review article charge-coupled devices as particle tracking detectorsjklay/rsi01549.pdf ·...

REVIEW OF SCIENTIFIC INSTRUMENTS VOLUME 69, NUMBER 4 APRIL 1998

REVIEW ARTICLE

Charge-coupled devices as particle tracking detectorsC. J. S. Damerella)

Rutherford Appleton Laboratory, Chilton, Didcot OX11 0QX, United Kingdom

~Received 15 September 1997; accepted for publication 12 December 1997!

Following a few years after the invention of the charge-coupled device~CCD! in 1970, thediscovery of charmed particles established the need for very high precision~few mm! detectors fortracking high energy charged particles. This review describes the work which has evolved over thepast 20 years from these disconnected events, both as regards the application of increasingly refinedCCDs to particle tracking~in particular as vertex detectors for identifying heavy flavor quarks andtau leptons!, and also the advances in CCD detector design stimulated by these requirements. Thelessons learned in this work should provide guidance for the construction of large arrays of CCDsor active pixel devices in the future in a number of areas of science and technology. ©1998American Institute of Physics.@S0034-6748~98!00104-X#

chehissnred

b

inc

alhliitnv

e

om

evto

ins

entr-ultofrsintn--

benal

ion, butc-jorn-

th-

of

s.

0ck-

Theonhe

lelec-e

c-

I. INTRODUCTION

There has been a long historical link between the tenology of optical imaging and charged particle tracking dtectors. Indeed, the adaptation of photographic film for tapplication~nuclear emulsions! provided tracking detectorfor minimum-ionizing particles~hereafter referred to aMIPs!, with few mm precision.1 Such detectors have had aillustrious history in particle physics; for example, they weused exactly 50 years ago in the first observation of thecay of pi to mu mesons.2

Over the years, the technologies of optical imaging~stilllargely based on photographic film! and particle tracking~in-creasingly using electronic detectors such as spark chamand multiwire gaseous chambers! drifted apart. However, theinvention of the charge-coupled device~CCD! in 19703,4

started a revolution which is still having profound effectsthe fields of optical imaging, particle tracking, x-ray detetion, analog storage devices, etc. Once again, one is dewith a technology with multidisciplinary applications, witconsequential benefits as ideas generated from one apption area find uses in others. However, at the time ofinvention, the potential value for particle tracking went urecognized. The emphasis in particle physics was for elarger area coverage~tens of square meters! and the typicaldrift chamber precision of;100mm was believed adequatfor all applications.5

The discovery of charm, tau leptons, and beauty/bottparticles during the period 1974–1977~particles with life-times in the range 10213– 10212 s! profoundly changed thepicture. Suddenly it was seen to be important to achimicron-level tracking precision in small detectors closethe interaction point in order to recognize events containheavy flavor particles, and ideally to reconstruct the tree

a!Electronic mail: [email protected]

1540034-6748/98/69(4)/1549/25/$15.00

--s

e-

ers

-ing

ca-s-er

e

gof

vertices, primary, secondary and possibly tertiary~PV, SV,TV!, assigning the tracks unambiguously to their true parvertices. In Fig. 1 is a sketch of the typical topological infomation contained in high energy jets of particles that resfrom the production of heavy quarks, and the possibilityextracting this information with the aid of two or more layeof high precision tracking detectors near the interaction po~IP!. Each jet of particles in this example includes one cotaining ab quark, whose decay~SV! releases a charm particle which subsequently decays~TV!. While Fig. 1 shows atwo-layer detector, in practice at least three layers woulddesirable in order to provide redundancy and an interalignment capability.

What technology could be used for such high-precisdetectors? Nuclear emulsions made a partial comebacktheir lack of electronic readout and their inability to seletively register triggered events were major handicaps. Maefforts were made to improve the precision of the thepopular tracking detectors~bubble chambers, multiwire drifchambers, etc.! but with only limited success. These tecniques were fundamentally unable to meet the challengethe ‘‘new physics.’’

The way forward was shown to lie with silicon detectorGermanium and silicon detectors~in the form of reverse-biased diodes with extensive depletion regions! had beenused for the detection of ionizing radiation for over 3years.6 Even as position sensitive detectors for particle traing, albeit with spatial resolution of;1 mm, silicon stripdevices were already in use as beam hodoscopes, etc.development of high-precision microelectronic fabricatitechniques, the ‘‘planar process,’’ lent itself perfectly to tproduction of microstrip detectors with;5 mm tracking pre-cision for MIPs.7 A very readable account of the remarkabhuman stories associated with the development of microetronics is to be found in George Gilder’s book on thsubject.8 While the development of silicon microstrip dete

9 © 1998 American Institute of Physics

thelehi-g

ueucrgtim

trirswes

os

doeclyd

trito

t-

epl

deticlean

lyirenh

de

cith

fac-theand-ets;ts.ex--

t bysci-m-Dsrarchpedhein

e toc-

ainInthet infac-n-se,, byfordtoe-xityor-ur-

ingnts

h-ceion-icleits

r-fl-tor

-

-

te

1550 Rev. Sci. Instrum., Vol. 69, No. 4, April 1998 C. J. S. Damerell

tors was being pioneered by the CERN-Munich part ofACCMOR collaboration at CERN, the RAL part of the samcollaboration initiated the development of CCD partictracking detectors.9 It was quickly demonstrated that sucdetectors could perform with similar precision to silicon mcrostrip detectors10 but ~being pixel based rather than usinstrips! they had the further advantage of providing trspace-point information. As such, they could be placed mcloser to the IP with no significant problems of track meing. In contrast to these advantages, CCDs lack the fasting capability ~strobed coincidence logic! of microstrip de-tectors with independent readout electronics on every sIn the ACCMOR experiment NA32, a pair of CCD detectofollowed by six planes of microstrip detectors formed a poerful combination, allowing the determination of the shortcharm particle lifetimes ever to be measured.11 Over the en-suing decade, microstrip detectors have found the mwidespread applications, particularly for general purpob-quark tagging. However, CCD detectors have continueprovide state-of-art vertex detection where experimental cditions were appropriate. Microstrip detectors, being projtive devices, are limited to environments with a relativelow density of hits, while CCD detectors, being pixel basecan accommodate far higher hit densities. While microsdetectors, used mainly as particle physics tracking detecare dependent on scientific users, CCDs, being suited torecording of two-dimensional~2D! information such as optical images, have an enormous user base, including domtic still and video cameras, medical and dental x-ray apcations, astronomy~visible light and x rays!, synchrotronradiation, etc. Without this broad market, such complexvices would never have been developed for the small partracking/vertex detector community. While the assembdetectors used in this field are among the largest ofCCD-based instruments~up to 100 CCDs and 300 Mpixels!,the overall market provided by particle physics is relativeminute. However, due to the demanding technical requments in this field, the particle tracking application is of cosiderable interest to manufacturers of scientific CCDs. Tefforts to satisfy these requirements have raised the unstanding and thereafter the quality of the devices to the befit of all classes of scientific users.

In contrast with the symbiosis between the different sentific application areas, it should be pointed out that

FIG. 1. Sketch showing the principle of a two-layer pixel-based verdetector for reconstructing the topology~PV, SV, TV! of a b jet. Tracks arelabeled according to their parent vertices.

e

h-

-

p.

-t

reeton--

,prs,he

s-i-

-ledy

--e

er-n-

-e

overlap between the domestic and scientific CCD manuturers is almost nonexistent. The reason for this is partlyfact that the domestic device manufacturers are understably focused on the huge, highly competitive, mass markthey have no time to deal with ‘‘nonstandard’’ requiremenFurthermore, their entire design base is locked intotremely shallow active layers~in order to achieve sharp images even in the red region of the spectrum!. As such, manyof their device characteristics are opposed to those soughparticle tracking and x-ray detection users. The needs ofentific customers have been filled by highly specialized copanies set up for the production of specially designed CCin small volume, with full flexibility to adapt the designs foeach specific application area. In one or two cases, reselaboratories in which CCDs are heavily used have develotheir own integrated circuit processing capabilities to tlevel where they are able to produce their own deviceshouse, but for the most part users have found it preferabldo the design in conjunction with a commercial manufaturer having a sufficient overall customer base to maintthe extremely costly state-of-the-art processing facilities.fact even the best equipped manufacturers of CCDs forscientific community are lagging behind the state of the arfeature sizes, etc, when compared with the largest manuturers of computer memory chips, CCDs for domestic cosumer products, etc. This will always tend to be the cagiven the relative sizes of these markets. Neverthelesscomparison with silicon microstrip detectors, the marketscientific CCDs is relatively large and multidisciplinary, anthe production facilities are highly sophisticated. Thanksthis fortunate state of affairs, the particle tracking/vertex dtector community has access to devices whose complefar outstrips the home grown in-house detectors which fmerly characterized high energy physics experiments. Fthermore, the sophistication of available devices is evolvrapidly, carried along by the ongoing pace of developmein the planar technology of microelectronics.

II. CCD TRACKING DETECTORS: OPERATINGPRINCIPLES AND PERFORMANCE

A. MIP signals in thin silicon detectors

The advantages of solid state detectors for higprecision particle tracking have been vividly apparent sinthe days of nuclear emulsions. Electrons released by theization process are confined extremely closely to the parttrajectory. Silicon has the further advantage that, due tosmall band gap~1.1 eV!, MIPs generate prolific signals~;80electron-hole pairs permm of track length!. Since ~as weshall see in Sec. II D! the CCD output signal processing pemits noise levels below 100e2; a detector thickness o10–20mm is sufficient to achieve MIP detection with excelent signal-to-noise performance. This minimum detecthickness is far less than with other technologies~e.g., siliconmicrostrip detectors! and carries with it several distinct advantages:

~1! reduced probability of ejectingd electrons of range sufficient to spoil a coordinate measurement;

x

aks

1551Rev. Sci. Instrum., Vol. 69, No. 4, April 1998 C. J. S. Damerell

FIG. 2. Energy loss distributions for various thicknesses of silicon detector with~in each case! a Laudau distribution for comparison. The separate pecorresponding to 0, 1, 2... plasmon excitation are already merged by a thickness of 10mm.

ceis

he

IPs

th

iethut

lie

tic

iewssm

nar-the

ing

ita-

t

ea-the

c-ichuch

~2! reduced projected track length for oblique inciden~in thick detectors the precision for oblique tracksseriously degraded!;

~3! the opportunity to thin the entire detector down to tactive layer thickness if desired~thinning may be ex-tremely important in reducing multiple scattering!.

Around 1980, when CCDs were first considered for Mdetection, it was far from clear whether this would be posible even in principle, with high efficiency. The quantityW~the mean energy for electron-hole pair creation,;3.7 eV!was of course well known; what was far from clear wasexpected fluctuations~straggling! in the energy deposition bya MIP for such thin samples. Theoretical estimates varwidely, and some calculated distributions were so broadhigh efficiency MIP detection would have been ruled oFortunately the measurements10 for ;20mm detector thick-ness demonstrated that the more optimistic earestimates12 had been correct. There followed papers13,14

making detailed comparisons between refined theore

-

e

dat.

r

al

procedures and the new data, and finally a definitive revpaper15 which provided the full description of the energy loof charged particles in solid silicon. What emerged frothese calculations was that the straggling spectra, whilerower than some early estimates, are much broader thanwidely used Landau distribution, the discrepancy becomgreater for thinner detectors. For very thin samples~e.g., 1mm! the energy loss is typically characterized by the exction of a small number of plasmons forM -shell electrons~with mean plasmon energy 16.7 eV!. Such cases are bessimulated by Monte Carlo calculations;16 results for a rangeof detector thicknesses are plotted in Fig. 2.

It should be noted that what the detector physicist msures is not the energy loss, but the charge released indetector. These are related byW, with a statistical factorF~the Fano factor! which expresses the suppression in the flutuations in the number of pairs created below that whwould be given by Poisson statistics. There has been mrecent experimental17 and theoretical18 work onW andF for

rsthve

e

praiacenthmh

neioosrainmeoepwchrent

erst

orseon

a-

iers

ple-

le-.

tse ton.in-ed.rge

on

e

rsedD


silicon. It is sufficient here to note that for the thin detectowe are concerned with, the conversion to charge relealeads to a negligible further broadening with respect toenergy loss distributions shown in Fig. 2. It should howebe remembered that the precise value ofW depends weaklyon the temperature~see Fig. 3!, reflecting the temperaturdependence on the silicon band gap.

B. CCD operating principles and general performance

Having just discussed the advantages of silicon as acision tracking medium, it should now be pointed out thateven more compelling factor in the choice of this materfor detectors has been the development of the planar profor integrated circuits~ICs!. This has allowed the productioof a huge variety of detectors bonded to, or combined on,same wafer with sophisticated readout ICs. CCDs form soof the most advanced of the second class of detectors. Tunique advantage for tracking lies in their ability to confithe signal charge within an extremely compact collectvolume, and then to transfer the signal packet without lonto the gate of a very low capacitance on-chip sensing tsistor. Despite these attractions, this on-detector multiplexhas its disadvantages; the serial sensing of the signals frolarge number of pixels by a single output circuit takes timCCDs are correctly seen as ‘‘slow’’ detectors by compariswith those where the signals are sensed in parallel by a srate output circuit for every detector element. In this reviewe focus on various particle tracking applications for whiCCD detectors have proven their superiority; these are ain which the above attributes are particularly important, awhere special procedures have been devised to allowrequirements of long readout time to be accommodated.

Let us now understand how the very thin active layreferred to in Sec. II A can be achieved. A CCD consiglobally of a reverse-biased structure~usually n1 on ap-type substrate!, on which a metal–oxide–semiconduct~MOS! gate structure and other features are superimpoFigure 4 shows the cross section of a typical device. It csists of a lightly doped epitaxialp layer on a heavily doped

FIG. 3. Temperature dependence of the pair-creation energyW in silicon. Adetector operated cold will produce slightly smaller MIP signals than onroom temperature.

seder

e-nlss

ee

eir

nsn-ga

.na-,

asdhe

ss

d.-

p1 substrate, with the top;1 mm of the p layer doped byion implantation of phosphorus, followed by thermal activtion, to becomen1. The partial depletion of then1/p regioncreates a potential minimum for storage of charge carr~electrons in this case! just above the depth of then1/pedge, into which signal electrons generated within the detion region of thickness;5 mm will be transported by driftin the electric field. But this is not the full story. Minoritycarriers~electrons! generated in the undepletedp-type sub-strate will diffuse isotropically. Those which reach the deption edge will be drawn into the potential energy minimumHowever, those which diffuse to thep/p1 edge feel a poten-tial barrier due to the thin intrinsic depletion layer; this acas a perfect reflector, so even these electrons continudiffuse and rapidly find their way into the storage regioReference 19 is possibly the first paper in which the prciples of this general phenomenon were explicitly describ

The localization and transfer of this stored signal chais achieved by means of two structures, an imaging area~I!and an adjacent readout register~R!, sketched in Fig. 5.Charge is confined in depth by the potential distributi

at

FIG. 4. Charge collection within a buried-channel CCD structure.

FIG. 5. Upper right: Sketch of charge storage in a CCD detector traveby a number of ionizing particles. Lower left: Corner region of the CCshowing the principal structural features.

t

lyipd, ihinicllyftefe

t srghet

toinixithrmerntibreCfoin

imayinat

anlao

beeheva

l

f

s-

rec

ongn

4nsand

e

eryCDl-rstly

ane

b-byop-

mally

wal

toor

inhe-

.

icleigh

s-

sult

geonslec-deslf-lltec-ns

s


mentioned above. Transversely, charge confinement inimaging area is by means of verticalp1 implants ~channelstop regions! and horizontal polysilicon gates, electricalisolated from each other and from the substrate. By manlating the potentials on these If gates, charge is transferrein parallel down each column towards the output registersuch a way that the 2D image is preserved. On each I sthe charge from the bottom row of the image area is traferred into the R register. Then there is a pause of verttransfer in order to allow this row to be read out seriathrough the output circuit at the end of the R register, awhich the next row transfer is made. The horizontal transalong the R register is achieved by a similar, independenof Rf gates. With each horizontal transfer, the signal chafrom the first pixel in the R register is transferred onto toutput node, a diode implant which is directly connectedthe gate of an adjacent MOS field effect transis~MOSFET!. The resultant modulation of the drain currentthe FET is used to sense the signal associated with that p

CCDs used for typical optical imaging are supplied wsignals of order 105 e2 per pixel. The reduction by a factoof ;100 in CCDs used for particle tracking detectors iposes two major challenges on the CCD design and option. First, there is the requirement of extremely efficienoiseless transport of the tiny charge packet through possthousands of transfers within the imaging and readoutgions of the device. This is discussed in detail in Sec. IIThe second major challenge is the electronic noise permance of the output circuit and associated signal processthis is discussed in Sec. II D. There is a third challengeposed on CCDs used for particle tracking; they are alwoperated in a radiation environment. This implies that havfound solutions to the first two critical requirements, care hto be taken to ensure that these are not destroyed byoperating environment. This is discussed in Sec. V.

For clarity, what has been described in this section,what will be in the remainder of this review, is one particuclass of CCD architectures that has proved to be the mgenerally useful for particle tracking. It should howeverpointed out that there are numerous other options, somwhich might be of interest in future, for example, as texperimental requirements are changed. Examples of aable options are:

~1! signal storage: electrons~n channel! or holes~p chan-nel!;

~2! signal storage: buried channel~BC! or surface channe~SC!;

~3! substrate material: epitaxial or bulk~and in each case, owhat resistivity?!;

~4! phases per pixel: options range from 1~virtual phase! to4;

~5! ‘‘pinned’’ operation for reduced dark current;~6! charge sensing other than the ‘‘floating diffusion’’ sy

tem ~see Sec. VI A!.

The reader interested in exploring these options isferred to the extensive literature on the subject. Referen20–22 are books which provide comprehensive discussiRef. 23 includes a pedagogical description of CCD desi

he

u-

nft,s-al

rrete

or

el.

-a-,ly-.r-g;-sgshe

drst

of

il-

-ess,s

for particle tracking. As well as the papers of Refs. 3 anddescribing the invention of the device, important discussioof subsequent developments are to be found in Refs. 2425 ~invention of the buried-channel architecture!,26 ~chargedistribution in buried-channel devices!,27 ~general review!,28

~first use of epitaxial material!,29 ~‘‘pinned’’ operation!,30

~integration of high and low resistivity regions on the samwafer!,31 ~review of performance limitations!32 ~numerousnovel architectures!, and33 ~ideas for extending performanclimits!. As well as providing a record of the extraordinadevelopments that have taken place since the original Cprototype~a six pixel linear device in which 1% of the signacharge was lost in each transfer!, these papers provide possibly valuable lessons for the future. For example, the fiattempts to build CCDs on epitaxial material were widepredicted to fail due to the poorer crystalline quality thbulk silicon. In fact they worked much better due to thintrinsic gettering of impurities into the oxygen-rich sustrate. Also, the papers in Ref. 32 were largely motivatedthe tremendous interest in extending the performance oftical CCD imagers to high definition television~HDTV! ap-plications. While it remains true~as mentioned in Sec. I! thatthe manufacturers concerned are not interested in the s‘‘scientific’’ market, their extraordinary R&D programs mawell provide ideas that are applicable to this market.

A pioneering paper on the applications of CCDs to losignal levels34 includes the invention of a method of signprocessing called correlated double sampling~CDS! whichwill be discussed in Sec. II D.

The use of CCDs for particle tracking is closely alliedx-ray detection. In fact, x rays provide a valuable tool fdevice characterization, since they deposit their signalsvery compact clusters, allowing charge collection from tdifferent regions of the CCD~Fig. 4! to be explored separately. Useful papers from the x-ray community~which pro-vide pointers to many others! are to be found in Refs. 35–38

C. Charge transfer efficiency

Large area CCDs are the key to state-of-the-art parttracking detectors. However, this implies phenomenally hcharge transfer efficiencies~CTEs! as the signal packet isshifted from pixel to pixel.~Note also the term charge tranfer inefficiency; CTI512CTE.! With modern devices, val-ues of charge transfer inefficiency~CTI! ,1025 areachieved, compared with;1022 with the original designs.This improvement by three orders of magnitude is the reof great ingenuity and hard work by many people.

In considering the factors leading to imperfect chartransfer, one should distinguish between signal electrwhich are free and those which are trapped. For free etrons, charge motion on the manipulation of gate electrois in general due to a combination of thermal diffusion, seinduced drift and fringing field drift. In the case of smacharge packets such as one is concerned with in MIP detion, only the third of these is important. For gate dimensio7–10mm typical of three or two phase CCDs with 20mm2

pixels, fringing field drift can provide CTE>99.999% atclock frequencies up to;100 MHz. High speed operation i

oeltin

ltrrntaitc

thedehe

ti

ufo-g-llnl

ewr-

othan

rfth

t oree

ean.p

tin7sein4in5051nar

,ect

or-fulllly

ca-ockip-

eckareaat

de-

a-

vedI

e

su-ct.u-ol-thethear-

ouldgategenelct

fermehe

d topu-


dependent on strong fringing fields, which provided onethe original motivations for moving from the surface-channto buried-channel architecture, under the name peristaCCD.39 ~Note also the term bulk channel, which is agasynonymous with buried channel.!

Regarding charge trapping, transfer inefficiency resufrom atomic impurities, crystal defects, etc. within the stoage volume, which trap electrons from the signal packet fosufficiently long time that they are lost from their parepacket as this is transferred along the CCD register. Agthe early surface-channel devices were particularly subjecsuch losses, as the electrons of the signal packet interawith the continuum of ‘‘interface states’’ at the Si/SiO2 in-terface. These give rise to energy levels that populateentire band gap and can trap electrons with a huge rangtrapping time constraints. Switching to buried-channelvices greatly reduced this problem, although one was tdependent on the bulk properties of the material. Dueimpurities in the starting material and/or those introducedprocessing, bulk trapping in early CCDs was often serioPositively charged traps can have ‘‘giant’’ cross sectionscapturing signal electrons;40 gold was a particularly dangerous impurity. Trap-induced CTI is at its worst for small sinal packets, since in such cases the storage volume wiconstant~being defined by the electron thermal energy athe shape of the potential well! whereas for large signapackets the charge density will be constant~given by theconcentration of charged donors in then channel!.

Trap densities in modern CCDs have been reducedextremely low levels so that, even for small packets of a fhundred electrons, CTE>99.999% is attainable under favoable operating conditions. The quantitative discussioncharge trapping is deferred to Sec. V, since in practiceonly traps that pose serious problems in well designedfabricated modern tracking detectors are those inducedradiation damage.

The progress to the present state of excellence iscorded in a number of important papers. Charge transferfree electrons was first considered in Refs. 41 and 42, incontext of surface-channel devices. In Ref. 43 the effecbulk traps is buried-channel CCDs was considered theocally, and experimental techniques for CTE measuremwere established which are still used. The theoretical trment in terms of the Shockley–Hall–Read generatiorecombination theory44,45 ~SRH theory!, is taken up in SecV. The potentially disastrous effect of interelectrode gawas treated in Ref. 46, and the vital role of temperature~bothas a diagnostic aid and in determining the optimal operaconditions! regarding bulk traps was discussed in Ref. 4The specific case of CTE at low signal levels was discusin Ref. 48, and the effects of carrier freeze out, whichcrease dramatically below 90 K, were discussed in Ref.An important study of the theoretical modeling of CTEburied-channel CCDs at low temperature is found in Ref.the dependence on pixel clocking rate is treated in Ref.

As well as effects related to free charge transport asignal trapping, there is a third class of effect which cdegrade CTE in unirradiated devices. Due to design or pcessing defects, there may exist potential wells~also called

flic

s-a

n,toted

eof-nons.r

bed

to

fed

by

e-oref

ti-ntt-–

s

g.d

-9.

;.dno-

potential pockets in the literature! within or between gateswhich can act to inhibit complete charge transfer; their effcan be particularly severe~even catastrophic! for small signalcharge packets.

One most important lesson has been the great imptance of defining the buried-channel potential over thesensitive area of the device. This is achieved by carefuprocessing the gates to ensure full overlap at the edges~seeFig. 6!. However, excessive gate overlap increases thepacitance to be driven, and hence worsens the clfeedthrough problem. By using modern processing equment and procedures, it is possible to achieve;2 mm gateoverlaps safely.

The importance of finite overlaps, and the need to chthoroughly that these have been achieved over the entireof each production device, is illustrated by a problem tharose in testing large CCDs for the SLD upgrade vertextector. The device quality control~QC! included illuminationwith ;43106 x rays from an55Fe source~this yields the MnKa line with energy 5.9 keV, generating;1600e2 signalclusters!. Looking along every column of the CCD seprately and setting a threshold a little below theKa peak, thehit rate versus I address could be studied. For a well behaCCD column the rate would vary slowly over the range50 at the R register to I52000 at the remote end of thimage area. In some cases~see Fig. 7! one would observe anabrupt fall in hit rate at some I address, and this would ually affect only one column; the neighbors would be perfeThe problem went undiagnosed for a while, until a particlarly bad batch exhibited blockage of several adjacent cumns at the same I address. Microscopic examination ofarea revealed the cause. Due to localized overetching,nominal overlap between gates was transformed in theseeas into a ragged gap, as seen in Fig. 8. Such a feature wthen expose the buried channel to fixed charges in theoxide, interface oxide, or polyimide passivation. This charcould of course create a local disturbance to the chanpotential, sufficient to block or trap small signals. The fathat CCDs normally work so well in terms of charge transof signals as small as a few electrons is due, not to sonear-magical uniformity of the channel potential, but to t

FIG. 6. Gate structure of a modern three-phase CCD register designeavoid potential wells due to radiation-induced charge buildup or other srious charge in the dielectric or surface-passivation layers.

lri-re

onssh

en

ige

ec

re-

al-w-

eadviorver

aener-llying

thetby

dal

vi-rtingol-D.toc-er-

tes.of

tlec-er-ut

hon

henach

vinha

ra


Poole–Frenkel effect52 which ~through quantum mechanicatunneling! effectively drags electrons out of small irregulaties between the gates of either the imaging or readoutister. It is in fact quite reassuring to observe the scaledevice imperfections that do cause problems for small-sigoperation; such processing faults can certainly be addreand largely eliminated. There will inevitably be mucsmaller potential variations~e.g., due to slight fluctuations inthe thickness of the gate oxide! on all devices, but these arevidently well below the threshold required to cause aCTE loss.

Such experiences demonstrate the importance of hstatistics test data in establishing the performance of thdevices for particle tracking. CCDs provide data of unpr

FIG. 7. Hit rate as a function of the I address along a few columns hasignificant traps. The trap location can be precisely determined by the sfall in hit rate.

FIG. 8. Photomicrograph of the CCD imaging area in a region of high tdensity. The ragged gate edge due to overetching is clearly visible.

g-f

aled

y

hse-

edented granularity for an electronic detector~comparable tonuclear emulsions, and in some ways superior, e.g. asgards spatial precision!, but this granularity carries with itthe potential for microscopic defects which can cause locized performance problems. For the most part, it is not hoever necessary to study intensively each pixel. Being rout in a column-based architecture, studies of the behaalong the length of each column generally suffice to uncosignificant problems.

D. Output signal processing

The aim with any tracking detector will be to achievesingle layer efficiency>99%. This is particularly the casfor a vertex detector, since the measurement from the inmost layer is especially valuable and, if lost, cannot be furecovered by the outer tracking, due to multiple scattereffects.

Achieving such a MIP efficiency with a 20mm activelayer thickness is a significant challenge. Measurement ofmean signal (;1600e2) would be quite straightforward, buas discussed in Sec. II A, this is considerably broadenedstraggling, on top of which a MIP cluster is typically dividebetween three or four pixels, even for tracks with normincidence to the CCD surface.

Assuming the CTE complexities discussed in the preous sections are under control, one may take as a stapoint the noiseless, lossless transfer of the originally clected pixel signal charge to the output node of the CCThe function of the output circuit is to convert this chargea voltage that can be digitized in the front-end off-CCD eletronics. In tracking detectors, the requirements for noise pformance are~by the standards of scientific CCDs! modest~&100e2 rms, whereas astronomers require,10e2!, butthe readout needs to be fast~R register clocking rates in theregion 10–100 MHz!. These atypical requirements creasome complications but make it possible to bypass other

Nearly all currently used charge sensing circuits arethe ‘‘floating diffusion’’ type, shown in Fig. 9. The circuicomprises an output gate, an output diode for charge coltion, a reset transistor T1, and an output transistor T2 opated as a source follower. Conventionally, the ‘‘outpnode’’ ~output diode plus T2 gate! is reset to a referencevoltage VRD by pulsing T1 into conduction prior to eactransfer of signal charge. The change in the node voltagesignal charge transfer, sensed by the source follower, tprovides a measure of the charge. This is repeated for e

grp

p

FIG. 9. Schematic diagram of a two-stage output circuit.

n

een

tseo

allydeo

tel

th

t2

hize

icis

no

iz

ely. 5cette-

t.ilyit

ti-th

nc

t-

cle

urceace

ci-

offea-

atage,

m-icsn toin

lidal

dtor

cu-for

eer-59

osed

tonceuitthens

ta.lid


pixel in the array. Johnson~and other! noise in the circuitleads to random fluctuations in the output voltage, athereby sets a limit on the minimal detectable signal.

A major potential noise source arises from ‘‘resnoise,’’ fluctuations in the voltage on the output node whthe reset transistor is turned off. In magnitude this amouto

S kT

CnD 1/2

rms V,

where Cn is the node capacitance, and can amount.100e2, which would be quite unacceptable. This noisource is conventionally eliminated by the technique of crelated double sampling~CDS!,34,53 which essentially con-sists of measuring the voltage level before and after etransfer, and recording the difference. This is intrinsicaslow, and can be avoided completely in particle trackingtectors, where the data density is always so sparse thatcan integrate the signal charge from the entire R regisresetting the node only at the end of each row. The signaany pixel is then simply the voltage difference betweensample for that pixel and its predecessor.

The second~and now unavoidable! noise source is thaassociated with the drain current flowing in transistor Twhich generally has both white and 1/f components. As willbe referenced at the end of this section, much progressbeen made in optimizing the MOSFET design to minimthe 1/f and other undesirable noise sources~e.g., due to hotelectrons creating avalanche current near the drain in devoperated in saturation!. For tracking systems, the problemagain simplified. Modern on-CCD FETs have 1/f noise cor-ner frequency,200 kHz, so at the readout rates we are cocerned with, only the white noise floor due to the Johnsnoise in the FET channel resistance is important.

The CCD node capacitanceCn can be divided into twocomponents,Cd the detector capacitance andCg the FETgate capacitance. The FET is usually designed to minimthe equivalent noise charge~ENC! which implies makingCg5Cd ; see Ref. 33. This is an example of a completgeneral noise optimization theorem, as discussed in RefPutting in typical experimental values for FET performanparameters from Ref. 33, the rms noise associated withoutput transistor, in the case where the signal is integrathrough the interpixel period, with a pixel clocking frequency of f c MHz, is given by ENC'0.91f c

1/2Cn1/2 e2,

whereCn is in fF. Below ;5 MHz, this formula becomesincreasingly inaccurate as the 1/f noise becomes significan

The procedure for noise minimization is thus primarto reduceCn . Vast progress has been made in this area, wvalues of 25 fF being currently available and 10 fF ulmately possible. For a readout frequency of 50 MHz,former value would give ENC532e2 rms, more than ad-equate for MIP detection with full efficiency.

However, the ongoing reduction in detector capacitamatched by the output FET capacitance~by reducing thechannel widthWF! carries with it a serious penalty; the ouput impedance is proportional to 1/WF . In order to achievethe required bandwidth for high sample rates, the loadpacitance must be correspondingly reduced. This prob

d

tnts

o

r-

ch

-ner,ine

,

as

es

-n

e

4.

hed

h

e

e

a-m

has been solved by integrating a second stage on-chip sofollower as shown in Fig. 9, with transistor T3 havingchannel width approximately 10 times that of T2, and henwith a sufficiently low impedance to drive the load capatance connecting to the front-end amplifier.

The procedure of resetting the node only at the endeach row has permitted an important noise-suppressionture, called extended-row filtering~ERF!,55 which can be un-derstood by reference to Fig. 10. The principle is thwhereas a valid signal creates a step in the node voltpickup spikes or rare~many standard deviation! noise fluc-tuations will normally be restricted to one sample. The coplete row of data is digitized on the front-end electronboard, and the signal associated with each pixel is takebe the minimum of two overlapping samples as shownFig. 10. These two samples will be closely similar for vadata, but one of them will be;0 ~and hence cause the globestimate to be discarded as below threshold! in the case ofnoise. This procedure has recently been further refine56

leading to extremely low noise rates even from a detecsystem of.300 Mpixels.

Key papers in the long evolution to the present spectalar noise performance of CCDs that should be consultedmuch more detailed discussions include Ref. 57, a pioning review of all CCD noise sources. References 58 andestablished the advantage of the buried channel as oppto the surface-channel MOSFET as regards 1/f noise, includ-ing the limits ofVDS andI D necessary if that advantage isbe maintained. Valuable general reviews of the performalimitations in CCDs, especially as regards output circnoise, are Refs. 31 and 33. In general, while much ofliterature is related to noise optimization for applicatio

FIG. 10. Extended row filter operating on a row of CCD digitized daWhereas both overlapping samples register a hit when ‘‘on time’’ for vadata, this never happens for the pickup/noise spike.

iffepgI

asla

udt eec-tt i

enthse

br

lsetahheeapk-

hele

ingS5s

rob-t-e ofheCD. Alsesebe

-ur-

hes toof

int

had

5.5er-

ek-tru-

lsf

cs.,ford ava-

singCDs

al

gese-al

le-outreas

lleriapu


such as astronomy where the requirements are rather dent, the progress made has frequently been directly apcable in extending by a factor of 200 the frequency ranover which one can meet the noise requirements for Mtracking detectors. The first efficient MIP detection wachieved at an R clocking frequency of 50 kHz. Siminoise performance can now be achieved at 10 MHz.

E. Device architectures

The simple postage stamp sized CCDs with single oput, as sketched in Fig. 5, have long been supersedemore advanced architectures. Figure 11 shows a recenample of particular interest for tracking. It is 10 times largthan early devices~ideal for constructing a large area detetor! but compensates for this~in terms of required readoutime! by being subdivided into quadrants that are read ouparallel. The placement of the bond pads~along the shortedges of the device! was selected to permit a full-coveragmulti-CCD geometry, and the dead regions along the loedges were minimized for the same reason. In short,availability of affordable fully customized designs has tranformed the possibilities for developing detectors truly tunfor particular requirements.

As well as the general layout, the devices can alsocustomized as regards the gate architecture. Generally thphase clocking~as in Fig. 5! is preferred for the imagingarea, but two-phase operation, with symmetrical clock puto minimize feedthrough to the analog-output, is advangeous for the R register. The noise performance and off-cdrive capability of the output circuit can be designed with tpossibilities outlined in Sec. II D. This device was designwith four output ports, but a feature developed for otherplications that will certainly be valuable in future CCD tracing detectors is that of multiport readout registers~with noloss of active area!. The experimenter can thus balance tdesired readout time against the volume of front-end etronics, which scales with the number of channels.

One vital topic will need to be addressed as clockspeeds are further increased. Despite the comments inII D, it has so far proved impossible to come close to aMHz readout rate with,100 e2 rms noise. The reason habeen a serious level of interference~crosstalk! between the R

FIG. 11. Basic architecture of a modern particle-tracking CCD. The pararegister~I register! shifts signal packets to each end of the device. A seregister ~R register! at each end shifts signal packets to a pair of outcircuits.

r-li-eP

r

t-byx-

r

n

ge-d

eee-

s-ip

d-

c-

ec.0

drive pulses and the sensing of the output voltage. The plem is partly on-chip but much more importantly in the fronend electronics. Sensing mV-level signals in the presenc;10 V fast clock pulses has proved extremely difficult. Tway forward probably relies on some developments in Carchitecture and changes in the front-end electronicspromising approach appears to be to generate the R pulocally to the register of the CCD, taking care with wirbonds and the trace layout on the CCD itself. It may evenpossible to operate with balanced~opposite-phase! sinusoidalclock signals to minimize higher harmonics. This will involve some special features to inhibit R register transfer ding the I-to-R shifting between rows.

In general, an increasingly intimate link between tCCD design and the front-end electronics design promiseopen up important new horizons in the overall capabilitythese detectors in the near future.

F. Tracking performance

By the late 1970s, CCD development was at the powhere their quality~uniformity, noise performance, etc.!made them valuable tools for astronomy. Particle signalsbeen observed in a related device~photodiode arrays!,60 al-though limited to the enormous charge disposition fromMeV a particles; the noise performance would not have pmitted MIP detection.

The first observations of MIP detection in CCDs camfrom astronomers61,62who saw cosmic ray signals as a bacground present particularly when they operated their insments in telescopes at high altitude~thus ruling out back-ground radioactivity as the source of these signa!.Following a theoretical evaluation,9 the first measurement oefficiency(.98%) and precision (;5 mm) for MIP detection wasmade in a CERN test beam.10 Figure 12 from Ref. 10 openedthe eyes of the HEP community to the potential for physiWith 17 hits over 1 mm2 and no problem of cluster mergingit was clear that such devices could provide the basispowerful vertex detectors. This development encouragenumber of other pioneering experimental studies with ariety of CCD architectures,63–67all using off-the-shelf opticalimaging devices available at the time.

The features of narrow band-gap and planar proceshave already been mentioned as advantages of silicon Cfor MIP tracking. There are in addition:

~1! use of a relatively low-Z material;~2! intrinsically single-sided processing~unlike some track-

ing detectors!, hence they can be thinned by mechaniclapping;

~3! adequate signals from only 20mm of material.

These features represent particularly important advantafor vertex detectors, in which multiple scattering in the dtector layers is frequently a limiting factor in the topologicevent reconstruction.

In order not to be swamped by dark current, partictracking CCDs are generally cooled. Depending on readconditions, quite modest cooling from room temperatu~e.g., to 0 °C! might be adequate in some applications, but

llt

at

icesslon

rereehes

ow

is

coe

gafd

ing

thesnu-

therststers-hethatter;tor.tal

.tercytialmsnals

eg-rely

and

kersomntion

ta

ne,ion

toren-for-theion

e


will be discussed in Sec. V, radiation effects can be gresuppressed by much more substantial cooling~and trackingCCDs are always operated in a radiation environment!. AllCCD tracking detectors so far used have been operated intemperature range 150–200 K, which of course has impltions for the mechanical and other design details, as mtioned in Secs. III and IV. Despite some early misgivingthis has not caused significant problems; it has in all cabeen possible to engineer simple cryostats of extremelymass which have been entirely adequate for the experimerequirements.

Early MIP signals, while generally confirming the mooptimistic theoretical estimates of charge deposition referto in Sec. II A, suffered from tails on the low side of thdistribution which were identified with process defects in tdevices available at the time. The improvements in proceing quality since then have been impressive. Figure 13 shthe MIP signals from a large CCD~of the layout shown inFig. 11! being read out at 5 MHz, where the scale;30 e2/, analog-to-digital converter~ADC! count. Theelectronic noise in this detector was;60 e2 rms, or 2 ADCcounts. The now-standard signal processing proceduresists of effectively a two-pass approach. The locally digitizsignals are transmitted in real time~by multiplexed opticfibers operating at 1 Gbit/s! to a rack of signal processinelectronics adjacent to the overall detector. In the first p~also carried out in real time during readout! addresses opixels which satisfy a low threshold are provisionally store

FIG. 12. One mm2 of a CCD in a MIP test beam with every pixel of 20mm2

being read out. No threshold: Gray scale indicates the signal amplitudeach pixel.

ly

thea-n-,eswtal

d

s-s

n-d

ss

,

along with the pixel signals from a generous surroundregion of interest~836 pixels total for every trigger pixel!.The aim is to efficiently find clusters even in cases wheresignals are split uniformly over a 333 area, as sometimehappens. Microprocessors look at the data from all thesemerous ‘‘regions of interest,’’ assemble clusters usingdata from pixels neighboring the one which triggered the fipass filter, and accept the cluster if a more robust cluthreshold of;250 e2 is satisfied. Figure 13 shows the cluter signal distribution for MIPs from such a detector. Tentries in Fig. 13 are guaranteed MIP signals by the factthe clusters have been subjected to a further offline filthey are located on fitted tracks from a three-layer detecIn this way the prevalent backgrounds in the experimenconditions~due to x-rays, etc.! are excluded from the plotThis two-pass approach with final adjudication by a clusthreshold permits essentially 100% MIP detection efficiento be achieved with very high noise immunity, as is essenin systems consisting of several hundred Mpixels if probleof data storage are to be avoided. In the absence of sigfrom ionizing particles, this signal processing leads to,50accepted noise clusters per CCD, of 3 Mpixels, a totally nligible level, and some of these accepted clusters are su‘‘real’’ ~due to background radioactivity, etc.!. Details of thissignal processing procedure are described in Ref. 56,references therein.

III. FIXED TARGET EXPERIMENTS

A. Detector design

As discussed in Sec. I, the purpose of the CCD trac~vertex detector! within an experiment is to resolve trackfrom secondary and tertiary vertices from those coming frthe primary interaction vertex. As noted in Sec. II F, curreCCD tracking detectors give a point measurement precisof ;5 mm ~by taking a simple centroid of the digitized dafrom a cluster of 20320mm2 pixels!. Of course, what isrelevant for physics is not the precision at the detector plabut that achieved after extrapolation to the interaction reg~IR! close to the primary vertex~PV!. By arranging CCDdetector planes approximately at distancesd and 2d from thePV, the typical precision at the IR will be;7 mm. For lowmomentum tracks, multiple scattering in the first detecplane will degrade the precision, whereas for high momtum tracks the precision may be better than this, since inmation from more remote detectors can further refineextrapolation. At first glance, an impact parameter precis

in

FIG. 13. MIP cluster signals from a large area CCD~data from the SLD!.

f

estwtc

e

alil

gon

iheerda

eckra

-

aecta

he

Cei-

or-

isationtorthiswDsnts

di-a-s.n.

athethatery-

ec.hat

byionthe

eionse-tionofsttheizeamf aon-nlythe

onalOne-hetes a

1/ceentsingwseryalids inwas

e


of <10mm appears to be more than adequate. Other thanlow momentum parent particles with Lorentz boostbg&1,the mean value of the impact parameter of a decay track~theamount by which the extrapolated track misses the IP! isapproximately 0.7ct; 220 mm for B decays and 100mm forcharm. However, the desirable precision is far below thvalues, since some tracks will by chance pass close topossible vertices, all distributions have significant tails, eIn fact the experience to date~and this applies equally to thcollider experiments discussed in Sec. IV! is that the physicsreach is substantially increased as the vertex detector quevolves; even today one falls short of anything like overkin this aspect of the experiment.

Due to one particular advantage, high energy fixed tarexperiments provided the first environment in which silicvertex detectors~both microstrip detectors and CCDs! wereable to make a contribution to heavy flavor physics. Thisthe simple fact that in the Lorentz transformation from tcenter-of-mass~c.m.! to the lab system, the majority of thsolid angle is compressed into a relatively small forwacone. Thus general purpose multiparticle spectrometerstypically long detector systems~tens of meters! with modest(;2 m) transverse dimensions. By using a well focusbeam on a thin solid target, and by placing the silicon traing detectors close to the target, very modest area covewas required. Thus, the two-CCD vertex detector usedACCMOR experiment NA3268 covered the entire spectrometer aperture with only 0.4 cm2 of CCD area even in thedownstream detector plane~see Fig. 14!.

As well as concentrating the relevant solid angle inlimited forward cone, the Lorentz boost provides a furthbonus, plenty of space to the side of each active deteplane for frames, readout boards, cables, etc. With the msive local electronics required for the early detectors, tspace was an essential requirement. Thus for the NA32periment, the only significant material generated by the Cdetector in the aperture of the experiment were the two layof silicon. Even without thinning the silicon below the orig

FIG. 14. CCD vertex detector for a fixed target experiment~NA32!. Dataare fast shifted into the quiet regions above and below the spectromaperture for CCDs 1 and 2, respectively, prior to readout.

or

eo.

ityl

et

s

re

d-gein

rors-

isx-Drs

nal wafer thickness~350 mm! this implied a negligible mul-tiple scattering penalty, again thanks to the effect of the Lentz boost on the momenta of the decay particles.

The final advantage of the fixed target environmentthe easy access to the detectors. In these conditions, radidamage is not a major problem. The entire CCD detecsystem could easily be replaced once or twice a year;was the practice followed in the NA32 experiment. In vieof their favorable characteristics, the only reason why CChave not been more widely used in fixed target experime~particularly for charm photoproduction, where the contions would be extremely clean! seems to have been the reltive unfamiliarity of these devices for tracking application

CCD tracking is not only of interest for vertex detectioTheir very high precision can be exploited in constructingcompact high resolution spectrometer, in cases whereangular and momentum range happen to be small, solarge area coverage is not required. An example is a vhigh resolutione1e2 pair spectrometer currently being prepared as part of a SLAC experiment.69

B. Readout architecture

In contrast to the physical advantages noted in SIII A, a disadvantage of the fixed target environment was tthe central region of the CCD active area was traversedthe high intensity primary beam. Consequential radiatdamage problems could be minimized by defocusingbeam in the horizontal dimension~a beam profileH3V of830.3 mm2 was used in this experiment!, but the hadronicdamage resulting from this beam~typically 106 particles/spilland a spill rate of 0.083 Hz! necessitated exchanging thCCDs every six months or so. The nature of the radiatdamage effects will be discussed in Sec. V. Another conquence of the beam traversal was the continual generaduring the spill of a huge number of hits in the active areathe CCD. Given the nonavailability of a true fast clear or falatch feature on a CCD, these hits threatened to obscurewanted data. A procedure was therefore devised to minimthis problem; the CCDs were clocked throughout the bespill in the so-called fast clear mode. This consisted osequence of I shifts at 1.4 MHz so that the image was ctinuously shifted into the R register and dumped. Since othe remote one-third of each CCD was needed to coverspectrometer aperture, it was possible to use the additiheight of the device as a parallel analog storage region.receipt of a trigger, the fast shifting continued until the rquired image region was in a ‘‘parking area’’ adjacent to tR register. Figure 15 illustrates the idea; an event generapattern of hits in the readout area~marked3!; these storeddata are shifted vertically upwards/downwards in CCDCCD2 into the parking areas. The CCD drive sequenwould then be changed to normal read mode, until the evhad been read, thereafter reverting to fast clear and releathe deadtime. During the relatively slow readout, the rotraversing the beam position vertically would encounter vhigh occupancies, but such regions could never contain vdata due to the system deadtime. In fact to keep conditionthe parking areas even cleaner, a small kicker magnet

ter

de

xtiorc

-inthhi

tenroolainr tmn

isenew

herergethein

iple

thaten-areingeadr.

asi-

ries

tec-ers-ci-, soalge at

x--g

70.meew

ull56,rest

ceasde-,allone

leasith

cal

men

3


used to dump the beam far upstream of the ACCMORtector during readout, but this was barely necessary.

C. Physics performance

The NA32 experiment with the CCD/microstrip vertedetector provided some of the cleanest charm reconstrucof any detector system. Figure 16 shows one of the ficharm decay events seen in the CCD-based vertex deteThe event-related hit density in the upstream detector~only12 mm from the IP! was;200/mm2; no other class of tracking detector could come close to resolving hits at anythlike this density. As can be seen, the beam-related hits inexperiment made a negligible additional contribution to toccupancy.

Given the harmonious conditions for CCD-based verdetectors in the fixed target environment, the only reasomove out of this arena was the physics importance of pcesses at c.m. energies that could only be accessed at cing beam facilities, and the need for heavy flavor identifiction in studying such processes. It was the overwhelminterest in such high energy processes that accounted forequirement to develop vertex detectors able to face thejor additional challenges of the colliding beam environme

FIG. 15. CCD readout scheme for experiment NA32. Shown is a beaeye view of the two CCDs of Fig. 14, with a relative sideways displacembetween the CCDs for clarity.

FIG. 16. Tracks from the IP and from a nearby charm decay in the NAvertex detector. The frame size is 131 mm2.

-

nssttor.

giss

xto-

lid--ghea-t.

This was an onerous undertaking, involving~in the case ofCCD detectors! around six years of intensive R&D work.

IV. COLLIDING BEAM EXPERIMENTS

A. Detector design

The main advantage of colliding beam experimentsthe availability of c.m. collision energies far beyond threach of fixed target experiments, and hence access tophysics processes. Against this overriding advantage, tare two main disadvantages. The first is the need for lasolid angle coverage in all detectors, and the second isfact that particle momenta are typically much lower thanthe boosted fixed target environment and hence multscattering in tracking detectors is serious.

Regarding these problems, it seems at first glancethe ideal vertex detector would consist of a series of conctric thin spherical shells. But the fact that the beamscontained in a cylindrical pipe, and the advantages of placthe first detector layer as close as possible to the IP, linevitably to a cylindrical geometry for the innermost layeFrom time to time there has been interest in using quspherical shells further out~e.g., the‘‘lampshade’’ geometry!but optimized mechanical designs generally lead to a seof nested cylindrical barrels~made as thin as possible! withstable mechanical supports at the ends. For microstrip detor systems there may be an argument for short cylindplus ‘‘endcap’’ detectors~planes normal to the beam direction! since the track precision is degraded for oblique indence. CCD detectors are essentially free of this problemin this case, ‘‘long barrels’’ generally constitute the optimdesign, unless one had some special reason for coveravery small polar angles.

Both CCD vertex detectors constructed for the SLD eperiment~called VXD2 and VXD3! have consisted of concentric barrels composed of ‘‘ladders,’’ the basic buildinelements of these detectors. For VXD2~for which workstarted in 1984! available CCDs~area;1 cm2! limited thecoverage severely. This detector is described in Ref.Given the rapid progress with CCD developments, it becapossible in 1994 to start design and construction for a ndetector, VXD3, using much larger (;12 cm2) full customdevices of the general architecture shown in Fig. 11. A freport on this detector has recently been published Ref.so the details can be omitted here. Points of general inteare as follows.

First, the CCD design included features of importanfor the mechanical construction. The on-chip circuitry wspecially arranged so as to permit the sawn edge of thevice to be within 300mm of the edge of the active areaalong the long edges of the CCD. The wire bonds werelocated along the short edges, and displaced away fromedge to permit the ladders~Fig. 17! to be tiled in each barrewith small azimuthal overlap between adjacent active ar~Fig. 18!. The active surface of each CCD was overlaid wevaporated aluminum squares of 60360mm2 on a pitch of 3mm, which were used as fiducials for purposes of optialignment.

’st

2

bial

ya

t

em

do

fucol-

ad’’teimtoo

easin

notndle,inrly

amt ofhhe

p-oc-ull

ol-ces--s,eringlife

m-tex-tly

nt,

llyents at

cu-

heeedsim-h

L


The very thin, flexible ladders were mechanically stalized by being firmly clamped at each end in the overberyllium support structure~see Fig. 19!. Due to the need toallow for controlled movements as a result of thermal ccling, the mechanical design incorporated a judiciousrangement of stress-relief features~sliding joints and elasto-meric adhesives!.

Since the total power dissipation in the cryostat duethe readout of the 307 Mpixels detector was only;15 W,the detector could easily be maintained at its operating tperature (;180 K) by a gentle flow of cold nitrogen gas.

B. Readout architecture

Having established the desired general mechanicalsign for collider vertex detectors, let us consider the readrequirements. Unlike the fixed target applications, theCCD area must be used as active detector; there is no sfor a ‘‘parking area’’ outside the active region. As the colider energy is increased, the need for high luminosity leto high trigger rates. This is particularly true for ‘‘discoveryphysics, where energy deposition in the calorimeter sysmay be small, and the energy threshold for a trigger is lited only by noise in the calorimeters. Therefore any detecneeds to operate in a deadtimeless or short deadtime mIn cases of high background rates, it is obviously desirablincorporate a fast gating capability so as to latch the dassociated with the triggering event. This requirement, eamet in most high energy physics detectors where the sig

FIG. 17. Two CCD ladders that formed the basic elements of the Svertex detector VXD3.

FIG. 18. Cross section~end view! of the VXD3 detector.

-l

-r-

o

-

e-utllpe

s

m-r

de.totalyal

charge is collected promptly on external preamplifiers, isan option available in a CCD detector. Thus all backgrouduring the readout is integrated with the signal. In principan equivalent of the kicker magnet used to kill the beamfixed target experiments could be used. This is particulasimple if the background comes mainly from beam–beinteractions; one needs only to displace the beams oucollision during readout. This is of course ruled out in higtrigger rate conditions by considerations of deadtime. Tmore widely used approach is to permit full-luminosity oeration throughout the readout and, if a second triggercurs, to simply continue reading until the corresponding fframe has again been acquired~deadtimeless operation!.

In practice, CCD detectors are ruled out at hadron cliders for two reasons. First, instantaneous rates are exsive. Despite the largepp total cross section, and correspondingly the enormous rate of ‘‘minimum bias’’ eventthese colliders have to run with very high luminosity in ordto have a reasonable rate for the tiny fraction of interestevents. Second, CCD detectors would have inadequateexpectancy due to their sensitivity to hadronic radiation daage ~see Sec. V!. Note that this situation is quite differenfrom the fixed target hadron beam experiments, wherechanging the CCDs every few months is neither too cosnor inconvenient. Nested barrels of detectors of;1 m2, bur-ied inside thousands of tons of other delicate equipmeneed to be far more robust.

In e1e2 colliders, the hadronic backgrounds can usuabe reduced to a very low level, so the radiation environmis generally tolerable. However, the instantaneous ratehigh-luminosity machines such as theB factories now underconstruction would also rule out CCDs on grounds of ocpancy. For the high energye1e2 colliders~LEP and SLC, atthe present!, the hit rates can easily be accommodated. Thuge beam pipe at LEP was a major drawback, and indall vertex detectors there have been constructed with thepler silicon microstrip technology. SLC, with its muc

FIG. 19. Cross section~side view! of the VXD3 detector.

D

hityauluw

toub

uinv

eeinin

oa

teaigse

a

iu

d

g

tI

fob

itvew

i

reer

ichted.2.be

om

asers

yiconreasides, itsub-

pic-CD

mo-

to


smaller beam pipe~2.5 cm, compared with 10 cm in LEPoriginally, subsequently reduced to 5 cm!, provided a muchmore hospitable environment for vertex detectors, and wita greater physics potential than LEP for a given luminosIn fact, the SLC luminosity has been considerably lower that LEP, and for this reason the heavy flavor physics resfrom the two machines have been of quite similar quality,to the present time. What was really needed for physicsthe LEP luminosity with the SLD vertex detector.

At SLC, the time structure of beam bunch crossings ams intervals lent itself to a convenient CCD detector readarchitecture. Note that the front-end electronics could alllocated outside the cryostat~Fig. 19! using thin copper–kapton flex circuits for the connections to the CCDs. Defaoperation consisted of fast clearing between beam cross~analogous to the fast shifting used in the fixed target enronment!, so that the CCDs were relatively background fron receipt of a trigger, followed by standard readout durwhich the background of the succeeding 25 beam crosswas accumulated.

The generation of CCD drive pulses, the amplificationthe analog signals, and the digitization of those signalstook place in the front-end electronics boards mounwithin the aperture of the SLD central drift chamber, at smpolar angle below the region of tracking. The digitized snals were multiplexed onto fast optical fibers and procesremotely, as discussed in Secs. II D and II F.

C. Physics performance

Key features of the VXD3 detector that placed it indifferent category from the competition were

~1! a small SLC beam pipe, hence small inner-barrel rad~28 mm!;

~2! thinned CCDs on beryllium substrates, giving thin laders~0.4%X0!;

~3! a good lever arm from first to second layer, allowinprecise extrapolation to the IR.

The performance of such a detector is best described byimpact parameter precision for tracks extrapolated to theas function of momentum. Results for this detector andits predecessor VXD2 are plotted in Fig. 20, and canparametrized approximately as

s rf514.0%33

p sin3/2 umm,

s rz526.5%33

p sin3/2 umm.

In each projection, the impact parameter precision wVXD3 is at least a factor of 2 better than has been achiewith competing microstrip detectors at LEP. Given the lomomentum of tracks in thee1e2 collider environment, themultiple scattering term in the impact parameter formulaof the greatest importance.

Looking at the data fromZ0→hadron decays in the SLDexperiment, the impact parameter distribution is compawith Monte Carlo~Fig. 21!. The remarkable agreement ov

it.ntspas

8t

e

ltgsi-

ggs

flldll-d

s

-

heRre

hd

s

d

four orders of magnitude demonstrates the degree of whthe performance of this detector is understood and simula

The performance for a typical event is shown in Fig. 2When the IR region is magnified, displaced vertices canseen clearly; this was an event of typeZ0→bb.

A great deal of heavy flavor physics has emerged frthis experiment; see Ref. 71 for recent reports.

V. RADIATION DAMAGE IN CCD TRACKINGDETECTORS

A. Introduction

CCD tracking detectors will inevitably be operated inradiation environment, a situation also encountered by uof imaging CCDs in industry~nuclear, x-ray and electronmicroscopy, for example!, for space-based optical and x-ratelescopes, etc. Radiation damage in these complex sildevices is therefore relevant to numerous application aand has been studied for many years. Reference 72 prova particularly valuable review. Despite being 17 years oldremains the most comprehensive general paper on thisject.

Despite these extensive studies, there is no simpleture that summarizes radiation effects of concern to all C

FIG. 20. Measured impact parameter resolution as a function of trackmentum for tracks at cosu50 for VXD2 and VXD3 compared with theMonte Carlo simulations.

FIG. 21. Data~points! and Monte Carlo~histogram! distributions of theimpact parameter with respect to the IP inZ0→hadron decays~VXD3 de-tector in the SLD experiment!. The tails on the positive side are dueheavy flavor decays.


FIG. 22. End view of a detector with a reconstructed event of typeZ0→bb.

vic

IP

yin

ouo

eetota

on

uon

eaon

gh

se

/ofera-

nge

eaceef oftedtral

r-ichn

ears

users for two reasons. First, the uses made of these deare highly variable. To a particle physicist~who is interestedin the tracking precision given by the centroid of a Mcluster! a 10% loss of signal~as long as it be slowly varyingacross the detector area! would not be serious. To an x-raastronomer, using the cluster signal amplitude to determthe x-ray energy, such a degradation would be disastrSecond, the radiation sensitivity depends strongly on theerating conditions, such as integration time, readout spetc. These conditions may be imposed by external facpeculiar to a specific application. For example, the limitions on operating temperature and power dissipationspace-based systems are likely to be more restrictive thaterrestrial applications.

In this section, an attempt is made to focus on the issrelevant to the particle tracking/vertex detector applicatileaving aside issues of great importance to other users.

B. Surface damage

Let us consider the typical case of ann-channel CCDwith dielectric consisting of equal thicknesses of SiO2 andSi3N4. The oxide layer has a band gap of 9 eV, and a menergy for electron-hole generation of 18 eV. The electr

es

es.p-d,rs-fin

es,

ns

and holes generated by ionizing radiation will~for a biasedCCD! drift in opposite directions. The electrons have himobility and are rapidly drifted into the bulk silicon.~Notethat the drift direction would be opposite to this in the caof a p-channel device.! The holes~at temperatures*150 K!have sufficient mobility to drift within minutes to the oxidenitride interface, where they have a significant probabilitybeing trapped. For devices operated at much lower temptures~e.g., liquid nitrogen temperature! the holes may remainessentially static within the oxide, but occasionally warmithe detector would permit them to drift to the oxide/nitridinterface or~if not trapped there! to be neutralized at theCCD gate electrodes.

In addition to fixed positive charge buildup within thdielectric, ionizing radiation causes an increase in the surfstates~interface states! at the oxide/silicon interface. Thesinterface states are primarily acceptorlike in the upper halthe band gap, and donorlike in the lower half. In a deplen-channel device, the Fermi level leaves the donors neuand the acceptors belowEf negatively charged. So the inteface charge partially neutralizes the delectric charge, whis one reason whyn-channel devices have higher radiatiotolerance~for ionizing radiation! than dop-channel ones.

Numerous procedures have been devised over the y

io

th

e

e

sis

n-

aneD

x

krseeheTh

eth

n;eria

hiiftp

c

Einr

ndina

onil

he-

e

nal-n-

theu-

s

rgeinor-i-

eith

thetput

theh

bederthe

ent

x,redselkofaltedis-c-

ureraps

by

he

ure

tely’’st a

elltore

xt orof


for reducing the radiation-induced charge accumulat~‘‘hard oxide’’ technology!. For a typical modernn-channelCCD with oxide and nitride layer thickness each 80 nm,radiation-induced shift in flat-band voltageDVFB is;100 mV/krad for devices irradiated under bias. Devicwith reduced dielectric thickness~but still good yield andstable operation! are available withDVFB;10 mV/krad, andanother factor of 10 reduction has been achieved in expmental devices.

The effect of the flat-band voltage shift is to progresively raise the potential of the CCD buried channel. If thshift exceeds;2 V ~i.e., after a dose of 20 krads with stadard devices!, charge transfer to the output node~whose po-tential, set byVRD , does not drift up! will become ineffi-cient. VRD can be raised to compensate, implying alsoincrease inVDD to preserve the output circuit gain. This cacontinue untilVDD reaches the limit set by overall voltagbreakdown. With the development of the pyrogenic CChard oxide process,73 the practical limit for ionizing radiationis .1 Mrad, which is entirely adequate for all CCD vertedetector applications contemplated to date.

The flat-band voltage shifts are greatly reduced~typi-cally by a factor of 10! if the CCD power is off duringirradiation. In collider applications, where the main bacground is accumulated during machine tuning, this at filooks like an attractive option. However, as we shall sbulk damage effects are much more dangerous and toccur regardless of whether the CCD is biased or not.inner layers of a vertex detector~sitting just outside a virtu-ally radiation-transparent beryllium beam pipe! are uniquelyvulnerable; no external radiation monitor will provide a usful dose measurement. It is therefore prudent to keepdetector operating at all times during accelerator operatioshould be run in fast-clear mode outside of data taking pods, using the induced signal current to monitor the radtion, and have an interlock to shut off the accelerator if texceeds a safe level. The increase in flat-band voltage sha small price to pay for the vital protection against unacceable bulk damage.

As well as causing flat-band voltage shifts, the interfastates produced by ionizing radiation act as sourceselectron-hole generation, i.e., increased dark current. In Happlications, there is no reason not to design the trackdetector for operation at cryogenic temperature, therebyducing the dark current to completely negligible levels.

Reference 74 provided the first insight into radiatioinduced surface damage in CCDs. Other papers whichscribe the important early progress in dielectric hardentechniques are Refs. 75–81. Reference 82 provides a vable review and report on recent developments.

C. Bulk damage

The term bulk damage refers to permanent radiatiinduced changes to the crystalline structure of the bulk scon. Most relevant for CCD particle detection are tchanges to then-type channel within which the electron signal charge is stored and transported. Bulk damage cancaused by electromagnetic radiation, for electrons abov

n

e

s

ri-

-

n

-t,see

-eiti--

sis

t-

eofPge-

-e-glu-

-i-

bea

threshold energy of;200 keV, determined by the minimumenergy transfer of;20 eV required to displace a silicoatom from the crystal lattice. Electromagnetic radiationways results in ‘‘point defects,’’ simple complexes of vacacies (V) and interstitial atoms (I ). The interstitials are be-nign and immobile, but the vacancies diffuse throughcrystal until they form a complex with some atomic inclsions. In the phosphorus-dopedn layer, they nearly alwaysform a Si-E center, an acceptorlike P–V complex which lie;0.45 eV below the conduction band edge.

Bulk traps can adversely affect the dark current, chacollection efficiency, and charge transfer efficiency. Evenheavily irradiated CCDs, the excess dark current can nmally be dealt with by modest cooling. Given the thin eptaxial layer (;20 mm) from which the MIP signal is col-lected, the requirements made on minority carrier lifetimare not severe, and there is essentially no problem wcharge collection into the potential wells. However, onceelectron charge packet starts its long journey to the ounode~possibly several centimeters,;2000 pixels!, the situ-ation is far more dangerous. At every location wherecharge packet is momentarily stored~and there are three suclocations for every pixel of a three-phase CCD! there is afinite probability that some of the signal charge maytrapped, leading to loss of charge transfer efficiency. In ornot to seriously degrade the signal-to-noise performance,average CTI of a tracking detector in a large instrumshould typically not exceed;1024.

For hadronic irradiation, the situation is more compleso let us first consider the case of a CCD that has suffebulk damage from electromagnetic radiation, and whon-channel is randomly populated with a single type of butrap. These acceptorlike defects have a high probabilitycapturing signal electrons which come within their electricsphere of influence. This situation is described by a restriccase of the general SRH theory of carrier capture and emsion from traps, in which only capture and emission of eletrons from/to the conduction band play a part. Hole captand emission are irrelevant since we are concerned with tin depleted material. This situation has been consideredvarious authors.43,47,50

Let us first take a qualitative look at the situation. As tcharge packet is transported from gate to gate~within a pixelor between neighboring pixels!, vacant traps that lie withinthe storage volume of the charge packet will tend to captelectrons. If the traps are already filled~either fortuitously,due to the passage of an earlier signal packet, or deliberafor this purpose by the injection of an earlier ‘‘sacrificialcharge packet!, they will permit the signal electrons to pasundisturbed. Also, if the signal packet is transported asufficiently high clock rate that the dwell timetg under anygate is small compared to the trapping time constanttc , thesignal electrons will pass. Also, if the trap emission timconstantte is small compared with the clock pulse rise/fatime t r , the trapped electrons will be reemitted in timerejoin their parent charge packet. Only if electrons atrapped and held long enough to be redeposited in the nelater potential well, does the process contribute to a loss

e

n

le

Fo

-rgrly

tutrothow-

cC

iv

lib

isD

tine

tse

bgc

f

ra

ch

theve auf-ly

nts.heyam-

ys-ies,

tg;e

nto

at-

rat-


CTE. This is evidently a multiparameter problem with somroom for maneuver.

Let us now look at the process quantitatively.Assuming all traps to be initially empty, the CTI is give

by

CTI5(j 51

NF

F j3Ntr

NsF12expS t r

teD G .

NF is the number of phases per pixel~3 for a three-phasestructure!.

F j is the fill factor for phasej , i.e., the probability that atrap in the charge packet storage volume will become filduring the dwell time.

F j512exp~2tg /tc!.

For most cases of practical interesttc is ;10 ns andF j

may be taken to be unity.Ntr is the trap density andNs , thesignal charge density, is a function of the signal size.very large charge packets, (*105e2) it is approximatelyequal to then-dopant concentration, but for signalsNe in theMIP range one findsNs}1/Ne since the signal electrons occupy a constant volume determined by their thermal eneand the three-dimensional potential well in which they astored. The CTE for small signals is correspondingreduced.82

Now

te5exp@~Ec2Etr!/kT#

snXnvnNc.

The terms in the denominator are in turn the electron capcross section for that trap type, an entropy factor, the electhermal velocity, and the effective density of states inconduction band. The numerator tells us that for shalltraps~or high temperature! te is likely to be short and, conversely, for deep traps and/or low temperatures,te is likelyto be long. In fact, for deep traps and appropriate clotimes, by reducing the temperature, one can sweep thethrough its full range from approximately zero~since thecharge is reemitted into the parent pixel during the drpulse risetime! to 3Ntr /Ns ~for a three-phase CCD! and backto zero, as all traps are filled by some long preceding deerate or accidental charge packets to have been clockedof the device. Figure 23 nicely illustrates this point. Thdemonstrates the growth in CTI due to irradiation of a CCwith a radioactiveb source. The density of Si-E centers in-creases, but the effect on CTI can be minimized by operaat or below 190 K, where the trap emission time becomadequately long. For this trap, the emission time constan210, 190, and 170 K are 69 ms, 1.06 s, and 31 s. The dradation in CTI below 160 K~even before irradiation! is notseen in later CCDs from the same manufacturer. It probarepresents an artifact of the register design or processinthis particular device. In practice, one can normally reduthe operating temperature to;85 K before the CTI rises to;1024 at the onset of ‘‘carrier freeze out,’’ the trapping osignal electrons by the phosphorus donor ions.50 This effectsets an effective lower limit to the useful operating tempeture of n-channel CCDs.

d

r

ye

ren

e

kTI

e

-out

gsatg-

lyofe

-

For hadronic irradiation of CCDs, because of the mugreater nonionizing energy loss~NIEL! factor,83,84 the dam-age rates are greatly increased, as is the complexity ofdamage process. Charged hadrons such as protons halarge cross section for Coulomb–nuclear scattering with sficient energy transfer to the struck silicon atom to not ondisplace it but to generate a cluster of further displacemeNeutrons have smaller interaction cross sections, but tinduce nuclear disintegration which creates even larger dage clusters~having dimensions typically hundreds of a˚ng-stroms in longitudinal and transverse dimensions!. Theseclusters constitute highly disordered regions within the crtal, and may be a source of mobile vacancies, divacancetc. In the heavily doped CCDn channel, after hadronicirradiation the majority of active defects are again Si-E cen-ters, but there are additional significant traps atEc20.4 eV,believed to be the divacancy (VV),85 and shallower traps aEc20.30 and20.12 eV.85,86Protons are the most damaginFig. 24 shows the CTI resulting from an irradiation with thvery modest dose of 3.63109 10 MeV proton/cm2. The ef-fect of the lower level traps is to extend charge trapping i

FIG. 23. Effect of ionizing radiation damage on CTI as function of opering temperature~90

Sr b source! ~from Ref. 82!.

FIG. 24. Effect of hadronic radiation damage on CTI as function of opeing temperature~10 MeV protons! ~from Ref. 85!.

T

oyhewnsinc

hrethen

I

wtutieet

l-D.orof

er

blntt inna

e

meeFdi

s

aoteanwitu

.th

ag-cu-

ry

o-to

y.tionndi-

D-indve-I.

sofen-ass

,

ar-an

atu-ndionica-ofysi-

ofrtantas

o, asra-tson

hi-par-e

entg

ofm-


the low temperature region, so the problem of degraded Cno longer has a clear solution by cooling.

While these proton damage results are of great imptance for their particular application area~space-based x-racameras!, they probably give a pessimistic impression for tconditions relevant to particle detection systems for tmain reasons. First, these results refer to very low sigdensities, so the benefits of the long trap emission timelow temperature are not exploited to the extent possibleparticle physics experiment. Second, the only hadronic bagrounds likely to be significant at ane1e2 collider are neu-trons leaking through the shielding. There is evidence tneutrons may be much less harmful than would be inferfrom these proton data. Taking the standard NIEL factor,data of Fig. 24 correspond to an equivalent dose of 1 Mneutrons of 331010 n/cm2. Yet there are measurements on-channel CCDs~buried channel!, which demonstrate CT;1024 for 331012 n/cm2 at a temperature of 140 K.87 Inthese measurements the time between charge packetsonly 10 ms, but there is evidence that at this temperathere was little CTE degradation by increasing this time10 s. There is the further difference that the neutron studhave all been made with large signal packets, but from R82 the degradation in CTE going from larger signals downpackets of;1000 e2 was only a factor of 2 in materiairradiated to 60 krad~beta source!. Therefore the sparse measurements of the effect of hadronic damage on CTE in CCare not internally consistent to better than a factor of 100is possible that the simple application of the NIEL fproton/neutron comparison is not valid. A low densitymassive damage clusters may be less harmful~for CTE! thana high density of small damage clusters. Improved expmental data are urgently needed.

D. Discussion

Due to their long readout time, CCDs are not applicaas vertex detectors in continuous high flux environmesuch as the LHC. They would also be completely ruled ousuch an environment due to the high hadronic backgrouCCDs have a proven record in fixed target experime~where the incident beam can be interrupted during the reout! and in thee1e2 linear collider environment, where thinterval between bunches~or between bunch trains! allowstime for readout. In both these environments, radiation daage effects have so far been modest. In the fixed targetvironment, given the small number of CCDs required, thcan simply be exchanged at intervals of six months or so.the e1e2 collider, with reasonable care over beam contions, the detector lifetime can be many years.

For the futuree1e2 linear collider, the backgroundmay be substantially higher. The dumps for secondarye1e2

pairs, for beamsstrahlung, and for the residual main beamall significant sources of neutrons. At this stage, it is nclear if any of these could cause problems for a CCD verdetector. As we have seen, there is an apparent discrepbetween the radiation damage data with neutrons andprotons as regards charge transfer efficiency, so the acperformance limits for a CCD detector are far from clear

What is long overdue is a comprehensive study of

E

r-

oalatak-

atdeV

asreosf.o

sIt

i-

esnd.tsd-

-n-yor-

retxcy

thal

e

radiation effects in one CCD design, comparing electromnetic, neutron, and charged hadron irradiation, with partilar attention to the operating conditions~clocking, chargeinjection interval, and temperature!, covering the region ofinterest for particle detection. It should be noted that vehigh clocking rates for the readout register (;50 MHz) areenvisaged for this environment in the future. This will prvide a significant suppression of CTI in this register duethe fact thattg will no longer be much larger thantc , so thefill factor discussed in Sec. V C can be far from unitEqually important as these systematic studies of radiaeffects is a serious evaluation of neutron background cotions likely to be encountered at the futuree1e2 linear col-lider ~the next major application area for a large scale CCvertex detector!. This work will reveal if there are any problems with the continued use of currently available CCDsthis field. Should there be difficulties with the anticipateneutron fluxes, there may be considerable room for improments in the CCD design, as will be discussed in Sec. V

VI. FUTURE PROSPECTS

A. Developments in silicon CCDs

Globally, the market for CCDs for scientific applicationwill continue to grow, and to underpin the developmentlarger, faster, thinner, more radiation-resistant devices. Geral developments in the silicon planar process for the mmarket~e.g., to feature sizes of 0.1mm and below! will filterthrough to CCD processing~first to mass market devicesand finally to the scientific market!. Those interested inbuilding CCD tracking detectors will of course profit fromall these developments. Despite the small scale of this mket it is likely that its role can continue to be much more ththat of passive ‘‘users’’ due to various factors.

First and foremost, as physicists, these users have a nral interest in the operating principles of their detectors aare always thinking of novel architectures and operatmodes, some of which have proved to be of general applbility. Second, working in a field that pushes the limitsanalog and digital electronic processing, high energy phcists are well placed to advance the readout possibilitiesthese detectors. During the past 20 years, the most impocontribution by CCD tracking detectors to other fields hbeen the speeding up of readout with,100 e2 rms noisefrom 10 kHz to 10 MHz. Work is already underway tachieve another factor of 10 speed enhancement. Thirdphysicists with an excellent knowledge of the effects ofdiation on materials, this community is very active in efforto improve the radiation hardness of a wide range of silicdetectors including CCDs.

This review article has concentrated on the device arctecture that currently has provided the most successfulticle tracking detectors. However, this picture will continuto evolve, and it may be useful to at least mention currR&D which could open up new opportunities for trackindetectors.

Regarding spatial precision, the current values;4 mm are normally adequate, particularly since the IP i

rinfos

ivshof

ios

niu

ono

r isinra

d,

ly-hinillf

a

sish

s

-siisenfreproa

a

mthd

-a

a

tun-

iseably.tlye

toe-

a

bebe

t is

nlichchpli-

ndoing

ftereseingthee

ig-


pact parameter precision is dominated by multiple scattefor most tracks. Should higher precision be requiredsome particular applications, there are good prospectachieving it. Reductions in readout noise~discussed later inthis section! and/or increased signal through greater actlayer thickness are feasible. It has long ago been establithat, as regards sensitivity, the pixel-to-pixel uniformityCCDs is generally so good that submicron (;0.1mm) pre-cision in centroid finding is achievable. Submicron precishas been demonstrated both with defocused star imagesatellite guidance systems88 and with x rays,89 and could un-doubtedly be developed for MIP tracking if required.

Regarding radiation hardness, the performance for ioing radiation is generally more than adequate, but the bdamage picture for neutron irradiation is somewhat cfused. This needs to be carefully studied with standard mern devices, plus variants that have been developed focreased hardness to proton irradiation. The most promivariants are those which reduce the signal charge stovolume, by higher levels ofn-channel doping and/or narrowor supplementary channels.90,91 Such variants have reducecharge storage capacity compared with standard devicesMIP signals are a factor of 100 below these levels~typically105e2!, so this is certainly not an issue. More speculativeone can consider whetherp-channel devices would have improved hardness with respect to neutron bulk damage. Thave been shown to have good resistance to ionizradiation,80 and their hardness to hadronic irradiation wsoon be measured, given the expanding marketsradiation-resistant CCDs.

As well as n- and p-buried-channel CCDs~plus thesurface-channel options which are probably not of usetracking detectors!, the junction orpn CCD offers anotherinteresting category of devices.92,93 For applications such ax-ray detection, they have some advantages and some dvantages~see Refs. 38 and 94 for recent reviews of eac!,and it may well be that there is a niche for these devicesome tracking detector applications.

Finally ~and most important! on the shopping list of desirable CCD upgrades is the question of improved responity of the output circuit, and hence improved signal-to-no~S/N!. This interest is driven not so much by the requiremof more precise signal measurement at current clockingquencies, but by the need to achieve similar performancmuch higher speeds. There is an ongoing impressivegram of node capacitance reduction within the present fling diffusion plus source-follower output architecture, withlimit of ;10 nF being achievable.33 Figure 25 illustrates theimprovement in high speed noise performance with sorecent realizations of this architecture. Beyond this,‘‘floating surface detector’’ of Brewer95 has been refined annamed the ‘‘double gate floating surface detector.’’96 In thislatter paper, spectacular responsivity (220mV/e2) and noiseperformance (0.5e2) are reported at a pixel clocking frequency of 3.6 MHz. In this beautiful architecture, the signcharge is stored beneath ap-type conducting channel ofsurface-channel~enhancement mode! output FET. There aresome design issues regarding implementing such a strucon CCDs with the relatively thick active layers conventio

grof

eed

nfor

z-lk-

d-n-g

ge

but

,

eyg

or

s

ad-

in

v-et

e-ato-t-

ee

l

re

ally used for particle tracking purposes, but with such noperformance, one could reduce the thickness considerand still expect adequate S/N at high clocking frequency

It should also be noted that the possibility has recenemerged of incorporating a fast-clear capability in framtransfer CCDs, without loss of MIP efficiency. The idea isbuild gated antiblooming drains, allowing clearing times blow 1 ms, in devices with reasonably thick~20–30mm! ep-itaxial layers. A MIP traversing the drain region would losesmall fraction of the signal~that in the depletion region! but~by reference to Fig. 4! could still be detected with 100%efficiency by virtue of the signal collected by diffusion fromthe undepleted epitaxial material. This option would onlyreally useful in conditions where the background couldswitched off~e.g., by a kicker magnet! during readout.

B. CCDs in new materials

Despite their many advantages for particle tracking, iclear that silicon CCDs have certain limitations~speed,radiation-hardness! which rule them out in some areas. Isome cases~see Sec. VI C! conditions may dictate radicanon-CCD solutions, but there may be situations in whcurrently available CCDs are ruled out, but in which muhigher speed, more radiation-resistant CCDs would be apcable.

In these cases, GaAs CCDs with GHz rate capability aexcellent radiation hardness can be considered. The ongprogress with band-structure engineering97 could in principlelead to some very favorable device characteristics. Asome remarkable early progress, reviewed in Ref. 98, thdevelopments seem to have lost momentum. Even if imagdevices with the required performance become available,application to particle tracking will not come easily. Threquirements of large area coverage, minimal thickness~inradiation lengths!, and mechanical robustness all raise snificant barriers to the use of this material.

FIG. 25. Noise spectra~LH axis! of ~a! surface-channel and~b! buried-channel devices as used in the SLD experiment.~c! The performance of acurrently available state-of-the-art design.~d!–~f! The corresponding CDSnoise performance~RH axis!.

re-rouxe

dcatemetefi-ub

nga

osyg-e,tooarli

icar

nsmthig

rommd

aridb

eag

wli-

,

od

tt

ul

nateisy

ion-

sed

inein

r.eir

-deeningeells

gain-

no-

f atheareing

-lleled

de-

lnd

tionc-

ch

tec-ighex-ng-IC

S


C. Active pixel sensors

In particle tracking applications where silicon CCDs aclearly ruled out~for example, at the LHC where beamcrossing rates and hadronic radiation levels would be phibitive!, the needs for pixel-based tracking detectors mbe met by a radically different approach; the active pisensor~APS!.

APS devices have for many years been suggested toplace CCDs as the preferred technology for optiimaging.99 While it is clear that every technology has a finilife expectancy, the dominant technology at any given tiis very difficult to displace. This situation was nowhere bter illustrated than in the history of CCD imagers. Since thinvention in 1970, they consistently failed to find a signicant market in video cameras for 15 years; the vacuum tdevices were continually refined and remained aheadCCDs throughout this period. While there are very excitidevelopments taking place with MOS APS devices, andready they are the technology of choice for certain low c~and low image quality! applications, they have a long wato go to catch up with modern CCDs for high quality imaing purposes. This may well happen over the next decadwhich case APS particle tracking devices will be ableprofit from that large market, as silicon CCDs currently dBut in the meantime, the particle tracking APS devicesbeing developed specifically for this very challenging appcation area.

The key characteristic of these devices which will permthem to function in areas where CCDs are prohibited is locharge sensing circuitry in every pixel. This naturally pemits fast gating of the signals; and the freedom from traporting the signal charge through large distances in the seconductor eliminates the overriding radiation softness ofCCD technology. There are two main classes of APS desmonolithic and hybrid. In the monolithic design~and this isthe class pushed for imaging applications! the detector vol-ume is integrated on the same IC with the readout electics, while in the hybrid approach, a detector chip is bubonded to a readout IC. The latter architecture is in sorespects simpler to engineer, and all the pioneering APStectors currently in use or envisaged for particle trackingbased on the hybrid technology. APS detectors fall outsthe scope of this review, but should be mentioned brieflyway of explanation of the complementary application arbetween these devices and CCDs as pixel-based trackintectors, now and in the medium-term future.

Numerous monolithic APS structures have been shoto have capability for optical imaging, including the ampfied MOS imager ~AMI !,100 charge modulation device~CMD!,101 bulk charge modulated device~BCMD!,102 basestored image sensor~BASIS!,103 static induction transistor~SIT!104 and a variety of complementary MOS~CMOS! sen-sors, e.g., in Refs. 99 and 105. Typically, the unit cellthese devices consists of a photodetection volume couplea few transistors~for readout, selection, and reset!. All thesedevices have the advantage with respect to CCDs thatsignal charge does not need to be transported throughsilicon, but the disadvantage that some or all of the caref

-stl

is-l

e-ir

eof

l-t

in

.e-

tl

--i-en,

n-pee-eeysde-

n

fto

hehely

crafted circuitry built once on the periphery of a CCD~withessentially no space constraints! now occupies real estate ievery pixel and is repeated maybe a million times. To dCMOS imagers have established a role as relatively nodevices in low cost imaging applications, where integratwith signal processing circuitry is more important than image quality~e.g., in robotic control systems!. However, sincethe minimum photolithographic feature size has decreasince the invention of the CCD from;10 to ;0.1mm, thepossibility for greatly increased integration of logic witheach pixel is rapidly advancing. It may well be that thboundary line between these classes of imager will movefavor of APS devices in the future, but this is not yet clea

Much of the problem with APS devices relates to thinferior noise performance. One idea for dramatically improving this is by integrating an avalanche photodio~APD! into a pixel imager. The drawback of APDs has bepoor gain stability, but there are procedures for introducnegative feedback106 which could solve this problem. Therremain serious issues of crosstalk between neighboring cdue to light generation in the avalanche process; here athere are possible solutions.107 This is an area with vast potential, but it could be a slow development process.

There have also been pioneering developments of molithic APS devices for particle tracking/x-ray detection.108

However, the APS systems currently used109,110 and thoseunder development111,112for particle tracking follow a ratherdifferent approach. Being hybrid devices, they consist osimple array ofpin diodes as the detector elements, usingsame technology as silicon microstrip detectors. To thisbonded readout chips with relatively complex processlogic in each pixel, and consequently large area pixels~cur-rently up to 105 mm2, aiming for ;15 000mm2 in the fu-ture, compared with 400mm2 for CCDs!. These devices, designed for very high rate operation, use a column paraarchitecture. A hit pixel sends its row address to a clockshift register at the edge of the active area~one register percolumn!, in which this information is shifted, to be latcheand transmitted off-chip in the event of a level-1 trigger bing received at a standard delay (;1 ms) relative to the trig-gering event. Major challenges~compared with the opticaAPS devices! are presented by the small MIP signals ahigh rate operation~25 ns bunch crossing interval!. As aresult, present designs have relatively high power dissipa~;1 W/cm2, approximately 100 times that of a CCD detetor!.

The combination of relatively large area pixels, mugreater thickness, and high power dissipation~which neces-sitate additional material for cooling! makes APS devicesrelatively unattractive, compared to CCDs, as vertex detors. They should and will be used in areas where the hcrossing rate and/or radiation background completelyclude CCD vertex detectors, such as at the LHC. In the loterm future, it is certain that ongoing developments intechnology ~particularly in photolithographic feature size!will lead to major improvements in the performance of APvertex detectors~hybrid or monolithic! even in these harshexperimental conditions.

beareviowetoend

eth

rgedig

th

ole

ndapa-et

atownin

op-ectorhe

xre-ave

eas;

ben atthisam-ri-

ehit

thte


D. Discussion

The use of CCDs in fixed target experiments canexpected to continue in miscellaneous application arwhere high precision, and/or high granularity are requiover a modest area. Tricks such as the division of the deinto an active area and a parking area will continue to allclean operation in apparently hostile conditions. Howevwith the high energy frontier having passed irrevocably inthe arena of colliding beam machines, the scope for nfixed target experiments is likely to continue to diminish, awith it support for advanced detector developments.

In the collider environment, APS devices will providthe only possible route to pixel-based vertex detectors inhostile environment of hadron colliders, but the high enee1e2 linear colliders will continue to provide a natural nichfor CCD-based detectors. Both for precision Standard Mophysics and for discovery physics at such a machine, hquality vertex detection will be essential, and probably

FIG. 26. Conceptual two CCD ladder design for a vertex detector atfuture International Linear Collider. The beryllium omega-beam substrano thicker than that for VXD3, but it is over 50 times more rigid.

es

dce

r,

w

ey

elh

e

key to the discovery of new heavy particles~SUSY or oth-erwise! just as vertex detectors at the CDF played a vital rin the discovery of the top quark.113 For the future LinearCollider ~which could be running on a timescale of arou2005! an ambitious R&D program is expected to lead toCCD-based vertex detector with spectacular physics cability.114 The key to achieving this physics potential is to greally close with the inner layer~radius ;10 mm! and toreduce~to ;0.12% X0! the layer thickness even below thachieved at the SLD. The conceptual ladder design is shin Fig. 26, and an isometric view of the overall detectorFig. 27.

The first of these aims depends on ongoing close coeration between the accelerator designers and the detphysicists. At the present time, fully detailed studies of tflux of background radiation at this machine115 support thepossibility of achieving this ambitious goal for the vertedetector radius; the physics payoff is impressive. Theduced layer thickness relies on CCD developments that hmostly been made in recent years for other application arsee Ref. 114 for a full discussion.

The major background seen by this detector wille1e2 pairs produced by the beam–beam interaction. Evethe smallest radius, the radiation damage effects frombackground are modest. Therefore as regards radiation dage this machine will be quite benign, provided that the vaous sources of neutron background~which emanate from theradiation dump areas! can be controlled. The main challengin moving so close to the IP is sorting out the background

eis

mall-angle
FIG. 27. Cutaway isometric drawing of the suggested five-barrel vertex detector. All associated electronics are external to the cryostat, in the sregion below the limit of tracking.

bya

fotg

sea

gh

v-

uipripvee

nurausthy

sm

r-o

mni

m

riedwn

t

ntdeepen-best

o.

asest

re-vy


rate in the event reconstruction. This will be minimizedrunning the readout electronics faster by a factor of 5 thhas been achieved to date~pixel clocking rate of 50 MHz!.As discussed in Sec. VI A, there are several optionsachieving this goal, although it will involve a significanR&D effort over the next few years. The most promisinapproach is probably to generate the balanced two-phadrive pulses locally to the CCD, possibly with a sinusoidwave form to minimize higher harmonics, and to use a hisensitivity ~three-stage! output circuit. There are ideas~stillto be proven! for achieving the I-to-R transfer while presering the sinusoidal R clocking.

Figure 28 illustrates the area coverage of varioleading-edge tracking detectors in the silicon microstrCCD, and APS technologies as a function of time. Microstdetectors remain in the lead, with APS detectors set to otake CCDs on the LHC timescale. Seen in terms of numbof channels~Fig. 29! the picture is different. CCDs are far ithe lead and are likely to remain so in the foreseeable futAs regards vertex detectors, one valid rule is ‘‘small is betiful,’’ at least as regards the inner layer radius. This is illutrated in Fig. 30, which shows that ongoing pressure onlayer thickness and beam-pipe radius have paid off in phics reach in thee1e2 linear collider environment, and thitrend is set to continue in the next generation of thesechines.

In short, CCD tracking detectors with;109 pixels and;0.1% X0 /layer will provide state-of-the-art heavy flavoidentification well into the next millennium. Only in environments where backgrounds are excessive will they need tdisplaced by less precise alternative technologies.

ACKNOWLEDGMENTS

In this review article, an attempt has been made to sumarize a large body of work undertaken by physicists aengineers from a variety of disciplines within the academcommunity and the semiconductor industry. As regards

FIG. 28. Active area of the silicon vertex/tracking detectors as functiontime. Microstrip detectors retain the capability of largest area coverage

n

r

Rl-

s,

r-rs

e.-

-es-

a-

be

-dcy

own work with CCD-based detectors, this has been carout over the past 20 years with many colleagues dralargely from the ACCMOR collaboration~CERN!, the SLDcollaboration~SLAC!, and the CCD group led by David Burat the General Electric Company~GEC! ~UK!. Too numer-ous to mention by name, their comradeship and brilliaideas have made this work an adventure and a source ofsatisfaction. I thank them sincerely, hope they have alljoyed it as much as I have, and can assure them that thehas yet to come!

NOMENCLATURE

ACCMOR: Collaboration working at CERN in the’70s and ’80s

AMI: amplified MOS imagerAPD: avalanche photodiode

fFIG. 29. Number of channels in the silicon vertex/tracking detectorsfunction of time. CCD-based pixel detectors retain the capability of fingranularity, but APS detectors may come close in the long-term future.

FIG. 30. Multiple scattering term in the formula for impact parameter pcision as function of time. For efficient topological reconstruction of heaflavors including charm, only the very best performance~as measured bythis parameter! is adequate.

r-

r

li-

ar

D

ut-

n

er

lk


APS: active pixel sensorb, b: beauty/bottom quark, antiquarkB: hadron containingb quarkBASIS: base stored image sensorBC: buried channelBCMD: bulk charge modulated deviceCCD: charge-coupled devicec, c: charm quark, antiquarkc.m.: center-of-massCDF: Collide Detector Facility at FermilabCDS: correlated double samplingCMOS: complementary metal–oxide–

semiconductorCMD: charge modulation deviceCTE: charge transfer efficiencyCTI: charge transfer inefficiencyENC: equivalent noise chargeERF: extended row filterFET: field effect transistorHDTV: high definition televisionHEP: high energy physicsI: imaging region register, gates, etc. or inte

stitial silicon atomIC: integrated circuitIERF: improved extended row filterILC: Internationale1e2 Linear Collider~future!IP: interaction point~i.e., primary vertex!IR: interaction regionIf: imaging register gatesK: temperature~kelvin!K,L,M ,...: atomic electron shellsLEP: Large Electron Positron collider at CERNLHC: Large Hadron Collider at CERNMIP: minimum-ionizing particleMNOS: metal–nitride–oxide–semiconductorMOS: metal–oxide–semiconductorMOSFET: metal–oxide–silicon field effect transiston: neutronNA32: CERN experiment in the early ’80sNIEL: nonionizing energy lossn2, n, n1: lightly, moderately, heavily dopedn-type

materialp: protonPV: primary vertex~i.e., interaction point!p2, p, p1: lightly, moderately, heavily dopedp-type

materialQC: quality controlR: readout register, gates, etc.Rf: readout register gatesSC: surface channelSi-E: phosphorus-vacancy complex in bulk si

conSIT: static induction transistorSLC: Stanford Linear ColliderSLD: SLAC Large DetectorSRH: Shockley–Read–Hall~theory of genera-

tion recombination!SUSY: supersymmetry theory of elementary p

ticles
-
SV: secondary vertex~e.g.,B decay point!T1, T2, T3: on-CCD transistorsTV: tertiary vertex~e.g., charm decay following

B decay!t, t: top quark, antiquarkV: vacancy in silicon crystalVXD2, VXD3: initial and upgrade vertex detectors of SLZ0: neutral vector boson1/f : low frequency transistor noisec: speed of lightCd : detector capacitance~node-substrate!Cg : gate-source capacitance of first stage o

put transistorCn : overall detector node capacitance (Cd

1Cg)e1, e2: positron, electrone2: charge on electronE: energy level within silicon crystalEf : Fermi energy levelEc : energy level of conduction band edgeEtr : energy level of trapEv: energy level of valence band edgeF: Fano factorf c : pixel clocking frequencyF j : fill factor for trapping under CCD phasej

( j 51...NF)H: horizontal dimensionI D : drain currentk: Boltzmann constantNc : effective density of states in conductio

bandNe : signal size~number of electrons!NF : number of phases of CCD register p

pixelNs : signal charge densityNtr : bulk trap densityP: particle momentumT: absolute temperatureV: vertical dimensionVDD : drain–substrate voltageVDS: drain–source voltagevn : electron thermal velocityVFB : flat-band voltageVRD : reset voltageDVFB : change in flat-band voltageXn : entropy factor for bulk trappingW: mean electron-hole pair creation energyWF : channel width~of output FET!X0 : radiation lengthZ: atomic numberb: particle velocity/velocity of light or radio-

active ~electron emitting! sourceg: total particle energy/masssn : cross section for electron capture by a bu

traps rf : impact parameter resolution inrf plane

~normal to beam direction!s rz : impact parameter resolution inrz plane

c-

ure

si

ch

.

ir

tur

ron

ta

tr

tro

ys-

gy

ter,

hes

ci.

l.

D-

ci.

cuits

De-


~containing beam direction and track diretion!

u : particle direction~polar angle relative tobeam direction!

tc : signal trapping time constantte : trap emission time constanttg : dwell time of signal under gatet r : clock rise–fall time

1C. F. Powell, P. H. Fowler, and D. H. Perkins,The Study of ElementaryParticles by the Photographic Method~Pergamon New York, 1959!.

2C. M. G. Lattes, H. Muirhead, C. F. Powell, and G. P. Occhialini, Nat~London! 159, 694 ~1947!.

3W. S. Boyle and G. E. Smith, Bell Syst. Tech. J.49, 587 ~1970!.4G. F. Amelio, M. F. Tompsett, and G. E. Smith, Bell Syst. Tech. J.49, 593~1970!.

5G. Charpak, Proceedings of the EPS Conference on High Energy PhyPalermo, 1975, p. 1291.

6G. K. McKay, Phys. Rev.84, 829 ~1951!.7J. Kemmer, Nucl. Instrum. Methods169, 499 ~1980!.8G. Gilder,Microcosm: The Quantum Revolution in Economics and Tenology ~Touchstone, 1989!.

9C. S. J. Damerellet al., Nucl. Instrum. Methods Phys. Res.185, 33~1981!.

10R. Baileyet al., Nucl. Instrum. Methods Phys. Res.213, 201 ~1983!.11ACCMOR collaboration, Phys. Lett. B236, 495 ~1990!.12H. Bichsel and R. P. Saxon, Phys. Rev. A11, 1286~1975!.13G. Hall, Nucl. Instrum. Methods Phys. Res.220, 356 ~1984!.14H. Bichsel, Nucl. Instrum. Methods Phys. Res. A235, 174 ~1985!.15H. Bichsel, Rev. Mod. Phys.60, 663 ~1988!.16H. Bichsel, Nucl. Instrum. Methods Phys. Res. B52, 136 ~1990!.17P. Lechneret al., Nucl. Instrum. Methods Phys. Res. A377, 206 ~1996!.18G. W. Fraseret al., Nucl. Instrum. Methods Phys. Res. A350, 368~1994!.19J. R. Hauser and P. M. Dunbar, Solid-State Electron.18, 715 ~1975!.20C. H. Sequin and M. F. Tompsett,Charge Transfer Devices~Academic,

New York, 1975!.21Charge-Coupled Devices and Systems, edited by M. J. Howes and D. V

Morgan ~Wiley, New York, 1979!.22J. D. E. Beynon and D. R. Lamb,Charge-Coupled Devices and The

Applications~McGraw-Hill, New York, 1980!.23C. J. S. Damerell, Vertex Detectors: The State of the Art and Fu

Prospects, Rutherford Lab Preprint RAL-P-95-008~1995!.24G. F. Amelio, W. J. Bertram, and M. F. Tompsett, IEEE Trans. Elect

DevicesED-18, 986 ~1971!.25R. H. Waldenet al., Bell Syst. Tech. J.51, 1635~1972!.26W. H. Kent, Bell Syst. Tech. J.52, 1009~1973!.27D. F. Barbe, Proc. IEEE63, 38 ~1975!.28T. I. Kamins and G. T. Fong, IEEE Trans. Electron DevicesED-25, 154

~1978!.29N. S. Saks, IEEE Electron Device Lett.EDL-1, 131 ~1980!.30H.-Y. Tsoi et al., IEEE Trans. Electron DevicesED-32, 1525~1985!.31D. J. Burt, Nucl. Instrum. Methods Phys. Res. A305, 564 ~1991!.32IEEE Trans. Electron Devices38 ~1991! ~entire issue!.33D. J. Burt, GEC J. Res.12 ~1995!.34M. H. White et al., IEEE J. Solid-State CircuitsSC-9, 1 ~1974!.35J. R. Janesicket al., IEEE Trans. Nucl. Sci.NS-32, 409 ~1985!.36G. R. Hopkinson, Nucl. Instrum. Methods Phys. Res.216, 423 ~1983!.37G. R. Hopkinson, Nucl. Instrum. Methods Phys. Res.216, 431 ~1983!.38A. D. Holland, Nucl. Instrum. Methods Phys. Res. A337, 334 ~1996!.39L. J. M. Esser, Electron. Lett.8, 620 ~1972!.40M. Lax, J. Phys. Chem. Solids8, 66 ~1959!.41J. E. Carnes, W. F. Kosonocky, and E. G. Ramberg, IEEE J. Solid-S

Circuits SC-6, 322 ~1971!.42J. E. Carnes, W. F. Kosonocky, and E. G. Ramberg, IEEE Trans. Elec

DevicesED-19, 798 ~1972!.43A. M. Mohsen and M. F. Tompsett, IEEE Trans. Electron DevicesED-21,

701 ~1974!.44W. Shockley and W. T. Read, Phys. Rev.87, 835 ~1952!.45R. N. Hall, Phys. Rev.87, 387 ~1952!.46R. W. Brodersen, D. W. Buss, and A. F. Tasch, IEEE Trans. Elec

DevicesED-22, 40 ~1975!.47M. G. Collet, IEEE Trans. Electron DevicesED-23, 224 ~1976!.

cs,

-

e

te

on

n

48M. D. Jack and R. H. Dyck, IEEE Trans. Electron DevicesED-23, 228~1976!.

49M. Kimata et al., Jpn. J. Appl. Phys., Part 122, 975 ~1983!.50E. K. Bangartet al., IEEE Trans. Electron Devices38, 1162~1991!.51G. J. Yateset al., Proc. SPIE2549, 172 ~1995!.52J. Frenkel, Phys. Rev.54, 647 ~1938!.53G. R. Hopkinson and D. H. Lumb, J. Phys. E15, 1214~1982!.54V. Radeka, Annu. Rev. Nucl. Part. Sci.38, 217 ~1988!.55C. J. S. Damerellet al., IEEE Trans. Nucl. Sci.37, 305 ~1990!.56K. Abe et al., Nucl. Instrum. MethodsA400, 287 ~1997!.57J. E. Carnes and F. W. Kosonocky, RCA Rev.33, 327 ~1972!.58S. P. Emmonset al., Naval Research Lab Report AD-A015 365~1975!.59R. W. Brodersen and S. P. Emmons, IEEE Trans. Electron DevicesED-

23, 215 ~1976!.60M. Algranati et al., Nucl. Instrum. Methods164, 615 ~1979!.61R. W. Leach and H. Gursky, Harvard–Smithsonian Center for Astroph

ics preprint 1196~1979!.62J. Wright, Dept. of Astronomy Cambridge University~private communi-

cation, 1980!.63A. Bross, Nucl. Instrum. Methods Phys. Res.201, 391 ~1982!.64A. H. Bhuiya, J. Appl. Phys.56, 2362~1984!.65M. Bocciolini et al., Nucl. Instrum. Methods Phys. Res. A240, 36 ~1985!.66A. D. Bross and D. B. Clegg, Nucl. Instrum. Methods Phys. Res. A247,

309 ~1986!.67C. Akerlof et al., Nucl. Instrum. Methods Phys. Res. A260, 80 ~1987!.68C. J. S. Damerell,Proceedings of Physics In Collision IV, 1984~Editions

Frontieres, Gif-Sur-Yvette, France, 1985!, p. 453.69C. Bula, SLAC-R-494~1997! p. 495.70G. D. Agnewet al., Proceedings of the 26th Conference on High Ener

Physics, Dallas, 1992~World Scientific, Singapore, 1992!, Vol. 2, p.1862.

71See, for example, SLD collaboration, Stanford Linear Accelerator CenSLAC-PUB Nos. 7117, 7170, 7266, 7278~1996/7!

72J. Killiany, Top. Appl. Phys.38, 147 ~1980!.73C. P. Chang and K. P. Aubuchon, Report No. N00173-77-C-0158, Hug

Aircraft Co. ~1978!.74J. M. Killiany et al., IEEE Trans. Nucl. Sci.NS-21, 193 ~1974!.75J. M. Killiany, N. S. Saks, and W. D. Baker, IEEE Trans. Nucl. S

NS-22, 2634~1975!.76R. A. Williams and R. D. Nelson, IEEE Trans. Nucl. Sci.NS-22, 2639

~1975!.77C. P. Chang, IEEE Trans. Nucl. Sci.NS-24, 2190~1977!.78C. P. Chang, IEEE Trans. Nucl. Sci.NS-25, 1454~1978!.79J. M. Killiany, IEEE Trans. Compon., Hybrids, Manuf. Techno

CHMT-1 , 353 ~1978!.80N. S. Sakset al., IEEE Trans. Nucl. Sci.NS-26, 5074~1979!.81N. S. Saks, IEEE Trans. Nucl. Sci.NS-27, 1727~1980!.82M. S. Robbins, Ph.D. thesis, Brunel University, UK, 1992.83E. A. Burke, IEEE Trans. Nucl. Sci.NS-33, 1276~1986!.84G. P. Summerset al., IEEE Trans. Nucl. Sci.NS-33, 1282~1986!.85S. Watts, A. Holmes-Siedle, and A. Holland, ESA Report BRUCR

ESACCD-95-IR~1995!.86N. S. Saks, IEEE Trans. Nucl. Sci.NS-24, 2153~1977!.87J. R. Srour, R. A. Hartmann, and S. Othmer, IEEE Trans. Nucl. S

NS-27, 1402~1980!.88P. M. Salomon, Proc. SPIE203, 130 ~1979!.89K. H. Schmidt et al., Nucl. Instrum. Methods Phys. Res. A359, 634

~1995!.90J. Janesicket al., Proc. SPIE1447, 87 ~1991!.91A. Holland, A. Abbey, and K. McCarthy, Proc. SPIE1344, 378 ~1990!.92M. Kleefstra, Solid-State Electron.21, 1005~1978!.93L. Struder et al., Nucl. Instrum. Methods Phys. Res. A257, 594 ~1987!.94H. Soltauet al., Nucl. Instrum. Methods Phys. Res. A377, 340 ~1996!.95R. J. Brewer, IEEE Trans. Nucl. Sci.NS-27, 401 ~1980!.96Y. Matsunaga, H. Yamashita, and S. Ohsawa, IEEE J. Solid-State Cir

26, 652 ~1991!.97F. Capasso, Science235, 172 ~1987!.98R. E. Colbethet al., Proc. SPIE1071, 108 ~1989!.99E. R. Fossum, Proc. SPIE1900, 2 ~1993!.100O. Yadid-Pecht, R. Ginosar, and Y. Diamand, IEEE Trans. Electron

vices38, 1772~1991!.101M. Ogataet al., IEEE Trans. Electron Devices37, 964 ~1990!.

ce

ice nd

ys-


102J. Hynececk, IEEE Trans. Electron Devices38, 1011~1991!.103N. Tanaka, T. Ohmi, and Y. Nakamura, IEEE Trans. Electron Devi

36, 31 ~1989!.104J. Nisizawa, T. Tamamushi, and T. Ohmi, IEEE Trans. Electron Dev

26, 1970~1979!.105S. K. Mendiset al., Proc. SPIE2172, 19 ~1994!.106G. A. Sokol and V. A. E. Shubin, Bull. Lebedev Phys. Inst.7, 3 ~1991!.107H. Komobuchi and T. Ando, IEEE Trans. Electron Devices37, 1861

~1990!.108C. J. Kenneyet al., Nucl. Instrum. Methods Phys. Res. A342, 59 ~1994!.

s

s

109P. Middelkampet al., Nucl. Instrum. Methods Phys. Res. A377, 532~1996!.

110K. H. Beckset al., Nucl. Instrum. Methods Phys. Res. A386, 11 ~1997!.111ATLAS inner detector technical design reports CERN/LHCC/97-16 a

CERN/LHCC/97-17~1997!.112CMS technical proposal CERN/LHCC/94-38~1994!.113F. Abeet al., Phys. Rev. Lett.74, 2626~1995!.114C. J. S. Damerell and D. Jackson, New Directions for High Energy Ph

ics: Snowmass 96, Stanford Linear Accelerator Center~1997!, p. 442.115S. Hertzbachet al., in Ref. 114, p. 178.

review article charge-coupled devices as particle tracking detectorsjklay/rsi01549.pdf ·...

Documents