I!I l~

Sky on a Chip: The Fabulous CCDJames Janesick, Caltech, and Morley Blouke, Tektronix, Inc.

i

TELESCOPES today are no biggerthan those in use two generations ago.Yet, curiously, astronomers can nowstudy objects that are hundreds or eventhousands of times fainter th;3.ll thoseaccessible in the 1930's.

• The explanation is simple: the last fewdecades have brought an unprecedentedimprovement in the sensitivity of lightdetectors. The changes in our own era areas significant as those caused by thedevelopment of the telescope in the 17thcentury or the photographic plate in the19th century.

Of all the new astronomical tools, onein particular stands out because it is analmost perfect radiation detector. Knownas a charge-coupled device, or CCD, it hasthe ability to register almost every photon(particle of light) that strikes it overwavelengths ranging from the near infra-red to X-rays. In contrast, the mostsensitive photographic emulsion used byastronomers records two or three percentof the light striking it, and over a muchnarrower range of wavelengths.

The CCD revolution came about indi-

rectly. It began in the late 1960's whentwo researchers at Bell Telephone Labora-tories, Willard S. Boyle and George E.Smith, sought to invent a new type ofcomputer memory circuit. In particular,they wanted to construct an electronic

analog to magnetic bubbles.While Boyle and Smith conceived the

CCD as a memory element, it rapidlybecame apparent that the tiny chip ofsemiconducting silicon they first demon-strated in 1970 had many other applica-tions, including signal processing and im-aging (the latter because silicon respondsto visible light).

In recent years the CCD's initial prom-ise as a memory element has vanished. Ithas, however, established itself as thepremier imaging device for scientific ap-plications, especially astronomical ones.Indeed, CCD's are at the forefront of anongoing revolution in all types of elec-tronic photography. They are alreadycommon in lightweight video systems andauto-focus cameras.

This picture of Uranus is thought to be thefirst astronomical image made with acharge-coupled device, or CCD. It wasobtained in 1975 by scientists from the JetPropulsion Laboratory and the Universityof Arizona, using the 61-inch telescope inthe Santa Catalina Mountains near Tucson.Recorded at a wavelength of 8900 ang-stroms in the near infrared, it shows aregion of enhanced methane absorption(dark area) near Uranus' south pole.

l""'""-

\

\\

: \\ \ '

\ \ \ \

\

Determining the brightness distribution in a celestial object with a charge-coupled devicecan be likened to measuring the rainfall at different points in a field with an array ofbuckets. Once the rain has ceased, the buckets in each row are moved horizontally acrossthe field on conveyor belts. As each one reaches the end of the conveyor, it is emptied intoanother bucket on a belt that carries it to the metering station where its contents' aremeasured. Artwork by Steven Simpson ..

238 Sky & Telescope, September, 1987

PIONEERING USE

Astronomers were among the first torecognize the extraordinary imaging capa-bilities of the CCD. In 1972 researchers atthe Jet Propulsion Laboratory (JPL) estab-lished a program to develop CCD's forspace astronomy. Three years later theJPL team, working with scientists fromthe University of Arizona, took what iprobably the first astronomical image ob-tained with a CCp (see the photographabove).

Today every major telescope in theworld has a CCD camera as part of itsobserving arsenal. Such a device can beused either for direct imaging or as thedetector in a spectrometer or other instru-ment. In addition, the Vega, Giotto, andSuisei probes to Halley's comet, and manyfuture space projects - including theHubble Space Telescope, Galileo, and theAdvanced X-ray Astrophysics Facility -will have CCD's at the heart of theircamera systems.

The operation of a CCD is quite simplein principle. An elegant analogy byJerome Kristian of the Carnegie Insti-

I!!I l••.•••.

Sky on a Chip: The Fabulous CCDJames Janesick, Caltech, and Morley Blouke, Tektronix, Inc.

t

TELESCOPES today are no biggerthan those in use two generations ago.Yet, curiously, astronomers can nowstudy objects that are hundreds or eventhousands of times fainter th;lll thoseaccessible in the 1930's., The explanation is simple: the last fewdecades have Qrought an unprecedentedimprovement in the sensitivity of lightdetectors. The changes in our own era areas significant as those caused by thedevelopment of the telescope in the 17thcentury or the photographic plate in the19th century.

Of all the new astronomical tools, onein particular stands out because it is analmost perfect radiation detector. Knownas a charge-coupled device, or CCD, it hasthe ability to register almost every photon(particle of light) that strikes it overwavelengths ranging from the near infra-red to X-rays. In contrast, the mostsensitive photographic emulsion used byastronomers records two or three percentof the light striking it, and over a muchnarrower range of wavelengths.

The CCD revolution came about indi-rectly. It began in the late 1960's whentwo researchers at Bell Telephone Labora-tories, Willard S. Boyle and George E.Smith, sought to invent a new type ofcomputer memory circuit. In particular,they wanted to construct an electronicanalog to magnetic bubbles.

While Boyle and Smith conceived theCCD as a memory element, it rapidlybecame apparent that the tiny chip ofsemiconducting silicon they first demon-strated in 1970 had many other applica-tions, including signal processing and im-aging (the latter because silicon respondsto visible light).

In recent years the CCD's initial prom-ise as a memory element has vanished. Ithas, however, established itself as thepremier imaging device for scientific ap-plications, especially astronomical ones.Indeed, CCD's are at the forefront of anongoing revolution in all types of elec-tronic photography. They are alreadycommon in lightweight video systems andauto-focus cameras.

This picture of Uranus is thought to be thefirst astronomical image made with acharge-coupled device, or CCD. It wasobtained in 1975 by scientistS from the JetPropulsion Laboratory and the Universityof Arizona, using the 61-inch telescope inthe Santa Catalina Mountains near Tucson.Recorded at a wavelength of 8900 ang-stroms in the near infrared, it shows aregion of enhanced methane absorption(dark area) near Uranus' south pole.

l-.....,

\

\\

: \\

\ \

Determining the brightness distribution in a celestial object with a charge-coupled devicecan be likened to measuring the rainfall at different points in a field with an array ofbuckets. Once the rain has ceased, the buckets in each row are moved horizontally acrossthe field on conveyor belts. As each one reaches the end of the conveyor, it is emptied into

. another bucket on a belt that carries it to the metering station where its contents aremeasured. Artwork by Steven Simpson.

238 Sky & Telescope, September, 1987

PIONEERING USE

Astronomers were among the first torecognize the extraordinary imaging capa-bilities of the CCD. In 1972 researchers atthe Jet Propulsion Laboratory (JPL) estab-lished a program to develop CCD's forspace astronomy. Three years later theJPL team, working with scientists fromthe University of Arizona, took what isprobably the first astronomical image ob-tained with a CCp (see the photographabove).

Today every major telescope in theworld has a CCD camera as part of itsobserving arsenal. Such a device can beused either for direct imaging or as thedetector in a spectrometer or other instru-ment. In addition, the Vega, Giotto, andSuisei probes to Halley's comet, and manyfuture space projects - including theHubble Space Telescope, Galileo, and theAdvanced X-ray Astrophysics Facility -will have CCD's at the heart of theircamera systems.

The operation of a CCD is quite simplein principle. An elegant analogy byJerome Kristian of the Carnegie Insti-

posed High-Resolution Solar Observatory.Today the largest CCD is made by

Tektronix, Inc. (see left and next page)and incorporates over 4,000,000 pixels(2,048 by 2,048). Its active image areameasures over 2.5 inches across, making itthe world's largest integrated circuit.

Impressive as these numbers sound,they are small compared to the size andresolution of photographic emulsions.Even the coarsest-grain 35-millimeter filmmight have a resolution equivalent to25,000,000 pixels. Fine-grained astronomi-cal plates can do a lot better.

However, CCD images have other ad-.vantages that render them far superior tophotographs. For example, the bottomleft diagram on page 241 shows how thequantum efficiency of a typical CCDexceeds that of several other detectors

used by astronomers.The eye, the first astronomical "detec-

tor," has a quantum efficiency of aboutone percent. In other words, of every 100

This image of the planetary nebulaNGC 6781 in Aquila was made by combin-ing three CCD images, each made with adifferent color filter. Like all CCD images,it is built up from an array of tiny squares(inset), each representing the data collectedby one pixel. The more pixels that make upthe image, the finer the detail visible. Thisimage was obtained with a CCD having512 rows and 320 columns of pictureelements. Courtesy Rudolph Schild.

should be in such a form that the bright-nesses of the objects viewed can easily bemeasured with high precision in absoluteunits, such as magnitudes.

• Linearity - the output should beuniform; if one object is twice as bright asanother, its image should be twice asstrong.

Unlike many other detectors currentlyin use, the CCD excels in everyone ofthese areas.

A CCD's resolution is determined pri-marily by the number of picture elements,or pixels, making up the imaging area.More pixels mean better resolution, asevident in the photograph at right.

The first large image-forming CCD'swere introduced by Fairchild Semiconduc-tor in 1973. They contained 10,000 pixelsarranged in 100 rows and 100 columns. Ashort time later, RCA Corp. made ver-sions achieving the standard televisionresolution of 512 rows by 320 columns(163,840 pixels). Detectors like this aretypical of the tCD's now in use atobservatories around the world.

The CCD's made especially for Galileoand the Hubble Space Telescope by TexasInstruments are now becoming availableto astronomers in limited numbers. These

have 640,000 pixels arranged in an 800-by-800 array. Texas Instruments has alsomade chips with over one million pictureelements for the cameras of the pro-

l=gest charge-coupled device currently available is made by Tektronix, Inc.}rising an array of 2,048 rows and 2,048 columns of light-sensitive picture elements,

- drip measures over 2.5 inches square. All illustrations supplied by the authors unless- otherwise.

understand why a CCD is so usefulpowerful, we need to understand

.:;2: a good astronomical detector must=.::. ::s most important requirements are:

• 31gh resolution - it must be able to_

photons entering the pupil, only one isdetected. In contrast, for every 100 pho-tons hitting a CCD, 50 to 70 or even moreare registered. In addition, the range ofwavelengths to which the eye can respondis much smaller than the CCD's. The

same limitation is true of photocathodes,film, and vidicon-type television detectors.. It has recently been shown that the next

generation of CCD's will respond usefullyfrom I angstrom in the "soft" X-rayregion of the spectrum to greater than10,000 angstroms in the near-infrared (seethe box on page 242). This huge wave-length span is far greater than that of anyother detector known; it is one of theproperties that makes CCD's useful forsuch diverse missions as the AdvancedX-ray Astrophysics Facility and HubbleSpace Telescope.

A Tektronix 2,048-by-2,048-pixel charge-couplet:device shown life size.

This monster chip, theworld's largest, is cur-rently under developmentfor astronomical applica-tions. Also shown here fursize is the 800-by-800-pix,CCD (inset) made espe-cially by Texas Instru-ments for the HubbleSpace Telescope and theCalileo Jupiter probe.While perfect examplesare worth dozens of timestheir weight in gold, cos-metically defective modelshave already been madeavailable to the astronomi-cal community.

How a Charge-Coupled Device Works

~1PiX~I-

r:: Electrons =1~ ~ ~

.~ .•••

CL

••• 1nTTT:TTI.t

•••

••• ! I• •• II: II,..• • •• ··1II •e. ••S.JI I:·: 1I1 : IHOIII: I11;·:111=Gates

Output amplifier

The operation of a CCD is illustrated schematically by this simplified chip consisting ofnine pixels, an output register, and an amplifier. Every pixel is subdivided into threeregions, or gates; each is an electrode whose voltage can be varied. Left: During anexposure the central gate of each pixel is "on" (yellow areas) and its neighbors are "off"(green areas). This creates "electron buckets" under the middle electrodes and barriersbetween adjacent pixels. Right: At the end of the exposure the gate voltages are changedand electrons shifted one gate to the right as new potential wells are created and old onesare destroyed.

~ ~ ~

~I

III· III: II1m)• •••• • ••••• · •• •••• •• • •••• ••

I

•I

• •••• I"•• • ••• 1I~)• ••

Left: As the voltages are cycled again, electrons flow.from the right-most gate of one pixelto the left-most gate of its neighbor. The electrons in the right-hand pixels are transferredto the output register. Right: Before the pixel array can be shifted again, charge must betransferred, one pixel at a time, through the output register and amplifier. When thisregister has been completely emptied, another cycle of pixel-array transfers is executed.These steps continue in a systematic way until every charge packet has moved horizontallyalong its row, vertically down the output register, and into the amplifier, where it ismeasured. In a real astronomical CCD, this process can take as long as IO seconds.

240 Sky & Telescope, September, 1987

A charge-coupled device must performfour tasks to generate an image. Theseoperations are:

• Photoelectron generation (rain).• Electron collection (buckets) .• Charge transfer (conveyor belts) .• Readout (metering station).The first operation relies on a physical

process known as the photoelectric effect- when light hits certain materials,electrons are generated. When a patternof light from, say, a distant galaxy fallson a CCD, the photons that make upthe incoming image are absorbed withinthe chip's silicon lattice, where they giverise to free electrons.

In the second step these so-calledphotoelectrons are collected in the near-est discrete collecting sites, or pixels(picture elements). These collection sitesare defined by an array of electrodes,called gates, formed on the surface ofthe CCD. In our rainfall analogy, thebuckets represent the pixels .

The third stage, charge transfer, isaccomplished by changing the voltageson each gate in a systematic way so thatthe electrons move horizontally fromone pixel to the next, conveyor-beltfashion .

At the end of each row is the so-called

output register, which is in effect aconveyor running perpendicular to allthe others. It transports the charge pack-ets one by one to the output amplifierwhere, in the final operating steps, theelectrons are counted and converted to a

form suitable for storage in a computer,display on a television monitor, andanalysis by astronomers.

j"

To overcome this non uniformity problem,observers use a technique called flat-field-ing. During an observing run an expo-sure is made of a uniform source such asthe dawn sky or inside of the telescopedome to map the variations among theCCD's pixels. This image represents theflat-field frame- corresponding to the sim-ulated raw image of the Black-eye galaxyshown above.

Once the response of each pixel is known,observers can scale the data by an amountdetermined from the flat-field exposure andthereby correct for each pixel's individualresponse. The result is the same as thatwhich would be registered by a CCD withpixels of identical sensitivity; it can now beanalyzed by astronomers. Image 'courtesy

Rud~lph Schild.

This simulated "raw" image from a charge-coupled device does not give a realisticrepresentation of the well-known Black-eye galaxy, M64 in Coma Berenices, be-cause individual pixels have varying sensi-tivities to light. This effect is shown hereschematically by dividing the -image in-to nine regions of grossly different sen-sitivities. Images such as this are obviouslynot very useful to astronomers or anybodyelse.

Even the most sensitive detector is essen-

tially worthless to astronomers if it can'tbe used to determine reliably the bright-ness of what is being observed in somesort of absolute units such as magnitudes.This requires that the detector be stableover long periods of time so that it alwaysproduces the same output signal for agiven light input.

Because they are solid-state devices;CCD's are inherently stable. Once a chiphas been calibrated by observing standardstars, photometric measurements made

during the course of a night can be held toan absolute accuracy of 0.5 percent (equiv-alent to an uncertainty of about 0.005magnitude).

The CCD also exhibits remarkable line-

arity. In other words, the number ofelectrons collected in a pixel is preciselyproportional to the number of incidentphotons (see the diagram above). This isin sharp contrast to photographic filmand vidicon-type television detectors. Theformer has a very complicated response to

100,000

10

RELATIVE EXPOSURE -

CCD's are exceptionally linear; doublingthe exposure yields a signal exactly twice asstrong. This holds over an enormous expo-sure range, enabling CCD's to detect simul-taneously very bright and very faint objectsin a single image. This behavior contrastsstrikingly with that of photographic emul-sions, which are only linear over a narrowrange of exposure. Emulsions must alsoovercome a threshold before they start toform an image at all. Worse, beyond acertain point emulsions saturate; their abil-ity to detect light actually decreases if theexposure is too long.

cause of the large brightness differencescommonly found between, say, the faintouter parts of galaxies and their brilliantnuclei. The greater the dynamic range, thegreater the CCD's ability to image bothbright and dim objects in 'the same field.In stark contrast, the best photographicplates can reproduce a brightness range ofonly 100 to 1.

MORE THAN PRETTY PICTURES

10,000-J~C9

i/j 1,000f--::J

:::

is 100

0; 0.12.000 4,000 6,000 8,000 10,000

WAVELENGTH (angstroms)

The superiority of the charge-coupleddevice (CCD) over almost all other image-forming detectors is illustrated here. In thered and near-infrared portions of the spec-trum, CCD's are more efficient at detectingradiation than any other type of device. Inthe ultraviolet, however, the performanceof older CCD's illustrated here is muchworse than television-type detectors andeven photographic film.

But a high quantum efficiency andbroad spectral response are useless if thedetector introduces false signals thatswamp those from a faint astronomicalsource. The electrons produced in eachpixel of a CCD are counted by a singletransistor built into the chip's outputcircuitry. The accuracy of the counting isdetermined by the noise introduced by thetransistor. Today's CCD's have amazinglylow noise and can reliably detect chargepackets containing as few as 10 electrons(generated by the capture of 15 to 20visible-light photons).

Small as these numbers seem, astrono-my . demands even greater sensitivity.Counting errors are typically comparableto the number of photons collected everysecond when a 24th-magnitude object isviewed with a 6O-inch telescope. Sub-stantial effort is now being expended toenable CCD's to count charge packetscontaining fewer than four electrons.

One of the most significant advantagesa CCD has over other detectors is its largedynamic range. For CCD's this importantproperty is defined as the ratio of thelargest charge packet that can be meas-ured to the smallest..

The larger the size of a pixel, the moreelectrons it can hold before overflowing.For example, the 15-micron-square pixelson Texas Instruments' 800-by-800 CCDcan each hold about 75,000 electrons.Picture elements twice as wide have a

capacity of nearly 1,000,000 electrons.These numbers reveal that current CCD's

can have dynamic ranges up to 100,000(1,000,000 electrons in a full pixel dividedby 10 electrons of noise).

. Large dynamic range is particularly im-portant for an astronomical detector be-

:;;;

2

PHOTON ENERGY (electron volts)10,000 1,000 100 10

The All-Seeing CCD

James Janesick is head of the advancedimaging sensors group at the Jet PropulsionLaboratory. He has developed CCD's forNASA space missions and ground-based astron-omy for the last 14 years. Morley Blouke isprincipal scientist in the CCD engineeringgroup at Tektronix, Inc. He is currently in-volved in designing, fabricating, and testinglarge CCD's for scientific applications.

CCD's FOR EVERYBODY?

Even though CCD's are almost perfectdetectors, they are not suitable for allastroilOmers. They are very expensive; a2,048-by-2,048-pixel chip presently listsfor over $80,000, or 19 cents per pixel!Devices with standard TV resolution rangefrom $2,000 to $10,000 (1.2 to 6 cents perpixel), depending on their quality.

CCD's generally remain beyond thereach of most amateurs, though the in-

creasing popularity and decreasing priceof lightweight video cameras is changingthis somewhat. Several Japanese manufac-turers sell CCD's for about $200 each.

(Cosmetic rejects go for about half thisprice.) But even chips this cheap are farmore expensive than photographic film.

Perhaps the most important disadvan-tage of CCD's is the amount of ancillaryelectronics needed to make them work.

Prime among these accessories is a power-ful computer to analyze the tremendousamount of data these tiny silicon chipscreate. A single image from a 2,048-by-2,048 CCD will contain 10 megabytes ofinformation. The amount of data gath-ered during a single observing run couldrapidly grow to one billion bytes or more.This must be stored, calibrated, and ana-lyzed - a very expensive enterprise.

Lastly, a CCD, like a racing car, re-quires a full-time engineering supportteam if it is to work at peak performance.

Nevertheless, an increasing number ofamateur astronomers are taking theplunge and experimenting with CCD's(S&T: January, 1985, page 71, and April,1986, page 407). This trend is sure tocontinue as CCD's and home computers

become cheaper and more powerful. Foramateurs and professionals alike, theCCD is truly a dream coming true. Withperformance improving continually it isdifficult to imagine what further impact it

will have on the astronomical com- ~munity in the next decade. ~

determined by directly measuring theamount of charge generated. This raisesan interesting possibility: in a single image(actually, many very short exposures super-imposed), a CCD can provide an X-rayastronomer with simultaneous high-reso-

lution spatial and spectroscopic data. No, other detector can do this.

BEYOND THE VISIBLE

In the past decade more and moreastronomers have turned their attention tothose parts of the electromagnetic spec-trum invisible to our eyes. Increasingly,observations are being made at ultraviolet(1200-3500 angstroms), extreme ultraviolet(120-1200 angstroms), and soft-X-ray (1.2-120 angstroms) wavelengths.

Photons in these short-wavelength por-tions of the spectrum are more energeticthan their visible counterparts. When ab-sorbed in the CCD's silicon lattice these

high-energy photons generate many elec-trons, the exact number depending on thewavelength (energy) of the radiation. Forexample, a 2.1-angstrom X-ray photonwill spawn, on average, 1,620 electrons.

For wavelengths less than 100 ang-stroms, therefore, a single photon can bedetected and its energy (or wavelength)

light and at best can achieve a photo-metric accuracy of only 5 percent over asmall portion of its already narrow dy-namIC range.

Vidicons are also highly nonlinear. The

relationship between their input andoutput signals must be approximated by acomplex mathematical expression that isdifferent for each point on the detector.This makes the extraction of photometricinformation difficult and time consuming.Even when the required procedures arefollowed, vidicon photometry is only accu-rate to about two percent.

In contrast, CCD's are linear 'to 0.1percent over most of their dynamic range,far superior to any other device. Thismakes it relatively simple to remove theeffects of pixel-to-pixel variations in sensi-tivity. In practice this is achieved bymeans of a technique known as "flat-fielding" (illustrated on page 241).

Vidicons and related detectors haveother serious flaws that limit their astro-nomical usefulness. They all form images

on a light-sensitive medium that isscanned by a beam of electrons just as ina television camera. The first problemencountered is to know exactly where thebeam is at any instant, for its positioncan't be measured directly. Thus the rela-tive positions of points in the image maynot correspond exactly to those in theobject viewed.

Compounding these positional uncer-tainties are the distortions introduced bythe magnetic fields used to focus thereadout beam. Taken together, these ef-fects compromise the fidelity with whichvidicons reproduce a given scene. Theseproblems do not apply to a CCD. Theposition of each pixel is fixed rigidly whenthe chip is manufactured.

\\ \ \'-

ExtremeultravioletX-ray

-Measured--Expected

10 100 1,000 10,000WAVELENGTH(angstroms)

Recent technological advances enablecharge-coupled devices to respond toradiation across the spectrum fromnear-infrared to soft-X-ray wavelengths.No other detector so efficiently respondsover such a broad spectral range.

100t'

-

~ 60lU