— 23 —

CCD and CMOS sensors

Nick WalthamI

Abstract

The charge-coupled device (CCD) has been developed primarily as a compactimage sensor for consumer and industrial markets, but is now also the preeminentvisible and ultraviolet wavelength image sensor in many fields of scientific researchincluding space-science and both Earth and planetary remote sensing. Today’sscientific or science-grade CCD will strive to maximise pixel count, focal planecoverage, photon detection efficiency over the broadest spectral range and signaldynamic range whilst maintaining the lowest possible readout noise. The relativelyrecent emergence of complementary metal oxide semiconductor (CMOS) image sen-sor technology is arguably the most important development in solid-state imagingsince the invention of the CCD. CMOS technology enables the integration on asingle silicon chip of a large array of photodiode pixels alongside all of the ancillaryelectronics needed to address the array and digitise the resulting analogue videosignal. Compared to the CCD, CMOS promises a more compact, lower mass, lowerpower and potentially more radiation tolerant camera.

The charge-coupled device

The concept of the charge-coupled device (CCD) emerged from the search fora silicon-based electrical equivalent of the magnetic bubble memory (Boyle andSmith 1970). In its simplest implementation, the CCD structure consists of a seriesof closely spaced electrodes separated from an underlying semiconductor substrateby a thin insulating oxide layer (Figure 23.1a). When a bias voltage is applied to anelectrode, a depletion region is formed in the semiconductor immediately beneathit. The depletion region is in effect a potential well which can store an electricalcharge packet. By pulsing the electrodes in an appropriate sequence the potentialwell, and hence its charge packet, can be transferred through the semiconductor(Figure 23.1b). A shift register can be formed by adding circuits for the insertionand detection of charge packets.

Although the CCD was originally conceived as a device to store digital infor-mation, it was evident that because a potential well could store variable quantitiesof charge it could also convey analogue signals. As a result, the CCD concept has

ISpace Science and Technology Department, Rutherford Appleton Laboratory, Harwell Scienceand Innovation Campus, Didcot, UK

391

392 23. CCD and CMOS sensors

Figure 23.1: (a) Charge-coupled device structure. (b) Charge transfer operation.

(a) (b)

been applied to analogue signal processing functions including simple delay linesand transverse filters. Without question, however, the greatest impact of the CCDhas been in the application of solid-state image sensors. Today the CCD appearsthroughout the consumer, industrial, medical, security and scientific imaging sec-tors. Only the relatively recent emergence of CMOS sensors has challenged theCCD in the consumer markets of mobile telephones and digital cameras whereminiaturisation and lower power consumption are key requirements. This begs thequestion of why the CCD has become the image sensor of choice for nearly allvisible light imaging systems and in particular scientific applications. There al-ready exist a number of texts that describe the concepts, physics and operationof the basic CCD structure, including its variations and refinements for particularapplications— see for example: Sequin and Tompsett (1975), Beynon and Lamb(1980), Janesick (2001) and Holst and Lomheim (2007). Only a brief overview canbe presented here, summarising the types of arrays and technologies that have beendeveloped and optimised for scientific imaging applications and in particular spaceinstrumentation.

Prior to the development of the CCD, electronic imaging relied mostly on the useof camera tubes. As a solid-state device the CCD offered the immediate attractionsof compactness, ruggedness and low-voltage operation. Of importance is that siliconis very responsive to visible light. At the peak of its sensitivity a back-illuminatedCCD is capable of absorbing and sensing very nearly all incident photons. Withappropriate fabrication and optimisation, silicon can also respond well to extremeultraviolet light and soft X-rays. In its simplest form, the basic structure of aCCD image sensor is formed from an array of electrodes running orthogonallyto a series of isolated charge transfer channels (Figure 23.2). The electrodes areusually connected together in groups of two, three or four phases. Biasing of theelectrode phases creates an array of isolated potential wells, or pixels, that collectthe photon-generated electrons. Following an exposure, the imaging area electrodes

393

Figure 23.2: Full-frame CCD array architecture.

are pulsed or “clocked” to transfer the integrated image charge pattern down thearray one line at a time. The lowest line is transferred into a serial readout registerthat runs orthogonally to the imaging area transfer channels or columns. Thisregister is clocked separately and at a higher rate allowing each pixel to be readout sequentially through a charge detection amplifier. Once the complete line hasbeen read out the imaging area electrodes can be clocked again so that the nextimage line is transferred into the readout register. This sequence is repeated untilall image lines have been read.

The CCD architecture described above is commonly referred to as a full-frameCCD array, the most popular for scientific imaging applications in space andground-based astronomy. However, an obvious limitation is that the image willbe smeared if the CCD remains exposed while being read out, particularly if theframe readout time is a significant fraction of the exposure time. A common solu-tion is to incorporate a mechanical shutter in front of the CCD so that an integratedimage can be read out in darkness. An alternative solution is to add a light-shieldedstorage array to which the integrated charge pattern can be transferred rapidly atthe end of the exposure period. The storage array can then be read out while anew image accumulates in the imaging array. There are two basic formats: theframe-transfer CCD (Figure 23.3a) and the interline-transfer CCD (Figure 23.3b).

In the frame-transfer CCD the storage array is added beneath the imaging array.This effectively doubles the size of the silicon chip and is unattractive in commercialcost-sensitive applications. In the interline-transfer CCD the storage array is incor-porated as light-shielded columns adjacent to the imaging columns. The weaknessin this approach is that the sensitivity is effectively halved. Unsurprisingly, thisarchitecture has seen little interest in the scientific community but rapidly gainedpopularity in the digital camera and camcorder markets. Two further CCD ar-chitectures applicable to Earth and planetary remote sensing are the linear CCD(Figure 23.4a), and the time-delay-integration (TDI) CCD (Figure 23.5a). The lin-ear CCD consists of a row of photodiodes from which photon-generated charge is

394 23. CCD and CMOS sensors

Figure 23.3: (a) Frame-transfer CCD architecture. (b) Interline-transfer CCD ar-chitecture.

(a) (b)

Figure 23.4: (a) The linear CCD. (b) Push-broom imaging.

(a) (b)

transferred into a serial readout register. The acquisition of a two-dimensional im-age relies on the sensor being scanned across the target scene. Linear CCDs with12 000 pixels or more are readily available. The technique of push-broom imagingis used to acquire very high resolution imagery (Figure 23.4b).

The TDI CCD is advantageous when there is insufficient illumination to obtaina useful signal from a conventional linear array. Its architecture is essentially thesame as a full-frame CCD array except that the x-y image format is typically verymuch larger in x than in y (Figure 23.5a). To avoid image smear, push-broomimaging requires the integrating CCD charge packets to be clocked down the arrayat the same rate as the projected image scans over the array (Figure 23.5b). TheTDI provides gain because the target scene is now integrated over the numberof lines in the array. TDI CCDs have been used successfully to obtain very highresolution imagery of Mars (Ebben et al 2007).

395

Figure 23.5: (a) The TDI CCD. (b) TDI CCD push-broom imaging.

(a) (b)

The scientific CCD

CCD fabrication technologies have undergone continuous development to im-prove performance. The development of large-format CCD arrays for scientific ap-plications was driven largely by the aspirations of the ground-based astronomicalcommunity and the requirement for a solid-state image sensor to replace the vidi-con camera tube in space applications. In the late 1970s a number of astronomicalresearch groups developed camera systems using the first commercially availablesensors from Fairchild— see for example: Marcus et al (1979) and Leach et al(1980). At the same time the CCD was being proposed for what were to becomeNASA’s HST and the Galileo mission to Jupiter. The earliest sensors, for examplethe Fairchild CCD201, had just 100 pixels × 100 pixels. RCA rose to the challengeof designing a CCD with 320 pixels × 512 pixels, the minimum needed to meet therequirements of the US TV standard. Texas Instruments collaborated with JPLunder NASA sponsorship to develop arrays of 400 pixels × 400 pixels and later800 pixels × 800 pixels — the dawn of the scientific CCD. This quest for a large-area, high-sensitivity and low-noise solid state image sensor resulted in many of thedevelopments now seen in today’s scientific or science-grade CCD catalogue. Thescientific CCD typically has several million pixels, high quantum efficiency over abroad spectral range, low readout noise and wide dynamic range.

Charge-transfer efficiency

The first significant development in CCD fabrication addressed the poor charge-transfer efficiency (CTE) of the simple surface-channel CCD architecture illustratedin Figure 23.1a. Interface “traps” at the silicon-silicon dioxide interface were foundto absorb and release charge with differing time constants, resulting in the smearingof charge packets being transferred through them. The CTE was found to be so pooras to make large-format image sensors impractical. The problem was overcome withthe introduction of the buried-channel CCD in which an additional layer of n-typesilicon is formed at the top of the p-type substrate and immediately below thesilicon dioxide insulating layer. The n-type silicon forces the potential wells to formdeeper within the substrate and away from the traps at the silicon-silicon dioxide

396 23. CCD and CMOS sensors

Figure 23.6: (a) CCD charge detection amplifier. (b) Correlated double sampling.

(a) (b)

interface. Signal charge can now be transferred through the substrate without itcoming into contact with the traps. Transfer efficiencies better than 99.999 % andas much as 99.9999 % are now achieved, enabling thousands of transfers withoutsignificant signal loss or smear.

Readout noise

Another important and on-going development has been to reduce readout noise.The output circuit of a CCD is a charge detection amplifier consisting of an out-put diffusion which has an associated parasitic node capacitance, a reset transistorto recharge the node capacitance and an output transistor operating as a source-follower to sense voltage variations on the node (Figure 23.6a). Pixel readout re-quires that the output node capacitance is pre-charged to a “reset” potential (Vrd)by pulsing the reset transistor “on” and “off” with a reset clock (∅r) as illustratedin Figure 23.6b. In practice there is a small discharge of the output node resultingfrom feed-through of the reset clock as it falls low. Signal charge from the last gateof the serial readout register (Vg) is then transferred onto the output node resultingin further discharge. The output transistor presents a video output signal (Vos). Asthe reset clock falls low, Johnson noise from the channel resistance of the resettransistor is frozen on the output node capacitance (C) and results in RMS pixel-to-pixel reset voltage fluctuations of

√kBT/C. This noise is commonly referred to

as kBTC noise, or reset noise, and would be the dominant source of CCD readoutnoise without a cancellation technique known as correlated double sampling (CDS).CDS acquires two samples of the CCD output, one before (S1), and one after (S2)the signal charge has been transferred onto the output node (Figure 23.6b). Sub-traction of the two samples cancels the kBTC noise as it is remains “correlated”or unchanged between the two samples. After reset-noise cancellation, transistornoise in the source-follower output transistor becomes the dominant noise source.The resultant readout noise will depend on the signal measurement bandwidthand thus the CCD readout rate. Higher readout rates require greater bandwidthin the readout electronics and result in higher readout noise. The best of today’sCCDs achieve a readout noise of ≈ 6 electrons RMS when read at ≈ 1 MHz, oras little as ≈ 2 electrons RMS when read at ≈ 100 kHz. A compromise is required

397

Figure 23.7: (a) The electron multiplication CCD. (b) Avalanche breakdown gain.

(a) (b)

between a low readout rate to minimise readout noise and a high readout rate tominimise the overall frame readout time. High frame readout rate is clearly de-sirable to maximise observing efficiency. One solution to the dilemma is to addadditional readout amplifiers to the CCD array such that a number of sub-sectionscan be read in parallel. Low readout rate for minimal noise is now possible withoutcompromising observing efficiency. An added benefit is that the additional readoutamplifiers provide a degree of redundancy in the case of a failure of a single read-out amplifier. A disadvantage, however, is that additional calibration of the datais required to correct for gain mismatch between the multiple readout amplifiers.

Electron multiplication readouts

To reduce readout noise still further, two CCD manufacturers have developedelectron multiplication readouts to provide on-chip gain (Hynecek 2001). A multi-plication register of typically several hundred stages is added to the end of the serialreadout register (Figure 23.7a). One electrode phase in the multiplication registeris clocked at significantly higher voltage (typically up to 40 V in amplitude) whichenables a probability of avalanche multiplication (Figure 23.7b). The word “prob-ability” is key as the net gain of a single stage might only be ≈ 1.01 but after500 stages the net gain is 1.01500 or ≈ 145. Because the gain is applied before thecharge detection output amplifier, the effective readout noise is the output noisedivided by the gain. This enables very low readout noise to be maintained at thehigh readout rates of TV cameras (≈ 10 MHz).

The electron multiplication CCD provides a solid-state equivalent of an inten-sified camera tube and with sufficient gain a solid-state photon counting detector.Not to be overlooked is an additional noise, introduced by the discrete nature ofthe multiplication process, which effectively increases the shot noise by a factor

√2

(Robbins and Hadwen 2003). The net gain is highly dependent on the amplitudeof the high-voltage clock and so appropriate control of its stability is necessary.Despite these and further issues concerned with spurious charge and ageing of thegain stage, the electron multiplication CCD provides a new and powerful tool forlow-light imaging.

398 23. CCD and CMOS sensors

Spectral range

A third continuing development of the CCD has been the challenge to maximiseits sensitivity over the widest spectral range possible. The first CCDs were “front-illuminated” requiring the incident photons to penetrate through semi-transparentpolysilicon electrodes to reach the underlying substrate. Unfortunately, the absorp-tion depth of silicon is such that a large fraction of the light is absorbed within theelectrodes, particularly at the shorter “blue” wavelengths. On the other hand, thereis a greater chance that the longer “red” and near-infrared wavelength photons willpenetrate deeply into the silicon substrate below the depletion regions. Electronsfrom these photons are free to drift laterally in a field-free region with the result thatimage resolution is degraded or blurred. A solution is to manufacture the CCD onepitaxial silicon which consists of a thin layer of the nominally-doped silicon, typi-cally 10 µm to 20 µm thick, on top of a very highly-doped bulk substrate. Electronscreated in the bulk substrate are likely to recombine before they drift to the deple-tion regions. Image resolution is maintained but at the cost of sensitivity to “red”wavelengths. To solve the lack of sensitivity to “blue” wavelengths researchers andmanufacturers started to develop the thinned “back-illuminated” CCD. This relieson the bulk of the silicon substrate being removed by chemical etching, and theCCD being illuminated through the back surface rather than through the polysil-icon electrodes. In principle, the sensitivity, or quantum efficiency (QE), is verymuch increased. However, a bare untreated silicon surface will contain a high den-sity of recombination centres or trapping sites. Any photoelectrons which makecontact with this surface will re-combine with holes and be lost. Surface passiva-tion is therefore necessary to reap the full benefits of the thinning and to maximisethe QE, particularly for the shorter “blue” wavelengths which are absorbed closeto this surface. An early method, referred to as back-surface charging, relied onthe CCD being held in a vacuum and subjected to a flooded exposure of ultra-violet light. This caused the native oxide to hold a negative charge that repelledphoto-generated electrons away and towards the depletion regions. Experimentsdemonstrated very high QE but the process was unstable and required periodicrecharging. A more reliable and stable method, referred to as back-surface doping,relies on a thin layer of dopant, typically boron, being incorporated in the back sur-face to form a p+-layer (Figure 23.8a). The change of doping concentration givesrise to a small potential step that repels photo-generated electrons towards thedepletion regions (Figure 23.8b).

Dopant is generally introduced by ion implantation but does not become activeunless in substitutional lattice sites. The lattice needs to reach a high temperaturefor this to occur but, because the thinned substrate will not tolerate furnace tem-peratures, only the thin doped surface layer is heated, using a very short pulse froman ultraviolet laser which heats the surface silicon to ≈ 1500 ◦C whilst increasingthe temperature of the remaining silicon by only 5 ◦C to 10 ◦C. With the addi-tion of an anti-reflection coating, excellent QE can now be obtained over a broadspectral range. The advantages of rear-illumination are clear from the characte-ristic QE curves of front- and rear-illuminated CCDs illustrated in Figure 23.9.However, multiple internal reflections within the thinned silicon substrate can giverise to interference fringing, typically at wavelengths in excess of ≈ 750 nm to

399

Figure 23.8: (a) The thinned back-illuminated CCD. (b) Potential profile of theCCD.

(a) (b)

Figure 23.9: Quantum efficiency curves for front- and back-illuminated CCDs.

800 nm, and rising in amplitude to ≈ 20 % at longer wavelengths. An exampleframe is reproduced in Figure 23.10. Their calibration and removal in subsequentimage processing can prove problematic in some applications such as narrow-bandimaging and spectroscopy. One method of suppressing the fringing is to apply ananti-reflection coating peaked at the fringing wavelengths that minimises internalreflections. A second method is to etch a λ/4 (≈ 80 nm) deep groove over each halfpixel so any fringing effect over the two half pixels is complementary.

Focal plane coverage

The final significant developments of the science-grade CCD addressed the re-quirement for greater focal plane coverage than provided by the first CCDs whichwere limited by the maximum reticle size of the manufacturing photolithographyto ≈ 20 mm × 20 mm. The technique of “stitching”, as illustrated in Figure 23.11,was developed to allow a large-format CCD to be constructed by precision litho-graphy from a selection of reticles, each containing a different part of the overallCCD design. CCD size is now limited only by the size of the silicon wafer and

400 23. CCD and CMOS sensors

Figure 23.10: Interference fringing in a thinned back-illuminated CCD.

Figure 23.11: Fabrication of large-format CCDs by stitching.

the more practical aspects of defect-free yield and hence cost. For even greatercoverage, an obvious solution is to populate the focal plane with more than oneCCD, accepting that there will be gaps between the chips due to the chip packag-ing and necessary electrical connections. To minimise these gaps, CCDs have beendesigned with minimal non-imaging on-chip circuitry and chip-carrier packagingalong one or more edges such that two or more chips can be closely butted upagainst each other. Gaps of only ≈ 0.2 mm to 0.5 mm are achievable. By wayof example, photographs of two HST Wide Field Camera-3 (WFC3) CCDs, firston a five-inch (12.7 cm) silicon wafer and then butted up together within the focalplane assembly, are reproduced in Figure 23.12. The basic CCD design is a stitchedarray of ≈ 4096 pixels × 2048 pixels with 15 µm pixel pitch such that the twochips butted together provide an overall coverage of ≈ 4096 pixels × 4096 pixelsor ≈ 61.4 mm × 61.4 mm.

Dark current

Depending on the application, an important CCD characteristic is its leakagecurrent, otherwise known as dark current. All semiconductors suffer from leakagecurrents that result from electrons having sufficient thermal energy to break free

401

Figure 23.12: HST WFC3 CCDs on a five-inch wafer and integrated in the focalplane assembly (© e2v technologies plc and Ball Aerospace & Technologies Corp.).

from the lattice. The surface traps at the silicon-silicon dioxide interface are largelyresponsible for the dark current generated in a CCD. At 20 ◦C, 1 nA cm−2 is typical,but this decreases by a factor of two for roughly every 7 ◦C to 8 ◦C reductionin operating temperature. Dark current has associated shot noise and a relativestatistical pixel-to-pixel non-uniformity of roughly 3 % to 10 % RMS. At 20 ◦C,dark current may well be the dominant contributor to the overall readout noiseof a CCD, even in moderate frame rate applications like TV sensors. However,with sufficient cooling, typically in the range −60 ◦C to −100 ◦C, dark currentcan be reduced to negligible levels even for the hour-long exposures often requiredfor astronomy. Liquid nitrogen has been popular for ground-based applications,whereas passive radiators have been used to cool sensors in space instrumentation.

An important consideration is the need for cleanliness because contamination,including materials out-gassing from the immediate surroundings, can condense onthe cold CCD (Kimble et al 1994). The issue is particularly serious for CCDs ima-ging in the extreme ultraviolet as even the smallest layer of surface contaminationcan significantly reduce the detector’s sensitivity. One solution to the problem isto provide a cold-trap on which contamination is condensed before it reaches theCCD. Another solution is to add a decontamination heater to the CCD package toallow periodic warming and out-gassing of contaminants. One operating techniqueand a further development have enabled CCDs to be operated with significantlyreduced dark current. Biasing the CCD such that the low level of the CCD clocksis significantly below the substrate potential causes holes to accumulate at the sil-icon surface, fill surface traps, and so suppress the dominant surface dark signalgeneration. The holes effectively change the surface from n-type to p-type and thesilicon can be considered as “inverted”. The technique of “dither-clocking” relieson a confined signal charge being periodically shifted back and forth between ad-jacent electrodes and ensures that all of the silicon is periodically inverted. Thedark current suppression is a function of operating temperature and the frequencyof the “dither” and can result in more than two orders of magnitude of reduction.A further development was to add an implant under one of the electrode phases to

402 23. CCD and CMOS sensors

define fixed potential wells under which charge could be collected with the wholeof the surface inverted. Dither clocking was no longer necessary. This CCD variantis referred to as a multi-pinned phase (MPP) CCD, or sometimes as inverted-modeoperation (IMO). Disadvantages of the MPP CCD are firstly a reduction in thecharge storage capacity, or full-well capacity, of roughly 20 % and secondly a signi-ficantly increased dark signal non-uniformity.

Dynamic range

Finally, two of the most important metrics of a science-grade CCD are its full-well capacity and readout noise. Together these determine the dynamic range ofthe CCD with ≈ 1 × 105 or more being obtainable from today’s best devices.Associated with these parameters is the relative linearity, which is typically 0.5 %to 1 % and dominated by the characteristics of the charge detection amplifier.Full-well capacity is a measure of the maximum signal charge that a pixel canstore and transfer, and is a function of the isolated electron storage area which isdetermined by both pixel size and the number of electrode phases used to definethe pixel. A 4-phase pixel holds and transfers charge beneath half of the pixel areawhilst a 3-phase pixel can only work with one third of the pixel area. A 2-phasepixel operates beneath half of the pixel area but the storage density is limited bya greatly reduced potential well depth. By way of example, a 12 µm × 12 µm pixelwill typically store ≈ 60 000 electrons (2-phase), ≈ 100 000 electrons (3-phase) and≈ 200 000 electrons (4-phase). A larger 24 µm × 24 µm 4-phase pixel could holdmore than 106 electrons.

This necessarily brief introduction to the science-grade CCD concludes by di-recting the reader to the literature for more in-depth discussion, recommendingJanesick (2001) in particular as a comprehensive source.

The CCD in space

The first CCDs to be flown in space were on the two Russian Vega probeslaunched in 1984 to image the nucleus of comet Halley in 1986. ESA followedwith CCD cameras on board the Giotto spacecraft, also to comet Halley. Thedevelopment of the CCD for space applications can be traced back to 1974 when,under the sponsorship of NASA, JPL undertook a programme to develop large-format CCD arrays for the Galileo mission to Jupiter, launched in 1989, and theHST, launched in 1990. Today the CCD is the preeminent visible- and ultraviolet-wavelength image sensor in space science and in both Earth and planetary remotesensing.

Radiation damage

One of the most important issues for operation in the space environment isradiation damage. Radiation changes the operational and performance character-istics of CCDs, and the primary concerns are ionisation, displacement damage andtransient effects.

403

Ionising radiation results in the accumulation of trapped charge in the CCD’soxide and the generation of traps at the silicon-silicon dioxide interface. The trappedcharge results in changes to the effective bias voltages applied to the CCD and arereferred to as flat-band voltage shifts. The accumulation of radiation is quantifiedby a parameter called the total ionising dose (TID). A biased CCD will typicallyexhibit a flat-band voltage shift of ≈ 0.08 V krad−1 (Si).1 Shifts of up to ≈ 2 Vcan usually be tolerated by careful optimisation of the CCD bias voltages priorto launch and/or by enabling software-controlled adjustments to be made duringthe mission. The second effect of ionising radiation is a significant increase in thesurface dark current. Besides operating at a reduced temperature, inverted-modeoperation, either by dither-clocking or the use of an MPP CCD, will generate holesthat will suppress the additional dark charge.

Displacement damage results from energetic particles, for example protons andneutrons, which collide with the silicon atoms and displace them from their latticesites, creating vacancy-interstitial pairs. Many of these recombine but some vacan-cies can link with phosphorous atoms to form a trap (e-centre) that degrades CTE.Particle-induced lattice damage can also give rise to an increase in dark current andits non-uniformity, and introduce a new type of pixel noise referred to as “randomtelegraph signals” (RTS). A good review of the subject can be found in Hopkinsonet al (1996). The observable degradation in CTE is very much application depen-dent but is particularly serious in the field of spectroscopic X-ray astronomy. Signalcharge trapped by a proton-induced defect is likely to be later released into trailingpixels with resulting image smear. This arises because the time for an empty trapto capture charge is very short whereas the time constant of the release is verymuch longer. The release time constant is exponentially temperature dependentand so a complex interdependency exists between operating temperature and thereadout rate of the CCD. The observable CTE degradation is also a strong functionof the general background signal present on the CCD with the smearing of smallcharge packets being more pronounced in the absence of a background signal tofill the traps. CTE is often assessed by the degree of smearing observed in X-rayevents from an Fe55 source (Janesick 2001). Events will normally be confined toone or two pixels but will become increasingly smeared with increasing radiation-induced CTE degradation (Figure 23.13). The same effect has been observed in theimages of star fields on HST (Kimble et al 2000) and has significant implicationsfor the accuracy of any radiometric calibration. The second result of displacementdamage is an increase in the contribution of dark current and its non-uniformityfrom the bulk silicon. Individual pixels may become localised regions of high darkcurrent, often referred to as “hot pixels”, and the number can be expected to in-crease during the mission life. Some pixels exhibit RTS noise, an unstable increasein dark current which jumps randomly between well-defined levels (Hopkins andHopkinson 1993, 1995). The final aspect of displacement damage to address is thefinding that a good proportion ≈ 60 % to 85 % of CTE damage can be repairedby annealing. This involves periodically heating the CCD to ≈ 120 ◦C, or higher,for tens of hours at a time (Holland 1991). Similar studies have also demonstrated

10.008 V Gy−1

404 23. CCD and CMOS sensors

Figure 23.13: Smearing of Fe55 X-ray events after CTE degradation in an irradiatedCCD.

partial annealing of hot-pixel damage at elevated temperatures (Holland et al 1990;Holland 1991).

Transient radiation effects occur when a particle, for example a cosmic ray,passes through the active volume of the CCD. Ionisation creates charge along theparticle’s path and results in a track of charge that may traverse many pixels.Although the events are transient and cause no lasting damage, they can resultin significant noise within an image. The identification and rejection of cosmic-ray events in subsequent image processing requires that the observer acquires twoimages of the same field and accepts only those features that are common to bothimages.

Protective measures

The importance of the CCD in space instrumentation has led to considerableresearch being undertaken to understand radiation damage, how best to protectagainst it and the development of radiation-hardening technologies. The first de-fence has been to ensure that the CCD is optimally shielded, typically with upto ≈ 25 mm of aluminium. Significantly more shielding worsens matters becauseincident high-energy particles create an excess of secondary events from the shield-ing itself. The effects of ionising radiation can also be reduced by manufacturingthe CCD with a modified gate dielectric layer but has the penalty of some loss inproduction yield. Minimisation of CTE degradation has seen the development ofthe supplementary buried channel or notch. The concept is to confine the signalcharge to a narrow channel, and thus a smaller volume, such that it is exposed toa reduced number of trapping sites. This can be effective for the transfer of smallsignals but the advantage is lost once the signal spills out of the notch and intothe wider channel. Finally, although CCD manufacturing has focused almost ex-clusively on the buried n-channel CCD, p-channel devices are now being exploredfor their greater resilience to CTE degradation, principally because of the absenceof phosphorous atoms with which to form e-centres (Hopkinson 1999).

405

Figure 23.14: (a) 3-Transistor CMOS Active Pixel Sensor. (b) Array architecture.

(a) (b)

CMOS

The emergence of CMOS image sensor technology is arguably the most impor-tant development in solid-state imaging since the invention of the CCD. CMOS im-age sensors, of which today most can be referred to as CMOS Active Pixel Sensors(APS), exploit the same silicon chip technology used in microprocessor systems.The attraction is mostly due to the nature of CMOS technology in that many mil-lions of transistors can be integrated on a single silicon circuit. This presents theopportunity to integrate a large array of pixels, each with its own photodiode andreadout transistors, alongside all of the ancillary electronics needed to address thearray, buffer the analogue video signal and even digitise it ready for processing,storage or display.

CMOS sensors, like CMOS integrated circuits, operate at considerably lowervoltages than CCDs, typically between 1.8 V and 5 V depending on process formand geometry. The large-scale functional integration and low-voltage operation areclearly very attractive in consumer markets that demand compactness, long batterylife and low production cost. As a consequence, many digital cameras and mobiletelephones now exploit CMOS rather than CCD sensors.

The question arises as to how CMOS compares to CCD technology for scientificimaging applications and in particular space instrumentation. CMOS sensors aredesigned in many architectural forms and a good introduction can be found inHolst and Lomheim (2007). In its simplest form the CMOS APS pixel consists ofa photodiode and three transistors: one to pre-charge the photodiode, one to sensethe signal voltage on the photodiode and one to select the row (Figure 23.14a).A pixel array will typically be x-y addressed by shift registers (Figure 23.14b)although some implementations employ address decoders to enable random pixelaccess. The APS pixel array is accessed one row at a time by enabling all therow-select transistors within a single row of pixels. At the bottom of the array, theindividual pixels within the row are selected and read out column-by-column. Thevideo signal may be fed through a multiplexer to an analogue output amplifier, aserial analogue-to-digital converter (ADC) or an array of column-parallel ADCs.

406 23. CCD and CMOS sensors

Figure 23.15: Layout of a typical CMOS pixel.

A disadvantage of the basic three-transistor pixel is that it is subject to kBTCnoise, which is typically the dominant source of readout noise. CDS has been foundto work with varying degrees of success by storing a “reset” frame off-chip prior toan exposure and later subtracting it from the “signal” readout frame. To enableon-chip CDS, researchers and manufacturers have developed four-transistor pixelsin which the new transistor is used as a gate between the photodiode and the sensetransistor. The concept is that the output node can be sampled after reset andbefore signal from the photodiode is switched onto the output node. This appearsanalogous to the CCD output amplifier circuit but requires the photodiode tobe “pinned” at appropriate potential. This ensures that all the signal charge istransferred to the output node, thus avoiding image lag, and that the kBTC noiseremains correlated between the two CDS samples. The disadvantage of the pinnedphotodiode is that the pinning significantly reduces the charge storage capacity ofthe pixel and thus also the usable linear dynamic range.

Quantum efficiency

The quest for a scientific CMOS sensor is today still in its infancy. The quan-tum efficiency of front-illuminated CMOS sensors is compromised by the in-pixelelectronics and aluminium bus tracks reducing the “fill factor”, a measure of thefraction of the pixel’s area that is actually sensitive to light. The layout of a typicalCMOS pixel is shown in Figure 23.15. In this example, the photodiode occupiesonly 19 % of the pixel area (as defined by the dotted line) but will typically yield afill factor ≈ 30 %. The losses arise from the reflection of light from the aluminiumbus lines and photon-induced electrons in the substrate being absorbed and lostwithin the in-pixel transistor electronics. An obvious solution to the problem is tothin and back-illuminate the sensor and several research groups have now demon-strated back-illuminated CMOS sensors that achieve quantum efficiencies in linewith their CCD counterparts (Waltham et al 2007; Janesick et al 2007).

407

Readout noise and dynamic range

A second area of research is to minimise readout noise and maximise chargestorage capacity and linearity. A limitation is the linear voltage swing that canbe obtained outside the transistor threshold regions in modern low-voltage CMOSprocesses. The linear dynamic range of today’s best CMOS sensors is ≈ 5000,considerably less than a CCD. Several approaches to overcoming the problem arebeing investigated including the concept of pixels that deliberately behave in a non-linear fashion and sensors that allow individual pixels to have varying exposureperiods. More complex five- and six-transistor pixels are also being investigatedin the pursuit of increased dynamic range (Janesick et al 2006, 2007). We cananticipate further progress in the future as researchers adapt to exploiting theadvantages of CMOS technology rather than attempting to emulate the eleganceof the CCD.

CMOS in space

CMOS sensors are already used in space, having applications in satellite businstrumentation such as star trackers and inspection cameras. CMOS is yet to havea significant impact in scientific payloads for which the CCD remains dominant.The principal advantages of CMOS over the CCD for space instrumentation arecompactness, low mass, low power and radiation hardness. The CCD remains un-challenged in dynamic range and photometric accuracy. The effects of radiation inCMOS and CCD sensors are similar in that both suffer from ionising radiation anddisplacement damage. However, the key advantage of the CMOS APS is that thereis no degradation of CTE. A new effect, unseen in the CCD, is a susceptibility tosingle-event latch-up (SEL). SEL is a potentially destructive condition, triggeredby an energetic particle, in which parasitic circuit elements form the equivalent of asilicon-controlled rectifier. The result is a large and potentially damaging increasein supply current that can only be recovered from by temporarily removing power.CMOS sensors will also exhibit the other effects of displacement damage such ashot pixels and RTS noise. CMOS sensors will undoubtedly play an increasing rolein space instrumentation, but today they are most likely to appear in those applica-tions that have extreme size, mass or power constraints, applications which requirecomplex operational modes such as random pixel access, or in those missions forwhich the radiation damage to a CCD would prove insurmountable.

Bibliography

Beynon JDE, Lamb DR (1980) Charge-coupled devices and their applications.McGraw-Hill

Boyle WS, Smith GE (1970) Charge-coupled semiconductor devices. Bell Syst TechJ 49:587

Ebben TH, Bergstrom J, Spuhler P, Delamere A, Gallagher D (2007) Mission toMars: the HiRISE camera on-board MRO. Proc SPIE 6690:6690B-1–14

Janesick J (2001) Scientific Charge-Coupled Devices. Proc SPIE Press Book, ISBN9780819436986

408 23. CCD and CMOS sensors

Janesick J, Andrews J, Elliott T (2006) Fundamental performance differences be-tween CMOS and CCD imagers: Part I, Proc SPIE 6276:62760M-1–19

Janesick J, Andrews J, Tower J, Grygon M, Elliott T, Cheng J, Lesser M, Pinter J(2007) Fundamental performance differences between CMOS and CCD imagers;Part II. Proc SPIE 6690:669003-1–23

Holland AD (1991) Annealing of proton-induced displacement damage in CCDsfor space use, Proc. of 10th Symposium on Photoelectronic Image Devices, IoP,33–40

Holland A, Abbey A, Lumb D, McCarthy K (1990) Proton damage effects in EEVcharge coupled devices. Proc SPIE 1344:378–395

Holst GC, Lomheim TS (2007) CMOS/CCD Sensors and Camera Systems, Belling-ham, Wash.; The International Society for Optical Engineering; JCD Publishing

Hopkins IH, Hopkinson GR (1993) Random Telegraph Signals from Proton-irradiated CCDs. IEEE Trans Nuc Sci 40:1567–1574

Hopkins IH, Hopkinson GR (1995) Further measurements of random telegraphsignals from proton-irradiated CCDs. IEEE Trans Nuc Sci 42:2074–2081

Hopkinson GR (1999) Proton damage effects on p-channel CCDs, IEEE Trans NucSci 46:1790–1796

Hopkinson GR, Dale CJ, Marshall PW (1996) Proton effects in charge-coupleddevices, IEEE Trans Nuc Sci 43:614–627

Hynecek J (2001) Impactron – a new solid state image intensifier. IEEE Trans onElectron Devices, 48:2238–2241

Kimble R, Brown L, Fowler W (plus 11 authors) (1994) CCD detector for the SpaceTelescope Imaging Spectrograph. Proc SPIE 2282:169–181

Kimble RA, Goudfrooij P, Gilliland RL (2000) Radiation damage effects on theCCD detector of STIS, UV, optical and IR space telescopes. Proc SPIE 4013:532–544

Leach RW, Schild RE, Gursky H (plus three authors) (1980) Description, perfor-mance and calibration of a charge-coupled device camera. PASP 92:233–245

Marcus S, Nelson R, Lynds R (1979) Preliminary evaluation of a Fairchild CCD-211and a new camera system. Proc SPIE 172:207–231

Robbins MS, Hadwen BJ (2003) The noise performance of electron multiplyingcharge-coupled devices. IEEE Trans on Electron Devices 50:1227–1232

Sequin CH, Tompsett MF (1975) Charge transfer devices. Academic Press, NYWaltham NR, Prydderch M, Mapson-Menard, H (plus three authors) (2007) De-

velopment of a thinned back-illuminated CMOS active pixel sensor for extremeultraviolet spectroscopy and imaging in space science. Proc SPIE 6690:669007-1–14