effects of radiation damage on scientific charge coupled devices

EFFECTS OF RADIATION DAMAGE ON SCIENTIFIC

CHARGE COUPLED DEVICES

by

Timothy D. Hardy 5

B.A.Sc., Simon Fraser University, 1994

THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE

i' REQUIREMENTS FOR THE DEGREE OF

MASTER OF APPLIED SCIENCE

in the School

of

Engineering Science -

O Timothy D. ~ a r d y 1997

SIMON FRASER UNIVERSITY

August 1997

All rights reserved. This work may not be reproduced in whole or in part, by photocopy or other means,

without permission of the author.

~ational Library u+u of Canada . Biblioth&que' nationale

du Canada

Acquisitions and Acquisitions et Bibliographic Services sewges bibliographiques ,

,395 Weltington Street 395. rue Wellington OttawaON K I A ON4 OmwaON K l A O N 4 Canada Canada

I . I

The author has granted a non- "

exclusive licence allowing the National Library of Canada to reproduce, loan, distribute or sell copies of tlus thesis in microform, paper or electronk formats.

L'auteur a accorde une licence non exclusive permettant a la Bibliotheque nationale du Canada de reproduire, prgter, distribuer ou vendre des copies de cette these sous la fonne de microfiche/film, de reproduction sur papier ou sur format electronique.

The author retains ownership of the L'auteur conserve la propriete du copyright in tlus thesis. Neither the droit d'auteur qui protege cette these. thesis nor substintial extracts fiom it Ni la these ni des extraits substantiels may be printed or othenclise de celle-ci ne doivent Etre imprimis reproduced without the author's ou autrement reprdduits sans son permission. autorisation.

Your h l 8 Volre reference

Our hle Notre reference

Name: Timothy D. Hardy

Degree: Master of lied Science (Engineering Science)

Title of thesis: Effects of radiation damage on scientific ,charge coupled devices

Examining committee:

Chair: Dr. Kamal Gupta, Associate Professor

Senior supervisor: Dr. M. Jamal Deen, Professor

- Supervisor: Dr. Andrew Rawicz, Professor

L ' // - Examiner: Dr. Glenn Chapman, ksociate Professor

Date approved: August 20, 1997

ABSTRACT

Charge coupled devices (CCDs) have had a revolutionary impact on the field of electronic imaging

and are in wide use today in areas from home video to optical astronomy. In some scientific appli-

cations, for example space missions, particle detection and some forms of medical imaging. the

CCD detector is subjected to harmful radiation present in the operating environment. We have

investigated the effects of damage to CCDs by high-energy proton radiation, of the type encoun-

.L+

tered by satellites in high earth orbit. The work was motivated by a satellite -.. astronomy mission

(Lyman Far Ultraviolet Spectrographic Explorer) requiring a CCD as a tracking device. ri

Our investigations encompassed three main performance characteristics of CCDs: dark current,

charge transfer efficiency (CTE), and read noise. We examined these characteristics and developed

theoretical models of them as functions of temperature, operating speed, and radiation dose. The

experiments were camed out on a thinned, backside-illuminated buAed channel CCD and a num-

ber of buried-channel MOS transistors similar to the output transistor of the CCD. These-devices

had been damaged by protons from a particle accelerator in such a way as fO simulate the radiation

environment of a high earth-orbit. The proton fluence was varied in order to investigate the pro-

gression of damage with time in orblt.

We- found that in general the damage was directly proportional to the proton fluence. The dark cur-

rent increased, the CTE decreased and the read noise increased proportionally to the radiation dose

received. These characteristics also showed an increased sensitivity to temperature. The dominant

factor in all c a e s was seen to be the introduction of bulk trapping states in the active region of the a

silicon. Models based on bulk trapping showed excellent agreement with the experimental data.

I would like to sincerely thank my supervisors, Prof. Jamal Deen of Simon Fraser University, and

* Rick Murowinski of the National Research Council, Dominion Astrophysical Observatory (DAO),

for their technical support, comments and suggestions. and for their assistance throughout this

project. They have made a very interesting research project for me, and have also supported my

attendance i t a conference to present part of this research. I am also grateful to the Canadian Space

Agency and Dr. John Hutchings, DAO for their support of this work, and to the National Research

Council, the Natural Sciences and Engineering Research council (NSERC) and Simon Fraser

University for partial financial support. I would also like .to thank the members of the Integrated

Devices and Circuits Laboratory at Simon Fraser University for their comments and suggestions

during our research meetings, and their support of my work. In particular, I ar;l grateful to Plamen

Kolev for assistance with the transistor measurements and for thqesults of his Deep Level Tran-

sient Spectroscopy (DLTS) experiments, and to Dr. Sergei Rumyantsev for a very helpful discus-

sion on measuring the low frequency noise in transistors. I also thank Professors Andrew Rawicz

and Glenn Chapman for being on my examining committee.

TABLE OF CONTENTS

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Approval ... 11

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abstract 111

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements iv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . List of Tables vii

. . . ' I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . List of Figures viii

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 1 . Introduction 1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . CCD development and current status 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Research motivation 12

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 2 . Device Structure and Operation 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charge generation 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charge collection 16

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chargk transfer 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charge detection '. 30

-+%

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 3 . Radiation Damage 35 Ionizationdamage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displacement damage 37 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bulk Trap Levels 39

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DLTS measurements . : 40 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Annealing 42

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . FUSE radiation environment 44

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 4 . Dark Current 49 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experimental results 57

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 5 . Charge Transfer Efficiency 66 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simple physical model 68 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Measurement techniques 73

. . . . . . . . . . . . . . . . . . . . . . . . . . Experimental results ; . . . . . . . . . . . . . . 79

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 6 . Read Noise 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Noise sources 90

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlated double sampling 93 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experimentd results 102

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 7. Conclusions. 1 17

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References 121

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . List of Abbreviations and Symbols. 127

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

I _ TSUPREM input file 13 1 Y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . MEDICIinputfile 133

LIST OF TABLES

Table 1 : Radiation-induced trap levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39

Table 2: DLTS measurements of irradiated buried-dhannel MOS transistors. . . . . . . . . .43 ,

Table 3: Integrated displacements for a three-year mission.. . . . . . . . . . . . . . . . . . . . . . .48

Table 4: Radiation induced traps. . . . :'. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 1

Table 5: Characteristics of test devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I03

Table 6: Generation-recombination noise parameters at room temperature. . . . . . . . . . 107

Table 7: Bulk trapping levels observed in proton damaged CCDs . . . . . . . . . . . . . . . . . I 17

Table 8: FUSE FES CCD performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,120

vii

LIST OF FIGURES

4 1

Figure 1 : Simple CCD analogy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 2: Typical quantum efficiency (QE) curves.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I 2 - Figure 3: Photo-electric effect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I 5

B

. . . . . . . . . . . . . Figure 4: Cross section of a MIS capacitor. a) depletion b) inversion. . I 7 'd

Figure 5: Potential well at the surface of an MIS capacitor. . . . . . . . . . . . . . . . . . . . . . .17

Figure 6: (a) P-N junction under reverse bias (b) charge concentration (c) electric field magnitude ir (d) potential distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .20

Figure 7: Buried channel CCD cross section and corresponding potential distributions for several applied gate voltages.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 1

. j Figure 8: CCD channel cross sections and potential disiributions (a) across the channel (b) along

thechannel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Figure 9: CCD cross section and potential distribution showing backside charging effects25

Figure 10: Potential well alteration by collected charge . . . . . . . . . . . . . . . . . . . . . . . . .26

Figure 1 1: Three phase charge transfer sequence showing the charge packets and potential wells under the CCD gates.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .27

B

Figure 12: Fabrication sequence of a three layer polysilicon process for CCD gate structures.29 !

Figure 13: Potential contours for a 15 mm pixel, three-phase CCD. . . . . . . . . . . . . . . . . 3 1

. . . . . Figure 14: Charge concentration contours for a 15 mm pixel, three-phase CCD. .32 ,

Figure 15: Diagram of a CCD output circuit showing a cross section of the last few gates and a schematic representation of the output source follower amplifier. . . . . . . . . . .33

Figure 16: CCD output sequence showing the waveforms for the phase three gate (p3). the reset . . . . . . . . . . . . . . . . . . . . . . . . . . gate (RG) and the output source node ( 0 s ) . .34

Figure 17: Displacement damage i n silicon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38

Figure 18: CR-DLTS spectra of radiation-damaged buried channel MOS&TS compared with the

... V l l l

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . spectrum of an undamaged device [40] 42

Figure 19: Arrhenius plot using the CR DLTS data of the device which received 2 . 7 ~ 1 0 ~ pro- 7

tons/cm'[40] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Figure 20: Total proton flux over a three-year mission . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Figure 2 1 : Total displacements over a three-year mission . . . . . . . . . . . . . . . . . . . . . . . . 47

Figure 22: Generation and recombination of carriers through trap levels . (a) electron capture; (b) . . . . . . . . . . . . . . . . . . . electron emission; (c) hole capture; (d) hole emission 50

Figure 23: Charge confinement in an MPP device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

Figure 24: Radiation damage of the CCDs used in our experiments . . . . . . . . . . . . . . . . 58

Figure 25: Experimental setup for CCD measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Figure 26: Dark current as a function of temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

Figure 27: Dark current noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

Figure 28: Effect of poor charge transfer efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Figure 29: Three phase clock timing with transition points marked for determining the trap emis- siontimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

Figure 30: Pulse train CTI measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Figure 3 1 : X-ray event plot for an undamaged CCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

Figure 32: X-ray event plot for a radiation damaged CCD . . . . . . . . . . . . . . . . . . . . . . . 77

. . . . . . . . . . . . . . . . Figure 33: Parallel CTI as a function of temperature (fast clocking) 80

. . . . . . . . . . . . . . . Figure 34: Parallel CTI as a function of temperature (slow clocking) 81

. . . . . . . . . . . . . . . Figure 35: Parallel CTI model as a function of c1oc.k period at 223 K 83

P Figure 36: Simulated charge concentration profiles under one gate of a 27 rnm pixel device . (a) . . . . . . . along channel . (b) across channel. (c) vertically from gate to substrate 85

Figure 37: Simulated filled trap profiles under one gate of a 27 mm pixel device at 155 K86 %

e=

Figure 38: Deferred charge vs . charge packet size at 155 K . . . . . . . . . . . . . . . . . . . . . . . 87

-.

Figure 39: CTI vs. charge packet size at 155 K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .88

Figure 40: Clamp-and-sample CDS processor schematic.. . . . . . . . . . . . . . . . . . . . . . . .94 "

Figure 4 1 : CCD output waveform and CS-CDS sample timing . . . . . . . . . . . : . . . . . . .95

Figure 42: Transfer function of a clamp and sample (CS) CDS processor . . . . . . . . . . .97

Figure 43: Schematic of a dual slope integration (DSI) CDS processor.. . . . . . . . . . . . .98

Figure 44: CCD output waveform with the output of a DSI processor . . . . . . . . . . . . ,100

Figure 45: Transfer function of a DSI processor versus normalized frequency. . . . . . ,101

Figure '46: Experimental setup for measuring the low frequency noise of the transistors. I04

Figure 47: Noise s p e p at room temperature for three buried channel MOSFETs with varying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . amounts of rad~at~on damage. . I 05

. . . . . . . . . . . . . . . . . . . . ~ i ~ u r Z 4 8 : Noise x frequency for the high radiation device. ,108

Figure 49: Noise x frequency for the high radiation device over a range of temperatures 1 10

Figure 50: Arrhenius plot of the g-r time constants over the temperature range 200 - 300 K I I 1

Figure 5 1 : Noise spectra for the three transistors in the range 208-2 13 K . . . . . . . . . . . I 12

Figure 52: .Total input referred noise vs. sampling rate the highly damaged transistor (2.7~10' protonslcm2) over a range of operating temperatures. . . . . . . . . . . . . . . . . . ,114

Figure 53: Total equivalent input noise at 2 10 K for damaged and undamaged devices from simulated spectra for a transistor in the forward active region.. . . . . . . . . . . . . . . . I 16 .

Introduction

' 1. INTRODUCTION

evlces (CCDs) were first introduced by Boyle and Smith [ I ] in 1970. Several dif- Charge couple~a-'--

/ F

ferent applications have been explored for the devices, including digital memories and analog sig-

d'

nal processing, but CCDs are best known today as optical detectors. The phenomenal success of

these devices in electronic imaging has spawned a great deal of research and many advances have

b e e n made over the past three decades. Due to their sensitivity and precision. CCDs have made a

particularly large impact in the field of scientificl)maging, from optical astronomy to medical

research. These applications continue to push the limits of CCD performance

' A. Background

CCDs operate in the charge domain. The electrical signals which propagate through the device are

small bundles of charge carriers, either electrons or holes. These "charge packets" are created,

stored and moved around inside the device to perform the operations required. The fundamental

unit of a CCD is a metal-insulator-semiconductor (MIS) ~ a ' ~ a c i t o r on which the charge packets are

stored. If two of these capacitors are placed close enough together, the charge can be transferred

from one to the other by manipulating the voltages on their gates. Charge transfer between capaci-

tors is the key operation performed by a CCD and is the origin of the term charge coupled device.

I f we make a whole string of these closely spaced capacitors in a row we can form a serial shift

register. Charge can be injected into the capacitor at one end of the register through an adjacent

diode, transferred down the line of capacitors, and read out at the other end with a charge detection

amplifier. The first CCD made by Boyle and Smith was composed of 24 capacitors in a configura-

tion like this and was first used as an 8-bit serial shift register. A binary " I " wa. represented by the

Introduction

presence of a charge packet, and "0" by the absence of a packet.

Originally, the CCD was envisioned as a memory device and there was much activity in this area

in the first few years after its invention. The first generation of commercial CCD memory devices

appeared in 1975 in the form of 8 kilobit and 16 kilobit memories [ 2 ] . These were serial-access

memory devices implemented as long circular shift registers. Eventually, however, CCDs were

outmoded by other technologies with faster access times or larger capacities. Another application

area explored in the early years of CCDs was analog signal processing. A CCD is inherently an -

analog device, since the charge packets can be of arbitrary size, and it lends itself naturally to

time-sampled signal processing. In these applications, each charge packet represents one sample

of the analog signal. A delay line is the most obvious possibility, since i t is basically an analog

shift register, and CCD delay lines were successfully used to perform such tasks as re-synchroniz-

ing video signals [3]. Analog transversal filters were also built. but when numeric processors,

became fast enough, it was more convenient to perform most signal processing tasks in the digital

domain. CCDs are still used in some cases, foiinstance to capture a transient signal which is too

fast for an analog-to-digital converter (ADC) to sample adequately. The CCD :an then be read out B

at a slower rate for the ADC to digitize. This approach is used in some digitizing oscilloscopes.

It was in imaging, however, that the CCD found an enduring niche. Over the past three decades, its

impact on the field of electronic image capture has been nothing short of revolutionary. The shift

from memory and signal processing into the optical detector field was simple for CCDs because

-the silicon from which they are made is naturally light sensitive. Light incident on a volume of sil-

icon will generate charge (electrons and holes) through the photoelectric effect. So in an imaging

CCD, instead of charge being injected electrically at one end of the register, i t is created by the

Introduction

incident light along the whole length and the size of the charge packets detected at the output (bar- C

ring complications to be discussed later) will be in direct proportion to the light intensity at each

point: This property enabled Tompsett, Amelio and Smith [4] to use the first CCD also as a simple

line imager. CCD line imagers have been used with success in such devices as spectrographs and

facsimile machines. The extension from a line imager to an area imager is straight forward: a

series of &D shift registers is placed side-by-side to form a two-dimensional array. If an image is

b focussed on this array, the photo-generated charges in the capacitors wal form an electrical ana-

packet corresponding to a single picture element (pixel). The charge

along the shift registers to a detection circuit and the image can be

recorded or displayed. Figure I shows% simple analogy.

In this figure. the buckets represent the CCD capacitors, the raindrops represent the photons $

light and the water collected in the buckets represents the generated charge packets. The conveyor

belts represent the CCD shift registers and the measuring station represents the charge-sensitive

output amplifier. After collecting rain water (charge) for a certain period of time (the exposure or

integration time) the side-by-side conveyor belts that form the array (the parallel registers of a+' . .

CCD) shift one unit and load the transverse conveyor (the serial register), which then conveys each

of its buckets one at a time to the measuring station. When the row has been completely trans-

ferred. the parallel registers shift again and load a new row into the serial register. This continues '

unt i l the entire array has been read out and the distribution of rainfall over the array of buckets can . . b- ---- .--

be reconstructed from the data. In the same way. the distribution of light intensity (the image) inci-

dent on a CCD can be reconstructed from the measurements of the charge collected in its capaci-

tors.

Introduction

' ' I // $/I / V / / I , buckets = cmacitors ! / / / I .

- measuring station = output amplifier

Figure I : Simple CCD analogy: ~nc~den t light is represented by rainfall and the CCD registers are rcpre- sented by conveyor belts. The "image" (rainfall distribution) is acquired by transferring the rain- water in each bucket one st a time to the measuring station.

B. GCD development and current status

The initial research efforts into CCD imaging arrays were aimed at producing devices for the large

markets of broadcast television. home video, surveillance, and closed circuit television (CCTV)

systems, which at the time were largely dominated by vidicon tubes. The idea was to make

replacements for the vidicon tube that incorporated the CCD's advantages in size. weight, reliabil-

Introduction

i ty and low power requirements. It turned out to be quite difficult, however, to produce arrays of

any appreciable size and it was many years before researchers were able to produce CCDs of wffi-

cient m a y size? (around 500x500 pixels for standard television) that could also match the vidicon

in frame rate and cosmetic quality. The first commercially available CCD camera was only a small

100x1 00 array unveiled by Fairchild in 1973 [5]. Nevertheless. driven by the mass market possi-

bilities and aided by advances in integrated circuit manufacturing technology. several manufactur-

ers were able to produce fully television-compatible cameras by the mid 1980s and CCDs soon

completely replaced vidicons in moSt applications. Today, CCD-based hand-held home video cam-

eras about the size of a paperback novel are widely available.

CCD imagers also generated considerable interest in the scientific community because of their low

noise, high linearity, large dynamic range, good geometric accuracy. and broad spectral response.

Astronomers. who are always interested in detecting fainter aAd more distant objects. were partic-

ularly impressed with the sensitivity of CCDs, which is approximately 100 times greater than pho-

tographic film. Frame rate was not a major issue, since astronomical exposures typically last from

a few minutes to several hours. Once arrays of reas~nable size were available, CCDs rapidly

became the detector of choice at all major astronomical observatories. NASA also commissioned

CCDs for several space missions and in 1980 Texas Instruments, Inc. managed to fabricate

800x800 pixel imagers of which four were used in the first Wide Field and Planetary Camera

(WF/PC) of the Hubble Space Telescope (HST) [6]. Another 800x800 pixel device made by Texas

Instruments with a different architecture designed to reduce susceptibility to radiation damage and

enhance spectral responsivity was sent on the space probe Galileo to take pictures of Jupiter and its

moons (71. Despite technical problems with other parts of both spacecraft (the flawed main mirror

of HST and the failed high-gain antenna of Galileo), the CCDs produced stunning results.

Introduction

Array size

As the capabilities of CCDs as scientific instruments became known, scientists pu~hed~manufac-

turers for devices with larger and larger array sizes, broader spectral response and lower noise. In

1983 Texas Instruments bettered their 800x800 array with a 1024x1024 device [8], and a year later

Tektronix, Inc. had produced a 2048x2048 array. In 1989, Ford Aerospace Corp. managed a

4096x4096 array [9], and this remains the largest commonly available format. Larger arrays have,

however, been built for special applications. A US Navy project is currently underway to build a

reconnaissance instrument for which Loral Aerospace is building CCDs of 9216x9216 pixels.

each pixel 8.75 pm square. This device, the largest CCD array ever built in terms of pixel count, is

80.6 mm on a side and each one takes up an entire 5 inch silicon wafer. At an astronomical CCD

conference in October 1996 [ I 01 it was reported that the effort had produced 3 or 4 working arrays,

with more being fabricated. These first devices, however, had numerous defects and the images

they produced were not cosmetically good enough for scientific work. In another effort, research-

ers at the Steward Observatory in Arizona are currently evaluating the scientific suitability of large

CCD arrays manufactured by Phillips Imaging Technology, Inc. These devices are built on 6 inch

wafers and consist of a 7 168x921 6 array of 12 pm pixels, making them the largest integratet cir-

cuits ever built [I I].

There are two difficulties with building these extremely large arrays: the first is reducing losses in

the large number of transfers each charge packet has to undergo to reach the output and the second . is avoiding defects in circuits of such enormous physical size and density. Charge transfer effi-

ciency (CTE) is a measure of the percentage of charge that is successfully transferred from one

pixel to the next during readout of the array. The CTE must be very high in order to get reasonable

Introduction

output from a large array. For example, an average packet in a 4096x4096 array undergoes around

4000 transfers to reach the output, and even with a CTE of 0.9999 it would arrive with only

two-thirds of its charge. The original CCDs of Boyle and Smith had CTEs of about 0.98, so i t was

fortunate the charge packets had only 8 transfers to undergo!

s

An important cause of poor CTE is trapping by mid-gap energy states. The first CCDs experienced 2

a lot of charge loss because the charge was stored and transferred at the interface between the sili-

con and the insulating layer where there are a large number of these trapping states. So early on,

researchers experimented with adding an implanted layer just below the surface to create "buried

channel" devices in which the charge was stored and transferred away from the surface states [I 21.

This made a dramatic improvement in CTE over "surface channel" operation and is the standard

device structure used today. Buried channel devices still encounter trapping states due to impuri-

ties in the silicon, but silicon manufacturing has improved so much in the last 30 years that impu-

rity levels are now very low and CTEs of up to 0.9999998 (six nines) have been achieved [I 31.

The great advances in integrated circuit fabrication that have accompanied the improvements in

silicon purity and crystal quality have also reduced the problem of circuit defects in making very

large arrays, but i t is still expensive to fabricate devices in which a single defect, such as a short

between two clock phases, can ruin an entire wafer. To circumvent this problem, many manufac-

turers and research groups have opted for a less demanding. and hence less costly solution to creat-

ing large area detectors: making several smaller arrays and tiling them together on a single focal

plane. With the most common geometries, four devices can be fabricated on one wafer, so a single

point defect will only affect one quarter of the devices. This approach has, however, required con-

siderable effort into producing CCDs in an appropriate format and devising means of butting the

7

Introduction

arrays together in such a manner as to minimize the dead space between them while maintaining

stringent optical flatness across the entire plane. At least three manufacturers are now producing

2048x4096 pixel arrays for this purpose which can be butted on three sides to produce conglomer-

ate arrays 8 192 pixels wide and any number of pixels long. Another advantage of these CCD

"mosaics" is that a much larger focal plane can be covered because it is no longer limited to the

size of a single silicon wafer. In two current MEGACAM camera projects, one being planned for

the Multiple Mirror Telescope (MMT) in Arizona and the other for the Canada France Hawaii

Telescope (CFHT) in Hawaii, a total of 32 2048x4096 CCDs is used to cover an approximately

240x240 mm focal plane [I 41.

Despite the impressive advances in array sizes and densities, CCDs are only beginning to match

the resolving power of photographic film, which combined with film's low cost, has largely pre-

vented CCDs from making inroads into the huge 35mm still camera market. At least two compa-

nies (Canon an&-Nikon) have high-resolution CCD-based digital still cameras available, both

based on 4096x4096 CCDs made by Eastman Kodak, but at roughly $20,000 each, they are aimed

mainly at professional news photographers. There are numerous digital cameras being marketed to

the consumer, but they are based on CCDs of small pixel counts (typically 640x480, a standard

resolution for computer displays) and produce images of relatively poor quality. Better products

are continually appearing, however, and there can be little doubt that CCDs will soon be compek

ing successfully with film. Of course, with film, one can always change the optics and use a larger

piece of film if greater resolution is required, and it will be a long time before a CCD can produce

images of the same quality as large format film cameras.

One of the major technological hurdles of building cameras with very large pixel counts, apart

Introduction

from producing large cosmetically perfect CCDs, is developing the supporting equipment to han-

dle the vast amounts of data produced. For example, a 4096x4096 array generates 32 Mbytes of

data for each image (with 16 bits per pixel digitization). and an 8192x8192 mosaic produces over

128 Mbytes per image. Efficiently dealing with this volume of data, especially in a portable device

like a still camera, is an area of ongoing development effort.

Noise

The sensitivity of a CCD is determined in large part by the noise, which imposes a fundamental

limit on the minimum detectable signal. There are many sources of noise, including photon noise,

thermal noise, and electrical noise in the readout circuit.

Photon noise is a reslrllt of the fact that photoelectric charge generation is a random process, gov-

erned by Poisson statistics. Therefore, due to the very natyre of the detection mechanism in a

CCD, there is noise present with a root mean square (rms) value equal to the square root of the sig-

nal level in electrons. However, this source of noise is not a serious detriment to the sensitivity of a

CCD because i t is proportional to the signal level and is lower for low signals.

The dark current is the amount of charge generated by thermal energy in the device, which pro-

duces charge in the pixels even when the CCD is not exposed to light (hence the term dark cur-

rent). Thermal generation is also a random process and therefore dark current also adds noise. In tb

this case, the noise is proportional to the amount of dark current and not the signal level, so it can

reduce the ability to resolve faint objects and is particularly harmful in the extremely long expo-

sures typical of astronomical imaging. Dark current is highest in regions where there are a large

number of mid-gap levels, so the largest contribution to the dark current comes from the surface.

with its high density of mid-gap states. However, CCD researchers have devised a couple of tech-

9

Introduction

niques which very effectively reduce or eliminate the surface dark current (see Chapter 4). The

remaining contribution to the dark current comes from mid-gap states in the bGlk silicon away

from the surface, and this has been significantly reduced by the same improvements in silicon

purity that have increased the CTE. Typical values in modem CCDs are about 20 p ~ / c m Z at room

temperature. These low levels have reduced the need for an elaborate cooling apparatus to lower

the dark current.

Read noise is the amount of noise introduced by the charge detection circuit at the output of the

CCD. This has been the ultimately limiting source of noise for most of the CCD's history. The first

CCDs exhibited input-referred noise levels of around 30 e- rms. Much effort has been expended to

lower the read noise by signal processing techniques and by optimizing the geometry of the tran- .k

sistors used in the output circuit [I 51. This has resulted in read noise levels in the best devices of *

around 2 e- rms at slow (50 kHz) readout rates. Another possibility which a few researchers have

pursued is devising output circuits that can sample the charge packet non-destructively [ 9 ] . It is

then possible to average multiple samples of the same charge packet, which reduces the noise by

the square root of the number of samples, though at the cost of a slower readout rate. Such a device

has produced output with a noise of less than 1 e- rms by averaging 64 samples per pixel [9].

Quantum Efficiency

Scientists have also been pushing the limits of CCDs in the area of quantum efficiency (QE). QE is

a measure of how accuraiely the charge generated in a pixel represents the actual intensity of light

incident on the CCD at various wavelengths. Typically there are losses due to reflections from and

absorptions in non-active layers of the device, and certain wavelengths may pass through the

device undetected. A great deal of loss at short wavelengths occurs in the insulating and gate layers

Introduction

of a CCD and three methods of overcoming this have been pursued. The first is to use special

implants to eliminate one or more of the gate phases so that a portion of each pixel has only a thin

oxide layer over it to interfere with incident light. This type 6f device was built by Texas Instru-

ments in 1981 and used in the Galileo spacecraft [7]. The second approach is to flip the device

over, etch away the substrate and illuminate i t from the backside. This clears the entire surface of

interfering structure. The thinning process, however, is difficult and therefore costly. It took ten

years of work by Texas Instruments to perfect the thinning process used in the production of the

CCDs for WFPC [I 71. Reticon, Inc. introduced a commercial thinned, backside-illuminated CCD

in 1987, and many manufacturers today offer thinned versions of their CCDs. The third and much

simpler approach to enhancing responsivity at short wavelengths is to use a phosphor coating

(such as lumigen, the material used in fluorescent yellow highlighting pens) which converts short

wavelength photons to longer wavelengths that pass more easily through the frontside surface lay-

ers. This was the approach used for the CCDs of the camera upgrade to the HST (WFlPC2) after

an unexpected problem with the thinned WFPC CCDs emerged shortly before launch, resulting in

a 5 million dollar emergency fix.

P To further improve QE performance, an anti-reflection (AR) coating is often applied to the devices

after manufacture. AR coatings reduce the losses at certain wavelengths due to reflection from the

surface of the CCD and can make a substantial improvement to the QE. Current CCDs can have a

QE that peaks at over 90% and depending on the type of AR coating. reasonable (>30%) perfor-

mance can extend down to wavelengths of 200 nm (ultraviolet) or up to I000 nm (near infrared). A

phosphor coating can extend the useful QE range down to 50 nm (far ultraviolet). Figure 2 shows a

set of typical QE curves for three different devices. These curves were measured for CCDs used at

the National Research Council's (NRC) Dominion Astrophysical Observatory (DAO).

I I

Introduction

0 backside-illuminated, AR coated A frontside-illuminated + frontside-illuminated, lumigen coated

SO0 400 600 800 1000 1200 Wavelength (nm)

F~pure 2: Typical quantum efficiency (QE) curves.

C. Research motivation .-

As the limits of CCD performance are pushed further, the devices become more and more sensitive

to small amounts of damage. One cause of damage encountered by CCDs in certain applications is

nuclear radiation. Radiation can cause charge buildup in the insulating layers or even damage to

the atomic crystal structure of the silicon and result in serious degradation of the sensor's perfor-

mance.

f

The aim of this thesis was to examine the effects of radiation damage on the operation of CCDs

and to discover, if possible. means by which the effects can be minimized. The motivation behind

Intrbduction

this investigation was a satellite astronomy project called the Far Ultraviolet Spectrographic

Explorer (FUSE). One of the instruments aboard the satellite is the Fine Error Sensor (FES) which

is used to keep the satellite pointed in the desired direction. The FES uses a CCD to track guide

stars and provide feedback to the attitude control system of the satellite. To achieve the necessary,

pointing accuracy for the telescope, the specification for the FES states that it must be able to

determine the centroid of a guide star image to within 0.08 of a pixel, which corresponds to an

angular deviation of 2 arcseconds.

In the original FUSE mission design, the satellite was to travel in a Highly Elliptical Orbit (HEO)

and twice each orbit would have traversed a belt of protons trapped by the earth's magnetic field.

In this orbit, the FES would have received a very large dose of energetic protons, and therefore

theae was considerable concern over the effects of this type of radiation on the instrument. After

initial studies, the mission wa5 revised with a lower orbit to reduce the amount of radiation

encountered, and the lifetime of the mission was reduced. However, radiation damage is still a

concern. Our investigations focussed on three characteristics of CCDs which are susceptible to

radiation damage: dark current, charge transfer efficiency (CTE), and noise in the output circuit.

We discovered the rate of degradation in these three areas under the level of radiation expected for

the FUSE mission and made recommendations for various means of reducing the effects.

The next chapter describes the basic structure of CCDs and the theory of their operation. Chapter 3

gives a brief outline of the physics of radiation damage in semiconductor materials. After this we

describe the experiments we performed to investigate radiation effects on the above three perfor-

mance characteristics of CCDs. Chapter 4 deals with dark current, Chapter 5 with charge transfer

efficiency, and Chapter 6 with read noise. In Chapter 7 we summarize our conclusions.

Device Structure and Operation

2.*DEVICE STRUCTURE AND OPERATION

3

The operation of a CCD can be broken down into four steps: charge genkration, charge collection,

charge transfer, and charge detection. Charge generation is how the external signal we desire to

detect (light intensity) is converted into an internal electrical signal in the form of electronic

charge. Charge collection is the next step, in which the generated charges are gathered into discrete

packets. It has a two-fold purpose: to allow integration of the signal over a long period of time and

to spatially localize the signal to get a two-dimensional signal distribution. Charge transfer is the

process whereby the integrated and spatially localized charge signals (charge packets) are moved

to a single detector. The detector performs the last step, which is to convert the charge signal into a

more convenient electrical signal, namely voltage, for further processing.

A. Charge generation

Charge generation occurs through the photo-electric effect, in which a photon of light interacts

with an electron in the valence band of a semiconductor and imparts enough energy for the elec-

tron to jump to the conduction band, creating an electron-hole pair (Figure 3). The energy required

is equal to the bandgap of the semiconductor (the energy difference between the conduction and

valence bands). Silicon has a bandgap of 1.12 eV and therefore any photon with energy greater

than 1.12 eV is capable of boosting electrons into the conduction band in silicon. Photons with

greater energy may cause more than one electron to jump to the conduction band. It has been

determined empirically that photons with energy greater than about 5 eV will generate I electron-

hole pair for every 3.65 eV of energy they possess [I 81.

The energy of a photon is related to its frequency by


I I

I hv electron-hole pair 1.12 eV

I

1 I I I

I / Figure 3: Photo-electric effect. An incident photon of energy hV excites an electron into the conduction

band, creating an electron-hole pair.

where h is Planck's constant, v is the frequency, c is the speed of light and h is the wavelength.

From this equation we can see that the upper end of the useful spectral range of silicon as a detec-

tor of electromagnetic radiation is about 1100 nm, which is in the near infrared. Above this wave-

length the photons do not have enough energy to excite electrons into the conduction band. Other

semiconductors with smaller bandgaps can be used if the detection of longer wavelengths is

desired. Germanium, for example, has a bandgap of 0.6 eV and germanium CCDs have been built

with good spectral response to 1600 nm [I 81. The responsivity of silicon also tapers off at very

short wavelengths because there is a reduced probability of inieraction (due to the lower absorp-

tion coefficient). The practical lower limit is about 0.1 nm, or a photon energy of 10 keV, which is

in the x-ray region of the spectrum. So, ideally, the useful range of a silicon detector like a CCD

extends over the near infrared, visible, ultraviolet, extreme ultraviolet and soft x-ray portions of the


electromagnetic spectrum. However, other effects, like reflection from the surface or absorption in - non-active regions of the device can significantly reduce the sensitivity of a CCD in certain spec-

tral ranges.

6. Charge collection

Once the charge is generated, it must be collected and stored. As stated in the introduction, the

basic element of a CCD is a MIS capacitor, upon which the photo-generated charge can be stored.

The device is built on a p+ type substrate on which is grown an epitaxial p-type layer of about 10-

20 microns in thickness. The resistivity of the epitaxial layer can be varied depending on the-appli-

cation, but a typical value is around 30-50 ohm-cm. On top of the epitaxial layer is an insulator

layer, which can be a simple oxide or a double layer of oxide and nitride. The nitride layer assists

in ensuring a uniform insulator thickness throughout the repeated oxidations and oxide etchings

involved in creating the multilayer gate structure of the CCD. The insulating layer is usually about

1000 angstroms thick in total. On top of the insulator the gate material is deposited to form the

metal plate of the MIS capacitor. Figure 4 shows a cross section of the capacitor.

If a positive voltage is applied to the gate of the capacitor, a depletion region is formed in the sili-

con below the gate as the majority carriers (holes in ptype silicon) are pushed away (Figure 4a).

The charge on the gate is balanced by the space charge in the depletion region and the resulting

potential profile will create a potential well at the surface (Figure 5). The steady-state minority car- 1

rier (electron) concentration is given by


gate

oxide

P-tY Pe silicon

Figure 4: Cross section of a MIS capacitor. a) depletion b) inversion.

gate ' 'v I

depletion edge

oxide ( ~ d )

Figure 5: Potential well at the surface of an MIS capacitor. Free electrons will gather just below the oxide at the potential maximum.

.. Device Structure and Operation

where yl is the intrinsic Fermi potential and 0 is the Fermi potential. If the applied gate voltage is

high enough, the surface potential yl, will exceed the Fermi potential by a sufficient amount to

allow a significant minority camer buildup in the potential well at the surface. The additional

charge on the gate then begirk to be balanced by collected minority carriers instead of the fixed

charge of a widened depletion region (Figure 4b). The surface is <aid to be inverted and the layer

of collected minority camers is called the inversion k ~ y e r . If a voltage greater than the threshold

voltage VT is applied to the gate, the minority carrier concentration at this point will exceed the

steady-state majority camer concentration and the surface is said to be in strong inversion.

What we have not addressed in the abo;e description is where the minority carriers in the inversion

layer come from. The capacitor is isolated, so the minority carriers can only come from carrier

generation processes in the depletion region, which takes time. These processes, including thermal

generation and the photoelectric effect, are described in Chapter 4. CCDs are operated in a tran-

sient state called deep depletion. In a deep depletion state, the gate voltage is sufficient to invert the

surface, but enough minority camers have not yet been collected to do so, and therefore the gate

. charge must be balanced by a wider depletion region. The depletion region and the resulting poten-

tial well in this state extend deep into the silicon substrate. If the other carrier generation mecha-

nisms are slow enough. the majority of the carriers collected on the capacitor will be those

generated by the photoelectric effect. and therefore the charge stored on the silicon plate of the

capacitor will be proportional to the incident light intensity.

Buried channel operation

As mentioned in the introduction, the surface is a poor place to store and transfer charge because

of the large number of trapping states there, so most modem CCDh are buried-channel devices. In


buried-channel devices, a shallow layer at the surface is implanted with n-type impurity atoms,

similar to a depletion-mode metal-oxide-semiconductor field effect transistor (MOSFET), and the

n-layer is biased positively with respect to the p-type substrate. This arrangement alters the poten-

tial distribution so that the charge on the capacitor is collected below the surface in the bulk sili- I

con.

To understand the potential distribution in a buried channel CCD we first consider a reverse biased

p-n junction as in Figure 6. For simplicity we assume a step junction, with a space charge distribu-

tion in the depletion region as shown. This fixed charge gives rise to an electric field which is

calculated by integrating the charge density. Integrating the electric field gives the potential distri-

bution, which has a maximum at the far edge of the n-type region. Free electrons will be swept by

the electric field along the potential gradient toward this maximum and out through the metal con-

tact.

We can now alter the picture to resemble a buried channel CCD by adding an oxide layer and a - . &

gate (Figure 7). If a potential VG lower than the reference potential Vre1is applied to the gate, the

potential at the surface will be pulled down, forcing the maximum of the potential distribution

deeper into the silicon. As VG is lowered further, the maximum moves deeper until the surface

potential is lower \ t n the substrate potential, as in the curve for VG2 in Figure 7. At this point,

holes are attracted from the surrounding ptype material and invert the surface, "pinning" the sur-

face potential to just below that of the substrate. If VG is lowered still further (e.g. to VG3), the

extra gate potential will be balanced by an increase in the inversion charge. so that the potential

within the silicon remains fixed. The potential well still exists in inversion, and although its height

can no longer be adjusted, i t is capable of collecting charge as before, a feature which is exploited


A

electric field

)x -0

A

potential

4

X

Figure 6: (a) P-N junction under reverse bias (b) charge concentration (c) electric field magnitude (d) potential distribution


Figure 7: Buried channel CCD cross section and corresponding potential distributions for several applied gate voltages.

in certain devices to reduce the dark current (see Chapter 4).

The above describes the potential variation along the vertical (gate to substrate) axis. In order to

spatially localize the charge signal into an array of pixels, the charge must be confined along both

of the remaining axes as well. Figure 8 shows two cross sections through a CCD channel. Figure

8(a) is the cross section across the channel, perpendicular to the direction of charge transfer. The

charge in this direction is confined by the channel stops, which are electrically connected to the


gate

potential t

gates oxide

potential 1

Figure 8: CCD channel cross sections and potential distributions (a) across the channel (b) along the channel.


substrate, and therefore at the substrate potential. They are thus biased negatively with respect to

the channel, and create a potential gradient as shown in the figure. Along the channel, parallel to

the direction of transfer, the charge is confined by the potentials applied to adjacent gates. In Fig-

ure 8(b) the collecting gate is set to VGI from Figure 7, and the adjacent gates are set to VGZ. This

creates a potential profile as shown.

Charge spreading

The collection of charge into discrete packets in a CCD is remarkably efficient. One of the features ,

of a CCD which makes it an attractive image sensor is that it has a 100% fi l l factor. This means

that there is no insensitive dead space between pixels. If a photon is absorbed within the channel

stop region or in the volume beneath a non-collecting gate, the generated electron will still be col-

lected by the potential well. If, however, the photon is absorbed deep in the device beyond the

depletion region, the generated electron will drift randomly until it recombines or until it enters the

depletion region and is swept by the electric field into the potential well. It is possible that the ran-

dom motion of the photoelectron will take it away from the pixel in which i t originated before it is

swept into a potential well, and thus the spatial resolution of the device is compromised. This

spreading effect is particularly problematic for long wavelength photons, which tend to penetrate

deeper into the device and are more likely to be absorbed in the field-free region. The spatial reso-

lution can be improved by reducing the number of electrons which are collected from the field-free

region [ I 61; however, this improvement comes at the expense of reduced quantum efficiency.

Backside illumination

In addition to the problem of photons which penetrate too far, there is a problem with photons

which do not penetrate far enough. Clearly, a CCD cannot detect photons which are absorbed in


the gate and oxide layers of the device and never reach the depletion region, which occurs most

frequently for short wavelength photons. Backside-illumination is a technique that was developed -* '

to improve the quantum efficiency of CCDs at shorter wavelengths. In a difficult, low yield pro-

cess, the CCDs are flipped over and the substrate is etched away right up to the epitaxial layer. The

device is then illuminated from the substrate side, and the photons enter the active region directly

without having to pass through the gate and oxide layers. The spectral dependence of the spatial

resolution is reversed for backside-illuminated CCDs, because in these devices the shorter wave-

lengths are the ones absorbed far from the frontside depletion regions.

An unanticipated problem with backside illumination was revealed when the first devices were

tested [18]. A thin native oxide layer grows on the backside surface when i t is exposed to air and it

turns out that holes can be trapped at the interface, creating a layer of positive charge. This charge

layer deforms the potential distribution so that a second maximum is created at the backside sur-

face (Figure 9). Therefore, short wavelength photons absorbed near the surface are swept back by

the electric field and become stuck at the backside until they recombine. There are several methods

which have been developed to create the necessary electric fields to negate the effect of the trapped

holes and drive the photoelectrons toward the potential wells at the frontside [19]. but we will not

discuss these here. With these techniques, i t is possible to eliminate the field-free region and

achieve a 100% internal QE; in other words, all electrons generated in the device are collected and

none are lost to recombination, so that &ery photon which is not reflected from the surface or

allowed to pass right through the device is detected.

Full well

As charge is collected in the potential well, the shape of the potential distribution is altered. The


gate ,

ox~de oxide

potential I

I I

I I I

F~gure 9: CCD cross section and potential distribution showing backside charging effects

peak is reduced. and the well becomes flatter and broader (Figure 10) until eventually it is no

longer capable of containing additional charge. The maximum amount of charge which can be

held in the potential well is called t ha fu l l well" charge. This level can be defined in several ways.

The two most common definitions in buried channel CCDs are referred to as surface full well and

bloomed full well. These levels are indicated in Figure 10. Surface full well occurs when the col-

lected charge begins to interact with the surface, and is manifested by a significant increase in trap-

ping phenomena, due to the surface states. Bloomed full well is the level at which the potential

equals that under the non-collecting gates and the charge is no longer confined (the potential pro-

file of Figure 8b is flat). The charge will spread up and down the channel, an effect known as

collecting phase

barrier phase


' potential

surface full well

F~gure 10: Potential wel l alteration by collected charge

"blooming". The optimum full well is achieved when the non-collecting gates, the "bamer

phases". are in inversion ( VGZ) and the collecting gate potential (VGl ) is set so that the surface and

bloomed full well levels coincide. If VGl is too low, blooming will occur first; i f i t is too high. the

collected charge will reach the surface first.

C. Charge transfer IC -

Once the charge has been collected into the pixels, the packets must be transferred to the output. e'

The charge contained In the potential well beneath a CCD gate is moved to the next and following

gates by what is usually described as a simple process of phased. or peristaltic, clocking. The most


common scheme has three-phases, such as the one in Figure 1 1, in which three adjacent gates form

Figure 11: Three phase charge transfer sequence showing the charge packets and potential wells under the CCD gates.

one pixel. The charge is collected under one phase (p l for example), which is held at a positive

voltage, while the other two phases (p2 and p3) are held at negative voltages (Figure 1 l(a)). The

adjacent phase in the desired direction of motion, for example p2, is then also made positive caus-

ing the charge packet to become distributed under p l and p2 (Figure 1 l(b)). A short time later p l

is set to the negative voltage level, forcing the entire charge packet to collect under p2 (Figure

27


1 !(c)). The next transfer begins when p3 is set high (Figure 1 I(d)) and ends with the packet under

p3. Repeating a similar sequence with p3 and p l will move the charge packet under the next p l

gate, completing a one-pixel transfer for the three-phase CCD.

The actual movement of the charge from one gate to the next occurs by three basic mechanisms

[20]: thermal diffusion, drift due to the fringing field between gates. and self-induced drift due to

the mutual electro&atic repulsion between charges. Thermal diffusion is simply the random ther-

mal motion of the electrons, which tends to move them from regions of high concentration to

regions of low concentration. Drift is the motion caused by an electric field. Because of the cou- /

piing between the gates and the resulting potential gradients. a "fringing field" exists which

sweeps electrons into the shifting potential wells. An electric field is also created by the electrons

themselves which causes them to repel each other. This self-induced drift is most effective for

large packets at the beginning of charge transfer. when the concentration of electrons in the start-

ing well is high. Thermal diffusion and fringe field drift are not dependent on packet size, and e

hence are important near the end of the transfer and at the small signal limit. All three transfer

mechanisms are sensitive to temperature through the thermal velocity and electron mobility. The

transfer proceeds quite rapidly, with a time constant on the order of a few nanoseconds, so it will

cause noticeable delay in only very high-speed devices.

To achieve efficient transfer, the gates of a CCD must be placed close together, closC enough, in

fact, that their depletion regions are coupled together. The first devices to be fabricated used a sin-

gle layer of aluminum for the gates, which was etched to form the individual phases. In order to

get good coupling between gates, they had to have gaps of less than 3 pm between them. Achiev-

ing this without any shorts between gates proved a considerable challenge and yields of properly

Device Structure and Operation i

functioning devices were low. Later, researchers discovered that the yield could be improved by

using multiple gate layers and overlapping them [ 2 I ] . This arrangement created good coupling

between gates, while reducing the chance of a short.because of the insulating layer between them.

Although several variations of the gate structure are in use, the most common is a 3 layer configu-

ration. Each layer is etched from a doped deposition of polysilicon. After etching. the wafer is oxi-

dized before the next layer of polysilicon is deposited in order to insulate the layers from each

other. The process for a typical CCD shift register is shown in Figure 12.

oxide - implanted n layer -

i p substrate

First polysilicon deposition

Polysilicon etch (phase one gates)

Oxidation

r 3

Second polysilicon layer 10 (phase two gates)

U

Th~rd polysilicon layer (phase three gates) and pass~vatlon ox~de

Flpure 12: Fabrication sequence of a three layer polysilicon brocess for CCD pare htructures


Simulations

As part of our investigations, we used simulation software to simulate a segment of a CCD shift

register. First, we used a two-dimenhonal process simulation package, TSUPREM4 [ 2 2 ] , to calcu-

late the doping concentrations and create the simulation mesh for six gates of a three-phase buried

channel CCD with 15 pm pixels using the three layer polysilicon process shown above. This

mesh was then fed into a two-dimensional device simulation package. MEDIC1 [23], which itera-

tively solves Poisson's equation and the charge-continuity equation to calculate the potentials and

charge concentrations in the device. The results are shown in Figures 13 and 14. Figure 13 shows a

set of potential contours, while Figure 14 shows the corresponding charge concentration contours.

D. Charge detection

Charge detection is the last stage of CCD operation in which the charge packets collected in the

device and transferred to the output are converted one at a time into voltage signals which can be

processed by external electronics. The usual method is to use a floating gate amplifier, shown sche-

matically in F ip re 15. The floating gate amplifier consists of two transistors: the output transistor

and the reset transistor. The output transistor is connected in a source fo l lok r configuration with a

load resistor RL from source to ground. The gate of the output transistor is connected through the .. I

reset transistor to the reset drain voltage, but when the reset transistor is off, the output gatelreset

source node is "floating". The output sequence begins with the reset transistor turning on and

resetting the output gate voltage to a fixed value ( V R D ) in the linear region of the output tnnsistor

The reset transistor is then turned off and the gate node is allowed to float. Then the 'next charge t

packet is transferred to the gate node through the last gate (LG) of the CCD serial regiher The last

gate is set to a fixed value between the high and low levels of the serial register gates so that i t


Distance (Microns)

Figure 131 Potential contours for a 15 pm pixel, three-phase CCD

forms a half-height potential barrier between p3 and the floating node. It is not clocked. in order to

avoid spurious signals caused by capacitive coupling between the last gate and the floating node. 4

Some CCDs have a special, separately clocked gate in the place of p3 called a "summing well".

The summing well (SW) allows one to combine, or "sum", several pixels in a row before trmsfer-

ring the combined packet to the output. The summing well gate is usually larger than the regular

gates in order to increase its full well capacity. When p3 (or SW) goes low, the charge packet is

spilled over the barrier onto the gate node and a voltage is induced there which is proportional to


F~gure 14: Charge concentration contours for a 15 pm pixel, three-phase CCD

the charge and inversely proportional to the capacitance of the node. The voltage at the output

source is the gate node voltage multiplied by the gain of the source follower. which is close to

unity. From the output source, the signal is applied to an external pre-amplifier and then to the rest

of the signal processing circuitry. The timing of the output sequence is shown in Figure 16. The AV

indicated on the output waveform (0s) in the figure is the difference between the output after the

reset pulse (the reset level) and the output after the charge packet has been dumped to the floating

node (the s~gnal level). The difference between these two levels is the value for that pixel. Note


last gate reset gate reset drain P3

Figure 15: Diagram of a CCD output circuit showing a cross section of the last few gates and a schematic representation of the output source follower amplifier.

that the OS waveform shows the feedthrough of the reset pulse, which is a result of the parasitic

gate-source capacitances of the reset and output transistors (CI and C2 in Figure 15).

The capacitance of the floating node is an important parameter because it determines the sensitiv-

ity of the output circuit. The smaller the capacitance, the greater the voltage induced by a given

amount of charge. The main component of the capacitance is that between the gate and channel of

the output transistor. The gate-channel capacitance is proportional to the area of the gate, so it is

desirable to make the output transistor as small as possible. There are also parasitic capacitances


reset feedthrwgh

os n t

AV

4 Figure 16: CCD output sequence showing the waveforms for the phase three gate (p3). the reset pate (RG)

and the output source node (0s).

such as the gate-source capacitances CI and C2 mentioned above. All other parasitics can be repre-

sented by a lumped capacitor C3 to ground. The gate-source and gate-drain capacitances can be

)

reduced by employing a lightly doped drain (LDD) structure, which reduces the overlap between

the gate and the source and drain implants (151. The sensitivity of these LDD-type output transis-

tor is around 1 pV/e-.

Although the single-stage floating gate amplifier is the most common, other forms of the output

circuit exist, such as the non-destructive charge sensing scheme mentioned in the introduction.

Recent devices manufactured by English Electric Valve (EEV), Inc. [24] have two-stage output

amplifiers. In this configuratibn, the second stage transistor is large to provide a high level of drive

capability while causing minimal loading to the first stage, which enables the first transistor to be

very small. increasing the sensitivity. These amplifiers exhibit an overall output sen4tivity of

4 pv/e- .

Radiation Damage

3. RADIATION DAMAGE

In many scientific imaging applications it is necessary to subject the detector to harmful radiative

environments. Such applications include almost any space mission, x-ray crystallography, certain

forms of medical imaging and energetic particle detection. Radiation deposits its energy In silicon

in various ways, some of which can result in permanent damage. At its most benign, the radiation

energy may simply be transferred to mechanical vibration of the silicon atoms and be manifested

a5 heat. Two of the more harmful e f f e c m s o f most concern in electronic devices: the first is ion-

ization and the second is atomic displacement.

A. Ionization damage

Charged particles, such as electrons or protons, lose most of their energy in Co~lombic scattering,

interacting with the silicon atoms through the electrostatic force. Because this is a long-range

interaction. the dominant effect is small energy transfers to the atomic electrons [ 2 5 ] . If enough

energy is imparted to the electrons, they+ill be ejected from the host atoms, creating free elec-

trons and positively charged ions (ionization). Photon radiation, such as x-rays or gamma rays, can

cause ionization in a similar way through Compton scattering. The ionization process is very simi-

lar to the photoelectric effect discussed in Chapter 2, and in the active silicon region i t merely

results in electron-hole pairs which will then diffuse and drift through the device, if they do not

immediately recombine. High energy photons (>I MeV) may also produce electron-positron pairs.

although the probability of this type of event is extremely low. Like photo-generated camers, the

holes will migrate toward the substrate or channel ctops. and the electrons toward the potential

wells where they will be collected as part of the signal. In nuclear particle detectors or x-ray imag-

3 5

Radiation Damage

ing, this is precisely the desired effect used to detect the passage of high-energy particles or x-rays.

In other applications the signal is spurious. A well known phenomenon of this sort which occurs

even in ground-based CCD astronomy is the appearance of spots or streaks in an image resulting

from the passage of cosmic rays (see part F for a description of cosmic rays). However, these spu-

rious signals can often be removed by image processing, for example, taking two images of the

same scene and eliminating any artifacts not present in both. In any case, in the active silicon

region, the effect is not permanent and is therefore of little concern. If the incident radiation causes

ionization in the oxide or other insulating material, however, the effect can be permanent.

The insulating layer has a much wider bandgap than the semiconductor, and therefore i t takes a

larger amount of energy to excite electrons to the conduction band (about 18 eV per electron i n

S i02 [26 ] i , and the mid-gap trapping states are correspondingly deeper. The existence of large

numbers of deep trapping centres in the oxide means that the electron-hole pairs created i n the

oxide layer which escape recombination can be trapped for long periods of time, or essentially per-

manently. In MOS devices under positive gate bias, the electrons are usually swept out of the oxide

very quickly by the high electric field, but the holes tend to be trapped near the semiconductor-

oxide interface [27]. The positive charge buildup due to the trapped holes alters the electric field in

the device, and results in a shift in the flat-band voltage. These changes to the flat-band voltage can

be compensated for by simply adjusting the operating voltages.

Ionizing radiation also creates trapping states at the semiconductor-oxide interface. These inter-

face states can have several effects. I f they are deep trapping states, holes or electrons can be held

semi-permanently at the interface, resulting in charge buildup as above. Interestingly, the negative

charge of trapped electrons can compensate for trapped holes and actually reverse the damage.

Radiation Damage

Shallower interface trapping states can severely degrade the charge transfer efficiency in a surface

channel CCD, so in radiative environments buned channel devices are invariably used. Ionization

damage can also provide mid-gap levels for carriers to thermally "hop" between the valence and

conduction bands, which means an increase in the dark current, but because this occurs at the sur-

face, it can be significantly alleviated by one of the techniques described in Chapter 3. Finally, the

interface traps due to ionization damage can affect the output transistor on the CCD. manifesting

itself as increased read noise due to trapping, though once again buried channel devices are used to

reduce the effects.

B. Displacement damage

The remaining, non-ionizing fraction of the energy deposited by the radiation goes into displace-

ments. Displacement damage occurs when the incident radiation interacts directly with the atomic

nucleus with enough energy to displace the atoms from their positions in the crystal lattice (Figure

17). The recoil atom from the initial collision may travel some distance through the silicon and

undergo further collisions or ionization interactions producing more recoil atoms and leaving a

trail of displaced or ionized atoms in its wake. The displaced atoms end up in interstitial positions.

leaving vacancies in the lattice, and the combination is called a Frenkel pair. Displacement damage

is pt-ima~ily caused by heavy panicles such as protons or neutrons, although electrons above a cer-

tain threshold ( - 180 keV) and even photons deposit a small fraction of their energy in displace-

ments [28]. Very high energy particles, especially neutrons because they are not subject to

Coulombic forces, may interact directly with the nucleus of the silicon atoms and create a cascade

of secondary particles. The ejected secondary particles can then also cause further displacements

or ionization.

Radiation Damage

Figure 17: Displacement damage in silicon.

After the initial damage. a rearrangement of the atoms occurs through thermal motion. Most of the

interstitial-vacancy pairs created by particle radiation recombine and have no permanent effect.

Typically 2% of the initially generated pairs remain [27]. The vacancies which do not recombine

are unstable and will migrate to more favourable positions in the lattice, often combining with

other vacancies or becoming trapped near impurities because of the stress these atoms cause to the

lattice. These vacancy-vacancy and impurity-vacancy complexes introduce new mid-gap energy

levels, which have the same effects as interface states, except that they occur in the bulk silicon.

/ They produce an increase in the bulk dark current due to thermal hopping and they produce I

increases in the CTI and read noise due to charge trapping. Large localized increases in dark cur-

rent, or "hot" pixels, are frequently observed, which may be due to clusters of defects. Because the

permanent effects of displacement damage are not confined to the surface as in ionization damage,

they are seen even in buried channel devices and the techniques for reducing the dark current at the

wrface are inadequate. Therefore. in modem scientific CCDs, the displacement damage is more

Radiation Damage

important than the ionization damage

C. Bulk Trap Levels

The most important of the radiation-induced defects in the CCDs we are studying is one which

introduces a bulk trapping state with m activation energy of about 0.4 eV below the conduction

band. Most CCD researchers [29-31 have identified this trap as being due to the phosphorus-

vacancy (P-V) complex (or E centre) because of the high concentration of phosphorus impurities

which is used to create the n-type buried channel of the CCD. However, other researchers attribute .

this energy level to a singly charged vacancy-vacancy (V-V) complex (or divacancy) [36] or a

combination of the two defects [37]. Benton [38] gives a comprehensive listing of silicon defects

and energy levels in which the E-centre and the singly charged divacancy are listed with distin-

guished levels of 0.34 and 0.41 eV, respectively. Table 1 summarizes the various trap levels

reported. I t should be noted that the energy level and cross-section are very difficult to resolve sep-

arately because their effects are closely coupled. This helps explain the divergence in the values

shown. Two other trap levels are commonly reported in radiation-damaged devices. These are the

Table 1 : Radiation-induced trap levels

Trap level EcE, (eV)

I

0.14

0.23

0.4 1

0.4

0.36 f 0.06

Trapping cross section

0,

-

Identification

0- V

V-V'

V-V + unknown

P-V

P-V

Researchers

N. S. Saks [29]

J . Janes~ck el al. 1301

K.C. Gendreau et al. [?4]

Radiation Damage

Table 1 : Radiation-induced trap levels


on

Trap level EcE, (eV) Identification Researchers

N. Meidinger and L. Striider [35]

1.i-1. Hopkins, G.R. Hopkinson. and 8 . Johlander [33]

A. Holland [32]

M. S. Robbins. T. Roy, and S. J. Watts [3 1 ]

J.L. Benton and^.^. Kirnerling 1381

S. Coffa er al. [36]

0 - V

v-v=

v-v + P-v

B.G. Svensson, C. Jagadish, and J.S. Will- iams 1371

oxygen-vacancy ( 0 - V ) complex ( A centre), which has an activation energy of 0.18 eV, and the

doubly charged divacancy (V-V=). which has an activation energy of 0.23 eV [37,38].

D. DLTS measurements

We performed deep level transient spectroscopy (DLTS) measurements [39,40] on a number of

Radiation Damage

buried channel MOS transistors which had been irradiated as part of an earlier investigation into

noise effects [41]. DLTS is a method of investigating trapping states by measuring the exponential

decay of the trapped charge. From the variation in the exponential time constant over a range of

temperatures the trap parameters can be determined. We used a constant resistance DLTS (CR-

DLTS) method in which the charge state of trapping levels is monitored through the change in the

a full description of the experimental method and setup, see

[39,40] 1

The test devices were lig ly-doped drain (LDD) depletion-mode n-type buried channel MOS- 9 FETs that were fabricated by Tektronix, Inc. as part of their CCD development program [42].

Three dies, each consisting of 15 independent transistors, were placed in 24-pin ceramic packages.

Each package contained transistors with width (W) to length (L) ratios *om 60 p d l O pm to 27

pm115 pm and LDD lengths varying from I to 4 pm. Two of the packages were placed in the beam

of the University of Western Ontario tandem accelerator and subjected to radiation by 1 MeV pro-

tons. The irradiations were performed at room temperature and all pins were grounded. One set of

devices received 5 .0~10 ' protons/cm2 and the other received 2 . 7 ~ 10' protons/cm2, as determined

by a previously calibrated event counting detector. The third set of devices was not damaged.

Figure 18 shows the DLTS temperature spectra measured for three different transistors of varying

proton dose. The ordinate is the change in threshold voltage measured over a fixed interval during

the exponential transient, in this case 7.44 ms. Five peaks are distinguishable in the spectrum of

the most damaged transistor, each corresponding to a different trapping level. An Arrhenius plot

(see Chapter 6 for an explanation) of the positions of these peaks for several different intervals is

shown in Figure 19. A linear least-squares fit to the data reveals the trap parameters, which are

Radiation Damage

E2

1 2 . 7 ~ 1 0 ~ protons/cm2 - t I

L i i / I 5 . 0 ~ 1 o8 protons/cm2

i

i I undamaged

E 3 I

- -

Figure 18: CR-DLTS spectra of radiation-damaged buried channel MOSFETs compared with the spectrum of an undamaged device 1401.

summarized in Table 2. Again we see a dominant peaks at around 0.43 eV and 0.23 eV. suggesting

the divacancy and the phosphorus-vacancy complexes, and at 0.17 eV, which is very likely the

oxygen-vacancy complex.

E. Annealing

After irradiation, the defects caused in a device can be repaired through thermal motion of the

atoms in the lattice. This process is called annealing and i t is highly temperature dependent. In

fact, the temperature dependence of the annealing process is often used to identify the defects

introduced by radiation because i t is different for different defects. The divacancy levels show little

Radiation Damage

Figure 19: Arrhcnius plot using the CR DLTS data of the device which recelved 2.7x10y protondcm2 [30]

Table 2: DLTS measurements of irradiated buried-channel MOS transihtors.

Trap label Trap level

Ec - E, (eV)


0, (cm- 2 )

1

Radiation Damage

annealing below -300 "C. and the oxygen-vacancy complex is stable to -350 "C, whereas the

phosphorous-vacancy has a characteristic anneal temperature of - 150 "C [38].

P Holland [43] has performed tests of the effect of annealing on the proton damage in CCDs and

found that 85% of the detectable damage could be removed by annealing at 160 "C for 16 hours.

This suggests that the damage was due to the E centre. Robbins et rrl. [31] report an almost com-

plete elimination of the trapping effect from the radiation-induced level at -0.4 eV after annealing 1

at 150 "C, again suggesting that the P-V centre is responsible. Robbins also observed an increase -3

in the trapping at a shallower level after the anneal, which was attributed to an increase in the den-

sity of 0 - V centres. We did not perform any experiments to investigate the effect of high-tempera-

ture annpling, but these results indicate that i t could be a successful means of alleviating radiation

damage. I t may not be practical for a spacecraft mounted device, however. Provision must be made

for a high-power on-board heater or periodic reorientation of the spacecraft to make use of solar

heating, and the CCD package must be able to survive the elevated temperatures required. The

FUSE FES design does include heaters to maintain the target operating temperature, but they are

insufficient to raise the temperature of the CCD above about 30 "C. Solar heating would be a pos-

sibility. but the maximum temperature stated for the CCD package used in the design is about

60 "C. Therefore, annealing is not possible without significant design changes.

E FUSE radiation environment

Spacecraft such as the FUSE satellite which operate in low earth orbits (LEOS) are subject to three

major sources of radiation (441:

heavy ions trapped in the magnetosphere,

4

Radiation Damage

protons and electrons trapped in the Van Allen belts.

cosmic ray protons and heavy ions. and

protons and heavy ions from solar flares.

Cosmic ray particles originate outside the solar system and include ions of all elements from

atomic number I to 92 with energies from around I0 MeV to 100s of GeV, which makes them dif-

ficult to shield against. most radiation originating in space, they are able to penetrate the

P earth's magnetic fields devices on the ground. The heavy, highly energetic particles pro-

duce intense ionization as they pass through matter, however the flux level of these particles is low

even for LEO, so although they are a concern in terms of the spurious signals they generate. we

will not consider them in estimating the permanent damage. The heavy ions trapped in the mag-

netosphere are largely of such low energy that they are not able to penetrate a spacecraft to affect

the electronics and the trapped electrons cause only small amounts of damage, so neither are of

much concern. The trapped protons, however, along with the solar flares, are very difficult to

shield against and can be a significant source of damage. The protons in the Van Allen belts vary in

energy from keV to hundreds of MeV and in intensity from 1 to I X I 05 protons/cm2/sec. The actual

populations depend on the altitude and inclination of the orbit. the cyclic activity of the sun, geo-

magnetic storm perturbations and the gradual change in the earth's magnetic field. The solar flare

activity is random and difficult to predict with certainty, although several probabilistic models

exist. /'-- I

\

'%

We have made estimates of the expected damage to the FES CCD due to energetic protons encoun-

tered by the device in orbit. The radiation environment was taken from the calculations by Stusi-

nopoulos [ - IS ] for a 700 km. circular. 28 degree inclination orbit and scaled for the basiline FUSE

Radiation Damage

orbit (800 km, circular, 25 degrees).

Figure 20 shows the expected total proton flux for the three-year mission as a function of proton

energy. The flux for three different shield cases are shown: 5 mm aluminum. 0.4 mm aluminum,

Energy (MeV)

Figure 20: Total proton t l u ( n e r a three-year mission

and no sh~eld . Also shown i h the spectrum of unattenuated solar flare protons for four anomalously

large holar flares, the number recommended in [45]. I t should be noted, however. that at the incli-

natton of the FUSE orbit (<-I5 degree\). all of these solar flare protons will be stopped by the geo- I

magnetic field and are therefore of no concern [GI.

Radiation Damage

Once the total proton flux is known, it is necessary to take into account the varying penetration

depth of protons of different energies in order to predict the effective damage to a device. The most

harmful protons are those which are absorbed in the active region of the silicon, which is usually a

layer about 2 microns thick. We took the spectra from Figure 20 and calculated the number of sili-

con lattice displacements expected in the active region or the CCD. This was based on data gener-

ated by Janesick et al. [30] using TRIM [46] software, who assumed proton incidence on the

frontside with an oxide layer of 2 pm and a polysilicon (gate) layer of 0.2 pm above the active sil-

icon. The number of displacements as a function of energy is shown in Figure 21. From this figure,

- - - - flares

Energy (MeV)

Flpurc 7 I . Total dlsplacemcnt> w e r a threc-year rnlslon

Radiation Damage

i t is clear that the most damaging protons are those at around 250 keV. Those with lower energies

cause no significant damage because they do not penetrate to the active region of the device and

those with much higher energies pass right through.

The displacement spectrums were then integrated to find tKe total number of active region dis-

placements expected over the FUSE mission. Since the solar flare protons will be completely

attenuated as mentioned above, they were not included in the estimates. The results are given in

Table 3.

Table 3: Integrated displacements for a three-year rnisslon

I Shielding I ~ i s ~ l a c e m e n ~ c r n ~ (

I no shield I 3 . 5 9 ~ 10' I

These results indicate that shielding with aluminum has a significant effect and should be 'I 1

employed to minimize the damage caused by radiatiq.

Dark Current

4. DARK CURRENT

A. Theory

The dark current in a CCD is the charge generated in the imaging area which is not caused by the

photoelectric effect. It will accumulate in the wells under the pixels even when no light is

incident on the device, hence the name "dark" current. The dark current is a problem in scientific

imaging for two reasons. The first is that it introduces an unavoidable background signal which

builds up in the potential wells during an exposure until the device reaches full-well and saturates

the image. This limits the exposure length which can be used, a factor which is particularly impor-

tant to astronomical applications where exposures can last up to several hours in order to detect

very faint objects. The second problem with dark current is that because i t is a random process. i t

introduces noise.

There are three sources of dark current:

I. therkal generation in the depletion region due to bulk traps:

2. thermal neneration at the silicon-silicon dioxide interface due to interface states; and

3. diffusion currer?t at the edge of the depletion region.

Bulk generation

Thermal generation of carriers in semiconductor materials is the result of electrons possessing

enough thermal energy to be excited from the valence band into the conduction band. According to

Fermi-Dirac statistics, the probability of an electron having enough energy to do this at thermal

equilibrium is [47]

Dark Current

where Ec is the energy of electrons in the conduction band and Ef is the Fermi energy. At normal

temperatures, very few electrons will have the required energy to make this transition directly.

However, the process is assisted by the presence of energy states in the band gap (or traps). When

these states are present, electrons can be excited to the conduction band in two steps. first to the

midgap state and then to the conduction band. Each of these transitions requires less energy than

'he direct one and is therefore more likely to occur.

There are four transitions to be considered for a trap with level E,, shown in Figure 22: (a) an elec-

tron jumps from the conduction band to the trap level (electron capture), (b ) an electron jumps

from the trap level to the conduction band (electron emission). ( c ) an electron jumps from the trap

level to the valence band (hole capture), and (d) an electron jumps from the valence band to the

trap level (hole emission).

Figure 2 2 : Generat~on and recomhination of carriers through trap levels. ( a ) electron capture; ( b ) electron emission; ( c ) hole capture; (d ) hole ernisuon -

Dark Current

The rate at which process (a) occurs depends on the density of electrons in the conduction band n

and the number of empty traps. It is given by

where vrh is the thermal velocity, n is the free electron concentration. N, is the density of traps and

f is the probability that a trap is occupied. The rate of process (b) is proportional to the number of

,filled traps

Here the proportionality constant is the electron emission probability en:

( E , - E , ) / k T en = vr , ,onn,e

where n, is the intrinsic carrier concentration and E, is the intrinsic Fermi energy. Similar expres-

sions can be written for the hole processes (c) and (d). The four processes will drive the fraction of

filled traps f to an equalized point where the number of electrons entering the trapping states is

equal to the number of electrons leaving, i.e.

Using Equation 7 we can combine the above expressions to determine f at this point, and hence

determine the net rate of camer generation C :

o , o p v I , , ~ , ( n : - n p ) G = ( E , - E , ) / k T ( E , - E , ) / k T

o , [ n + nie 1 + o , [ p + n , e 1

Dark Current

( 9 )

where p is the density of free holes and op is the hole capture cross-section.

When a semiconductor is in steady-state, nj2 = n p and the net generation rate is zero. For a device

in deep depletion, such as a CCD pixel during exposure, n and p are much less than n, in the deple-

tio r! region and the net generation rate becomes

G = on"pvihNini - - - i ( E , - E , ) / k T ( E , - E l ) / k T

o , n , e + o p n i e 2 TO

where ro is called the "effective lifetime within a depletion region" and is given by

( E l - E , ) / k T ( E , - E l ) / k T One + o p e

To = 2 o n o p v , , N ,

The bulk dark current density (in nA/crd) can now be written

where .rd is the width of the depletion region. The dark current in electrons per pixel can be found i

by multiplying equation ( 1 2 ) by the pixel area A, and dividing by the electronic charge q

Dark Current

For states near the midgap. phich dominate the bulk dark current generation because of the expo-

nential dependence of the generation rate on the trap energy. ro is nearly constant and the primary

temperature dependence of the dark current comes from the intrinsic carrier concentration n,,

which is given by ,

where Nc and Nv are the effective densities of states in the conduction and valence bands.

Surface generation

At the interface between the silicon and the silicon dioxide, the periodic lattice structure of the

crystalline silicon is disrupted and results in a high density of midgap states. These interface, or

surface, states cause a high rate of thermal generation at the surface. The process is identical to

bulk generation, except that instead of isolated trapping states, there is a continuum of energy lev-

els distributed across the bandgap. Instead of a single trap concentration we now have an

energy-dependent trapping state density D(E,), and must integrate the generation rate over energy

to get the net generation rate G:

~ , o ~ ~ ~ , , ~ @ ( E , ) ( E l - E , ) / k T ( E , - E , ) / k T + o,,n,e

dE1 W ' , e

Again, because the generation rate is exponentially dependent on the activation energy E,, the

dominant generation centres are those within a few kTof the midgap energy E,. The distribution of

states D(E,) has been found to be fairly uniform across the bandgap, except for significant peaks

near the conduction and valence band edges where the con!ribution to thermal generation is

53

Dark Current

extremely low. Therefore, for calculation purposes, it is reasonable to use a uniform distribution,

D(E,) = D,. If we make the further simplification of replacing op and 0, with their geometric mean

0 = A v i [48], the integral in ( 1 5) can be evaluated analytically to give

2 and we can write the surface dark current J,, in nAkm . as

where so is the "surface generation velocity" and is given by

The dark current in electrons per pixel can be found by multiplying equation ( 17) by the pixel area

Ap and dividing by the electronic charge q

As with bulk dark current which is dominated by mid-gap states, the main temperature dependence

of the surface dark current comes from the exponential term in n,.

Surface dark current suppression

Because of the high density of states at the silicon-oxide interface, the surface generation forms (he

largest component of the dark current (up to 98%) in most CCDs. Fortunately, in buried-channel

devices, i t is possible to significantly reduce or eliminate the surface dark current by running the

54

A Dark Current

clocks in an inverted mode. If the gates of the device are biased sufficiently negative, an inversion

layer of holes will be created at the surface. These free holes will f i l l the interface states and pre-

vent them from capturing electrons from the valence band, thus halting the carrier generation pro-

cess. Therefore, while the surface is inverted, the surface component of the dark current is

essentially zero. To take advantage of this phenomenon, i t is necessary only to set the low level of

the transfer clocks to a sufficiently negative potential. However, for a three-phase device, one of

the phases must remain high during an exposure in order to create a potential well for collecting

the signal charge and the surface under this phase will still produce dark current. There are two

methods for dealing with this difficulty: placing an extra implant under one of the phases so that all

the phases can be inverted at the same time, or using a dithered clocking scheme.

The first method involves performing an extra implantation step during the CCD fabrication pro-

cess. The usual method for an n buried-channel device is to add a light boron implant under one of

the three phases. The negatively charged impurity atoms lower the potential under that phase, even

in inversion, and allow all three phases to be inverted while maintaining a potential barrier between

pixels so that the charge packets are still confined (Figure 23). Devices with this implant are

referred to as Multiple Pinned Phase (MPP) devices, since all the phases can be inverted, or

"pinned" at the same time. Because of the dramatic difference that surface dark current suppres-

sion can make, many CCD manufacturers offer MPP versions of their devices.

For devices without an MPP implant, i t is still possible to suppress the surface dark current under

all the phases of a device using a dithered clocking technique. It has been shorn [48] that when the

surface is switched from inversion to depletion. the surface generation rate remains low for a char-

acteristic period of time before recovering to its steady-state value. Therefore if a phase is only

Dark Current

v ~ 2 " ~ 2 " ~ 2 v ~ 2

gates oxide _ _ _ - - - - - - - - - - - _

buried channel - - - - - - boron implant

potential

-x

Figure 23: Charge confinement in an MPP device.

allowed to remain in depletion for an amount of time shorter than the recovery time before it is

inverted again, the surface dark current will be almost completely suppressed. To do this during

long exposures, it is necessary to rapidly switch the charge collecting potential well back and forth

between two of the phases, with each of the two phases being inverted while it is not collecting

charge. If the switching period is kept short enough, this will effectively suppress the dark current

through the entire period.

Diffusion current

The third source of dark current, the diffusion current, is caused by the minority carrier gradient at

the edge of the depletion region. At the depletion edge, the concentration of minority carriers is

zero and increases exponentially into the substrate. For a p-type substrate the minority carriers are

Dark Current

electrons and therefore there is an electron diffusion current flowing into the depletion region from

7 the substrate. This current is proportional to n,- (as compared to simply n , for thermal generation),

and at usual CCD operating temperatures (< 100 "C) it is negligible [49]

6. Experimental results

We made a number of measurements of the dark current in radiation damaged CCDs. The CCDs

we measured are TK5 12's, made by Tektronix, with 51 2 x 5 12 square pixels. each of 27 pm size.

These are three-phase, thinned, backside-illuminated buried-channel devices with two serial regis-

ters and outputs. The CCDs were subjected to bulk damage by means of a 3 MeV proton beam in a

i manner similar td that described in [41,51] and the DLTS measurements in Chapter 3. The imag-

jng area and serial registers were divided into three zones as shown in Figure 24. Section A was

shielded and the device was placed in the beam of the accelerator and subjected to 1 . 5 ~ 1 0 ~ pro-

tons/cm2, as measured by the calibrated event counting detector. The shield was then adjusted to

cover both sections A and B and the device was subjected to a further 4 . 5 ~ 1 0 ~ protonslcm2. Both

irradiations were conducted while the device was cooled to -80 "C. The cumulative result of the

irradiations was to produce three radiation levels of 6.0x 1 0~ro tons1cm2, 1 . 5 ~ 1 o9 protons/cm2

and no radiation in sections C, B, and A respectively. Based on calculations similar to those

described in Chapter 3 for the FUSE radiation environment, we estimated that this would result in cy.

the two damaged regions containing about 9 . 0 ~ 1 0 ~ and 2 . 3 ~ 10' displacementslcm2. The undam-

aged section A was used as a control region. In all of the experiments reported here, which were

conducted roughly 2 years after the irradiations were performed, the CCDs were always read out

of the undamaged output amplifier in section A

Dark Current

I I

serial - [ I 1 ID register I I

I I I

imaging area -

output

-

amplifier I I I I

I I

1 1 1 1

I

I

undamaged 1 1 .5x109 I 6.0x109 1 protons/cm2 ; pr~tons/cm2 I

Figure 24: Radiation damage of the CCDs used in our experiments.

The experimental setup is shown in Figure 25. The CCD is mounted in a liquid nitrogen dewar

with a quartz window. A low noise amplifier is located a short distance from the dewar to provide

some gain ( 5 to 10) and to buffer the CCD output. An electronic chassis is located within one

metre of the dewar, containing the clock drivers and bias supplies for the CCD. I t also houses the

analog signal processing chain which performs a correlated double sample with the dual slope

method [41] prior to passing the signal to the 16-bit analog-to-digital converter. The electronic

chassis is under the command of, and passes its data to, the data acquisition processor. Raw data

are saved as disk files in the FITS (Flexible Image Transport System) standard 1.501, and are subse-

quently analyzed.

Dark Current

electronic chassis preamplifier

I

CCD

cold f block clock

drivers

supplies bias 1 temperature controller I

host PC c Figurc 25: Expcrimental setup for CCD measurements

The temperature during the experiments is regulated with the help of a calibrated silicon diode

mounted in the cold block contacting the CCD. The diode voltage drop 3t a constant current is

used as input to a linear temperature regulator capable of depositing up to about 10 watts of heat at

the CCD mount. The thermal connection between the CCD mount and the 77 K station in the 1

dewar contains a variable resistance link to allow us to further extend our experimental tempera-

ture range. A platinum resistance temperature detector (RTD) is clipped directly to the CCD pack-

age and serves to indicate the temperature of the device.

The TKS 12 h a an implant allowing i t to be run in an MPP mode. so we measured the dark current

Dark Current

with the device running in this mode as well as in a normal inverted mode (two ~ h a s e s inverted).

Figure 26 shows the results over the range 180-280K. Above around 270K the dqrk current was so

high that the device would saturate during the readout period. Below 180K the dark current was * t;r

too small to be measured without extremely long exposures.

x undamaged

0 1.5 x 1 o9 protons/cm2 I

+ 1 5 x 1 n9 nrntnns/cm2 / , , , --: surface component , , , , 4 bulk component

A 82 2C0 220 240 260 Tempera t i i r e ( K )

F ~ p u r e 26: Dark current as a function of temperature. The symbols represent measurements in the three sections of the device. The solid lines represent the bulk generation model and match the MPP mode data. The dashed lines represent the surface generation model and match the non-MPP mode data.

The figure shows six sets of data. The upper three are the data obtained while running the device in

non-MPP mode, with one set for each of the three sections of the CCD: unirradiated, low radiation,

Dark Current

and high radiation. The lower three data sets were obtained with the device in MPP mode. The data

clearly show an increase in the dark current with increased radiation.

We used the models given in the above sections to fit theoretical curves to the data. The dashed

lines represent the surface generation model from Equation ( 1 9). which was fit to the non-MPP

data because this data should be dominated by the surface dark current. We found that our initial

data did not match the expected teniperature dependence. Since the theoretical dependence has

been so well established by numerous CCD researchers, we assumed that our temperature readings

must be wrong. We postulated that there was a temperature gradient between the RTD and the . ,

CCD due to an imperfect thermal connection so that the RTD reading deviated from the actual

device temperature by an increasing amount as the C ~ D - W ~ S cooled. We were able to correct fo? .

this by applying a linear transformation to our temperature data. A good f i t to the theoretical

dependence was found by adjusting the RTD readings by 7% of the difference from room temper-

ature. The data shown in Figure 26 have been adjusted by this factor.

For the unirradiated section of the device the plotted model in Figure 26 uses

D, = 1 . 6 ~ 1 o8 cmP2 ev-I. and o = 5x 1 0-l6 ern-?. In the high radiation case we evaluated the

16 7 model with D, = 4 x 1 0 ~ cm-2 ev-l , and o = 5x10- cm--, and for the low radiation case we used

exactly one quarter of this Dl value. Since we have no way of independently determining the a for

the surface states, there exists an ambiguity in the D, and o values. For our results we simply chose

o to be representative of typical values and found the corresponding D, which provided the best fit

to the data. The dark current for a CCD is usually quoted as the room temperature rate in nA/cm2.

Our measurements here correspond to room temperature currents of 0.15. 0.9, and 3.6 nA/cm2 for

Dark Current

the unirradiated, low radiation and high radiation sections of the device.

The solid lines represent the bulk generation model of Equation (13). We fit this model to the MPP

mode data because this data should have no surface component due to the total inversion of the

surface in MPP mode operation. The displayed curves were calculated using a depletion depth .rd

of 4.3 pm (based on device simulations), and assuming a trap at the midgap with E, = 0.55 eV

below the conduction band and on = op = 5 x 1 0 ' ~ cm-l. For the unirradiated case we used a trap

concentration N, = 4x 1 o9 ~ m - ~ , for the low radiation case N, = I x 10"' cm-', and for the high radi-

1 o ation case N , = 4x10 c m 3 . However, as in the surface current case above, there is some ambigu-

ity here between the trap concentration and the trapping cross section, and the o values were

simply chosen as typical. These results correspond to room temperature rates of 0.03 1.0.076, and

0.3 1 nNcmZ for the unirradiated, low radiation and high radiation sections of the device.

f' We see from these results that the damage is directly proportional to proton fluence. The density of

radiation-induced traps in the high radiation section, for both surface states and bulk states, is

exactly four times greater than in the low radiation section.

Interestingly, the bulk dark current and surface dark current show an identical dependence on tem-

perature. This is because they are both dominated by trapping levels near the midgap. We origi-

nally expected the bulk dark current to show an activation energy of around 0.4 eV because this

(identified as the E centre) was the dominant level in both our CTI measurements (see Chapter 5 )

and our DLTS measurements (Chapter 3). as well as being the most widely reported trapping level -

in radiation damaged CCDs. To our knowledge, a bulk trap level of 0.55 eV has not been reported

before by CCD researchers. However, the result is not surprising. If a 0.55 eV level exists, because

C

Dark Current

i t is so close to the midgap, i t would dominate the dark current generation due to the exponential

dependence of thermal generation on the activation energy of the trap. A 0.55 eV trap will produce

roughly 160 times as much dark current as a 0.4 eV trap for the same trap concentration, so it

would be impossible to see the effects of the 0.4 eV trap on the dark current. The 0.55 eV level

would not have shown up in our CTI or DLTS measurements because it is too deep. Also, as Saks

[29] points out, the traps responsible for the dark current generation may be located in the depleted i

portion of the p-type substrate, whereas to contribute to the CTI they must be in the n-type buried

channel. The substrate does not contain the high concentration of phosphorous impurities which

go into creating the E centre. and therefore entirely different trapping states may dominate

Noise

Thermal generation is governed by Poisson statistics in which the variance is equal to the mean, so

the inherent noise in the process (the standard deviation) is proportional to the square root of the

mean level. However, in CCD images there is additional noise because the mean generation rate

varies from pixel to pixel, and the total noise shows a linear dependence on the global mean level.

This additional variation seems to be relatively 'k able over time, and therefore can be measured

and sy,stematically removed. Figure 27 shows measurements of the dark current noise in the high

radiation section of the TKS 12. The data indicated by crosses was obtained with the device oper-

ated in MPP mode, so that data is dominated by bulk dark current. The data indicated by diamonds

was measured in non-MPP mode, so that data is dominated by surface dark current. In Figure 27.

the upper two sets of data represent the raw dark current noise, which follows a linear dependence

on the dark current level. The solid lines represent linear least-squa;es fits to the data. The upper

(MPP) line has a slope of 0.32 and the lower (non-MPP) line has a slope of 0.048. The noise devi-

Dark Current

+ MPP mode 0 non-MPP mode

- - - residual noise

0 4 5.0~10 1 .Ox10 5 1.5~10 5

Mean dark c h a r g e ( e - )

Figure 27: Dark current noise. The upper two sets include spatial noise. the lower two sets have the spatial noise subtracted out. The lines are least-squares fitted curves.

ates from linearity at the high end, which is probably due to the saturation of the signal. The full

well for this device is around 200000 e- and as the mean approaches this level the noise (standard

deviation) will be lower than i t should be because the upper tail of the distribution will be cut off.

The lower two sets of data represent the noise with the spatial noise component subtracted out. To

obtain this data we took two images at each dark current level, and subtracted one from the other.

The fixed pattern noise was thus removed. The residual noise, adjusted for the noise increase from

the subtraction, displays the expected square root dependence in the non-MPP case. as indicated

by the lower dotted line. In the MPP case, the residual noise still follows a linear curve with a slope

Dark Current

It is interesting that the MPP mode (bulk) noise is greater than the non-MPP mode (surface) noise

for a given dark current level, both before and after bias subtraction. From the bias-subtracted data

it appears that the noise associated with bulk generation exceeds Poisson statistics. However, for a

given exposure length, the mean MPP dark current level is so much lower that the resultingdark

current noise is still less than for the non-MPP dark current.

We conclude that an effective means of reducing the noise in an image due to the radia-

tion-induced dark current, as well as eliminating the bias introduced by this dark current, is to sub-

tract a dark frame. A dark frame is an exposure of equal length at the same temperature with no

light incident on the device. Subtracting the dark frame will eliminate the spatially fixed noise

caused by the dark current, leaving a much lower residual noise. In the surface dark current case,

the residual noise is at the fundamental limit imposed by the Poisson statistics which govern the

generation process. We tried to improve the noise reduction of the dark frame subtraction by creat-

ing a "super dark frame", an average of several long dark exposures. This super dark frame was

then scaled to the length of each exposure and subtracted. We found that this did not improve the

noise performance - the best noise removal was obtained when the dark frame had the same length

of exposure as the image, and was taken at about the same time. This may be due to small fluctua-

tions in the temperature of the device over time as a result of imperfect temperature regulation.

Charge Transfer Efficiency

5. CHARGE TRANSFER EFFICIENCY

The charge transfer efficiency (CTE) of a CCD is the amount of charge in a given packet which is

su&essfully clocked past the series of gates from one pixel to the next. This factor is usually repre-

sented as a single fraction, typically 0.999990 for a modem CCD. Because the CTE is so high for

most CCDs, it is more conveniently expressed in terms of the charge transfer inefficiency

(CTI = I -CTE = 1 X I o - ~ ) . The use of such a number implies that the total signal loss is simply 'pro-

portional to the amount of charge in a packet and the clocked distance. However, these and other

experiments [I 3, 33, 5 1-54] show that the CTI is dependent on a number of other factors, includ-

ing temperature, clocking speed and even the history of charge packets clocked through a given

pixel.

The charge transfer efficiency is an important parameter in scientific CCDs, especially for the

demanding photometric applications of optical astronomy. Astronomers are frequently interested

in the precise brightness of a given object. and if an indeterminate amount of signal is lost, it is

impossible to estimate accurately from the CCD image. If the CTE is poor, some faint objects may

even be completely obliterated. Low CTE also has the effect of smearing the image, which will

affect the CCD's usefulness in assessing object morphology (shape) or position. The latter is of

importance to our work for the FUSE mission, since the CCD is to be used to track the position of

stars in order to maintain the pointing of the main telescope. If the centroid of a star image deviates

in unpredictable ways by even very small amounts (less than one tenth of a pixel), the performance

of the Fine Error Sensor will be compromised. Figure 28 shows an image taken by the damaged

TK5 12 CCD we used in our experiments (see Chapter 4) in which the smearing effect can be


clearly seen. The point images in the most damaged section (section C) have been smeared into

F~gure 28: Effect of poor charge transfer efficiency. This CCD image shows severe C E problems in the radiation damaged section of the CCD on the right side.

streaks, especially those in the upper right comer, which have undergone the most transfers (the

readout amplifier is at the lower left comer).


A. Simple physical model

All three of the charge transfer mechanisms mentioned in Chapter 2, and therefore the CTI, are

sensitive to temperature. Thermal velocity decreases with temperature, slowing charge transfer due

to thermal diffusion. On the other hand, the carrier mobility will increase as the temperature drops,

resulting in improved effectiveness of fringe-field and self-induced field drift. In addition, the effi-

ciency of transfer due to the self-induced field is dependent on the size of the charge packet. How-

ever, the time constants of all these processes are on the order of a few nanoseconds, and at the

clock rates used for astronomical CCDs (-Ips per transfer) the &ffects on the CTI are negligible.

In such slow-scan applications, two mechanisms which increase CTI through charge deferral are

k more important: potential pockets and bulk states or traps in the silicon. Potential pockets are

irregularities in the potential well shape in which signal charge can be caught during transfer. In

well-designed buried-channel CCDs these pockets are minimized or eliminated, and trapping by %

bulk states is the more important charge-deferral mechanism. Trapping is certainly the most

impoifant effect for any CCD which has suffered bulk damage from energetic particles. A trapping

state captures charges from a packet and emits them at a later time, which may be after the charge

packet has been transferred to the next pixel. The charge which is "deferred" in this way is lost

from the original packet and is therefore a source of CTI. The following analysis. after Kim [55] ,

can serve as the basis of a charge deferral model resulting from bulk states. A similar analysis is

given in [56] . +

The capture and emission time constants T, and T, fdr the bulk states are


(21)

where El is the trapping state energy level below the conduction band, a, is the trapping cross sec-

tion. vrh is the thermal velocity of carriers, n is the density of electrons in the conduction band and

Nc is the effective density of states in the conduction band. From Sze [57].

where A* is the eih-ective Richardson constant (252 ~ / c r n ~ / ~ ' for n-ty pe < 100, Si). T is the abso-

lute temperature a d y is the electronic charge. We can therefore express the emission time con-

stant as a function of temperature:

When a charge packet is present, the number of charges n, held in bulk states within the packet vol-

ume V , containing N, traps per unit volume is (assuming all traps are full)

When the charge packet ha been removed, 11, will decrease exponentially with emission time con-

stant r, so that


( 2 5 ) dn1 - 1 - - -- d r %

Consider one gate of the typical clocking scheme of a three phase CCD such as p l of Figure 29.

Figure 29: Three phase clock timing with transition points marked for determ~ning the trap emission times.

We assume that the charge packet has substantially arrived under this gate at time T I . Traps in the

volume occupied by the charge packet will be filled with time constant T,, and because this time

constant is relatively short, we will assume the states are filled at T I . At time T2 the charge packet

becomes shared with pZ (as in Figure 1 I(b)), the volume of the charge packet underpl dinminishes

and the traps underpl begin to emit. By time T3 the charge packet has moved on completely to p2. ,-

The traps underpl will continue to emit until T6, or, in the case of a sparsely illuminated CCD,

I Charge Transfer Efficiency

until the next non-empty charge packet arrives. At that time, the states which have emitted their

charges during the period since T2 will be filled froin the new packet, resulting, in a charge loss

from the new packet of

4

where T,,,,, is the total emission time from T2 to the arrival of the new charge packet. A portion of

this charge, however, is regained. Any charges emitted between T2 and T4 will rejoin their parent

packet due to the fringe field (see Figures 1 I (b) and (c)). In fact, some of the charges emitted

between T4 and TF (Figure 1 1 (d)) will also join the parent packet, but because of uncertainty of the

partition function, we will assume that the time period during which the charges can join the parent

packet is c,,, = T4 - T2. The amount of charge that is reintroduced to the packet by these emis-

sions is

and therefore the net loss from a charge packet for one transfer through a p l gate is

where T',,, is the emission time from the previous packet. A similar evaluation using appropriate

values for T,,,, rind T,,,,, can be performed for the other two phases. If more than one trap level


exists, additional terms of the form of the right side of (28) can be added, using the corresponding.

values of r, and N,. i

'A significant complication to the above model is the spatial distribution of the traps and the charge

packets within the bulk silicon. The volume of the charge packet will vary with the number of elec-

trons contained in it , and thus the packet will interact with a varying fraction of the traps within the

pixel. The equivalent two-dimensional effect in surface-channel CCDs is known as the "edge

effect" [58]. This three-dimensional "shell effect" means that each shell of traps within the pixel

will have a different emission period&,,, which will depend on the frequency of charge packets

large enough to interact with it. The traps 'in the outer shells, once filled by the amval of a large

charge packet, will continue to emit until a charge packet of equivalent size amves to refill them.

A further complication is that the packet volume does not have well-defined boundaries, but rather

the charge concentration falls off gradually from the centre of the packet. Our earlier assumption

that the capture time constant r, was short is not valid for small charge concentrations because r,.

is inversely proportional to the charge concentration. If the capture time constant is not short com-

pared to the time the sharge packet spends in contact with the traps, we cannot assume that all the

traps are filled and Equation (24) is no longer valid. In order to calculate the number of filled traps

in the less dense shoulder regions of a charge packet, i t

into account. Trapping theory [33] gives the number of

- -,/I, 1 2 , - n,, . ( 1 - e

is necessary to take this time dependence

filled traps as

where t i h time and n,, is the number of filled traps at steady-state, given by


and r, will vary across the charge packet accordin6 to Equation (20). Therefore, the number of

filled traps will depend on both the charge distribution in the packet and the amount of time t it

spends in contact with the traps (the "dwell time" underneath a single gate [33]). To include this

dependence in CTI calculations, the N,V, term in Equation (28) must be replaced by a volume inte-

gral of Equation (29) which gives

B. Measurement techniques

Pulse train technique

There are various methods which can be used to measure the CTI of a CCD. The pulse train tech-

nique for measuring the CTI was the first one to be used, and was naturally suited to the analog

shift registers that were the first devices to utilize the charge coupling concept. The technique '

involves electrically injecting a stream of identical charge packets into one end of a serial register

and then transferring them to the other end where they are detected by an output circuit. If there is

a significant CTI problem, the output will typically look something like Figure 30. The first, or

loading, pulse experiences charge loss due to the CTI. The remaining pulses do not, if charges are

not lost during transfer, but are simply deferred by the trapping mechanism described earlier. Since

each charge packet is roughly the same size, the packets will interact with the same traps within

the cells of the serial register, and the number of charges trapped will be the same. During each


leading pulse

n deferred charge

trailing pulses

Figure 30: Pulse train CTI measurement.

transfer, the charges emitted by the traps which do not rejoin the original packet will join the fol-

lowing packet and be used to refill the traps so that there is no net loss from the following packet.

The non-leading packets therefore can be used to estimate the original size of the leading packet,

enabling a calculation of the charge loss and the CTI. The charges captured from the last packet in

the train will join empty packets, and will not be able to interact with the traps available to the

larger packets from the train. Therefore, the traps in the outer "shells" will not be refilled, but will

continue to emit their charges into subsequent packets. These trailing pulses give a time profile of

the decay of the trapped charge. -

The pulse train technique is notssuitable for most imaging CCDs because they are not equipped

with an input structure for electrically injecting charge into the registers.

X-ray illumination . One of the more precise techniques developed for measuring the CTI in CCD imaging arrays

involves sparsely illuminating the CCD with monoenergetic x-rays. X-ray photons from nuclear

decay have a predictable amount of energy and because this energy is much greater than the band-


gap, when absorbed in the CCD each photon produces a small cloud of free electrons (see Chapter

2). The cloud contains a statistically consistent amount of charge with a mean value equal to the

photon energy divided by 3.65 eV and a variance equal to the mean value multiplied by the "Fano

factor", which is typically about 0.1 [59]. Because this cloud is smaller than the size of a pixel

(< 1 pm), all the charge is collected into a single packet and the result is a well defined charge

packet size. For example, the Mn decay resulting in ~ e " produces photons of 5.9 keV which cre-

ate charge packets with a mean size of 1620 e' and a standard deviation of 13 e-. By illuminating

the CCD with such a radioactive source and examining the resultant image, we can measure how

the charge packet changei as a function of clocking distance through the device, and thereby mea-

sure its CTI. This is commonly accomplished by taking a series of images and plotting on the same

graph the pixel values against the row number (for parallel CTI) for all the images. Each row

should have a cluster of pixel values at around 1620 e', corresponding to x-ray events, but the

8

mean value will decrease for rows farther from the output due to the CTI. The slope of the line fit

to these points is equal to the CTI. Figure 31 shows such a plot for an undamaged CCD. As one P

can see from this plot, the slope becomes difficult to accurately determine for CCDs with very low

CTI due to the noise in the signal. The minimum detectable CTI depends on the number of trans-

fers. in this case (512 row transfers) the limit is about 5 x 1 0 - ~ . It is possible to improve the resolu-

tion by shuffling a portion of the image back and forth numerous times, thus increasing the number

of transfers [13]. Figure 32 shows an x-ray event plot for a radiation damaged CCD. In this case,

the CTI is too high to be accurately determined with this method. Our investigations involved mea-

suring CTIs of this magnitude and higher, so we were not able to use the x-ray technique. How-

ever, we did make use of another very practical feature of x-ray photons: the uniform packets


0 100 200 300 400 500 600 R o w number

Figure 3 1 : X-ray event blot for an undamaged CCD

provide a simple and precise measurement of the system gain.

Fine spot illumination

A particularly versatile measurement technique is possible using a setup similar to [33], in which a ,

very small pinhole is imaged onto the CCD in such a way that all the light is absorbed in a single

pixel. This method requires a rather elaborate equipment setup because of the difficulty in creating

a stable pinhole image smaller than the size of a single pixel, which is usually around 20 pm. The

result is something similar to the x-ray technique, except that the size of the charge packet is


0 100 200 300 400 500 600 Row number

Figure 32: X-ray event plot for a radiation damaged CCD

adjustable by changing the intensity of the pinhole image. However, the size is not known a priori

and must be estimated from acquired images. Also, the charge is only generated in one pixel per

image, so a much larger number of images is required to obtain good statistics. 0

EPER

Another method often used to measure the CTI is the Extended Pixel Edge Response (EPER) tech-

nique [59]. In this technique, the CCD is illuminated uniformly and read out in such a way as to

obtain an image which extends over one or both of the edges opposite the output amplifier. It is


1 quite similar to the pulse train technique, as applied to all the parallel registers at once, except that - the registers are loaded in parallel by illumination instead of serially by electrical injection.

Because of this, the "leading pulse" only undergoes one transfer, so it cannot provide a useful mea-

surement of the charge loss, but we can use the "trailing pulses", the charge packets measured in

the extended regions. The size of these "extended pixels" indicates the number of charges that

have been captured from the image pixels and then re-emitted during a complete set of transfers

through the parallel registers. The charges emitted in the wake of a pulse indicatesthe number of

empty traps that would be available to capture charge from a subsequent signal packet, and there-

fore this number can be used to estimate the CTI. The first trailing pixel indicates the charge that

would be captured from packets that were two pixels apart, the second the loss from packets three

pixels apart, and so on.

EPER is perhaps more properly considered a technique for measuring charge deferral, since it

only gives an indirect measure of the CTI. Because it does not measure the actual diminished mag-

nitude of a transferred charge packet, it cannot detect losses due to charge which is completely -

absorbed or deferred for times much longer than the clocking periods. However, if the CTI is dom-

inated by trapping at a single energy level, which i t appears to be in radiation damaged CCDs,

EPER is a very useful tool. The EPER technique is particularly useful in probing how CTI changes

with charge packet size, since, like the fine spot technique, the illumination level can be easily var-

ied. However, EPER requires no special equipment beyond a diffuse light source for creating the

flat field image.

We used the EPER technique because of its simplicity, its ability to adjust the charge packet size,

and its ability to measure CTI over a very wide range. We calculated the CTI by dividing the total '


arxount of charge in the extended pixels (the charge loss) by the amount of charge in the last pixels

of each column (the original packet size) and then dividing by the total number of single-pixel

transfers involved in the read out. This gives a "worst case" CTI figure equivalent to that experi-

enced by an isolated feature in a completely dark field, averaged over the number of columns.

C. Experimental results

CTI as a function of temperature '!

For our CTI experiments we used the same setup as in Chapter 4, and tested the same device. Fig-

ures 33 and 34 show the parallel EPER CTI of the radiation damaged TK5 12 device as a function

of temperature. The two figures represent two different clock timing schemes. The three sets of

data plotted in each figure indicate measurements of the three damage regions of the CCD. We

note not only a general increase in CTI with increasing amount of damage, but also that the CTI is

more strongly dependent on temperature in the damaged sections. Unfortunately we could not get

data at higher temperatures because the dark current was so high, especially in the high radiation

zone, that it was distorting the results. This distortion can be seen in the upper few temperature

points of the high radiation section in Figure 33.

Since i t could be assumed that the additional CTI experienced by the TK5 12 from radiation dam-

age is the result of bulk trapping, we tried to fit the trapping model described by Equation ( 2 8 ) - h

the data. First, the CTI in the undamaged part was modeled. In Figure 33, i t was sufficient to sim-

ply model it as a constant ( 1 . 2 1 ~ 1 0 - ~ ) , which is shown as the lowest line. Then the model was fit to

the excess CTI in the high radiation data. The uppermost line drawn on the graph indicates the

model evaluated for a combination of three traps with the parameters given in Table 4, and making

use of clock timings measured from our CCD electronics.

79


A undamaged

Temperature ( K )

Figure 33: Parallel CTI as a function of temperature (fast clocking). The symbols are experimental data from the three sections of the device. The lines indicate model results evaluated for a combination of the three traps in Table 4.

In these experiments we used the three-phase non-MPP clocking scheme shown in Figure 29 for

the parallel clocks with a clock pulse width (T3 - T I ) of 96 ps. For Figure 33 the period of the

clocking cycle (T6 - T , ) was 200 ps for the first 500 cycles, during which the serial register was

simply flushed, and 22.2 ms for the final 13 cycles when the serial register was being read out

between parallel transfers. The lower temperature CTI maximum in Figure 33 is caused by Trap A

and the higher maximum is a caused by the combination of Tmps B and C.

3 -

In Figure 34 we modeled the undamaged CTI as a constant ( 2 . 8 8 ~ 1 0 - ~ ) plus a trapping level with \., .


0 1 Sx1 o9 protons/cm2

A undamaged + +

Figure 34: Parallel CTI as a function of temperature (slow clocking). The symbols are experimental data from the three sections of the device. The lines indicate model results evaluated for a combination of the three traps in Table 4.

Table 4: Radiation induced traps

E, = 0.48 eV, on = S X I O - ' ~ cm-2, and N,V, = 1.5. The radiation-induced trap parameters remained

the same but the clock periods used for this figure were 25.3 ms for 490 cycles during which the


output was digitized, and 400 ms for 23 cycles when the data was being written to disk. In

Figure 34, the maxima have shifted down in temperature due to the slower clock rate and only the

feature due to Traps B and C is visible.

Finally, assuming the density of trapping states is proportional to the proton dose, the model was

evaluated for one quarter of the N, values used in the high radiation cases, and those results were

plotted as the middle lines in both figures for comparison with the low radiation data. The excel-

lent fit indicates that the CTI increase is, in fact, proporti~nal to the proton fluence

When matching our model to CTI vs. temperature data, it is difficult to decouple the effects of the

trap energy and the trapping cross-section, so we qould not unambiguously determine these two

parameters. We chose values which resulted in a fit and were in line with previously pub-

L lished figures. The activation energy of the lower peak seems too high to beBttributable to the oxy-

gen-vacancy complex, which is reported to have E, = 0.18 eV [38]. therefore we decided to try

values corresponding to the divacancy, which introduces two levels that always appear in equal

concentrations [29]. The parameters for Trap A and Trap B correspond to these two levels. The

Trap C parameters mach the published figures for the phosphorus-vacancy complex, which is usu-

ally given as the dominant trap level in irradiated CCDs Energy; levels for this complex have been

reported in the range 0.4 - 0.44 eV [29-381. Traps B and Cgare not resolved in our data, and the

upper maximum could be due to a single trap at around 0.4 eV, but we introduced Trap B to see if

the data were consistent with a divacancy interpretation, which it appears they are.

I t is clear from comparing Figure 33 with Figure 34 that both the temperature and the clocking

speed can have a significant effect on the CTI. The model we have described can be used to choose

combinations of temperature and clocking speed to minimize the CTI for a radiation damaged


CCD. As an example, we used the model to calculate the CTI for various parallel clock periods at

the designed operating temperature of the FUSE FES (223 K). The results are shown in Figure 35.

10 - 6 10 -5 13 - 4 10 - 3

Paro i le c lock 2eriod ( s c c )

Figure 35: Parallel CTI model as a function of clock period at 223 K. The lines indicate model results evaluated for a combination of the three traps in Table 4.

For this plot, we used a clock pulse width equal to one half the period, which corresponds to a i

clocking pattern with no break between cycles ( i . e . T6 immediately follows T5) . This type of pat-

tern is used when unwanted portions of the CCD are being flushed out prior to reading out the

region of interest because in that case there is no need to wait for the read out of the serial register

between parallel clock cycles. The FES only reads out small sections of the CCD to minimize the


processing time, so most of the transfers a charge packet undergoes in the FES CCD will be with

this clock pattern. We find that the minimum CTI is achieved at a clock period of about 8 ps.

CTI as a Function of Signal

Simulations

In order to quantify the shell effect mentioned earlier, we examined the charge packet distribution

within the bulk silicon using the same s~mulation software that produced Figures 13 and 14. We

used TSUPREM4 [22] and MEDIC1 [23] to simulate one 27 pm pixel of a three-phase buried

channel CCD with an MPP implant under one of the phases (like the TK5 12 we measured). Using

MEDICI's photo-generation feature, we were able to create a range of simulated charge packets

with Increasing numbers of electrons and determine the corresponding charge concentrations in v

the potential wells. The input files for these simulations are included in the appendix. By perform-

ing simulations of two two-dimensional cross sections at right angles through the device. we con-

structed a three-dimensional model of the charge concentrations and potentials. We were then able

to perform the integration in Equation ( 3 1 ) numerically and see the relationship between signal

level and CTI due to bulk trapping. Figure 36'shows the simulated charge distnbution for eleven

different packet sizes ranging from around I00 to 270,000 electrons, with each subsequent packet

being two to three times the size of the previous one. Each of the three plots shows the charge con-

centration along one axis through the centre of the packet: (a) along the channel, (b) across the

channel and (c) vertically from gate to substrate.

The simulated charge distributions were then processed to determine the percentage of filled traps

throughout each packet. Using Equations (29) and (30) and a dwell time of 0.1 seconds, we calcu-

lated the percentage of filled traps for Trap A at a temperature of 155 K. The results are shown in


Distance (pm)

Figure 36: S~mulated charge concentration profiles under one gate of a 27 pm pixel device. (a) along channel, (b) across channel. (c) vertically from gate to substrate.

the three plots of Figure 37. A dwell time of 0.1 seconds was used because although in a three-

phase device the charge packet is present under a given phase for only about one third of a one

pixel transfer period, in an EPER measurement there are packets in every pixel, and the time

between charge packets is not long enough for the traps to empty again. Therefore we have

assumed that the eflective dwell time is roughly the entire readout period. As i t turns out, this dwell

time is so long compared with the capture and emission times that the traps always reach steady

state. As well, the emission time is so long that the steady state fraction n,, is always approxi-


0.0 0.5 1 .O 1.5 2.0

Distance (pm)

Figure 37: Simulated filled trap profiles under one gate of a 27 pm pixel device at 155 K for a trap with

E, = 0.21 eV, and on = 5x10-l6 ern-"?. (a) along channel. (b) across channel, (c) vertically from gate to substrate.

mately one for capture time constants corresponding to electron concentrations at the limit of the

accuracy of our simulation model (-1012 ~ m - ~ ) . Therefore. our data is not able to verify Equations

(29) and (30).

We then took the trap occupation data, multiplied it by the trap concentration, and performed a

numerical integration over the volume of the packet to find the total number of filled traps in each

packet. Finally, we inserted the number of filled traps into Equation (31.) to find the deferred

charge for comparison with experimental data.

86

Charge Transfe~ Efficiency

We again used the EPER tichnique and the same experimental setup. Figure 38 shows a plot of

deferred charge versus the packet size in electrons at a temperature of 155 K, with a logarithmic x-

axis to improve the legibility. The crosses represent thekeshlts of EPER measurements of deferred

S gnol pocket size (e - )

Figure 38: Deferred charge vsxharge packet size at 155 K. Symbols indicate eqperimental data from the damaged sections of the device. The lines indicate the simulated results with E, = 0.2 1 eV,

on = 5 x 1 0 - ' ~ cm-', and N, = 5 . 2 ~ 1 0 ' ~ cm-I (high damage) and 1 3 x 1 0 ' ~ cm-' (low damage).

,

charge in the high radiation section of the TK5 12, and the diamonds represent deferred charge in

the low radiation section. The lines represent the deferred charges calculated from the simulated


charge packets. The trap concentrations used for the calculations were 5 . 2 ~ 1 0 ' ~ ~ m - ~ for the high

radiation section and 1 . 3 ~ 1 0 ' ~ cm-3 for the low radiation section. so again we see that the trap con-

i centrations vary linearly with proton fluence. Figure 39 shows the corresponding estimate of CTI.

The two lowest signal points display a poor f i t due to the difficulty in accurately determining the

deferred charge at such a low level. The error is magnified in Figure 39 because of dividing by the

low signal level to get the CTI.

F~gure 39: CTI vs. charge packet slre at. 155 K. Symbols indicate expertmental data from the damaged sec-

t ~ o n s of the dcv~ce . The ltncs ind~cate the simulated results with E, = 0.21 eV, 0, = 5 ~ 1 0 - ' % m - ~ .

and N, = 5 . 3 ~ 1 0 ' ~ cm-' (hlgh damage) and 1 . 3 ~ 1 0 ' ~ cm-' (low damage).


With the stated trap parameters, the simulated results match the experimental data very well.

which indicates that the non-linear variance in CTI with signal level can indeed be explained by r

the variance in &e packet volume and density. These results agree qualitatively with previously a

published results [53,33] at low signal levels, although both of these papers show a linear depen-

dence at high signal levels. The specifics of device fabrication, such as pixel size and channel dop-

ing, will have a large effect on the shape of the curve, so a direcr comparison with these previous

results is not practical.

It is well known that a CCD exhibits larger CTI for small signals_, and we see here that this effect

increases markedly with radiation damage. The increased CTI at small signal levels can be --

explained by the fact that the volume over which a charge packet is distributed is not proportional

to the amount of charge contained in it . Charge packets with small amounts of charge have a lower

t

charge density. The bulk traps, however, have a fixed uniform density and therefore the smaller

packets interact with more traps per electron of signal. The effect in the trapping model in Equa-

tion (3 l ) above is to create disproportionately large CTI for smaller packets.

--(I

The importance of charge packet distribution to bulk trapping, especiaily for smalr"jmckets. has

been demonstrated clearly by the success of several researchers in using a nardw supplementary

buried channel, or "notch", to reduce the radiation-induced CTI [30,60]. The no!ch is a small cen-

tral section of the buried channel which has been doped higher than the rest, creating a narrow ..

trough in the middle of the potential well. Small charge packets are constricted inside this trough,

which reduces their volume and thus the number of traps they encounter. A corresponding reduc-

tion in the radiation-induced CTI is observed

Read Noise

6. READ NOISE

The noise in t"he output signal of a CCD dictates the minimum detectable signal ;.nd therefore rep-

resents the ultimately limiting factor in the CCD's sensitivity. There are many sources of noise in

CCDs. but for most of the CCD's history the dominant source of noise has been tk noise intro-

duced by the output amplifier (the read noise). Radiation damage has been found to cause an .

increase in the read noise [41,51] and we investigated this phenomenon further. 8,

A. Noise sources

There are three main sources of noise affecting the CCD output. These are the thermal, or Johnson,

noise associated with the resistances in the output circuit, the kset noise associated with the charg-

ing of the gate capacitance of the output transistor, and the flicker, or I l f , noise of the output tran-

sistor. In addition, the output transistor may exhibit excess generation-recombination noise from

isolated mid-gap trapping levels, especially if i t has been subjected to radiation.

SS

Thermal noise

Thermal noise is a result of the random thermal motion of the electrons in the device. It is indepen-

dent of frequency and is therefore a form of "white" noise. Thermal noise is directly proportional

to the temperature and the resistance. The noise spectral density (in A~NZ) is given by

where k is Boltzmann's constant, T is the absolute temperature and R is the resistance. Because

there is a resistance associated with the channel of a MOSFET, i t will exhibit thermal noise. Mod-

3 -- v'

Read Noise

eled as a current source in the channel of the transistor, the current noise spectral density can be .

written [61] I

where y is a constant of proportionality which accounts for the variation of the conductance along

the channel and gdo is the drain conductance of the transistor at zero drain bias. For a transistor in

saturation, y = y,,, = 213 and gdo = gm, the transconductance. This gives

f In a CCD, the main source of thermal-noise is the output source follower transistor. We do not

expect the thermal noise to be affected by radiation.

Reset noise 9

'> Reset noise is the noise resulting from the uncertainty in the charging of a capacitor due to noise in

the charging circuit. In a CCD output circuit, reset noise is seen when the voltage at the gate of the

output transistor (the floating node) is fixed by the reset transistor. The reset noise is essentially a

sampling of the noise in the reset transistor. The total noise (in v') can be found by integrating the

noise spectrum with the low-pas filter characteristic of the capacitor circuit. For the simple case

when the noise is white thermal noise. this is

where C is the capacitance of the gate node and R is the resistance in the reset circuit. Note that the

Read Noise

noise does not depend on the resistance. The reset noise in CCDs is usually eliminated by signal

processing, as we shall see later.

Generation-recombination noise /'

The interaction of carriers with trapping states causes a random fluctuation in the current due to the

appearance and disappearance of carriers. This random fluctuation adds another noise component.

The spectrum SN resulting from a fluctuating number of carriers N has a Lorentzian shape and is

given by [62]

- 1

where A N - is the variance of N, f is the frequency and T is the lifetime of the carriers. The lifetime

r is determined from the capture and emission time constants of the trap (see Chapter 5) by

If there are M separate trapping levels, each with its own time constant T,, their contributions to the I

noise spectrum can be combined as

where A, is the variance of the noise associated with the ith trap. We expect that generation-recom-

bination noise will be present in the output MOSFE'F of an irradiated CCD due to the bulk trap-

ping states introduced by the radiation damage.

92

b

Flicker noise

Read Noise 5

Flicker noise is oftemreferred to as Ilfnoise because its spectral density has a I l fa frequency

dependence with u close to unity. The origins of flicker noise are varied %id not weliunderstbod.

I t is generally accepted that in MOS devices the dominant source is the interaction of the channel

carriers with near-interfacial oxide traps in SiOz [63 and references therein]. Due to their varying

distances from the interface, these traps have a continuous distribution of trapping time constants.

The combined effect of these traps is to produce a noise power spectral density Sf of the form

where K F is the flicker noise coefficient, lo is the drain current, AF is a constant of about 2. and a

is a constant which varies between 0.8 and 1.2. KF is a device-dependent parameter and varies

widely.

In buried channel CCDs, the MOS transistors used at the output are usually also buried-channel

devices. Because the signal carrying channel is located below the surface in such devices, the car-

riers are kept away from the interface and the near-interfacial oxide traps. This has been'found to

significantly reduce the flicker noise [64]. Therefore, we expect that the flicker noise in buried

channel devices will not be greatly affected by radiation damage, even though radiation causes an

increase in the number of surface tnps and would seriously affect a surface-channel device.

6. Correlated double sampling

In order to reduce the amount of noise at the output, especially the reset noise, researchers devised

C. Read Noise r il

a signal processing technique known as correlated double sampling (CDS) [65]. The main idea

behind the technique is to sample the output twice, once just after the reset pulse, and once after

the charge packet hias been dumped to the gate of the output transishi, and subtract the two sam- \

ples. The subtraction'8f the two samples effectively eliminates the reset noise because this noise is

a constant after the reset transistor turns off. The same reset noise voltage will therefore be present

in both samples and will be subtracted out. CDS also reduces any other low frequency noise which

is correlat'ed between the two samples, including, significantly, the llf noise.

S \

Clamp and sample

One simple method of performing CDS is to use a clamp-and-sample circuit (CS-CDS). Figure 40

shows a schematic of a clamp-and-sample processor. Figure 4 1 shows the typidal output wavefonn

output dram I I

I

I

I

I I ?. sample swltch I c1 I

I

output source I I pre amp

I I I RL >I+%- yx T c 2

I

I - - I I - - - I -

F~pure 40: Clamp-and-sample CDS processor schemat~c.

a.

Read Noise

from a CCD showing the switching points for CS-CDS. Both switches are normally closed. At the

/--- reset pulse

reset level fl

CCD output I I pixel value I

- signal level

1 ; ,-- sample

Figure 3 1 : CCD output waveform and CS-CDS sample timing

clamp point, the clamp switch is opened, fixing the bias across C1 to the reset level. The voltage at

the output then becomes the preamp output minus the reset level. A short time later during the sig- . i

>

rial period of the output signal, the sample switch is opened and the output is held by C2 at a volt-

age equal to the signal level minus the reset level.

The CS-CDS processor producqgan output noise signal. N,,(r) equal to

1

Read Noise

N , , ( r ) = N ( t ) - N ( r - T , ) (40)

where N ( r ) is the input noise signal and T, is the time between samples. Note that due to the sam-

pling operations in the implementation of CS-CDS, the actual output does not continuously vary

with the input as suggested by Equation (40). However, this simplification does not affect the accu-

racy of the following analysis, since the equation does hold at the sample point, which is all we are

interested in. The Fourier transform of N J t ) is

where S& is the Fourier transform of the input noise. The magnitude squared of this quantity is

from which we can see that the transfer function of the CS-CDS processor is zero at DC and has

maxima at odd multiples of half the sampling frequencv. Usually a low pass filter is inserted before

the CS-CDS processor which cuts out the higher maxima and the overall transfer function

becomes essentially a bandpass filter centred around the first peak at&ll(2T5). If we use a sec-

ond-order prefilter with cutoff frequency f,, the overall transfer function of the CS-CDS processor

IS

A plot of !he tran&er function versus frequency, normalized to ]IT5, is shown in Figure 42. The

96

Read Noise

Figure 42: Transfer function of a clamp and sample (CS) CDS processor. Frequency is normalixd to the sampling frequency l/Tr The cutoff frequency of the sccond order prefilter is ]IT,.

cutoff frequency of the second order prefilter is ]ITs, which we found to be the optimum value

when the input contains flicker noise.

The total noise at the output is found by multiplying the input noise spectrum S& by the transfer

function and integrating over frequency. For the clamp-and-sample processor the output noise

voltage V,, in volts rms is:

4 sin (nf T ~ ) '

Read Noise

As in most measurements, what is really important is not the absolute magnitude of the noise, but

the signal to noise ratio (SNR). The prefilter will affect the signal as well as the noise, so the output

signal voltage will be less than the input. We can calculate the output magnitude V, from -

which is the response of a second order filter to a step input V,, like the one shown in Figure 4 1 .

The SNR is the ratio VJV,.

Dual slope integration

An elegant method for performing CDS processing that eliminates the need for the low pass prefil-

ter is dual-slope integration (DSI). Figure 43 shows a schematic of a DSI circuit. After the reset %.

output drain I I reset

1 I

output source I

pre amp I RL integrator I I

F~gure 43 Schemat~c of d dual slope Integration (DSI) CDS processor.

Read Noise L

pulse, the CCD output is applied to the input of an integrator where it charges a capacitor. When 8

the signal charge is dumped to the output transistor, the output signal from the CCD is inverted and

applied to the same integrator where it discharges the capacitor for the same amount of time. After

the second integration, the output of the integrator will be proportional to the difference between

the-two signals. When the output has been sampled, the integrator is reset by closing a switch

across the capacitor. Figure 44 shows the output of a DSI processor in response to a typical CCD

output waveform. u

The Fourier transform pair for the integration function I ( ( ) is

where T,,, is the time constant of the integrator. This transform is valid only for integration from

minus infinity. Since the integrator is reset each cycle, i t is effectively only integrating from zero to

T,,,. To account for this we must subtract the integration from minus infinity to zero:

The transform of this subtraction introduces a product term of the kind in Equation (4 1 ) to the

oberall transfer function. The resulting transfer function is

To get the overall transfer function of the DSI processor, we must add another of these product

99

Read Noise

CCD output

reset pulse ,

reset level

I

I

I

I

I

I

I I I \

signal level

integrator output

, .

F~pure 44: CCD output waveform w ~ t h the output of a DSI processor

s + I

terms for the subtmtion of the reset integration from the signal integration. The overall transfer

function Hdsl isthen

I I

I I

I I I

I I I

. I I

s. I I

I I 1

I

I 3 .

I I I I I

I I I I

reset I , I

I I 1 1

Read Noise . -

Normally, there,is no significant pause between the two integrations and T, is approximately equal

to T,,,. The magnitude squared of the DSI processor transfer function will then be

This transfer function is shown in Figure 45 versus normalized frequency, using r,,, = TJ2 .

F~gure 45: Transfer functlon of a DSI processor versus n o r m a l i d frequency (T,,, = T 4 2 )

The total noise at the output of the DSI processor (in volts rms) is again obtained by integrating the

Input noise spectral density S\.Cf) with the transfer function Hd,,:

101

Read Noise

And again we must consider the effect of the processor on the input signal. The signal V, at the

output of the integrator in response to a step input V,, is

from which we can determine the SNR. In Equations ( 5 1 ) and ( 5 2 ) we note that both V, and Vn are

inversely proportional to T,,,, and therefore the SNR is independent of this value.

C. Experimental results 4.

We measured the noise spectra of several transistors, some of which had been damaged by proton

radiation. These were the same devices we used for the DLTS experiments in Chapter 3. The

results presented here are based on the measurements of three representative transistors, one of

each radiation dose. Unfortunately, many of the transistors were damaged in handling and no

longer usable. so we chose from the available working devices three whose dimensions were simi-

lar to each other and typical of the transistors used in actual CCD arrays. These test devices are

summarized in Table 5 . The three devices had different LDD lengths, but according to the results

in [-I?] the variation in the characteristics is minimal (for LDD lengths greater than zero)

Curren t-voltage measurements

The current-voltage ( I - V ) char~cteristics of the transistors were measured with a curve tracer and

found to match typ~cal FET I-V c u r v e . Because these are depletion-mode buried channel devices,

Read Noise

Table 5: Characteristics of test devices

the threshold voltage is a large negative value. With the substrate-source voltage VBS set to - 12 V

Transistor

D4Q8

D6Q4

D9Q5

the threshold voltages VT were around -1 1 V. We used VBs = -1 2 V because this is the typical sub-

strate bias used in CCD arrays, which ensures that the conducting channel is well below the sur-

face. We note that in previous measurements [41,42,66], the substrate was shorted to the source.

'-,

Width (pm)

50

60

60

which may have caused an increase in surface-related noise due to the shallow channel. We note

LDD length (pm)

2

3

1

Length (pm)

11.5

13

10

also that if VBS is zero, these transistors cannot be shut off because the surface inverts before the

threshold is reached, pinning the surface potential and preventing any further reduction in c h a n n e T -

Proton fluence (protons/cm2)

undamaged

5 .OX 10'

2 . 7 ~ 10'

width. The "threshold vbltage" reported in [41] for these devices is actually the inversion voltage

We expect the threshold voltage to be affected by radiation due to charge buildup in the oxide. We

measured the threshold voltages for all the working devices on each die, but there was a large vari-

ance in our results, and no clear dependence on radiation damage could be discerned.

Noise measurements

The experimental setup used to measure the noise spectra is shown in Figure 46. We used the same

dewar and temperature controller as in our CCD measurements (see Chapter 4) , but with the quartz

window replaced by a solid aluminum disc for shielding from electromagnetic interference (EMI).

For ease of analysis, the transistor was connected in a common emitter configuration and biased in

the linear, or triode region. To minimize 60 Hz powerline interference, the transistor biases were

103

Read Noise

Metal enclosure shielded

I cab les

Liquid nitrogen

/ dewar

controller

Wavetek 5820A Spectrum Analyzer

SR560 preamplifier IEEE 488

b u s

F ~ p u r e 46: Experimental setup for measuring the low frequency noisx of [he transistors.

supplied by batteries which were shielded and connected to the dewar with coaxial cables. All

resistors were wire-wound to reduce excess flicker noise. The voltage signal at, the drain node was

boosted by a Stanford Research Systems SR560 low noise amplifier (also powered by batteries) ,r

before being applied to the input of a Wavetek 5820A autocorrelating spectrum analyzer. The

amplifier was set to AC coupling. using a second order band pass filter with c u t 0 f f . e Hz and

1 MHz, and the gain was set to 100. The gain and frequency response of the amplifier were charac-

. terized by using a test signal from the spectrum analyzer. In order to cover a large frequency range

with reasonable accuracy, we measured the noise spectra in three frequency ranges: 0-50 Hz (spec-

Read Noise

trum analyzer channel width P = 0.125 Hz), 50-2.050 Hz (P = 5 Hz),.and 0-50 kHz (P=125 Hz). A

Hanning weighting function was applied to all measurements. The data from the spectrum ana-

lyzer were recorded by a PC using an IEEE 488 interface bus.

The resulting spectra for each of the transistors operated in the linear region at room temperature is

shown in Figure 47 along with the system noise. We measured the system noise by replacing the

Figurc 47: Noise spectra at room temperature for three buried channel MOSFETs with varying amounts of radiation damage.

transistor with a reference resistor between the source and drain nodes, and subtracting the thermal + .

noise of the resistor calculated from (32). The smooth lines in Figure 47 are noise models

Read Noise

(described below). The noise is represented in this figure as the magnitude squared current, in

A'MZ, from a noise current generator between the source and drain. The data in the figure were

arrived at by taking the raw spectrum analyzer data, compensating for the gain and frequency

response of the amplifier, and then calculating the equivalent noise current. The noise current is

equal to the drain voltage vd divided by the parallel combination of the output resistance of the

transistor Ro and the load resistor RL:

Since the output resistance of the transistor varies with the operating point, we.measured Ro with a

curve tracer for each operating point and used this value to calculate the noise current. Figure 47

shows a marked increase in the low frequency noise with radiation damage. We expect from the

discussion of noise sources above that the excess noise will be dominated by generation-recombi-

nation noise from bulk traps. We f i t a noise model to the measured data, including thermal noise,

flicker noise, and generation-recombination (g-r) noise from one or two isolated trapping levels.

The model is simply a sum of equations (33). (39). and (38):

The thermal noise arises from the resistance of the channel, and in the linear region, where the

transistor acts as a voltage controlled resistor, the resistance of the channel is simply equal to the

output resistance Ro. In the measured spectra, we find that indeed the thermal noise component is

accurately modeled by Equation (32):

Read Noise

(55)

The thermal noise shows no dependence on radiation dose, as expected. The variation in the white

high frequency noise seen in the figure is due to the variation in the output resistance of the three

transistors.

At room temperature, we found that the low frequency noise is totally dominated by the g-r noise,

even for the undamaged device, and the dramatic increase in the damaged devices can be

explained by the introduction of new bulk trapping states. The flicker noise is insignificant, which

is what we expect for buried channel devices. The noise models plotted in Figure 47 are based on

Equation (54) and use the g-r parameters given in Table 6. The parameters were determined by

Table 6: Generation-recombination noise parameters at room temperature

I Transistor I I Tl I

multiplying the spectra by frequency to accentuate the g-r peak and then fitting the model to the

data, as suggested in [67, 681. An example of this for the high-radiation spectrum is shown in Fig-

ure 48.

D4Q8 (undamaged)

D6Q4

( 5 . 0 ~ I ~ ~ r o t o n s / c r n ' )

D9Q5

( 7 . 7 ~ lo9 protons/crn')

The data in Table 6 indicate that the proton damage introduces two trapping levels with character-

107

1 .24x10-'~

1x10-l7

2 . 6 ~ 10-18

5x10"~

6x10'"

0.0352

0.002

0.013

0.002

0.009

Read Noise

Figure 48: Noise x frequency for the high radiation device ( 2 . 7 ~ 1 0 ' protons/cm'), showing clear evidence of g-r noise peaks. r I and t? denote the centres of the two overlapping peaks.

e

istic time constants of about 0.002 seconds and 0.01 seconds at room temperature. The level of

trapping noise seems to be proportional to the proton fluence, at least for the 0.002 level. I t is very

difficult to get accurate values for A or r for these traps because the peaks overlap.

The g-r noise is dependent on temperature, because the trapping mechanism is thermally activated,

and we can use this dependence to characterize the traps. The peak noise power occurs when

f, = f,. . We can evaluate r at this point from Equation (23):

which can be rearranged to give 7

Read Noise

( 56 )

Therefore, if we plot ln(r~ ' ) versus IlkT (an Arrhenius plot like Figure 19 in Chapter 3 for the

DLTS data), we should get a straight line with a slope equal to the activation energy of the trap E,.

We measured the spectra of the transistors over a range of temperatures from 200 K to 300 K and

estimated the time constants for the g-r peaks in each one. Figure 49 shows the noise-frequency

plots for the highly damaged device over this range. The corresponding Arrhenius plot, including

data from the low damage device, is &own in Figure 50. Due to the inaccuracies involved in esti-

mating the time constants of the overlapping peaks, we cannot make any definite conclusions

about the trap energy levels from these measurements alone. The plotted lines are models based on

the trap energies found in Chapter 3 and represent traps with activation energies of 0.43.0.30, and

0.22 ev. The lines show that these results are consistent with the results from the DLTS study

(Chapter 3) and the CTI measurements (Chapter 5). We conclude that the g-r noise seen in the

damaged devices is probably due to the same bulk traps.

Scientific CCDs are usually operated at low temperatures to reduce the dark current to minimal

levels, so i t is important to examine the noise at these temperatures. The CCD for the FUSE FES

will be operated between 2 13 and 223 K. Figure 5 1 shows the noise spectra of the three transistors

T

Read Noise

Figure 49: Nolse x frequency for the high radiation device (2.7~10' protons/cm') over a range of temperatures. showing the variation in the g-r noise peak.

in this temperature range. From the figure we can see that in this temperature range the flicker

noise dominates. The g-r noise is insignificant except for a small peak in the highly damaged

device which appears to be due to the trap at 0.30 eV. The flicker noise does not show a strong

dependence on the radiation damage and since the white noise variations can once again be attrib- -.

uted to variations in the output resistance, i t appears that the noise at these low temperatures is

largely unaffected by radiation. This explains why other groups have not seen an increase in the

read noise of CCDs after radiation [60, 691. We do not know the origin of the observed flicker

noise, but Kandiah and Whiting [W], who observed similar spectra on undamaged CCD output

l lo

Read Noise

Figure 50: Arrhenius plot of the g-r time constants over the temperature range 200 - 300 K. The lines indicate linear fits with the activation energies indicated.

transistors, propose that i t is due to holes which have been trapped at the surface under the gate by

a poor cqnectiorl between this region and the source and drain diffusions. These holes could inter-

act with the surface states and cause flicker noise. If the flicker noise is due to surface states, we

would expect i t to increase with radiation damage because radiation causes an increase in the den-

sity of the surface states. We do observe an increased flicker noise in the damaged devices, but it is

a very minor effect.

Effect of CDS signal processing

In order to more completely determine the effect of noise increases on a CCD system, we calcu-

I l l

Read Noise

1 (208 K) b I

-26 i , , , , , , , ; 13 1 I 1 1 i I l l I I 1 1 1 1 1 1 , I I I I 1 1 , LLLL

35 1 6 - o2 1 o3 1 a4 1 1 1 1 1

1 c5 Frequency (Hz)

Figure 51: Noise spectra for the three transistors in the range 208-2 13 K. Solid lines correspond to the noise model.

lated the effect of correlated double sampling on the noise signal and arrived at a total equivalent

- - input noise.

The output transistor of a CCD is operated in a source follower configuration with a load resis-

tance R,, which is typically 15-30 k R We multiplied our measured drain current noise spectra by a

load resistance of 20 k R to get output noise voltage spectra and extended the frequency range by

assuming that the noise was constant (white) above 5OkHz. These extended voltage spectra were

then multiplied by the transfer function of a dual slope integrator (DSI) and integrated over fre-

qurncy (Equation (44)). divided by the gain of the DSI processor (Equation (45)). and finally

Read Noise

divided by the output sensitivity of the transistor. The output sensitivity So is the voltage at the out-

put per electron of charge deposited on the floating node and is equdl to the gain of the source fol-

lower G divided by the capacitance CF of the floating node:

where q is the charge on an electron. Kim et a1.[42] measured the sensitivity of CCDs fabricated in

the same lot as the devices we tested and using the same type of transistor for the output transis-

tors. They found So to be about 1.3 pV/e' and we used this value in our calculations.

Figure 52 shows the results of these calculations for the noise spectra of the most damaged transis-

tor over a range of temperatures and DSI sampling periods. The total input-referred noise ( in elec-

trons) is plotted as a function of the sampling frequency for each temperature. The sampling

frequency f, is calculated from the sampling period T, (see Figure 44) and is given by 1 /(2T,) xince

two samples are required for each output value.

We see from Figure 52 that for high sampling frequencies, the total noise drops as the sampling

rate is reduced. The noise at high sampling frequencies is due mainly to the high frequency white

noise and the reduction occurs because the bandwidth of the DSI processor is decreasing (the noise

is higher for the 201 K and 214 K curves in the high frequency range because these spectra had

higher thermal noise due to a lower output resistance at the operating point used). The bandwidth

varies linearly with sampling frequency, and we see the total noise varying linearly also. We can

use the linear bandwidth variation to make a simple prediction of the total noise variation by mul-

tiplying the noise spectrum by frequency. Note the similarity betwhn Figures 49 and 52.

..L

Read Noise 3.-

Id I I I

10 2 13 3 1 o4 I o5 1 c6 Sarnp ing f r e q u e n c y ( H Z )

Figure 52: Total input referred noise vs. sampling rate for the highly damaged transistor ( 2 . 7 ~ 10' pro-

tons/cm2) over a range of operating temperatures.

The fact that the noise drops at lower sampling frequencies has often been used by CCD operators

to improve noise performance by slowing down the read out rate, but the improvement is limited at

lower sampling frequencies. At low sampling frequencies, in the absence of g-r noise (i.e. at 214

K), the noise is dominated by the flicker noise and the total noise flattens out because the input

noise is rising as Ilf and the noise-bandwidth product is a constant. For transistors with significant

g-r noise (i.e. all the curves above 253 K), the total noise increases at lower frequencies as the

sampling frequency approaches the upper side of the g-r peak. The g-r noise has a Lorentzian spec-

kead Noise

trum and varies as 1 / f above the g-r peak, so the noise-bandwidth product is inversely propor-

tional to sampling frequency in this region. Therefore, for g-r noise dominated transistors, the

speedtnoise trade-off reverses in the frequency region just above the g-r peak. In this region, the

noise performance actually degrades with slower sampling rates. At sampling frequencies below

6 . the g-r peak, the noise will begin to improve again because the Lorentzian n h e spectrum flattens

out. This effect can be seen in the curve at 201 K, which shows a small bump at about 500 Hz due

to a small g-r peak in the spectrum caused by the 0.22 eV trap.

Finally, noting that the total noise is a strong function of the thermal noise level at high sampling

frequencies, we performed a calculation to find the total noise for a transistor operated in the for-

ward active region. In this operating region the thermal noise current is considerably lower due to

the higher effective channel resistance and therefore we expect the total integrated noise to be

lower. Because the output transistor of a CCD is usually operated in the forward active region, this

would give a better idea of the actual noise in a typical CCD system. For this purpose we created

simulated noise spectra based on the models plotted in Figure 51, but with the white noise reduced

to a level calculated from Equation (34). The transconductance 8, was estimated at 70 pS from

room temperature measurements. Figure 53 shows the results. In this figure, the noise is plotted as

a function of the pixel rate R, instead of the sampling frequency. The pixel rate is generally lower

*

than the sampling frequency and can be calculated from

where T, i \ the sampling tlme and To is a fixed overhead time. The overhead time includes allow-

Read Noise

i L - I 1 3 r

a t 3 'n -

1 0 1

L

- 0 &

i I " 5 1 - C

i 1 I I

i 1

L undamaged

F~gure 53: Toral equivalent input noise at 210 K for damaged and undamaged devices from simulated spectra for a transistor in the forward active region.

ances for things like capacitor settling times and the digitization time which are necessary for

proper operation of the CDS procehsor. In our calculations we assumed an overhead of I ps.

Including this time gives a more realistic picture of the achievable noise performance. Adding the

overhead means that the pixel rate asymptotically approaches 1/T,, at very short sampling periods

and thus the noise climbs faster near this value.

We conclude from Figure 53 , in agreement with previous findings [60. 691, that only a very minor

increase in read noise can be expected from radiation damage at the normal operating temperature

of moht CCDs, including the FUSE FES, and therefore i t is of little concern.

Conclusions

7. CONCLUSIONS

We have examined the effects of high-energy proton radiation on scientific charge coupled devices

and evaluated the seventy of the damage at various levels of proton fluence. For buried channel

devices which operate in the region below the semiconductor-oxide interface, we found that the

dominant factor in the damage effects was an increase in the number of bulk trapping states. We

have seen evidence for traps at several energy levels, summarized in Table 7.

Table 7: Bulk trapping levels observed in proton damaged CCDs

Energy Cross section (3, Introduction (Ec - E,) Observed trap effects

(cm--) rate" (cm-I )

0.22 10.15 1 8.7 I increased low temperature CTI

l s 6 I increased high temperature CTI 0.43

increased high temperature read noise

Because of the exponential temperature dependence of the time constants of the trapping mecha-

nism, the damage effects show a'very significant dependence on temperature.

Dark current \ Our measurements of the dark current indicate that the increase in thermal generation in radia-

tion-damaged devices ls directly proportional to the proton fluence. The increases exhibited in the

117

0 55

The ~ntroductlon rate I \ glben as the trdp concentration (trap\/crn3) per un l t flucnce (pro~ondcrn2) of 3 M e V protons

I0 l 5 6 7 ~ncreased bulk dark current

Conclusions

CCD we measured were equivalent to a room temperature dark current of 5.8x10-" nA/cm2 per 4-

unit fluence (3 MeV proton/cm2) for the surface component and 4.7~10-" nA/cm2 per unit flu-

ence for the bulk component (MPP mode). The variation of,the dark current with temperature indi-

cated that both the surface and bulk components were due to trapping states very close to the

midgap. The dark current showed a significant variation from pixel to pixel, but most of the varia-

tion was of a spatially fixed nature. We found that the spati'al noise introduced by this variation

could be largely eliminated by subtracting a bias frame of equal exposure length. With the fixed

pattern noise subtracted out, the residual noise of the surface current was equal to the Poisson

noise inherent in the thermal generation process. In the bulk dark current case, the residual noise

was greater than Poisson statistics, showing instead a linear dependence on the dark current level,

but the total noise had still been reduced by 96%.

Charge transfer efficiency

The charge transfer efficiency also displayed a linear dependence on the proton fluence and we

found that the CTE became more dramatically dependent on the combination of temperature and

clocking speed, due to its origin in trapping phenomena with temperature dependent time con-

stants. Our model of CTE based on trapping theory matched our experimental data very well. and

seems to be a useful tool for predicting the temperaturelclock rate variation. It can thus be used to

optimize these operating conditions if the parameters of the dominant traps are known. According

to our model, if the clock rate is optimized for the baseline FUSE FES operating temperature of

-50 O C , the increase in the CTI is equivalent to 5.1 x 1 0 - l ~ per unit fluence. P

Read noise

Once again in the noise case, the damage factor was proportional to the proton dose. We observed

Conclusions

an increase in the low frequency noise in the damaged devices which we attributed to genera-

tion-recombination noise caused by the bulk traps. The effect was quite sensitive to temperature.

and we found that at -50 "C the total noise increase was negligible.

Recommendations

Based on our research, we made the following recommendations to the FUSE FES designers,

which are applicable to any satellite mission employing CCDs.

The CCD should be shielded with the equivalent of Smm aluminum. This will reduce the

amount of proton radiation experienced by the device by a factor of about 30.

A MPP mode device or dithered clocking is highly desirable. This will dramatically

reduce the amount of dark current and dark current noise.

Use dark subtraction with dark frames of same exposure length taken as close as possible

to the image frdme in order to get the best reduction of the spatially fixed dark current

noise.

Make operating temperature and clock rate commandable in flight so that the CTI can be

optimized as damage progresses.

A notch-type device highly desirable to reduce the low signal CTI.

Make operating voltages commandable in flight. This will enable compensation for the

flat band voltage shift caused by trapped charge in the oxide.

Use a correlated dual sampling (CDS) processing scheme to reduce the low frequency

read noise.

Conclusions

Table 8 shows the expected performance of the FUSE FES CCD at different stages of the mission

if these recommendations are adopted. The results are based on an operating temperature of

Table 8: FUSE FES CCD performance

Characteristic

I Total damage (displaccments/crn') I 0 1 6 6 x 1 0 ~ ' I 1 . 3 ~ 1 0 ~

Launch

-50 "C, 24 pm pixels, exposure lengths of 1 second, a parallel transfer rate of 12.5 kHz, and a pixel

read out rate of 50 kpixels/s.

1/2 mission life (1.5 years)

Dark current (e-/pixel/s)

Dark current noise (e-) .

Parallel charge transfer inefficiency

End of mission (3 years)

0.7

0.008

2.9x10-6

1.2 1

0.014

5.1x10-~

1.7

0.0 19

- 7 . 4 ~ 10."

REFERENCES

W.S. Boyle and G.E. Smith, "Charge coupled semiconductor devices," Bell Systems Tech- nical Journal, Vol. 49, pp. 587-593 ( 1970).

W.F. Kosonocky and K. H. Zaininger, in Charge-coupled Devices and Systems. M.J Howes and D.V. Morgan, Editors, Wiley-Interscience, New York, p. 2 18 ( 1979).

D.D. Buss, A.F. Tasch. Jr. and J.B. Barton, in Charge-coupled Devices und Sysrems, M.J. Howes and D.V. Morgan, Editors, Wiley-Interscience, New York, p. 12 1 ( 1979).

M.F. Tompsett, G.F. Amelio, and G.E. Smith, Bell Sysrems Technical Journnl. Vol 49, p. 593 ( 1970).

A.L. Solomon, "Parallel-transfer-r lrter charge-coupled imaging devices," 1974 IEEE 7.. Inrercon Technicnl Pupers, Session Q, New York. NY. March 26-28. ( 1974).

M. Blouke. J. Janesick, J. Hall and M. Cowens, "Texas Instruments (TI) 800x800 charge- coupled device (CCD) image sensor." in Solid Srare lmugers for Asrronomv, J. Geary and D. Latham, Editors, Proc. SPIE Vol. 290, pp. 6-1 5 ( I98 1 ). , -

J. Janehick, J. Hynecek and M. Blouke, "Virtual phase imager for Galilee," in Solid Srure Imugersfor Asfronomy, J. Geary and D. Latham, Editors, Proc. SPIE Vol. 290, pp. 165- 173 (1981).

R.D. McGrath, J. W. Freeman and W.F. Keenan, "A 1024x 1024 virtual phase CCD imager," IEDM Technical Digest, December 1983, p. 739.

J. Janesick, T. Elliott, A. Dingizian. J. Gunn, R. Bredthauer. C. Chandler and J. Westphal. "New advancements in charge-coupled device technology - sub-electron noise and 4096x3096 pixel CCDs," in CCDs in Asrronomv, G.H. Jacoby, Editor, Astronomical Soci- ety of the Pacific Conference Series Vol. 8, Brigham Young University Press, Utah, pp. 18- 39 ( 1989).

R. Bredthauer, "The state of CCD technology at Lockheed Martin Fairchild Systems," Pmcevdings ofrhe ESO Workshop "Optical Drtrc-tors for Astronomy ", J.W. Beletic and P.Amico, Editors, Kluwer ASSL Series, (1997, in press).

A. Theuwissen, "Modular CCD concept for large area CCD imagers," Pmc-retiit~~ys ofrhc' ESO Workshop "Oprical Derecrorsjbr Asfronomy ", J.W. Beletic and P.Amico. Editors. Kluwer ASSL Series. (1997, in press).

R.H. Wilden rr. rrl., "A buried channel charge coupled device," Bell Sysrrrns Trchnictrl

Journal, Vol. 5 1 , pp. 1635- 1640 ( 1972).

R. Murowinski, M.J. Deen, T. Hardy, "Charge transfer efficiency in low temperature charge coupled devices," Proceedings of the Symposium on Lor$. Temperature Electronics and High Temperature Supercondimiviry Proceedings Vol. 95-9. C.Claeys, S. Raider, R. Kirschman and W. Brown, Editors, The Electrochemical Society Press, New Jersey, pp. 369-382 ( 1995).

0. koulade, "The MEGACAM project: a wide field imaging camera at CFHT." Proceed- ings of the ESO Workshop "Optical Detectors for Astronomy ", J.W. Beletic and P.Amico, Editors. Kluwer ASSL Series. ( 1997. in press).

H.E. Kim, M. Blouke, D. Heidtmann, "Effects of transistor geometry on CCD output sensitivity", in Charge-Coupled Devices and Solid State Optical Sensors, M. Blouke, Editor. Proc. SPIE Vol. 1242, pp. 195-204 ( 1 990)

M.M. Blouke and D.A. Robinson, "A method for improving the spatial resolution of frontside-illuminated CCDs." IEEE Transuctions on Electron Devices Vol. 28, pp. 25 1-256 (1981).

M. Blouke, J. Janesick, J. Hall and M. Cowens, "Texas Instruments (TI) 800x800 charge- ' coupled device (CCD) image sensor," in Solid Stare Imagers for Astronomy J. Geary and D. Latham, Editors, Proc. SPIE Vol. 290, pp. 165- 173 ( 198 1 ).

J. Janesick, "History and advancements of large area array scientific CCD imagers," in the notes from the short course An lntrodirction to Scient$c Charge Coirpled Devices, ( 199 1 ).

J. Janesick, T. Elliott, T. Daud and J. McCarthy, "Backside charging of the CCD," in Solid S t m Imuging A r r a ~ s , K . Prettyjohns and E. Dereniak, Editors, Proc. SPIE Vol. 570, pp. 46-79 ( 1985).

E. K. Banghart, J. Leyine, E. Trabka, E. Nelson and B. Burkey, "A model for charge transfer in buried channel charge-coupled devices at low temperature." IEEE Trunsuctions on Electron Detices Vol. 38, pp. 1 162-74 ( 199 1 ).

W.J. Bertram et cil., "A three level metallisation three-phase CCD," IEEE Trwruc!ions en Electron Detic,es Vol. 2 1 . pp. 758-767 ( 1974).

TMA TSUPREM-4, Th.0-Dimensiorrul Process Sirnulution Pro~rum. Version 6.1. Tech- nology Modeling Associates. Inc., Palo Alto, California (1994).

TMA MEDICI, TLwDimensionul Device Simitlution Pro~r(im, Version 2.0, Technology Modeling Associates, Inc.. Palo Alto, California (1994).

"EEV CCD47 Series", data sheet from English Electric Valve (EEV) Inc., Chelmsford,

United Kingdom ( 1995).

V.A.J. Van Lint, Mechanisms of radiation effects in electronic materials, Wiley-Inter- science, New York, p. 14 ( 1980).

V.A.J. Van Lint, Mechanisms ofradiation effects in electronic materials. Wiley-Inter- science, New York, p. 222 (1980).

V.A.J. Van Lint, "The physics of radiation damage in particle detectors." Nuclear Instrii- ments and Methods in Physics Research, Vol. A253, pp. 453-459 ( 1987).

V.A.J. Van Lint, Mechanisms of rudiation effects in electronic rnaterictls. Wiley-Inter- science, New York, p. 85 (1980).

N. S. Saks, "Investigation of bulk electron traps created by fast neutron irradiation in a buried channel CCD," IEEE Transactions on N d r a r Science, Vol. 24, pp. 2 153-57 (1977).

J. Janesick, G. Soli, T. Elliott, S. Collins, "The effects of proton damage on charge-coupled devices," in Charge Coupled Devices and Solid State Optical Sensors 11, Proc. SPIE Vol. 1447 pp. 87-108 (1991).

M.S. Robbins, T. Roy, and S.J. Watts, "Degradation of charge transfer efficiency of a buried channel charge coupled device due to radiation damage by a beta source," RADECS 91, IEEE Proceedings Vol. 15, pp. 327-332 (1992).

A.D. Holland, "The effect of bulk traps in proton irradiated EEV CCDs," Niiclerrr lnstrii- rnenrs and Methods in Phvsics Research. Vol. A326, pp. 335-343 ( 1993).

I.H. Hopkins, G.R. Hopkinson, and B. Johlander. "Proton-induced charge transfer degradation in CCDs for near-room temperature applications." IEEE Trunsactinns on Nuclear Science. Vol. 41 No. 6, pp. 1984-91 ( 1994).

K.C. Gendreau, G. Prigozhin, R. Huang and M. Bautz, "A technique to measure trap characteristics in CCDs using x-rays." IEEE Transactions on Electron De~ices , Vol. 42 No. 1 1 , pp. 1912-17 (1995).

N. Meidinger and L. Striider, "Radiation hardness of pn-CCDs for x-ray astronomy." IEEE Transac~ions on Nuclear Science, Vol. 42 No. 6, pp. 2066-73 ( 1995).

S. Coffa, V. Privitera, F. Priolo. S. Libertino, and G. Mannino, "Depth profiles of vacancy- /

and interstitial-type defects in MeV implanted Si." Journal of Applied Physics, Vol. 8 1 No. 4, pp. 1639-44 (1997).

B.G. Svensson, C. Jagadish, and J.S. Williams, "Generation of point defects in crystalline

silicon by MeV heavy ions: dose rate and temperature dependence," Physical Review, k t - t en , Vo1. 7 1 No. 12, pp. 1860-63 ( 1993).

[38] J.L. Benton and L.C. Kimerling, "Capacitance transient spectroscopy of trace contamina- tion in silicon," Journal of the Electmchemical Society, Vol. 129, pp. 2098-2 102 ( 1982).

[39] P. Kolev, T. Hardy, M.J. Deen and R. Murowinski, "The use of constant-resistance DLTS to study proton radiation damage in CCD output MOSFETs," Proceedings of the Sympo- sium on Diagnostic Techniques for Semiconducror Materials and Devices. Proceedings Vol. 97-12, P. Rai-Choudhury, J. Benton and D. Schroder, Editors. The Electrochemical Society Press. New Jersey, 12 pages ( 1997).

[40] P. Kdev, T. Hardy, M.J. Deen and R. Murowinski, "Constant-resistance DLTS in CCD output MOSFETs", submitted to Journal of the Electrochemical Sociery ( 1997).

[41] R. Murowinski, G. Linzhuang, and M.J. Deen, "Effects of space radiation damage and temperature on the noise in CCDs and LDD MOS transistors", IEEE Tr(rnsacrions on Nircktr Science. Vol. 40 No. 3, pp. 288-294 ( 1993).

H. Kim, M.M. Blouke and D.L. Heidtmann, "Effects of transistor geometry on CCD output sensitivity," in Charge-Coupled Devices und Solid State Optical Sensors, Proc. SPIE Vol. 1242, pp. 195-204 ( 1990).

A. Holland, A. Abbey, D. Lumb and K. McCarthy, "Proton damage effects in EEV charge coupled devices," in Photonics for Space Envimnments, Proc. SPIE Vol. 1344, pp. 378- 394 ( 1 990).

J. Banh and E.G. Stassinopoulos. "Ionizing radiation environment concerns." in Single Ecent Criticulih Analysis, NASA Goddard Space Flight Center report. (awaiting publication).

E.G. Stassinopoulos and J. Barth. "FUSE elliptical orbit radiation environment," NASA Goddard Space Flight Center publication X-900-91-09.

J. Ziegler. J. Biersack and U. Littrnark, The Sroppirlg crnd Range oflons in Solids, Perga- mon Press ( 1985).

Ch.-K. Kim, in Charge-coupled Derlices and Systems, M.J. Howes and D.V. Morgan, Edi- tors, Wiley-Interscience, New York, pp. 10- 14 ( 1979).

B.E. Burke and S.A. Gajar. "Dynamic suppression of interface-state dark current i n buried-channel CCDs," IEEE Trunsuctions on Electron Derices, Vol. 38 No. 2, pp. 285-290 (1991)

Ch.-K. Kim, in Charge-c-oirpletl Derices crnd Systems, M.J. Howes and D.V. Morgan, Edi-

tors, Wiley-Interscience, New York, p. 17 (1979).

D.C. Wells, E.W. Greisen and R.H. Harten, "FITS: A Flexible Image Transport System." Astronomv and Astrophysics Supplement Series, Vol. 44, pp. 363-370 ( I98 1 ).

R. Murowinski, G. Linzhuang. M.J. Deen, "Effects of space radiation damage and temperature on CCD noise for the Lyman FUSE mission," in Photonics for Space Environments, E. Taylor, Editor, Proc. SPIE Vol. 1953, pp. 71-81 ( 1993).

T. Hardy, R . Murowinski and M.J. Deen, "Charge transfer efficiency in proton damaged CCDs". Proceedings of the ESO Workshop "Optical Detectors for Astronomy", J.W. Beletic and P.Amico, Editors, Kluwer ASSL Series. ( 1997, in press).

A. Mohsen and M. Tompsett, "The effects of bulk traps on the performance of bulk channel charge-coupled devices," IEEE Trrtnsuctions on Electron Devices. Vol. 21 No. 1 1 , pp. 701-712 (1974).

I. Zayer, I . Chapman, D. Duncan, G. Kelly, K. Mitchell, "Results from proton damage tests on the Michelson Doppler Imager CCD for SOHO." Churge Coupled Devices and Solid State Optical Sensors Ill, M.M. Blouke, Editor, Proc. SPIE Vol. 1900, pp. 97- 107 ( 1993).

Ch.-K. Kim, in Charge-co~rpled Devices and Systems, M.J. Howes and D.V. Morgan, Edi- tors. Wiley-Interscience, New York. pp. 57-58 (1979).

C. Dale, P. Marshall, B.Cummings, L. Shamey and A.Holland, "Displacement damage effects in mixed particle environments for shielded spacecraft CCDs," lEEE Transcrcrions on N~rcIeur Science, Vo1. 40 NO. 6, pp. 1628-37 ( 1993).

S.M. Sze, Physics of Semiconductor Deiices, Wiley-Interscience, New York, p. 261 ( I98 1 ).

M.F. Tompsett, "The quantitative effects of interface states on the performance of charge- coupled devices", lEEE Transacrions on Electron Devices Vol. 20 No. 1 , pp. 45-55 ( 1973).

J. Janesick, T. Elliott, R. Bredthauer, C. Chandler and B. Burke, "Fano-noise limited CCDs," in X-Ray lnstrirmentarion in Astronomy 11, L. Golub, Editor, Proc. SPIE Vol. 982 pp. 70-95 ( 1988).

A. Holland, A. Holmes-Siedle, B. Johlander and L. Adams, "Techniques for minimizing space proton damage in scientific charge-coupled devices", lEEE Trnnscccrions on Nlrclrcrr Science. Vol. 38 No. 6, pp. 1663- 1670 ( 1991 ).

A. van drr Ziel, Noisr in Solid Stcrte Devices ~rnd Circlrits, Wiley-Interscience, New York. p. 76 ( 1986).

A. van der Ziel, Noise in Solid State Devices and Circuits. Wiley-Interscience. New York. p. 121 (1986).

D.M. Fleetwood. P.S. Winokur, R.A. ~ e b e r ; T.L. Meisenheimer, J.R. Schwank, M.R. Shaneyfelt and L.C. Riewe, "Effects of oxide traps, interface traps, and 'border traps' on metal-oxide-semiconductor devices," Journal of Applied Physics, Vol. 73 No. 10, pp. 5058-74 ( 1993).

K.Kandiah and F.B. Whiting, "Nonideal behaviour of buried channel CCDs caused by oxide and bulk silicon traps," Nuclear Instruments and Methods in Physics Research. Vol. A305, pp. 600-607 (1991 ).

M.H. White, D.R. Lampe, F.C. Blaha and I.A. Mack, "Characterization of surface channel CCD image arrays at low light levels," IEEE Journal of Solid Stare Circuits. Vol. 9 No. 2, pp. 1-14 (1974).

H. Kim and D.L. Heidtmann, "Characteristics of I/f noise of the buried-chann~l charge- coupled device (CCD)." in Optical Sensors and Electmnic Photogrcrphy, Proc. SmE -

Vol. 107 1 , pp. 66-74 ( 1989). E ' \

B.K. Jones, "Low-frequency noise spectroscopy," IEEE Trcrr~suctiorrs on Electron Devices, Vol. 4 1 NO. l I . pp. 2 188-97 ( 1994).

-. M.J. Deen, "Low frequency noise as a characterization tool for InP- and GaAs-based double-barrier resonant tunnelling diodes," Materials Science and Et~gineeritlg B, Vol. B20, pp. 207-2 13 ( 1993).

D. Murata-Seawalt, J.D. Orbock. W.A. Delamere, W. Fowler and M.M. Blouke, "Proton radiation effects on multiple-pinned phase CCDs," in Char,qe-Coupled Detices urd Solid State Optical Sensors, Proc. SPIE Vol. 1242, pp. 84-93 ( 1990)

LIST OF ABBREVIATIONS AND SYMBOLS

ADC

AR

CCD

CDS

CFHT

CR-DLTS

CS-CDS

CTE

CTI

DLTS

DS I

EPER

FUSE

FES

FET

g-r

H E 0

H ST

LDD

LEO

LC

MIS

MMT

MOS

MOSFET

MPP

OD

analog-to-digital converter

anti-reflection

charge coupled device

correlated double sampling

Canada France Hawaii Telescope

constant resistance deep level transient spectroscopy

clamp-and-sample correlated double sampling

charge transfer efficiency

charge transfer inefficiency ,. 3

deep level transient spe&@~~t3py

dual slope integrator

extended &jiel edge response

Far Ultraviol@ Spectrographic Explorer

Fine Error Sensor

field effect transistor

generation-recombination

highly elliptical orbit

Hubble Space Telescope

I~ghtly-doped drain

low earth orbit

last gate

metal-insulator-semiconductor

Multiple Mirror Telescope

metal-oxide-semiconductor

metal-oxide-semiconductor field effect transistor

multiple pinned phase

output drain

output source

oxygen-vacancy

phosphorus-vacancy

quantum efficiency

reset drain

reset gate

summing well

vacancy-vacancy (divacancy)

Wide-Field and Planetary Camera

A * 'effective Richardson constant (252 A / C ~ ~ ~ / K ~ for n-type <100> Si)

pixel area (cm')

c speed of light in vacuum ( 2 . 9 9 8 ~ I Ol" c d s )

C capacitance (F)

trapping state density (cm-')

electron emission probability

energy (eV)

energy in the conduction band (eV)

Fermi energy (eV)

intrinsic Fermi energy (eV)

trap energy (eV)

electric field (V/cm)

frequency (Hz)

flicker noise comer frequency

drain conductance ( S )

transconductance (AN

c m e r generation rate ( , - I )

Planck's constant ( 4 . 1 3 6 ~ 1 0-15 eV-s)

bulk dark current density (A/cm2)

stirface dark current density (AIcm2)

Boltzmann's constant (8.6 17x 1 o - ~ eV/K)

7 free electron concentration (cm-- )

3 intrinsic free electron concentration (cm-- )

number of occupied traps at steady state

number of occupied traps

acceptor concentration ( ~ m - ~ )

effective density of states in the conduction band (cm-')

3 donor concentration (cm-- )

3 trap concentration (crn-- )

effective density of states in the valence band (cm-')

3 free hole concentration (cm-- )

electronic charge ( 1 . 6 0 2 ~ 1 0 ~ ~ ~ C)

load resistance (R)

surfrlce generation velocity (cmls)

noise spectral density (A'IHZ)

tlrne ( s )

temperature ( K )

thermal velocity (crnls)

applied voltage ( V )

charge packet volume (crn3)

thre~hold voltage ( V )

deplet~on depth (cm)

wavelength (cm)

frequency (Hz)

electron trapping cross-section (cm-')

hole trapping cross-section t m - l )

carrier lifetime (s)

effective lifetime in a depletion region ( s )

electron capture time constant (s)

electron emission time constant (s)

integrator time constant (s)

intrinsic Fermi potential ( V )

surface potential ( V )

Ferm~ potential (VJ

APPENDIX

A. TSUPREM input file

$ CCD SUPREM4 - - - 27 micron 3 phase structure

OPTION DEVICE=x S X and Y mesh definition

LINE X LOC=O SPACING=2 TAG=LEFT LINE X LOC=9 SPACING=0.2 LINE X LOC=13.5 SPACING=l LINE X LOC=lS SPACING=0.2 LINE X LOC=22.5 SPACING=2 LINE X LOC=27 SPACING=2 TAG=RIGHT

LINE Y LOCATION=O TAG=TOP SPACING=0.05 LINE Y LOC=l.3 SPACING=O.l LINE Y LOCz2.0 SPACING=l LINE Y LOC=lO SPACING=2.0 TAG=BOTTOM

ELIMINATE COLUMNS X.MIN=B X.MAX=19 Y.MIN=2 REGION SILICON XLO=LEFT XHI=RIGHT YLO=TOP YHI=BOTTOM

BOUND EXPO XLO=LEFT XHI=RIGHT YLO=TOP YHI=TOP BOUNDARY REFLECT1 XLO=LEFT XHI=RIGHT YLO=BOTTOM YHI=BOTTOM INITIALIZE ORIENT=100 ROT.SUB=O.O RATIO=1.5 RESISTIV BORON=40 PLOT.2D X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 Y.OFFSET=2.0 T.SIZE=0.4 + L.BOUND=1 C.BOUND=l GRID L.GRID=l C.GRID=l

$ Oxide DEPOSITION OXIDE TYICKNES=.l SPACES=l CONCENTR PLOT.2D Y.MIN=-2 Y.MAX=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 + Y.OFFSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l

$ Implant buried channel (no maSk) IMPLANT PHOSPHOR DOSE= lel2 ENERGY= 160 PEARSON RP . EFF $ Deposit polysilicon and etch to create phase one gates DEPOSITION POLYSILI THICKNES=.4 SPACES=l CONCENTR PHOSPHOR=lel9 ETCH START X=O Y=O ETCH CONTINUE X=O Y=-2 ETCIl CONTIh'UE X=9 Y=-2 ETCH DONE X=9 Y=O ETCH START X=18 Y=O ETCH COXTINUE X=lf? Y=-2 ETCH COTJTINLJE X=27 Y=-2 ETCH DONE X=2? Y=O ?LOT.?D Y.MIN=-2 Y.KU=l X.SIZE=0.25 Y.SIZE=O.25 X.OFFSET=2.0 +

Y.OFFSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l $ Oxide DEPOSITION OXIDE THICKNES=O.l SPACES=l CONCENTR $ Deposit polysilicon and etch to create phase two gates DEPOSITION POLYSILI THICKNES=.4 SPACES=l CONCENTR PHOSPHOR=lel9 ETCH POLYSILI START X=O Y=O ETCH POLYSILI CONTINUE X=O Y=-2.5 ETCH POLYSILI CONTINUE X=17 Y=-2.5 ETCH POLYSILI DONE X=17 Y=O ETCH OXIDE START X=O Y=O ETCH OXIDE CONTINUE X=O Y=-2.5 ETCH OXIDE CONTINUE X=17 Y=-2.5 ETCH OXIDE DONE X=17 Y=O PLOT.2D Y.MIN=-2 Y.MAX=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 +

Y.OFFSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l S Mask for MPP barrier implant (under phase 3) DEPOSITION PHOTORES POSITIVE THICKNES=l SPACES=l CONCENTR ETCH PHOTORES START X=O Y=O ETCH PHOTORES CONTINUE X=O Y=-3 ETCH PHOTORES CONTINUE X=9 Y=-3 ETCH PHOTORES DONE X=9 Y=O PLOT.2D Y.MIN=-2 Y.MPX=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 +

Y.OFFSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l S Implant MPP barrier IMPLANT BORON DOSE=7ell ENERGY=60 PEARSON RP.EFF ETCH PHOTORES ALL PLOT.2D Y.MIN=-2 Y.MAX=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 + Y.OFFSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l

$ Oxide DEPOSITION OXIDE THICKNES=O.l SPACES=l CONCENTR $ Deposit polysilicon and etch to create phase three gates- DEPOSITION POLYSILI THICKNES=.4 SPACES=l CONCENTR PHOSPHOI$=lel9 ETCH POLYSILI START X=10 Y=O ETCH POLYSILI CONTINUE X=10 Y=-3 ETCH POLYSILI CONTINUE X=27 Y=-3 ETCH POLYSILI DONE X=27 Y=O PLOT.29 Y.MIN=-2 Y.MAX=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 +

Y.OFFSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l ETCH OXIDE START X=10 Y=-0.5 ETCH OXIDE CONTINUE X=10 Y=-3 ETCH OXIDE CONTINUE X=27 Y = - 3

ETCH OXIDE DONE X=27 Y=--0.5 PLOT.2D Y.MIN=-2 Y.f.lAX=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 +

Y.OFPSET=2.0 T.SIZE=0.4 L.BOUND=l C.BOUND=l $ Passivation

4 DEPOSITION OXIDE THICKPJES=0.7 SPACES=l CONCENTR PLOT.2D Y.MIN=-2 Y..?LA.X=l X.SIZE=0.25 Y.SIZE=0.25 X.OFFSET=2.0 +

Y.OFFSET=2.0 + T.SIZE=0.4 L.BOUND=l C.BOUND=l 5 Buried channel drive

DIFFUSION TIME=150 TEMPEI?AT=1075 INERT SELECT Z=boron TITLE="Boron profile" LABEL="Boron conc. (#/cmA3)" PLOT.1D X.VALUE=14 LINE.TYP=l COLOR=l LEFT=O RIGHT=2 X.SIZE=0.25 + Y.SIZE=0.25 X.OFFSET=2.0 Y.OFFSET=2.0 T.SIZE=0.4 SELECT Z=phosphorus TITLE="Phosphorus profile" +

LABEL="Phosphorus conc. (#/cmA3)" PLOT.1D X.VALUE=14 LINE.TYP=l COLOR=l LEFT=O RIGHT=2 X.SIZE=0.25 + Y.SIZE=0.25 X.OFFSET=2.0 Y.OFFSET=2.0 T.SIZE=0.4 SELECT Z=abs(doping) TITLE="Dopingn PLOT.1D X.VALUE=14 LINE.TYP=l COLOR=l LEFT=O RIGHT=2 X.SIZE=0.25 + Y.SIZE=0.25 X.OFFSET=2.0 Y.OFFSET=2.0 T.SIZE=0.4 PRINT.1D X.VALUE=14 $ Output to MEDICI SAVEFILE OUT.FILE=3x27m.med MEDICI POLY.ELE ELEC.BOT

B. MEDIC1 input file

TITLE CCD 2D cell from SUPREM4 $ Read in SUP-4 file MESH TSUPREM4 IN.FILE=3~27m.med RECTANGU SAVE OUT.FILE=3x27m.msh MESH "ASCII PLOT.2D BOUNDARY GRID N.SIZE=0.25 DEPLETIO FILL Y . W = 1 0 X.OFFSET=2.0 +

Y.OFFSET=2.0 DEVICE="XTERMU COMMENT Specify physical models to use MODELS CONMOB FLDMOB AUGER BGN SRH

$ Symbolic factorization, solve initial SYMBOLIC NEWTON CARRIERS=O METHOD AUTONR ITLIMIT=500 LOAD ÂSCII.IN IN.FILE=initial.sol SOLVE V1=3 V2=-8 V3=-8 N.BIAS=(lS) OUT.FILE=initial.sol PLOT.2D BOUNDARY Y.MAX=lO X.OFFSET=L.Ci Y.OFFSEi'=2.0 DEVICE="XTERMH CONTOUR POTENTIA NCONTOUR=ll LINE.TYP=l COLOR=l REGRID POTENTIA FACTOR=1.5 ABSOLUTE "CHANGE SMOOTH.K=O X.MIN=5 + X.MAX=22 Y.MIN=O Y.MAX=2 COS.ANGL=2.@-"ASCII OUT.FILE=regrid.msh PLOT.2D BOUNDARY GRID N.SIZE=0.25 DEPLETIO FILL Y.MAX=lO X.OFFSET=2.0 + Y.OFFSET=2.0 DEVICE ="XTERMm

$ Read in mesh file MESH ÂSCII.IN IN.FILE=regrid.msh RECTANGU PLOT.2D BOUNDARY GRID N.SIZE=0.25 DEPLETIO FILL Y.MAX=5 X.OFFSET=2.0 +

Y.OFFSET=2.0 COMMENT Specify physical models to use MODELS TEMPERAT=224 CONMOB FLDMOB SRH AUGER GATE1 GATE.GEN=l BGN $ S\ynbolic factorization, solve intial SYMBOLIC GUMMEL CARRIERS=O METHOD AUTONR ITLIMIT=500 $ LOAD ÂSCII.IN IN.FILE=empty.sol

SOLVE V1=3 V2=-8 V 3 = - 8 P?.BIAS=(15) OUT.FILE=empty.sol PLOT.1D POTENTIA X.START=O Y.START=0.55 X.END=27 Y.END=0.55 + HORZ.STA=O.O FIND.DIS=O.O TITLE="no charge" T.SIZE=0.4 X.SIZE=0.25 + Y.SIZE=0.25 C.SIZE=0.25 LINE.TYP=l COLOR=l PLOT.1D POTENTIA X.START=13.5 Y.START=O X.END=13.5 Y.END=2 + HORZ.STA=O.O FIND.DIS=O.O TITLE="no charge" T.SIZE=0.4 X.SIZE=0.25 +

Y.SIZE=0.25 C.SIZE=0.25 LINE.TYP=l COLOR=l PLOT. 2 D BOUNDARY CONTOUR POTENTIA DEL.VALU=l LINE.TYP=l COLOR=l

$ Generate photo-electrons SYMBOLIC NEWTON CARRIERS=l PHOTOGEN X.START=13.5 Y.START=O X.ENDz13.5 Y.END=l R.CHAR=O.O + A1=12e19 A2=0.0 A3=0.0 A4=0.0 "PC.LJNITS C1=0.0 C2=1.0 C3=0.0 C4=0.0 +

RECO=O.O N.INTEG=5 UNIFORM $ First solution with charge packet

SOLVE V1=3 V2=-8 V3=-8 TSTEP=2e-9 NSTEPS=l DT.MAX=0.25 N.BIAS=(15) + OUT.FILE=3x27m.sol PLOT.1D ELECTRON X.START=13.5 Y.START=O X.END=13.5 Y.END=2 + HORZ.STA=O.O FIND.DIS=O.O TITLE="Depth profile (phase 2 high)" + T.SIZE=0.4 X.SIZE=0.25 Y.SIZE=0.25 OUT.FILE=vel.dat PLOT.1D ELECTRON X.START=O Y.START=0.5 X.END=27 Y.END=0.5 HORZ.STA=O.O + FIND.DIS=2.0 TITLE="Mid-channel profile (phase 2 high)" T.SIZE=0.4 + X.SIZE=0.25 Y.SIZE=0.25 OUT.FILE=hel.dat

$ Second solution, larger charge packet SOLVE V1=3 V2=-8 V3=-8 TSTEP=4e-9 NSTEPS=l DT.MAX=0.25 N.BIAS=(15) +

OUT.FILE=3x27m.sol PLOT.1D ELECTRON X.START=13.5 Y.START=O X.END=13.5 Y.END=2 + HORZ.STA=O.O FIND.DIS=O.O TITLE="Depth profile (phase 2 high)" +

T.SIZE=0.4 X.SIZE=0.25 Y.SIZE=0.25 OUT.FILE=ve2.dat PLOT.1D ELECTRON X.START=O Y.START=0.5 X.END=27 Y.END=0.5 HORZ.STA=O.O + FIND.DIS=2.0 TITLE="Mid-channel profile (phase 2 high) T.SIZE=0.4 + X.SIZE=0.25 Y.SIZE=O L S OUT.FILE=he2.dat

effects of radiation damage on scientific charge coupled devices

Documents