Previous: INTRODUCTION
Up: INTRODUCTION
Next: Operational requirements
Previous Page: INTRODUCTION
Next Page: Operational requirements

What CCDs are and how they work

A CCD is an array of photosensitive elements, each one of which generates photoelectrons in response to light and stores them as a small bucket of charge. When requested, the elements form a bucket brigade; each row of charges is passed from element to element down the columns and horizontally along the final row to be measured in turn and recorded digitally. Each picture element (pixel) is typically 20 to 30 m square. CCDs in use as astronomical detectors for the past decade have about 200,000 pixels, usually arranged in a format of about 500 rows by 400 columns, and the active area is thus about 1 cm. Successful observations have now been made with a new generation of CCDs 5 times this size, with 1500 to 2000 pixels.

The semiconductor technology which allows all this to happen is sketched in Fig 1.1.

The device is a silicon integrated circuit consisting of an oxide-covered silicon substrate upon which an array of closely-spaced electrodes is formed. These electrodes are made from polysilicon (polycrystalline silicon) which is transparent to red light. Photons imaged onto the surface penetrate the electrode structure into the substrate where they generate hole-electron pairs in numbers precisely proportional to the numbers of incident photons. The holes are conveniently lost by diffusion down into the depths of the substrate, while the electrons migrate rapidly to the nearest biased electrode where they collect as a single charge-packet. The array of `horizontal' electrodes above `vertical' charge-transfer channels (columns) defined by column isolators, or channel-stop regions, is known as buried-channel architecture, a design which permits rapid and efficient charge-transfer. A group of three electrodes between column isolators forms a pixel, the central electrode carrying the (positive) bias voltage during exposure, and defining the pixel centre by virtue of this bias. Fig 1.2 is the schematic representation of a very small CCD - 5x4 pixels - to demonstrate the charge-coupled aspect of CCDs and how frame-transfer is organised. To read out the CCD after exposure, the charge packets are shifted down the channels (columns) by taking the voltage on the adjacent electrode to a high level and then reducing the voltage on the first electrode. The process is known as clocking, and a three-phase system of voltage waveforms and electrode triplets is needed to achieve this (Fig 1.2). Three is the minimum number of phases: with double-electrodes and two phases, the charge packet would not know which way to go.

As each row reaches the bottom of the array it enters the horizontal register. This is structured (Fig 1.2) with triplet electrodes as for the signal pixels; and the horizontal clocking waveforms transfer the packets along this register in the same way. On reaching the end of the line, each charge packet is detected as a voltage across a capacitance, the voltage is amplified by an on-chip amplifier, and digitised with the resulting numbers placed sequentially on a storage medium such as disk or tape. The next row is driven down into the horizontal register, and the horizontal readout repeated. With the CCD cleared of charge in this way, the final array of numbers represents the image.

This digital or slow-scan mode of operation requires one to several seconds to read out a frame, the digitization process taking the time. Analogue operation is used for CCDs in television cameras, the original motivation for CCD development in the 1970s. Here the voltage developed at output by each charge packet is amplified and constitutes the video signal which is recorded directly on video tape. TV frame rate is 30 images/second, and the charge-clocking is thus required to be some 100 times faster; digitization takes far too long.

The readout time for a CCD is a significant time which is `wasted', i.e. not used to gather photons from stars. It and the amount of tape used can be reduced by defining a window on the CCD; only within the window will the data be digitised and stored. Charge from the area outside the window must be clocked off the chip, but is sent to earth and not digitised.

The exceptional efficiency and wavelength response of CCDs are illustrated in Fig 1.3. In the case of a front-illuminated CCD (Fig 1.1) the loss of efficiency at the blue end of the spectrum is due to absorption of blue photons by the electrodes before they can penetrate to the substrate. Redder photons have no such difficulty, but penetration into the substrate increases strongly with wavelength. Generally electrons generated further into the substrate diffuse back up to be successfully collected in the potential wells under the electrodes. However, with very deep penetration by the reddest photons, photoelectrons are progressively lost, and indeed infra-red photons pass straight through the substrate without generating any photoelectrons. To improve the blue response, the thinned, back-illuminated CCD was developed, in which the device is turned around so that the substrate faces the light and the blue photons are not faced with the immediate hurdle of impenetrable electrodes. The substrate is thinned to maximise the collection of photoelectrons by the now backward-facing electrodes. The remarkable feature of the design is that the thinning of the substrate does not seriously impair the red response.

The preservation of spatial resolution with wavelength is also surprising, given that the photon colour defines the depth in the substrate at which the photoelectrons are generated. The active region of the substrate is 5 to 10 m thick, a dimension which is not small in comparison with the pixel size of 20 to 30 m. In front-illuminated (thick) chips, there is some slight degradation of resolution with wavelength due to lateral spreading of photoelectrons before collection, but the effect is small enough to be very difficult to detect in astronomical applications. A second cause of resolution-loss is imperfect charge-transfer efficiency.In clocking out, a small proportion of charge is left behind at each transfer, this residual adding itself to the next charge packet following through the array. The effect is minimised by buried-channel architecture, for which the residual charge from each transfer is < 10 times the signal. However, it is clearly visible at low signal levels for most CCDs used by astronomers.

Dark signal decrees that CCDs must be cooled to cryogenic temperatures for use as astronomical detectors. The dark signal arises from electrons thermally generated in the substrate close to the collection area. At room temperatures, the dark signal is such that left to their own devices most CCDs would saturate in < 1 second. Not only would integrations be hopelessly short for astronomical purposes, but the electrons generated from astronomical objects would be overwhelmed by the numbers of dark-signal electrons by orders of magnitude. The dark signal is fortunately very temperature-dependent, following the exponential diode-law. Cooling via temperature-controlled liquid-nitrogen dewars or by closed-cycle refrigeration provides the crucial reduction in dark signal. The mobility of electrons is somewhat impaired by cooling, so that a compromise is required to maintain adequate charge-transfer efficiency; this compromise occurs at a temperature of about 150.

Careful temperature control of the CCD has a second benefit in that it stabilises the efficiencies of the pixels which are slightly temperature-sensitive. Individual pixel responses can be calibrated accurately, and this results in improved imaging and in excellent photoelectric (flux measurement) properties for the device.

Temperature control does not, however, inhibit cosmic rays, which, as they pass through the diodes, deposit large amounts of energy and therefore electrons in small clusters of pixels on the CCD. Control of CCDs against cosmic rays is only possible by limiting exposures, keeping the cosmic rays to manageable numbers and by repeating frames to compare one with another.

The charge-transfer process can be essentially noise-free, and almost all the noise contributed to the signal by the CCD is from the output stage, the voltage detector. This readout noise is from the combined on-chip amplifier and the effective capacitance at the front of this capacitor over which the individual signal charges develop their voltages. For example, the electrode structure of (front-illuminated) GEC chips results in a much smaller effective capacitance than (back-illuminated) RCA chips. Q = CV; therefore given similar amplifier noise, the same charge signal develops a much higher voltage on the GEC chip, which thus has exceptionally low readout noise of some 5 electrons per pixel, compared with 50 electrons per pixel for the RCA chip. (The efficiency of the RCA chip is higher, however, particularly in the blue; the interplay of efficiency and readout noise in terms of signal-to-noise ratio is considered below.) There are various tricks to minimise the readout noise, one of which is very slow clocking to allow long integration of the voltage generated by each charge signal. However, 1 noise from the output amplifier provides a limit to how slowly the sampling can be done.

At the low signal levels being digitised, hum and other electrical pick-up may make a significant contribution to the readout noise, especially as cables and connectors age.

If the charge-transfer process is not noise free, this can show as a non-linearity at low light levels. For this reason, CCDs are sometimes preflashed to push the signal above the non-linear threshold. Sky images, especially through broad-band filters, are never dark, however, so this problem is negligible with the INT prime focus camera, and no preflash facility is provided for CCDs using it.

The third major factor in CCD performance for astronomers is : how large a charge signal can be held by a pixel. The peak capacity is typically 200,000 electrons, and CCDs are demonstrably linear devices up to within a half of this limit. The arrival of further photons continues to create photoelectrons, which then tend to spread in either direction along the column. Together with readout noise, the saturation level defines the dynamic range of a CCD, crudely, as (saturation-level)/(readout noise). This ratio is 200,000/50 or 4,000 for RCA CCDs, and 40,000 for GEC CCDs, the latter corresponding to 11.5 mag. This assumes working down to the 1 level, something which astronomers would never do. In practice, however, dynamic ranges as large as 8 mag have been used.