TWENTY years ago, two engineers, at Bell Laboratories in New Jersey invented a novel electronic device based on silicon that was to become an important scientific tool. Willard Boyle and George Smith created the charge coupled device, or CCD. This detector responds to invisible light and X-rays and, using the latest electronic gadgetry, it can record even the faintest image.
Today, scientists use CCDs in many scientific disciplines, for example, astronomers employ them to record the images of faint quasars, while physicists use them to detect exotic subatomic particles, and even chemists and biochemists are beginning to exploit CCDs in spectroscopy and to detect X-rays from diffraction patterns (see last week’s feature, ‘Particle detectors come out of the laboratory’). Now scientists are moving on, they are developing new silicon devices to improve on the CCD.
Advertisement
A typical CCD is a silicon chip, with an area of about 1 square centimetre. Under a microscope, it looks like a draughtboard of some 250 000 tiny squares surrounded by a few metal tracks to provide voltages. Each square corresponds to one picture element, or pixel. The surface layers of the CCD are constructed so that when you apply the appropriate voltages, each pixel attracts electric charge – usually ‘free’ electrons that have become detached from atoms in the silicon. Any electrons caught in a pixel (about 15 square micrometres) can remain there for a long time. So, in effect, the whole array stores the pattern of charges.
When light strikes a CCD, it releases atomic electrons which move into the pixels. The number of electrons depends on the amount of light arriving, so recording an image on a CCD is like forming an image on photographic film. The difference is that you can use a piece of film only once. The CCD, on the other hand, can be used again and again. This is because you can move the pattern of charge corresponding to one image out of the array of pixels, and record it more permanently while the next image forms. If you vary the voltages in the correct, regular way, the electrons move along, one pixel at a time, until they leave the array at a single point. The original two-dimensional array of captured electrons becomes transformed into a long chain of ‘buckets’ of electrons.
The buckets from the row nearest the outlet lead the way, followed by those from the next row, and so on, until the whole array is cleared. As the buckets leave the array, the charge in each of the buckets generates a small voltage at the output of a transistor. This process converts the image into a chain of voltage pulses. Each voltage pulse is proportional in size to the amount of light that arrived at each pixel on the CCD.
The first CCDs consisted of a single line of a few pixels; since then, researchers have developed longer linear CCDs for a variety of applications. But it is the two-dimensional CCDs with their capacity for storing large amounts of information that have probably had the most impact in science. By 1973, CCDs with large areas of more than 10 000 pixels had become available. Soon afterwards, researchers at NASA’s Jet Propulsion Laboratory in Pasadena, California, realised that these tiny chips might replace bulky, conventional TV cameras on spacecraft. But it was not only the small size of the CCDs that made them so interesting for astronomy.
Light arrives at a camera like a stream of bullets from a machine gun, with the bullets being tiny packets of energy, or quanta, known as photons. Photographic film records at best only a few per cent of the arriving photons; a CCD, by contrast, produces a signal for some 70 per cent of the photons. Such a high efficiency means that a CCD can record images much more quickly than photographic film. This is an important factor with telescopes, which are expensive to run, and where there is great competition for time.
A CCD can also record a wide range of light levels, from the very faint to the very intense: its ‘dynamic’ range is hundreds of times as great as that of photographic film. The top end of this range is limited by the number of electrons that the pixels can store – typically several hundred thousand electrons. The bottom end of the dynamic range is limited by ‘noise’ – the charge that collects in the pixels even if no light falls on the CCD.
This extra charge arises partly from the thermal ‘jiggling’ of the atoms, which shakes electrons free. One way to reduce the noise, and so extend the lower limit of the dynamic range, is to cool the CCD and reduce the thermal vibrations. Astronomers often cool their CCDs to temperatures of -100 Degree C or so.
The first CCD images in astronomy were of planets, and were recorded in 1976. Since then, the use of CCDs in astronomy has grown enormously, not only in optical astronomy (at visible wavelengths) but also in X-ray astronomy. In the optical region, CCDs can reveal faint objects, so allowing astronomers to peer more deeply into space. In X-ray astronomy, the strength of the signal from each pixel indicates the energy of the X-rays arriving at different parts of the image.
One problem astronomers face in making long exposures with CCDs on spacecraft arises due to the constant bombardment by energetic subatomic particles – cosmic rays – issuing from the Sun and other stars. Charged subatomic particles, such as protons and electrons, ionise the silicon, releasing atomic electrons which accumulate in the pixels.
The number of cosmic rays arriving at the tiny area of a CCD each second is small, but so is the number of photons when astronomers are trying to see a particularly faint object. The flow of cosmic rays, therefore, helps to set a limit on the faintest objects that astronomers can observe with CCDs.
Tracking particles in space
But what is a problem for the astronomer proves to be an advantage for the particle physicist. If CCDs respond to charged particles, then why not use them to detect subatomic particles, such as protons and pions? This is what a team, led by Chris Damerell at the Science and Engineering Research Council’s Rutherford Appleton Laboratory in Oxfordshire, succeeded in doing.
CCDs have an advantage over some particle detectors because they record true ‘space-points’ on the tracks of charged particles – that is, they record in two dimensions simultaneously. This contrasts with several other kinds of silicon ‘strip’ detector which record only in one dimension, corresponding to a row or a column of a CCD (see Figure 1). With these detectors, a point on the track of a particle can be calculated only by using two detectors arranged to give measurements in two dimensions, but at separate points in space. This can lead to ambiguous results when many closely-spaced tracks pass through the same detector. CCDs avoid this uncertainty and are, therefore, inherently more accurate.
In 1980, Damerell and his colleagues first proved the viability of using CCDs as particle detectors in tests at CERN, the European Laboratory for Particle Physics near Geneva. Their aim was to work with CCDs developed for TV cameras by GEC’s Hirst Research Centre in London. To succeed, Damerell’s group had first to overcome the major drawback in using CCDs as particle detectors: the time it takes to shuffle the information in all the pixels off the chip.
You can run the pixel ‘conveyor belt’ quite quickly but the electronic techniques needed to reduce noise in the signals as they come off the CCD tend to slow it down. These tricks are not necessary if the number of electrons per pixel is high enough, as is the case in TV images where a deluge of photons arrive at the CCD. But astronomers and particle physicists deal with a smattering of photons and particles, so the numbers of electrons released are very low. Cooling to reduce thermal noise helps, but the electronic trickery is also essential to reduce what is known as ‘reset’. This noise arises from the output transistor.
Reducing the noise involves subtracting the average voltage that it produces from the total voltage from each pixel. The result should be a voltage corresponding to the number of electrons deposited in the pixel by the photon or particle. Astronomers have made this technique, known as correlated double sampling, work very successfully. But, in a typical device, it takes as long as 40 microseconds per pixel, or 10 seconds to read all the pixels. This is far too long for experiments in particle physics, where beams from particle accelerators can produce thousands of interesting ‘events’ every second. However, Damerell’s team has succeeded in adapting the technique to make it much faster, so that they can read all the pixels on a CCD in about 50 milliseconds.
In 1985, Damerell and his colleagues introduced two CCDs into a large experiment at CERN, code-named NA32. The aim was to study the particles produced when a beam of high-energy particles struck a copper target, and to search for short-lived particles that decayed into other particles almost as soon as they had emerged from the target. The CCDs were the first detectors after the target, followed by planes of silicon strip detectors, and then by larger detectors to track the particles and to help to identify them.
CCDs reveal short-lived particles
The CCDs proved their worth in resolving the multitude of tracks emerging from the target, and especially in revealing sprays of particles originating beyond the target. Such sprays are the telltale signs of the decays of the short-lived particles that the experimenters were seeking.
Damerell’s team, working with Steve Watts and his colleagues at Brunel University, are building a complex detector containing 480 CCDs. In this detector, the CCDs will be built into four concentric ‘barrels’, each built from delicate ceramic ‘ladders’ which will support the CCDs. The barrels will form the central part of the much larger detector, which will be used to study particles produced in the collisions of electrons and positrons (antielectrons) at the Stanford Linear Accelerator Center (SLAC) in California.
Despite their many applications, silicon CCDs have drawbacks. As X-ray detectors, for example, they are most efficient only over a relatively small range of X-ray energies, from 1 to 5 kiloelectronvolts. As particle detectors, they produce only tiny signals, and this makes it difficult to read them out.
As detectors of infrared light, they are useless because silicon is transparent at these wavelengths. Nevertheless, researchers have successfully made infrared detectors by combining two kinds of pixel device. The conventional CCD made of silicon reads out signals from another pixel device built from a material that is sensitive to infrared radiation, such as silicon ‘doped’ with small amounts of arsenic. In such a hybrid device, the CCD lies face-to-face with the detector chip, and the two chips are interconnected pixel by pixel. This approach means that you can build the complex structure of the CCD array on a familiar material that is easy to work with – silicon – while keeping the architecture on the more exotic, less familiar material fairly simple. But there is the disadvantage of having to make about a million connections between the detector chip and the CCD.
Researchers in the US have taken this type of hybrid technology further by replacing the CCD with a ‘random-access’ chip to read out the underlying detector. With a CCD, all the pixels are read out in sequence, which takes time. Time may also be wasted reading out pixels that do not contain any information. Some applications where images change rapidly require a more rapid read-out. In this case, it would be useful to gain access to only those pixels that have registered a signal – in other words, the pixels that contain more electrons than the basic noise level.
At the University of California’s Space Sciences Laboratory, in Berkeley, John Arens – who pioneered the use of pixel arrays for imaging in infrared astronomy – has been working with infrared pixel detectors attached to random-access read-out chips which have been built by the Hughes Aircraft Company. (By far the most money for research into imaging at all wavelengths comes from budgets for defence R & D) These read-out chips contain an array of transistors that can be turned on by signals threading the array along ‘address’ lines. Once selected, a transistor links its detector pixel to the read-out circuitry. In this way, you read out only the pixels you address.
Arens and his colleagues built an infrared camera and an infrared spectrometer based on hybrid pixel devices. This led to an exciting result – the discovery of a new interstellar molecule, acetylene. But now, Arens and Garret Jernigan, from Berkeley, and Stephen Gaalema from Hughes, have joined forces with Steve Shapiro and William Dunwoodie from SLAC to try to build a similar device for detecting particles.
They want to build a detector to cope with the challenging conditions that will exist at the Superconducting Super Collider (SSC), the giant particle accelerator that the US plans to build in Texas. At the SSC, particle beams will collide more than 60 million times a second, and something interesting may happen as often as 60 000 times a second. This implies that all the information from an interesting collision must be recorded within about 15 microseconds, far less than the time its takes to read a CCD. However, particle tracks will often be clustered in relatively small regions. The ability to ignore most of the detector would greatly speed up the read-out, and so overcome one of the inherent problems with CCDs.
Arens and Shapiro also intend to improve on the CCD by using a structure that produces larger signals. Large signals are easier to handle, and if they are substantially greater than the thermal noise levels, there is no longer any need for cooling. To produce big signals, the pixels should capture as many as possible of the released electrons when a charged particle passes through the silicon. The problem with a CCD is that the electrons migrate to the pixels from only the top 10 to 15 micrometres of the device, even though the chip may be 200 to 300 micrometres thick.
This thin upper layer of a CCD is the so-called depletion layer. Here all the charge carriers normally present in the silicon have been cleared out by the electric fields that the applied voltages set up. When a charged particle passes through, only those electrons released in the thin depletion layer gather in the pixels. Electrons released in the remainder of the CCD, free electrons, wander randomly before recombining with silicon atoms.
The thickness of the depletion layer depends on the electrical resistance of the silicon; the higher the resistance the deeper the depletion. However, conventional integrated circuits such as CCDs are built on silicon with low resistance, hence the narrow depletion layer. This feature of CCDs is not a problem with photons of visible light, which release many electrons, but it does restrict the use of CCDs as X-ray detectors. At energies above about 5 kiloelectronvolts, an X-ray releases too few electrons in the depletion layer to provide a useful signal. The solution would be to make the depletion extend through the full thickness of the chip. Such a fully-depleted pixel detector be useful not only in particle physics, but also in X-ray astronomy and the many areas of science that use X-rays to study the structure of materials.
Arens and his colleagues hope to overcome these problems by using different chips for detection and read-out, connected by bumps of the element indium (which are also used in the infrared devices). The detector chip would consist of high-resistivity silicon – that is, silicon with a high electrical resistance – 300 micrometres thick, with appropriate structures to collect electrons. The read-out chip, by contrast, would consist of silicon with a low resistance, of the kind suitable for conventional transistors.
Some researchers hope to eliminate the complexity of hybrids by manufacturing devices on high-resistivity silicon, so that the detector and read-out circuitry could be sited on the same chip. One of the difficulties in such an approach lies in the need to minimise the impurities in the silicon, because these modify its conduction properties. Steve Holland at the Lawrence Berkeley Laboratory has built detectors with a process that involves ‘gettering’ or trapping the impurities. He provides a gettering layer on the ‘back side’ of the chip – the side without the structures that define the detector elements. This layer is made of polysilicon doped with phosphorus, both of which trap impurities.
David Nygren and colleagues at the Lawrence Berkeley Laboratory, again with their sights on the SSC, are building on Holland’s work in their bid to develop a fully integrated pixel device that combines read-out with detection on the same chip. They have also proposed an intriguing scheme whereby they can unscramble the ‘hits’ caused by particles from several different collisions (see the Figure below). This would allow the detector to record over consecutive time intervals until a signal from other pieces of apparatus told the detector which intervals contained the interesting information.
Meanwhile, in Munich, Gerhard Lutz, at the Max Planck Institute, and Josef Kemmer, at the Technical University, have proposed a new type of transistor that sits on a substrate of depleted silicon. In their design, the underlying structure ensures that the bulk of the silicon is fully depleted. The transistor itself behaves much like a normal transistor. The crucial difference is that the electric fields in the depleted silicon should guide electrons released by a charged particle to a location beneath the transistor. Once there, this accumulated charge increases the current that flows across the transistor when a voltage is applied to a central ‘gate’ electrode. The ‘depleted transistor’ is therefore a pixel that can read itself.
To create a pixel device, Lutz and Kemmer propose using an array of ‘depleted transistors’ each with two control gates. One gate of each transistor would be connected to all the others in the same row; the second gate to all the others in the same column. To turn on a particular transistor – to find out how much charge is trapped in a pixel – you would pulse the appropriate row and column gates at the same time.
In another variation of their pixel device, Lutz and Kemmer take the concept of charge storage literally into another dimension. Their basic idea is to create a three-dimensional grid of pixels. The bulk of the substrate, fully depleted for particle detection, would lie below this grid. The problems in making such a three-dimensional device will be severe. At present, the structures on chips are built in the top 10 micrometres or so of the silicon wafer. But the multi-layer pixel device of Lutz and Kemmer illustrates just how far the imagination can take the original concept of the CCD.
Remembering that the CCD took 10 years to come of age, it seems likely that a whole range of pixel devices will be available 10 years from now. The development of such detectors also shows that new technology can quite quickly make an impact on how fundamental research is carried out.