloscopes provide measurement precision adequate for a medium resolution display ( 8 bits), with good timing accuracy. Newer
oscilloscopes with data memory and computational ability can perform data averaging to
reduce noise and improve voltage accuracy. Probes and input circuitry designed for moderate precision can be a limiting factor in this type of system.
First, four high-impedance passive probes were tested on the same input channel of a
Hewlett-Packard 54532A oscilloscope which had not been calibrated in a number of years.
The response to an 0.8v, ms square pulse is shown in Figure A. 1. Three HP 10441A probes
produced similar results, and a probe from another manufacturer is clearly not matched to
the input characteristics of the oscilloscope.
Removing the worst of the group, the step responses of the remaining three probes is
shown in Figure A.2, zoomed in show high precision. An unusual transient behavior with
long settling time is seen.
The four input channels were sequentially connected to the same signal source and
probe. Figure A.3 shows the variation in the responses between different channels. Channel
two appears to have the fastest settling time of the four. Expanding the pulse width and time
base reveal that the performance is actually worse, as seen in Figure A.4. The settling time
is on the order of 2-3ms, and the previous waveforms were far from their final values.
Tuning the passive probe, or using a BNC cable connected to oscilloscope input did not
significantly improve the long settling time. Calibration was needed, however measurements were needed in a short amount of time. As a temporary solution a newer, recently
calibrated oscilloscope was borrowed. The newer scope was an Agilent 54832D mixedsignal oscilloscope.
Used with the same passive probes as before, the step response had
a faster settling time as shown in Figure A.5. Measurement performance was greatly improved compared to the previous oscilloscope, and this setup was used for most measurements presented in this document. Settling time for a large voltage step is a few hundred
microseconds for 10b (0.1%) precision. Critical measurements were performed at appropriate speeds for high measurement accuracy.
To summarize, both the oscilloscope and probes need to be calibrated periodically to
produce precision results. Long transient responses can result when equipment is mismatched out of calibration.
115
APPENDIXA. PRECISIONOSCILLOSCOPEMEASUREMENTS
116
probe1 .m hp54542a cal-out,probe overshoot CH1 19 aug 04 achen
0.
0.
o
C
0
o-0.
-a
-0.
_n
-v.
vu
0
0.2
0.4
0.6
Time milliseconds
0.8
1
FigureA.1: Step responses offour oscilloscope probes
probe1 .m hp54542a cal-out,probe
^ ^^^
-O.UUY
settling CH1 19 aug 04 achen
-0.01
o -0.011
o
( -0.012
c -0.013
0
a-
_-0.01
4
a,
no-0.015
-0.016
-_n
n 7
-v.v,,0
0
0.2
0.4
0.6
Time milliseconds
0.8
Figure A.2: Probe settling behavior
1
117
channels.m hp54542a cal. signal 19 aug 04 achen
-u.u.C
0
0.2
0.4
0.6
0.8
1
time milliseconds
Figure A.3: Four oscilloscope channels, same input
mlong1.m pulsegen to bnc to scope 19 aug 04 achen
1.93
1.925
1.92
1.915
,
1.91
c
° 1.905
-.
1.9
1.895
1.89
1.885
0
1
2
3
4
time milliseconds
Figure A.4: Longer observation time
5
6
APPENDIXA. PRECISIONOSCILLOSCOPEMEASUREMENTS
118
mmsocall .m hp54832d calibration 256x 19 aug 04 achen
1.04
.o
>
0
1.035
c
a
CL
1.03
in
1.025
0
0.5
1
1.5
2
2.5
time milliseconds
Figure A.5: Newer oscilloscope step response
3
Bibliography
[1] A. I. Akinwande, "Course notes for 6.976 flat panel display devices." Lecture 1,
2001.
[2] D. E. Mentley, "State of flat-panel display technology and future trends," Proceedings of the IEEE, vol. 90, no. 4, 2002.
[3] C. W. McLaughlin, "Progress in projection and large-area displays," Proceedings of
the IEEE, vol. 90, no. 4, 2002.
[4] C. G. Hwang, "Semiconductor memories for IT era," in International Solid-State
Circuits Conference, pp. 24-27, IEEE, February 2002.
[5] N. N. Chubun, A. G. Chakhovskoi, and C. E. Hunt, "Efficiency of cathodolumines-
cent phosphors for a field-emission light source application," Journal of VacuumScience and TechnologyB: Microelectronics and Nanometer Structures, vol. 21, no. 4,
pp. 1618-1621, 2003.
[6] C. E. Rash, ed., Helmet Mounted Displays: Design Issues for Rotary-Wing Aircraft. Fort Rucker, Alabama: United States Army Aeromedical Research Laboratory, 1998.
[7] C. H. Chen and A. R. L. Travis, "Wide field of view optics for flat panel 3D displays,"
in Society for Information Display 1999 International Symposium, pp. 1004-1007,
May 1999.
[8] C. Chinnock, "Microdisplay innovation driving optical switch development," Electronic Design, vol. 49, no. 5, p. 64, 2001.
[9] Brillian Corporation Annual Report on Form 10-K for the Fiscal Year Ended December 31,2003.
[10] W. J. Hossack, E. Theofanidou, J. Crain, K. Heggarty, and M. Birch, "High-speed
holographic optical tweezers using a ferroelectric liquid crystal microdisplay," Optics Express, vol. 11, no. 17, pp. 2053-2059, 2003.
[11] M. Bolotski and P. Alvelda, "Low-power miniaturized information display systems,"
in International Symposium on Low Power Electronics and Design, (Arlington, Virginia), pp. 279-281, August 1998.
119
BIBLIOGRAPHY
120
[12] J. Ammer, M. Bolotski, P. Alvelda, and J. T. F. Knight, "A 160x120 pixel liquid-
crystal-on-silicon microdisplay with an adiabatic dacm" in International Solid-State
Circuits Conference, (San Francisco, California), pp. 212-213, February 1999.
[13] M. J. Ammer, A Highly Integrated Adiabatic Energy Recovery Digital to Analog
Converter. PhD thesis, Massachusetts Institute of Technology, 1999.
[14] P. V. Kessel, L. Hornbeck, R. Meier, and M. R. Douglass, "A MEMS-based projection display," Proceedings of the IEEE, vol. 86, pp. 1687-1704, August 1998.
[15] W. E. Howard and 0. F. Prache, "Microdisplays based upon organic light-emitting
diodes," IBM Journal of Research and Development, vol. 45, no. 1, 2001.
[16] 0. Prache, "Ful color SVGA+ OLED on silicon microdisplay," in SID International
Symposium, pp. 514-517, Society for Information Display, 2001.
[17] W. Graupner, C. M. Heller, A. P. Ghosh, and W. E. Howard, "High resolution color
organic light emitting diode microdisplay fabrication method," Proceedings of SPIE
- The International Societyfor Optical Engineering, vol. 4207, pp. 11-19, 2000.
[18] Fraunhofer-Gesellschaft,
Research News 11, Topic 6, 11/2004, Vest-pocket beamer.
[19] M. H. Keuper, S. Paolini, and G. Harbers, "Ultra-compact LED based image projec-
tor for portable applications," in Society for Information Display 2003 Symposium
Digest, vol. 34, pp. 713-715, May 2003.
[20] V. M. Bove and W. Sierra, "Personal projection, or how to put a large screen in a
small device," in Society for Information Display 2003 Symposium Digest, vol. 34,
pp. 708-711, May 2003.
[21] H. Urey, "Optical advantages of retinal scanning displays," in Proceedings of SPIE -
The International Society for Optical Engineering, vol. 4021, pp. 20-26, April 2000.
[22] S. Marti, "TinyProjector." http://web.media.mit.edu/
stefanm/TinyProjector,
1/2005.
[23] R. S. Johnston and S. Willey, "Development of a commercial virtual retinal dis-
play," in Proceedings of Helmet- and Head-Mounted Displays and Symbology Design (W. Stephens and L. A. Haworth, eds.), pp. 2-13, 1995.
[24] R. Sainsbury and J. D. Clevenger, "High performance, versatile HMDs," in Cockpit
Displays X, vol. 5080, pp. 49-57, SPIE, April 2003.
[25] M. Billings, "CRT replacement for the next generations head-up display," in Pro-
ceedings of SPIE Vol.4022 - Cockpit Displays VII: Displays for Defense Applications, pp. 411-421, 2000.
[26] A. E. Gamal, "Trends in CMOS image sensor technology and design," in Electron
Devices Meeting, pp. 805-808, December 2002.
BIBLIOGRAPHY
121
[27] R. Newman, "Visible light from a silicon p-n junction," Physical Review, vol. 100,
no. 2, pp. 700-703, 1955.
[28] A. G. Chynoweth and K. G. McKay, "Photon emission from avalanche breakdown
in silicon," Physical Review, vol. 102, no. 2, pp. 369-376, 1956.
[29] L. J. Kabell and C. J. Pecoraro, "Silicon avalanche light sources for photographic
data recording," in Optical and Electro-optical Information Processing (J. T. Tippett,
ed.), Cambridge, MA: MIT Press, 1965.
[30] J. A. Kash and J. C. Tsang, "Dynamic internal testing of CMOS circuits using hot
luminescence," IEEE Electron Device Letters, vol. 18, no. 7, p. 330, 1997.
[31] R. H. Haitz, "Studies on optical coupling between silicon p-n junctions," Solid-State
Electronics, vol. 8, pp. 417-425, 1965.
[32] L. C. Kimerling, "Silicon microphotonics," Applied Surface Science, vol. 159, pp. 813, 2000.
[33] J. Kramer, P. Seitz, E. Steigmeier, H. Auderset, and B. Delley, "Light-emitting devices in industrial CMOS technology," Sensors andActuatorsA, vol. 37-38, pp. 527533, 1993.
[34] L. W. Snyman, H. Aharoni, and M. DuPlessis, "Indirect bandgap semiconductor
optoelectronic device," United States Patent #5994720, 1999.
[35] L. W. Snyman, H. Aharoni, M. duPlessis, and R. Gouws, "Increased efficiency of
silicon light-emitting diodes in a standard 1.2um silicon complementary metal oxide
semiconductor technology," Optical Engineering, vol. 37, no. 7, pp. 2133-2141,
1998.
[36] B. Zheng, J. Michel, F. Ren, and L. C. Kimerling,
"Room-temperature
sharp
line electroluminescence at lambda=1.54um from an erbium-doped, silicon lightemitting diode," Applied Physics Letters, vol. 64, pp. 2842-2844, 23 May 1994.
[37] W. P. Giziewicz and C. G. Fonstad, "Optoelectronic integration using aligned metalto-semiconductor bonding," J. Vacuum Science and Technology A, vol. 20, no. 3,
pp. 1052-1056, 2002.
[38] M. A. Green, J. Zhao, A. Wang, P. J. Reece, and M. Gal, "Efficient silicon lightemitting diodes," Nature, vol. 413, pp. 805-808, 23 August 2001.
[39] W. L. Ng, M. Lourenco, R. M. Gwilliam, S. Ledain, G. Shao, and K. Homewood,
"An efficient room-temperature silicon-based light-emitting diode," Nature, vol. 410,
pp. 192-194, 8 March 2001.
[40] J. Perkins, J. Rumpler, and C. G. Fonstad, "Magnetically assisted self assembly - a
new heterogeneous integration technique," in MIT Microsystems Technology Laboratories 2002 Annual Report, 2002.
BIBLIOGRAPHY
122
[41] H. Aharoni and M. du Plessis, "Low-operating-voltage
integrated silicon light-
emitting devices," IEEE Journal of Quantum Electronics, vol. 40, pp. 557-563, May
2004.
[42] G. F. Marshall, J. C. Jackson, J. Denton, P. K. Hurley, O. Braddell, and A. Mathewson, "Avalanche photodiode-based active pixel imager," IEEE Transactions on
Electron Devices, vol. 51, no. 3, pp. 509-511, 2004.
[43] A. Spinelli and A. Lacaita, "Physics and numerical simulation of single photon
avalanche diodes," IEEE Transactions on Electron Devices, vol. 44, pp. 1931-1943,
November 1997.
[44] C. Niclass and edoardo Charbon, "A single photon detector array with 64x64 resolution and millimetric depth accuracy for 3d imaging," in International Solid-State
Circuits Conference, vol. 48, pp. 364-365, February 2005.
[45] H.-S. Wong, "Technology and device scaling considerations for CMOS imagers,"
IEEE Transactions on Electron Devices, vol. 43, no. 12, pp. 2131-2142, 1996.
[46] P. Ecker, MCP Detector Systems (online documentation).
http://www.proxitronic.de/prod/omcp/eos.htm,
Proxitronic,
April 2002.
[47] L. E. Shea, "Low-voltage cathodoluminescent phosphors: A 20-year chronology of
low-voltage cathodoluminescence efficiency," Interface, pp. 24-27, June 1998.
[48] C. Bower, D. Shal6m, D. L6pez, et al., "A micromachined vacuum triode using a
carbon nanotube cold cathode," IEEE Transactions on Electron Devices, vol. 49,
pp. 1478-1483, August 2002.
[49] H. Luo, D. Ban, H. C. Liu, P. J. Poole, and M. Buchanan, "Pixelless imaging device
using optical up-converter," IEEE Electron Device Letters, vol. 25, pp. 129-131,
March 2004.
[50] S. H. Lo, D. A. Buchanan, Y. Taur, and W. Wang, "Quantum-mechanical
modeling
of electron tunneling current from the inversion layer of ultra-thin-oxide nmosfets,"
IEEE Electron Device Letters, vol. 18, pp. 209-211, May 1997.
[51] Y.-S. Lin, C.-C. Wu, and C.-S. C. et al., "Leakage scaling in deep submicron CMOS
for SoC," IEEE Transactions on Electron Devices, vol. 49, pp. 1034-1041, June
2002.
[52] K. Roy, H. Mahmoodi-Meimand,
and S. Mukhopadhyay, "Leakage control for deep-
submicroncircuits," Proceedingsof SPIE - The InternationalSocietyfor Optical
Engineering, vol. 5117, pp. 135-146, 2003.
[53] A. Agarwal, H. Li, and K. Roy, "A single-vt low-leakage gated-ground cache for
deep submicron," IEEE Journal of Solid-State Circuits, vol. 38, pp. 319-328, February 2003.
BIBLIOGRAPHY
123
[54] M. O'Halloran and R. Sarpeshkar, "A 10nw 12bit accurate analog storage cell with
10aa leakage," IEEE Journal of Solid-State Circuits, vol. 39, pp. 1985-1996, November 2004.
[55] K. R. Lakshmikumar, R. A. Hadaway, and M. A. Copeland, "Characterization
and
modeling of mismatch in mos transistors for precision analog design," IEEE Journal
of Solid-State Circuits, pp. 1057-1066, December 1986.
[56] P. G. Drennan and C. C. McAndrew, "Understanding mosfet mismatch for analog
design," IEEE Journal of Solid-State Circuits, vol. 38, pp. 450-456, March 2003.
[57] D. Johns and K. Martin, Analog Integrated Circuit Design, pp. 454-460. New York,
New York: John Wiley and Sons, Inc, 1997.
[58] K. Bult and G. Geelen, "An inherently linear and compact MOST-only current divi-
sion technique," IEEE Journal of Solid-State Circuits, vol. 27, pp. 1730-1735, Dec.
1992.
[59] C. Bastiaansen, D. Wouter, J. Groeneveld, et al., "A 10-b 40-MHz 0.8um CMOS
current-output D/A converter," IEEE Journal of Solid-State Circuits, vol. 26,
pp. 917-921, July 1991.
[60] A. Van den Bosch, M. Boremans, M. Steyaert, and W. Sansen, "A 10-bit 1-
GSample/s Nyquist current-steering CMOS D/A converter," IEEE Journal of SolidState Circuits, vol. 36, pp. 315-324, March 2001.
[61] S. R. Norsworthy, R. Schreier, and G. C. Temes, eds., Delta-Sigma Data Converters
- Theory, Design, and Simulation. Piscataway, NJ: IEEE Press, 1997.
[62] H.-S. Lee, D. A. Hodges, and P. R. Gray, "A self-calibrating 15 bit CMOS A/D con-
verter," IEEE Journal of Solid-State Circuits, vol. SC-19, pp. 813-819, December
1984.
[63] Y. Cong and R. L. Geiger, "A 1.5-v 14-bit 100-Ms/s self-calibrated DAC," IEEE
Journal of Solid-State Circuits, vol. 38, pp. 2051-2060, December 2003.
[64] A. Karanicolas, H.-S. Lee, and K. Bacrania, "A 15b Ms/s digitally self-calibrated
pipeline adc," in International Solid-State Circuits Conference, pp. 60-61, IEEE,
February 1993.
[65] B. Linares-Barranco, T. Serrano-Gotarredona, and R. Serrano-Gotarredona, "Compact low-power calibration mini-DACs for neural arrays with programmable
weights," IEEE Transactions on Neural Networks, vol. 14, pp. 1207-1216, September 2003.
1-66]R. J. van de Plassche, "Dynamic element matching for high accuracy monolithic d/a
converters," IEEE Journal of Solid-State Circuits, pp. 795-800, December 1976.
BIBLIOGRAPHY
124
[67] L. R. Carley, "A noise-shaping coder topology for 15+ bit converters," IEEE Journal
of Solid-State Circuits, vol. 24, pp. 267-273, April 1989.
[68] R. J. V. de Plassche and H. J. Schouwenaars, "A monolithic 14 bit A/D converter,"
IEEE Journal of Solid-State Circuits, vol. SC-17, pp. 1112-1117, December 1982.
[69] G. A. M. V. der Plas, J. Vandenbussche, W. Sansen, et al., "A 14-bit intrinsic accuracy Q2 random walk CMOS DAC," IEEE Journal of Solid-State Circuits, vol. 34,
pp. 1708-1718, December 1999.
[70] G. Wegmann and E. A. Vittoz, "Very accurate dynamic current mirrors," Electronics
Letters, vol. 25, pp. 644-646, May 1989.
[71] D. Wouter, J. Groeneveld, H. Schouwenaars, et al., "A self-calibration technique for
monolithic high-resolution D/A converters," IEEE Journal of Solid-State Circuits,
vol. 24, pp. 1517-1522, December 1989.
[72] W. Groeneveld, H. Schouwenaars, and H. Termeer, "A self-calibration technique
for monolithic high-resolution d/a converters," in International Solid-State Circuits
Conference, pp. 22-23, IEEE, February 1989.
[73] D. G. Nairn and C. A. Salama, "Ratio-independent current mode algorithmic analog-
to-digital converters," in International Symposium on Circuits and Systems, pp. 250253, IEEE, 1989.
[74] H. J. Schouwenaars, D. Wouter, J. Groeneveld, et al., "An oversampled multibit
CMOS D/A converter for digital audio with 115-db dynamic range," IEEE Journal
of Solid-State Circuits, vol. 26, pp. 1775-1780, December 1991.
[75] K. Waardenburg, K. O'Sullivan, H. Schouwenaars, and W. Groeneveld, "A dual
low-power continuous calibration d/a converter (cc-dac) for digital audio," in Inter-
national Conference on Consumer Electronics, pp. 290-291, IEEE, June 1991.
[76] A. R. Bugeja and B.-S. Song, "A self-trimming 14-b 100-MS/s CMOS DAC," IEEE
Journal of Solid-State Circuits, vol. 35, pp. 1841-1852, December 2000.
[77] T. Fiez, G. Liang, and D. Allstor, "Switched-current circuit design issues," IEEE
Journal of Solid-State Circuits, vol. 26, pp. 192-202, March 1991.
[78] C. Toumazou, J. Hughes, and N. Battersby, eds., Switched-Currents:
an analogue
technique for digital technology. IEE Circuits and Systems - Series 5, London,
United Kingdom: Peter Peregrinus Ltd., 1993.
[79] S. J. Daubert and D. Vallancourt, "Noise analysis of current copier circuits," in In-
ternational Symposium on Circuits and Systems, pp. 307-310, IEEE, 1990.
[80] C. Aust and D. S. Ha, "A low-power variable resolution analog-to-digital converter,"
in International ASIC/SOC Conference, (Arlington, Virginia), pp. 460-463, 2001.
BIBLIOGRAPHY
125
[81] A. T. K. Tang and C. Toumazou, "Novel self-calibrated high-speed D/A converter
using trimmable current sources," in Proceedings of the 1994 IEEE International
Symposium on Circuits and Systems, pp. 469-472, IEEE, 1994.
[82] J. Grinberg, W. P. Bleha, A. D. Jacobson, A. M. Lackner, G. D. Myer, L. J. Miller,
J. D. Margerum, L. M. Fraas, and D. D. Boswell, "Photoactivated birefringent liquidcrystal light valve for color symbology display," IEEE Transactions on Electron
Devices, vol. ED-22, no. 9, pp. 775-783, 1975.
[83] R. D. Sterling and W. P. Bleha, "Electronic cinema using ILA(r) projector tech-
nology," in Society for Information Display International Symposium, pp. 216-219,
May 1999.
[84] W. P. Bleha, "ILA(r) projector technology for electronic cinema applications," in
International Broadcasting Convention, IEEE, September 1997.
[85] J. B. Gao and J. W. et al., "Gallium arsenide photoaddressed infrared liquid crystal
spatial light modulator," in Proceedings of SPIE Vol4919 - Advanced Materials and
Devices for Sensing and Imaging, pp. 91-96, September 2002.
[86] A. Geissler, P. Gusssek, and F. Reichel, "Optically addressable liquid crystal spatial light modulator with single crystal cds photoconductor," Proceedings of SPIE
Volume3800 - Liquid Crystals III, pp. 130-135, October 1999.
[87] M. Stanley, M. A. Smith, and A. P. S. et al., "3D electronic holography display
system using a 100 mega-pixel spatial light modulator," in Proceedings of SPIE Vol.
5249- OpticalDesign and Engineering,pp. 297-308, 2004.
[88] C. Warde, A. D. Fisher, D. M. Cocco, and M. Y. Burmawi, "Microchannel spatial
light modulator," Optics Letters, vol. 3, pp. 196-198, November 1978.
[89] Hamamatsu Photonics K.K., Electron Tube Center, Proximity Focused Image Intensifier V7090 Series, March 1998.
[90] W. J. Smith, Modem Optical Engineering, 3rd Ed., pp. 222-223. New York, New
York: McGraw-Hill, 2000.
[91] Hamamatsu Photonics K.K., Electron Tube Center, Circular MCP and Assembly
Series, 1999.
[92] P. Jones, A. Tomita, and M. Wartenberg, "Performance of NCAP projection dis-
plays," SPIE Vol. 1456 - Large-Screen Projection, Avionic, and Helmet Mounted
Displays, pp. 6-14, 1991.
[93] J. L. Sanford and E. S. Schlig, "Direct view active matrix VGA OLED-on-
crystalline-silicon display," in Society for Information Display Symposium Digest,
pp. 376-379, 2001.
[94] Overview of lithium ion batteries, brochure by Panasonic Corporation, August 2003.
BIBLIOGRAPHY
126
[95] R. T. Smith, "Drive approaches for FED displays," in International Display Research
Conference, pp. F35-F41, Society for Information Display, 1997.
[96] E. H. Stupp and M. S. Brennesholtz, Projection Displays, pp. 27-37.
New York: John Wiley and Sons, Inc, 1999.
New York,
[97] G. B. Levy, W. Evans, J. Ebner, et al., "An 852x600 pixel OLED-on-silicon color mi-
crodisplay using CMOS subthreshold-voltage-scaling current drivers," IEEE Journal of Solid-State Circuits, vol. 37, pp. 1879-1889, December 2002.
[98] G. Landsburg, "Drivers for novel displays," Information Display, vol. 18, pp. 18-21,
August 2002.
[99] H. J. Schouwenaars, D. Wouter, J. Groeneveld, and H. Termeer, "A low-power stereo
16-bitCMOS D/A converter for digital audio," IEEE Journal of Solid-State Circuits,
vol. 23, pp. 1290-1297, December 1988.
[100] M. J. M. Pelgrom, A. C. J. Duinmaijer, and A. P. G. Welbers, "Matching properties
of MOS transistors," IEEE Journal of Solid-State Circuits, vol. 24, pp. 1433-1440,
October 1989.
[101] K. Nagase, S. ichi Ohkawa, M. Aoki, and H. Masuda, "Variation status in 100nm
CMOS process and below," in Proceedings of the 2004 International Conference on
Microelectronic Test Structures, pp. 257-262, IEEE, 2004.
[102] Maxim Semiconductor Application Note AN-693: IC's Boost Video Performance,
published March 2000.
[103] S.-Y. Oh, D. Ward, and R. Dutton, "Transient analysis of MOS transistors," IEEE
Journal of Solid-State Circuits, vol. SC-15, pp. 636-643, August 1980.
[104] F. Mu, A. Edman, and C. Svensson, "Digital multiphase clock/pattern generator,"
IEEE Journal of Solid State Circuits, vol. 34, pp. 182-191, February 1999.
[105] R. Ho, B. Amrutur, K. Mai, B. Wilburn, T. Mori, and M. Horowitz, "Applications of
on-chip samplers for test and measurement of integrated circuits," in Symposium on
VLSI Circuits, pp. 138-139, IEEE, 1998.
[106] Y. Zheng and K. L. Shepard, "On-chip oscilloscopes for noninvasive time-domain
measurement of waveforms in digital integrated circuits," IEEE Transactions on Very
Large Scale Integration (VLSI) Systems, vol. 11, pp. 336-344, June 2003.
[107] E. Lisuwandi, "Feedback circuit for organic led active-matrix display drivers," Master's thesis, M.I.T. Dept. of Electrical Eng. and Computer Sci., 2002.
[108] L. R. Carley, "Trimming analog circuits using floating-gate analog MOS memory,"
IEEE Journal of Solid-State Circuits, vol. 24, no. 6, pp. 1569-1575, 1989.
BIBLIOGRAPHY
127
[109] J. Hyde, T Humes, C. Diorio, M. Thomas, and M. Figueroa, "A 300-Ms/s 14-bit
digital-to-analog converter in logic CMOS," IEEE Journal of Solid-State Circuits,
vol. 38, pp. 734-740, May 2003.
[110] J. Kymissis, C. D. Dimitrakopoulos, and S. Purushothaman, "Patterning pentacene
organic thin film transistors," Journal of Vacuum Science and TechnologyB, vol. 20,
pp. 956-959, May/June 2002.
[111] C. Y. Hong, Intelligent Field Emission Arrays. PhD thesis, Massachusetts Institute
of Technology, 2003.
MITLibraries
Document Services
Room 14-0551
77 Massachusetts Avenue
Cambridge, MA 02139
Ph: 617.253.5668 Fax: 617.253.1690
Email: docs@mit.edu
http://libraries. mit. edu/docs
DISCLAIMER OF QUALITY
Due to the condition of the original material, there are unavoidable
flaws in this reproduction. We have made every effort possible to
provide you with the best copy available. If you are dissatisfied with
this product and find it unusable, please contact Document Services as
soon as possible.
Thank you.
Some pages in the original document contain
pictures or graphics that will not scan or reproduce well.