Random telegraph signals in charge coupled devices · Random telegraph signals in charge coupled...

ARTICLE IN PRESS

Nuclear Instruments and Methods in Physics Research A ] (]]]]) ]]]–]]]

*Corresp

189-525-872

E-mail a1 Present

Engineering

3PH, UK

0168-9002/$

doi:10.1016

Random telegraph signals in charge coupled devices

D.R. Smith*,1, A.D. Holland1, I.B. Hutchinson1

Department of Physics and Astronomy, Space Research Centre, University of Leicester, University Road, Leicester, LE1 7RH, UK

Received 6 October 2003; received in revised form 2 March 2004; accepted 15 March 2004

Abstract

An investigation of fluctuating pixels resulting from proton irradiation of an E2V Technologies CCD47-20 device is

presented. The device structure, experimental set up and irradiation methodology are described, followed by a detailed

analysis of radiation induced random telegraph signals, RTS. The characteristics of the observed flickering pixels are

discussed in detail and the proposed models explaining the mechanism behind the phenomena are viewed in light of the

collected data.

r 2004 Elsevier B.V. All rights reserved.

PACS: 85.60.G; 07.89

Keywords: Random telegraph signal; CCD; Charge coupled device; Radiation damage

1. Introduction

This paper deals with a specific type of radia-tion-induced bright pixel in charge coupled devices(CCDs); those observed to have a fluctuatingcharge level. The apparent random nature of thefluctuation period has resulted in such pixels beingcalled ‘flickering pixels’ which are said to exhibit‘Random Telegraph Signal’ (RTS) behaviour.The term RTS has also been applied to ‘Flicker’,or ‘1/f’ noise [1,2]. Flicker noise however resultsfrom electron and hole emission and capture from

onding author. Tel.: +44-189-527-4000; fax: +44-

8.

ddress: [email protected] (D.R. Smith).

address: Department of Electronic and Computer

, Brunel University, Uxbridge, Middlesex, UB8

- see front matter r 2004 Elsevier B.V. All rights reserve

/j.nima.2004.03.210

interface traps adjacent to the channel region ofany CCD field effect transistors, while the RTSphenomena under study in this paper are shown toresult from bulk traps, showing well-defined timeconstants and characteristics independent of thesurface conditions of the CCD. Little study hasbeen made of the RTS phenomenon, the mainreference sources for information being the pub-lished papers by Hopkins and Hopkinson [3,4]whose investigation of RTS was initiated as aresult of other authors reporting CCD pixelsexhibiting fluctuating dark current levels [5,6].

The period of amplitude fluctuation is observedto be proportional to the temperature of the CCD.As the operational temperature of a device isreduced, the mean time constants for the low andhigh amplitudes of the fluctuation are increased.CCDs used in X-ray applications are usually

d.

ARTICLE IN PRESS

Table 1

E2V Technologies CCD47-20 characteristics

Active image area 13.3� 13.3 mm

Image section 1024� 1024 pixels

Store section 1024� 1024 pixels

Pixel size: Image section 13.0� 13.0mm

Store section 13.0� 13.0mm

Readout register 13.0� 13.0mm

Epitaxial silicon thickness 20 mm

Resistivity 20–30O cm

Spectral range 400–1100 nm

Table 2

UDT Sensors PIN-3CD photodiode characteristics

Active area 3.2 mm2

Active thickness 27mm

Capacitance 10 pF (at 10 V/1 kHz)

Leakage current 2 nA

Rise time 15 ns (50O load)

D.R. Smith et al. / Nuclear Instruments and Methods in Physics Research A ] (]]]]) ]]]–]]]2

cooled to around �100�C where the mean timeconstants are of order several days and will notcause significant concern to the collection of data.RTS pixels are however still observed at such lowtemperatures, for example 5% of backgroundevents in the MOS 2 camera of XMM-Newton,operating at �120�C, are the result of flickeringpixels in 5 of the 7 CCDs [7]. At temperaturesabove �20�C the time constants become of orderhours and start to become more of a problem indata analysis. RTS has already caused significantproblems for the optical CCD detectors of theGOMOS instrument on the ENVISAT satelliteand may also become more significant in futureX-ray missions where the trend is for warmeroperation CCDs, for example the Demonstrationof a Compact Imaging X-ray Spectrometer(D-CIXS) instrument on the SMART1 missionto observe the Moon and its subsequent develop-ment for the Bepi Colombo mission to Mercury.

The mechanism behind RTS is still not wellunderstood and the work described in this paperwas initiated to improve on the current proposedmodels of RTS. This paper first describes theCCD47-20 device used for the study and theproton irradiations carried out before describingthe techniques used to characterise the resultingpixels exhibiting RTS behaviour.

2. The E2V technologies CCD47-20

The E2V Technologies CCD47-20 is a frontilluminated frame transfer device that can beoperated in inverted mode to suppress darkcurrent. The image and store sections of theCCD each contain 1024 pixels� 1024 pixels of13 mm2. The device characteristics are summarisedin Table 1.

3. Irradiation methodology

Irradiation of the CCD47-20, device number9211-5-3, was carried out using the BirminghamUniversity cyclotron facility. Prior to irradiationof the devices, the uniformity of the proton beamover the target region was examined by using a

photodiode in pulse counting mode. The spectrumanalyser used was a Nucleus Inc. PCA-II card andsoftware, the photodiode used was a UDT Sensorsdiode, part number PIN-3CD. The photodiodecharacteristics are summarised in Table 2. Thephotodiode was mounted on a support armattached to the inside face of the cryostat frontplate, which allowed the diode to be positioned ona locus passing through the centre of the beamline. The flux per cm�2 reaching the photodiode in1 min was measured several times both in thecentre of the beam and at a position 5 mm awayfrom the centre, covering an area representative ofthe CCD target area to be irradiated. Thevariation in beam uniformity across the targetarea was measured to be 715%. The mean energyof the proton beam was 6.5 MeV for the irradia-tions carried out. An example of a typicalspectrum is shown in Fig. 1.

The primary ionisation peak is clearly discern-ible from the noise peak allowing the accuratedetermination of the number of protons beingcounted in the active region of the photodiode.Protons that interact with the diode twice duringthe same shaping period are also seen in therecorded spectrum and are represented by a

ARTICLE IN PRESS

0

100000

200000

300000

400000

0 100 200 300 400 500Channel Number

Co

un

tsPrimary peak

Double hits

Analysisregions

Noise peak

Fig. 1. A PIN-3CD photodiode pulse height spectrum.

D.R. Smith et al. / Nuclear Instruments and Methods in Physics Research A ] (]]]]) ]]]–]]] 3

secondary ‘pile-up’ peak. The number of counts inthe secondary peak was doubled and added to thenumber in the primary peak to estimate the totalproton fluence. The analysis regions for each peakare indicated in Fig. 1.

During each irradiation the photodiode waspositioned B2 cm in front of the shielded sectionof the target CCD and used to accurately monitorthe proton fluence reaching the CCD in real time.The system live time, Tlive, and the actual elapsedtime, Telapsed, for each irradiation were bothmonitored and used to account for the dead-timein the system. Using the Non-Ionising Energy Loss(NIEL) function, the final 10 MeV equivalentproton fluence received in each irradiation,F10MeV, was calculated by

F10 MeV ¼N

A

� �Telapsed

Tlive

� �NIELð10 MeVÞNIELðbeamÞ

� �: ð1Þ

where N is the number of counts in the primaryproton peak plus two times the number ofcounts in the secondary peak, A is the area ofthe diode and NIEL(beam) is the energy of thebeam in MeV. The error associated with thedosimetry of each irradiation was taken to beB20%. The use of NIEL scaling for silicon deviceshas been described in detail by Burke [8] and VanLint [9]. The specific use of NIEL scaling in CCDs

is discussed by Srour et al. [10] and Ambrosiet al. [11].

During each irradiation the beam line and targetchamber were under vacuum to prevent loss ofprotons to ionisation with air and all CCD pinswere grounded to avoid potential static damage. A6.5 MeV proton beam was used to give a rawproton fluence of 8.8� 107 protons cm�2 to onethird of the CCD, equal to a 10 MeV equivalentproton fluence of 1.5� 108 protons cm�2. The restof the CCD was covered with an aluminium shieldto prevent the protons from damaging that part ofthe device. It should be noted that the store sectionof the CCD47-20 has its own aluminium shield,although this is not thick enough, B1 mm, to stopthe protons passing through it.

A second irradiation was carried out with twothirds of the CCD shielded with aluminium whilethe rest of the device was irradiated through100 mm of copper foil. The copper had the effect ofreducing the mean energy of the proton beam to2.0 MeV. The same flux of protons was given tothe CCD as for the 6.5 MeV irradiation. The rawproton fluence given to the CCD in this case was7.8� 107 protons cm�2, equal to a 10 MeV equiva-lent proton fluence of 3.6� 108 protons cm�2. Forthe second irradiation the photodiode was alsocovered with 100 mm of copper in order to measurethe same proton flux and mean energy as that

ARTICLE IN PRESS

Fig. 2. A schematic showing the proton irradiated areas of device number 9211-5-3 and the associated 10 MeV equivalent proton dose

received.

Fig. 3. The average pixel amplitude of 1000 rows in the image

section of the proton-irradiated CCD.


reaching the CCD. The central part of the CCDremained unirradiated as a control area. Theshielding regime and 10 MeV equivalent protondose received by each area of the device are shownin Fig. 2. The exposure time for each irradiationwas B80 s.

The two 10 MeV equivalent fluences given to theCCD are representative of mission fluences ex-pected to be received by typical Earth orbitingspacecraft. For example a 10 MeV equivalentfluence of 7.3� 108 protons cm�2 is expected tobe received by the CCD in the X-ray Telescope(XRT) of the Swift gamma-ray burst missionduring its initial three years of operation [12]. Forthe MOS CCDs in the two European PhotonImaging Cameras of XMM-Newton, the 10 MeVequivalent proton fluence expected after 10 yearsof operation in orbit is 5� 108 protons cm�2 [13].

4. Dark current changes

The mean dark current level increased in theareas of the CCD that were irradiated. Fig. 3shows the average pixel amplitude in each column

of the CCD which scales with the mean energy andthe fluence of the protons received by eachirradiated area. The dark current increase in thearea of the CCD irradiated through 100 mm ofcopper foil is far higher than the increase resultingfrom the direct irradiation due to the very largenumber of bright pixels generated by the 2.0 MeV

ARTICLE IN PRESS


protons increasing the average pixel amplitude inthese columns. The ‘curve’ of the line in the2.0 MeV irradiated region is due to the copper foilbeing positioned at a slightly off vertical angleacross the device. At a temperature of 24�C, theunirradiated region in the centre of the device hada dark current level of B2600 electrons, while thedark current levels for the 6.5 and 2.0 MeV meanproton energy irradiations were B3400 electronsand B14,000 electrons per pixel, respectively.

5. Characterisation of RTS properties

A dedicated CCD sequencer program was usedto obtain the data required for analysis of theradiation induced RTS pixels described below. Thesoftware allowed sampling of a given pixel’samplitude every 0.25 s. The short time betweensamples was to ensure that high-frequency transi-tions were adequately sampled. Previous studieshave shown that the time constants of RTS pixelsdecrease with increasing temperature [3]. The samenumber of switches from high to low charge statecan be observed in B1 h at 45�C as in 12 h at�10�C. The data collected from device 9211-5-3were therefore taken at temperatures in the range45–55�C to reduce the amount of time required toobtain large data sets. Fig. 4 shows an example ofa recorded CCD image taken at 55�C using theRTS analysis sequencer program.

Fig. 4. An image taken at 55�C using a sequencer program that

only reads out pixels in a selected row of the CCD. Each row in

the image is recorded at 0.25 s intervals revealing the change in

amplitude over time of RTS pixels.

The recorded data were input into RTS analysissoftware that characterised the number of distinctamplitude levels associated with a given pixel. Thecharacterisation was carried out by first convol-ving the measured amplitude spectrum with amatched Gaussian filter (the matched Gaussianwas generated by fitting a Gaussian function to theaverage amplitude spectrum of 10 ‘stable’ pixels).The amplitude levels were then identified bycounting the number of positive to negativetransitions in the gradient of the convolution(a detection threshold of 5s above the baselinenoise was used giving a false amplitude leveldetection probability of (1/(2� 106)). Pixels with asingle amplitude level were classified as ‘stable’while those with two or more amplitude levelswere classified as RTS pixels. For each selectedpixel, the software recorded the number ofamplitude levels, the mean ADC value of eachamplitude level and the period between amplitudeswitches.

The amplitude detection efficiency can becalculated by fitting the measured RTS transitionamplitude distribution at a given temperature witha Gaussian function and observing the point atwhich the detection threshold intersected the fit.The four panels of Fig. 5 illustrate the output fromthe RTS analysis software; the raw pixel amplitudevariation over time, the measured amplitudespectrum, the convolved amplitude spectrum andthe gradient of the convolution. The raw data usedfor the figure is that of a 2-level RTS pixel, the twodistinct amplitude levels being clearly resolved bythe analysis software.

5.1. General properties

From a sample of 1800 pixels in the 3.6 �108 protons cm�2 irradiated region of the CCD,the number of RTS pixels and the number ofdistinct amplitude levels in each RTS pixel wererecorded. The number of amplitude levels wasinvestigated to determine if the occurrence of RTSpixels with more than two amplitude levels scaledwith the statistical probability of more than oneRTS defect occurring in a given pixel. If themeasured number of 3 or 4 level RTS pixels was

ARTICLE IN PRESS

100

200

300

400

500

600

700

800

0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300

Time (seconds)

Pix

el A

mp

litu

de

(AD

C V

alu

e)

Raw Data

1

10

100

1000

100 300 500 700

Pixel Amplitude (ADC Value)

Co

un

ts

Amplitude Spectrum

1

100

10000

1000000

100 300 500 700

Pixel Amplitude (ADC Value)

Co

un

ts

Convolved Spectrum

-3000

-1500

0

1500

3000

100 300 500 700Pixel Amplitude (ADC Value)

Dif

fere

nti

al C

ou

nts

Gradient of Convolution

Fig. 5. Output from the RTS analysis software showing the sensitivity of the software to picking out the number of distinct amplitude

levels present in the raw pixel amplitude data. The raw data illustrated was recorded with the CCD operating at 55�C.

y = 11.6x-4.4

0

0.1

0.2

0.3

0.4

0.5

0.6

1 4 6

Number of Amplitude Levels

Fra

ctio

n o

f T

ota

l Pix

els

Statistically expected fraction of 3 or 4 level pixels

Power law fit to data

2 3 5

Fig. 6. The fraction of RTS pixels exhibiting 2, 3, 4 and 5

distinct amplitude levels.


the same as the number expected by the statisticalprobability, all fluctuating pixels should be theresult of one or more 2-level transitions within agiven pixel. If the observed number of RTS pixelswith more than two amplitude levels was signifi-cantly greater than the expected number it wouldindicate that RTS pixels with more than two levelsare the result of additional processes.

A histogram of amplitudes was obtained fromraw data collected at 50�C for each RTS pixel andthe RTS analysis software used for the detection ofdistinct amplitude levels. Fig. 6 shows the mea-sured distribution of RTS pixels with 2, 3, 4 and 5amplitude levels and the statistically expectedfraction that should be present if the explanationfor 3- or 4-level RTS pixels is simply that two2-level RTS phenomena are located in the samepixel. The observed number of pixels showingmore than two distinct amplitude levels is sig-nificantly below the expected value due to multi-level RTS exhibiting amplitude levels that liewithin the detection threshold of each other.

Detailed modelling of the distribution of ampli-tude levels in multi-level RTS pixels is currentlybeing investigated and will be the subject of a

ARTICLE IN PRESS

0.00

0.20

0.40

0.60

0.80

1.00

1 4Event Size (Pixels)

Fra

ctio

n o

f to

tal p

ixel

s

Horizontal Split

Vertical Split

2 3

Fig. 7. The distribution of event sizes from a total of 921 RTS

pixels.


future publication, which in particular, will con-sider how the choice of threshold detection leveleffects the observed amplitude distribution. Noclear evidence that additional processes are re-sponsible for the higher number of amplitudelevels observed was found, the most likelyexplanation for multi-level RTS being a numberof 2-level RTS phenomena residing in a singlepixel.

To deduce if the RTS phenomenon could belinked to the high field regions within a CCD pixel,the ‘event’ size of the bright pixels containing RTSwas investigated. If a high proportion of RTSpixels were found to be located in single pixelevents it would indicate that the defect causingRTS is located in the inter-electrode or the channelstop high field regions, where it is very difficult forcharge to diffuse into adjacent pixels. If a largeproportion of horizontal or vertical split eventswere observed, the defect causing RTS might beconcentrated in the lower field regions of a pixel,where the charge generated could diffuse intoadjacent pixels before being collected into thecharge storage region.

From a total of 921 RTS pixels observed in the1.5� 108 protons cm�2 irradiated region of theCCD, only a very small number of the RTS pixelswere located adjacent to another bright pixel.Fig. 7 shows the percentage of the observed RTSpixels having different ‘event’ sizes. The number of2 pixel events is consistent with the probability ofobtaining two single events in adjacent pixelproviding evidence for the location of RTS in theinter-electrode or channel stop high field regions.

Further evidence for the location of RTS in thehigh field regions of a pixel can be obtained byconsidering the physical structure of a pixel andthe charge storage and transport volumes. Fig. 8 isa diagram of the CCD47-20 3 phase pixelstructure, indicating the charge storage region,the inter-electrode and channel stop high fieldregions and the associated movement of charge forthe potential situation shown.

Assuming the depth and width of the chargestorage and high field regions in a pixel are thesame, the ratio of the volume of the charge storageregion in a CCD47-20 pixel during a single pixeltransfer, Vcs, to the volume of the high field region

within the pixel, Vhf, can be approximated by

Vcs

Vhf¼

Dtransfer

Dhfð2Þ

where Dtransfer is the distance travelled during asingle pixel transfer and Dhf is the distancetravelled through a high field region during thetransfer. Substituting in suitable values of 13 and0.2 mm, respectively, Vcs/Vhf is found to be 65. Ifthis value is comparable to Ntraps/NRTS, whereNtraps is the number of traps in a given sample ofpixels and NRTS is the number of pixels in thesample showing RTS characteristics, this providesevidence that the RTS phenomenon may be linkedwith traps located in the high field regions of aCCD pixel.

An earlier study by Smith et al. [14] found thatafter irradiation of a CCD47-20 device with3� 108 protons cm�2, from an area containing18,400 pixels B1.2% had a charge level greaterthan 5s of the mean dark current level. Of thisfraction, B45% will exhibit RTS characteristics,i.e. B100 of the sample pixels. An irradiationfluence of 1� 109 protons cm�2 results in a CTI ofB2� 10�4 electrons per pixel [15]. For a Mn KaX-ray this CTI value results in the loss of a singleelectron per transfer through 3 pixels. In a sampleof 18,400 pixels there will therefore be B6000traps. In this instance Ntraps/NRTS is found to be60, comparable with the Vcs/Vhf value 65 support-ing the high field location of the RTS phenomena.

ARTICLE IN PRESS

Column isolation

High fieldregions (grey)

Outline ofsingle

13 µm2 pixel

4.5 µm

13.0 µm

Direction ofcharge

movement

Chargecollection

area

13.0 µm φ2 at 0V

φ2 at 0V

φ3 at 0V

φ1 at 0V

φ1 at 0V

Fig. 8. The structure of a CCD47-20 pixel indicating the inter-electrode and channel stop high field regions and the movement of

charge into the charge storage region of each pixel.

Table 3

The mean RTS transition amplitude at different temperatures

Temperature (�C) Mean RTS transition

amplitude (nA cm�2)

45 0.60

50 0.75

55 1.05


5.2. Amplitude properties

The mean RTS transition amplitude over a 1 hperiod was recorded at 5�C intervals from 45�Cto 55�C for 85 RTS pixels. The transitionamplitude is the change in dark current level fromthe bright pixel pedestal level to the high RTSamplitude. In 1 h B80 RTS transitions areobserved at 45�C, the number increasing toB150 At 55�C. As the temperature is increasedthe mean RTS transition amplitude is also foundto increase, with the distribution of amplitudesbecoming more widely spread. The mean RTSamplitude at each measured temperature is shownin Table 3. Fig. 9 shows an example of anamplitude versus time plot for an RTS pixel atthe three different temperatures, highlighting thetransition amplitude change. Fig. 9 also showshow the bright pedestal amplitude increases with

increasing temperature as a result of the extra darkcurrent in the pixel.

A previous investigation into RTS pixels in aTH7895 M device, with pixels of 19 mm2, aimed tosee if there was a correlation between the RTStransition amplitude and the dark current pedestalamplitude [16]. No correlation was observed forthe data collected at 10�C. Fig. 10 shows therelationship between RTS transition amplitudeand dark current pedestal amplitude for 24 RTS

ARTICLE IN PRESS

1.50

2.00

2.50

3.00

3.50

4.00

4.50

5.00

5.50

0 25 50 75 100 125 150

Time (seconds)

Pix

el A

mp

litu

de

(nA

.cm

-2)

55oC

50oC

45oC

Fig. 9. The variation in transition amplitude and bright

pedestal amplitude with temperature of an RTS pixel.

y= 0.29x0.94

0.01

0.10

1.00

10.00

0.01 0.10 1.00 10.0Dark Current Pedestal Amplitude (nA.cm-2)

RT

S T

ran

siti

on

Am

plit

ud

e (n

A.c

m-2

)

55 oC

45 oC

10 oC (Bond 1996)

Fig. 10. RTS transition amplitude variation with dark current

pedestal amplitude.

-1.5

-1.0

-0.5

0.0

0.5

1.0

2.18E+20 2.21E+20 2.24E+20 2.27E+20 2.30E+201/kT

Ln

RT

S A

mp

litu

de

Eact = 0.53 ± 0.15 eV

Arrhenius Fit

Fig. 11. RTS transition amplitude activation energy.


pixels of device 9211-5-3 at 45�C and 55�C andalso displays the data recorded by Bond at 10�C[16]. As the temperature increases the spread in theobserved transition amplitudes becomes larger forhigher pedestal amplitudes. A power law trendlinecan be fitted to all three data sets indicating apower law relationship between the temperature ofthe device and the spread in the correlationbetween the transition amplitude and the pedestalamplitude.

The mean transition amplitude of 10 RTS pixelswas evaluated at 2.5�C intervals from 45�C to55�C. Plotting the natural logarithm of thetransition amplitude as a function of 1/kT, theRTS transition amplitude was found to follow anArrhenius relationship with a mean activation

energy of 0.5370.13 eV. The Arrhenius relation-ship is given by

1

Et¼ Ae�EA=kT ð3Þ

where Et is the RTS transition amplitude at a giventemperature, A is a non-thermal constant, EA isthe activation energy in eV, k is the Boltzmannconstant and T is the temperature in Kelvin.

Fig. 11 shows the variation in the mean RTStransition amplitude with temperature for each ofthe 10 RTS pixels evaluated. The Arrhenius fit toone of the data sets is shown by the dotted line inthe figure. The errors associated with each datapoint are indicated for one data set and arise fromthe temperature stability error and the noisevariation in the mean amplitude level of a givenRTS pixel. This mean activation energy value iscomparable with the activation energy of0.5770.03 eV found by Bond [16] and lies nearthe mid-band energy of 0.55 eV, indicating that theE-center or the J-center are the most likely defectsresponsible for RTS.

5.3. Period properties

The period of time spent in the high and lowamplitude states during each RTS transition wasrecorded for eight 2-level RTS pixels. A ‘switch’from one level to another was defined as anamplitude change above 5s of the mean darkcurrent pedestal level of the given pixel. Data was

ARTICLE IN PRESS

1.50

2.00

2.50

3.00

3.50

4.00

4.50

5.00

0 25 50 75 100 125 150

Time (seconds)

Pix

el A

mp

litu

de

(nA

.cm

-2)

55 oC

50 oC

45 oC

Fig. 12. The variation in high- and low-state period with

temperature of an RTS pixel.

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

2.18E+20 2.21E+20 2.24E+20 2.27E+20 2.30E+20

1/kT

Ln

Tim

e C

on

stan

t

Low State

High State 1

High State 2

Eact = 0.2 ± 0.1 eV

Eact = 0.1 ± 0.1 eV

Eact = 0.2 ± 0.1 eV

Fig. 13. RTS high- and low-state period activation energies.


recorded at 2.5�C intervals from 45�C to 55�C.Fig. 12 shows an example amplitude verses timeplot for an RTS pixel at the three differenttemperatures highlighting the high and low stateperiod changes.

As temperature is increased the distribution ofobserved high and low state periods becomesnarrower, the mean time spent in a high or lowstate becoming shorter. A w2 fitting routine wasused to determine the time constant for each stateat the five measured temperatures. In each case thebest fit to the mean time in the low state was asingle exponential. Plotting the natural logarithmof each measured low state time constant as afunction of 1/kT reveals an Arrhenius relationshipwith an activation energy of 0.270.1 eV. TheArrhenius relationship in this case is given by

1

t¼ Ae�EA=kT ð4Þ

where t is the RTS time constant at a giventemperature.

The high state data is better fitted by acombination of two exponentials, revealing onetime constant that varies with temperature in asimilar way to that of the low state and a secondtime constant that varies more markedly withtemperature. Fig. 13 shows the Arrhenius relation-ship of the two high state time constants that haveactivation energies of 0.170.1 and 0.270.1 eV andalso shows the low state data. The error associatedwith each data point is shown. Although the three

sets of data could be fitted using a straight linethrough the error bars, statistical analysis revealsthat in the case of the low state data and one of thehigh state data sets, a slope is the more statisticallysignificant fit to the data. All three measuredperiod activation energies are within the measurederror, indicating that an activation energy ofB0.270.1 eV is common to all three measuredtime constants. The RTS study by Bond [16] founda single time constant for the low state and alsoonly a single time constant for the high state. Ineach case the period activation energy was foundto be 0.970.1 eV, much larger than the observedvalue in this study.

5.4. Annealing

The possible link between the RTS phenomenonand the E-centre, suggested by the comparableRTS amplitude and E-centre activation energies,was further investigated by an annealing study.Device 9211-5-3 was subjected to an unbiasedanneal at a temperature of 120�C for a period of2 h. The characteristic anneal temperature of theE-centre is B120�C. If the mechanism behind theRTS phenomenon were linked to the E-centre, asignificant fraction of the RTS pixels observedbefore heating the device should be annealed wheninvestigated afterward.

Characterisation of 69 RTS pixels, of which 6had >2 distinct amplitude levels, took place both

ARTICLE IN PRESS


before and after the anneal. The characterisationinvolved recording the amplitude of the selectedRTS pixels at 0.25 s intervals for 5 min. All theselected RTS pixels were from the area of the CCDexposed to a fluence of 3.6� 108 protons cm�2.The data was collected at 50�C.

Table 4 summarises the state of the 69 mon-itored RTS pixels after the anneal. Of the totalsample, 28% of the pixels still showed RTScharacteristics, while 72% were completely an-nealed. This is comparable to the value of B80%obtained by Holland [17] when annealing brightpixels at 160�C for 16 h. The large fraction of RTSpixels annealed strongly supports the case for theunderlying mechanism behind the phenomenonbeing linked to the E-centre. It is also interesting tonote that of the 6 RTS pixels with more than 2distinct amplitude levels, three were annealedcompletely while only 2 of the amplitude levelsof the remaining 3 pixels were annealed. Thisobservation strongly supports the idea that multi-level RTS is the result of more than one bi-stabledefect occurring within a given pixel, a single RTSdefect being annealed and reducing the number ofobserved amplitude levels by two in each of theobserved cases. Fig. 14 shows the variation inamplitude over time for six of the monitored pixelsboth before and after annealing.

Previous work by Bond [16] observed changes inthe amplitude and period of 7 monitored RTSpixels during a stepped anneal study. The studyfound that RTS defects were gradually annealed,the time in the high-state amplitude becomingincreasingly longer until it eventually becameinfinite. In contrast, of the 50 RTS pixels thatwere annealed in the work carried out for thisstudy, 42 were characterised by amplitudes very

Table 4

Post-anneal characteristics of a sample of 69 RTS pixels

Post-anneal

classification

Number

of pixels

% of total pixels

RTS completely

annealed

50 72.5

RTS partially annealed 3 4.3

RTS still present 16 23.2

Total 69 100

close or below the pre-anneal low state amplitude.Of the remaining 8 annealed RTS pixels, only 1was annealed to an amplitude level comparable tothe pre-anneal high state amplitude, with theremaining pixels annealing to amplitude levelsbetween that of the pre-anneal high and lowamplitudes.

There were 16 pixels that still exhibited RTScharacteristics after annealing. In each case theflickering period and transition amplitude de-creased slightly, with the exception of 2 pixelswhere the RTS transition amplitude became muchlarger than it was before annealing.

6. Discussion and conclusion

The RTS pixels observed after the protonirradiation of a CCD47-20 device display verysharp amplitude transitions between distinct levelswith high time constants and well defined activa-tion energies. After a 10 MeV equivalent dose ofB3� 108 protons cm�2, approximately 45% of thebright pixels in the irradiated area show signs of adark current fluctuation, between two or moredistinct amplitude levels, above 5s of the meandark current. The spread of bright pixels andpixels exhibiting RTS characteristics is uniformthroughout the area of irradiation. Of the ob-served RTS pixels >90% were isolated eventsindicating that the mechanism behind the RTSphenomenon may physically lie inside the inter-electrode and channel stop high field regions of agiven pixel. Consideration of the physical structureand the extent of the high field regions within aCCD pixel also supports this hypothesis, with theratio of the charge storage volume to the volumeof the high field region in a pixel being comparableto the ratio of the number of traps to the numberof RTS defects observed within a pixel. Theseratios are 65 and 60, respectively.

The number of RTS pixels with >2 distinctamplitude levels is lower than the statisticallyexpected number if >2 levels is the result of two ormore 2-level RTS mechanisms residing within agiven pixel. This is due to multi-level RTSexhibiting amplitude levels that lie within thedetection threshold of each other. In this case an

ARTICLE IN PRESS

0 50 100 150 200 250 300 350 400 450 500Time (seconds)

Pix

el A

mp

litu

de

(arb

itra

ry s

cale

)

Pre-anneal Post-anneal

A, B = Complete anneal of RTS; C, D = RTS still present;

E = Complete anneal of multi-level RTS; F = Partial anneal of multi-level RTS

(A)

(B)

(C)

(D)

(E)

(F)

Fig. 14. The amplitude variation with time of RTS pixels monitored before and after annealing. The amplitude scale is arbitrary to

allow the presentation of the data, however the relative amplitude scale of each data set is the same.


amplitude detection efficiency of B0.88 wasobtained. This fact does however support the ideathat multi-level RTS is not due to a separatemechanism to that of 2-level RTS, the most likelycause being a number of 2-level RTS being presentwithin the same pixel. The observed partialannealing of multi-level RTS is also in agreementwith this hypothesis. RTS pixels exhibiting 3distinct amplitude levels and not a multiple of 2can be explained not only by the inability of thesystem to detect other amplitude levels above thenoise in the data, but also by considering the high-

and low-state amplitudes of each defect in thepixel. The observed pixel amplitude at any giventime is the superposition of the amplitude of eachdefect in the pixel at that moment. If there are twobi-stable defects within a given pixel the resultingpixel amplitude can show 4 amplitude levels or 3.The observation of 3 levels arises when thetransition amplitude of each of the bi-stable defectsis the same. A similar superposition argument canbe used to deduce that for three bi-stable defectswithin a given pixel, any number of amplitudelevels between 4 and 8 can be observed.

ARTICLE IN PRESS


The RTS transition amplitude does not show astrong correlation with the dark current pedestalamplitude, with transition amplitudes variations ofthe order of B1 nA cm�2 for a given pedestalamplitude at 45�C. As temperature is increased thedark current increases and the spread in theobserved transition amplitudes becomes largerfor higher pedestal amplitudes.

The above observed RTS properties indicatethat the likely mechanism behind RTS involvesdiscrete transitions between two states separatedby an energy barrier. A number of theoreticalexplanations of the RTS phenomenon have beenproposed and these models are described below.

6.1. Field enhancement

The defect must be field enhanced, accountingfor the large-transition amplitudes, low activationenergies and correlation with bright pixels. Thedefect responsible may be located in the inter-electrode or channel stop regions of a pixel wherethe electric field strength is larger. Electrons canclimb over a potential barrier lowered by theapplied electric field causing Poole-Frenkel fieldenhancement [18]. Another suggestion is thatcharge captured by a defect may create an electricfield around itself which then influences nearbydefects. However, the field created would not reachvery far and the resulting number of defectsinfluenced would not be enough to account forthe large transition amplitudes seen. This idea isalso statistically unlikely if there are not manydefects present in the silicon lattice.

6.2. Multiple defects

The high-transition amplitudes may be theresult of many bulk defects contributing chargeat the same time. Work has shown that around 50defects would be required to act together togenerate the amplitudes seen and so this theory isthought to be unlikely [19].

6.3. Multi-stable defect

The observed well-defined time constants sug-gest that a multi-stable defect, with two or more

states separated by an energy barrier, may beresponsible for the RTS phenomenon. The defectmust be common as it is widely seen after, and insome cases before, the proton irradiation of adevice. There are however, no known defects withthe appropriate activation energy and energystates. Normally the time constants of captureand emission of charge are thermally independent.This is not true for the observed RTS timeconstants, which show a strong temperaturedependence. The RTS switching mechanism there-fore involves a phenomenon that is independent ofsimple electron capture and emission probabilities.A proposed model is a multi-stable defect with thestable configuration dependent on the charge state:state A being stable for one charge state, and stateB being stable for another [20,21]. The configura-tion can be flipped over the potential barrier fromstate A to state B by thermal fluctuations. Theenergy level of each state, along with any fieldenhancement, will determine the level of thermallygenerated conduction band electrons and thereforethe dark current amplitude level. If one state isnearer the mid gap than the other there will be twoclear dark current levels observed after fieldenhancement. This model cannot explain themulti-level RTS pixels observed, if they are theresult of a different process to 2-level RTS, andalso does not account for the second high periodtime constant observed in the data. Fig. 15illustrates the proposed defect. In one charge state,state A has the lower energy, while in the othercharge state, state B has the lower energy.

6.4. Reorientation of the E-centre

This model was suggested by Bond [16] andinvolves the reorientation of the E-centre in astrong electric field. The correlation between RTSbehaviour and dark current spikes indicates thatthe defect responsible for the dark current spikesmay also be the cause of RTS. The E-centre is acommon bulk defect in proton irradiated siliconand is generated in numbers large enough toexplain the large fraction of pixels exhibitingRTS after irradiation. It has been shown thatthe E-centre in its neutral charge state has anextra positive charge on the P-atom, and a

ARTICLE IN PRESS

Fig. 15. Energy versus defect configuration for a defect with

two stable states, A and B.

Fig. 16. The silicon lattice containing an E-centre defect. The

vacancy can reorient itself from its normal position nearest the

phosphorous atom to a new nearest site by moving through the

lattice as shown.


corresponding extra electron orbital [22]. Thedefect has a resulting dipole moment, and thefield enhancement factor caused by the defect willdepend on its orientation within the appliedelectric field [18]. The E-centre has been observedto reorient its axis, the vacancy taking the place ofany one of the four silicon atoms closest to thephosphorous atom, moving through the siliconlattice by thermally overcoming potential barriers[22]. Fig. 16 shows the structure of the siliconlattice containing an E-centre defect and onepossible reorientation. The level of dark currentgeneration is dependent on the orientation of thedefect within the applied electric field. A move-ment of the vacancy from a small angle to a largeangle relative to the electric field vector will resultin large amplitude RTS signals and vice versa. Forthis model to be viable, RTS time constants shouldbe correlated to the kinetics of reorientation of thedefect. The measured activation energies forreorientation are 0.9370.05 eV, higher than theobserved 0.270.1 eV in the CCD47-20 study. Themodel also explains the lack of a correlationbetween RTS amplitude and the dark currentpedestal and lack of any direct evidence for fieldenhancement, since the model assumes that theamplitude is dependent on the defect orientationand not the electric field strength. Electric fieldshave however been observed to influence thereorientation kinetics of defects, which mayexplain the large variation in time constantsobserved [23].

6.5. Summary and future work

The models described above each provideexplanations for a number of observed RTScharacteristics, but not all. The reorientation ofthe E-centre provides the most detailed descriptionof a mechanism to explain the RTS phenomenon,but does not account for the two high state timeconstants observed in this study. The workpresented in this paper has shown that the mostlikely model for RTS involves the E-centre, thehigh field regions of a device and a single bi-stablemechanism. Future work will involve the use of aproton microprobe to ‘inject’ protons into specificregions of a CCD pixel allowing the directmeasurement of the location of bright and RTSdefects within the volume of a pixel and accuratecorrelation with the high field and charge storageregions [24]. Pixels exhibiting RTS are also foundin unirradiated CCDs, these RTS phenomenabeing generated during the manufacturing process.A deeper understanding of the mechanism causingRTS may lead to an improved manufacturingmethod that can prevent RTS pixels beinggenerated not only during the manufacturingprocess, but also prevent those generated by

ARTICLE IN PRESS


irradiation, resulting in devices with an improvedradiation tolerance.

Acknowledgements

The authors would like to thank Mike Smith atBirmingham University, UK, for his assistanceduring the experimental phase of this study andE2V Technologies for the CCD used in this work.

References

[1] K. Kandiah, Proc. Noise Physical Systems 1/f Noise (1985)

19.

[2] K. Kandiah, M.O. Deighton, F.B. Whiting, J. Appl. Phys.

66 (2) (1989) 937.

[3] I.H. Hopkins, G.R. Hopkinson, IEEE Trans. Nucl. Sci.

NS-40 (6) (1993) 1567.

[4] I.H. Hopkins, G.R. Hopkinson, IEEE Trans. Nucl. Sci.

NS-42 (6) (1995) 2074.

[5] J.R. Srour, R.A. Hartmann, K.S. Kitazaki, IEEE Trans.

Nucl. Sci. NS-33 (6) (1986) 1597.

[6] P.W. Marshall, C.J. Dale, E.A. Burke, G.P. Summers,

G.E. Bender, IEEE Trans. Nucl. Sci. NS-36 (6) (1989)

1831.

[7] J. Ballet, private communication, 2003.

[8] E.A. Burke, IEEE Trans. Nucl. Sci. NS-33 (6) (1986) 1276.

[9] V.A.J. Van Lint, Nucl. Instr. and Meth. A 253 (1987)

453.

[10] J.R. Srour, C.J. Marshall, P.W. Marshall, IEEE Trans.

Nucl. Sci. NS-50 (3) (2003) 653.

[11] R.M. Ambrosi, D.R. Smith, A.F. Abbey, I.B. Hutchinson,

E. Kendziorra, A. Short, A. Holland, M.J.L. Turner, A.

Wells, Nucl. Instr. and Meth. B 207 (2003) 175.

[12] R.M. Ambrosi, A.D.T. Short, A.F. Abbey, A.A. Wells,

D.R. Smith, Nucl. Instr. and Meth. A 482 (2002) 644.

[13] A. Holmes-Siedle, S. Watts, A. Holland, Final Report on

ESTEC Contract No. 8815/90/NL/LC(SC), Brunel Uni-

versity, UK, 1995.

[14] D.R. Smith, A.D. Holland, M.S. Robbins, R.M. Ambrosi,

I.B. Hutchinson, Proc. SPIE 4851 (2003) 842.

[15] A. Holland, A.G. Holmes-Siedle, B. Johlander, L. Adams,

IEEE Trans. Nucl. Sci. NS-38 (1991) 1663.

[16] I.H. Bond, Ph.D. Thesis, University College London,

1996.

[17] A.D. Holland, Ph.D. Thesis, University of Leicester, 1990.

[18] P.A. Martin, B.G. Streetman, K. Hess, J. Appl. Phys. 52

(12) (1981) 7409.

[19] J. Kirton, M.J. Uren, S. Collins, M. Shulz, A. Karmann,

K. Scheffer, Semicond. Sci. Technol. 4 (1989) 1116.

[20] A. Chantre, J. Appl. Phys. A 48 (1989) 3.

[21] G.D. Watkins, Semicond. Sci. Technol. 6 (1991) 111.

[22] G.D. Watkins, J.W. Corbett, Phys. Rev. 134 (5A) (1964)

1359.

[23] L.C. Kimerling, Inst. Phys. Conf. Ser. 46 (1979) 56.

[24] A. Simon, C. Jeynes, R.P. Webb, R. Finnis, Z. Tabata-

baian, P.J. Sellin, M.B.H. Breese, D.F. Fellows, R. van den

Broek, R.M. Gwilliam, Nucl. Instr. and Meth. B (2003) in

press.

Random telegraph signals in charge coupled devices · Random telegraph signals in charge coupled...

Documents

Transcript of Random telegraph signals in charge coupled devices · Random telegraph signals in charge coupled...