|
September 1999 |
Getting the Right CCD |
||||||
|
|
|
|
Recent innovations in the processing and design of charge-coupled devices (CCDs) have allowed some price-performance barriers to be knocked down. A review of the technology and selection criteria for these scientific imagers is timely. Selection of the right CCD requires a close look at electrical performance. Several parameters, such as those on the selection chart (Table 1), should be considered. Though this list is not exhaustive, it is representative of the variety of devices available. The values of some of these parameters have been measured under differing conditions: e.g., RMS noise for the FT18 was measured at 25 frames per second read rate, while for the SI502 it was measured at 0.2 frames per second. The process of reading the information from a CCD is subject to a limit on noise performance called readout noise. Because pixels are read sequentially, the reading of a new pixel requires clearing the charge from the one that precedes it. This reset operation never quite goes the same each time, so there is a small but still significant uncertainty. Generally CCDs designed to optimize noise performance will have noise floors on the order of 5 electrons (e-), while others designed with speed of readout in mind may have higher noise floors, from 15 to 35 e-. Applications requiring the utmost sensitivity would require these lower-noise devices. The drawback to them is that they must be read out relatively slowly, and this may make them unsuitable for applications requiring high temporal resolution. High frame rates require high-speed readout, but speed can come with a heavy penalty: higher read noise. A tradeoff between speed and noise may be required in the selection process. CCDs designed for very low-noise operation are among the slowest offered. To circumvent this limitation some are available with multiple outputs, allowing readout of the image information through parallel identical read electronics. This multiplies more than just the read rate, however: the price tag for the system can get multiplied quickly. But it is possible to compromise a bit and still get good performance at a good price. For example, the Orbis 2 from SpectraSource Instruments (SSI) can accommodate CCDs with 1, 2, or 4 outputs, allowing a device with fine noise performance to operate at effective read rates much higher than possible using a single-output design. On the other hand, if speed is the overriding consideration, there is a good selection of relatively fast CCDs available. Devices from several manufacturers have 1000 ´ 1000 resolution and read rates high enough to support 25 or even 30 frames per second. At these rates mechanical shutters are impractical, since few available can even achieve these rates, and those that do have short lifetimes. For example, a shutter with a lifetime of 1 million operations will last only about nine hours at 30 frames per second.
Dark current is a source of signal inside the silicon of the CCD
that is time- and temperature- The dark current is sometimes incorrectly considered to be noise, but since the signal part is repeatable, it can be subtracted from image data. If not subtracted, it does add its own perhaps noisy-looking contribution to the image, but almost all of this disappears when a dark field subtraction is performed on the image. If integration time is long enough, it is possible for this signal alone to saturate pixels, so for low-flux applications this is an important parameter. Within this signal there is noise in the form of shot noise (as there is in the photon signal we want to measure). The noise inside this signal is random and therefore does not repeat. So it cannot be subtracted away, rendering a perfect image. But its influence on image quality is not great, because the shot noise varies by the square root of the signal.
Figure 2 shows the performance of a SpectraSource Teleris 2 CCD camera equipped with the Kodak KAF0401E CCD. With a Peltier thermoelectric cooler, this system has excellent dark current performance, operating at a temperature that results in it having a dark-current rate of about 0.5 electron per second per pixel. The noise floor of this camera is plotted at a time-independent 13 electrons RMS, typical for these systems. Note that the signal shoots off the graph very quickly, but this would be subtracted from the image data anyway. The important thing to note is the point of the crossover between the shot noise in the dark signal and the read noise of the CCD. The length of time required for this to occur could be considered a figure of merit for the CCD camera, which tells us how long an integration time is possible before the shot noise dominates over the read noise of the camera. The longer this takes, the better the camera. For the Teleris with a Kodak KAF0401E, this time is more than 300 seconds. Another consideration is full-well capacity, a measure of the maximum signal that a pixel can hold. While the value for full well is highly dependent on a number of factors, the general rule of thumb is the bigger the pixel the bigger the full well. Applications requiring wide dynamic range or the best signal-to-noise ratio possible in high-illumination settings would tend to need CCDs with the highest possible full-well capacity. CCD technologies have evolved into three main design types that address quantum efficiency (QE), the ratio of incident photons to detected photoelectrons. These are: Front illuminated: the standard method of CCD fabrication. Photons enter the silicon by passing through the transparent overlying layers of the integrated-circuit elements from the front of the device. The advantages are low cost and good selection. The disadvantages are low quantum efficiency -- at best 30 or 35 percent peak, and low blue and no near-ultraviolet (NUV) response. Enhanced front illuminated: included are the Texas Instruments Virtual Phase and Kodak Blue Plus processes (see Figure 2). By enhancing the geometries and materials used in the normal front-process CCDs, some of the losses of front illumination can be reduced, thereby increasing the peak quantum efficiencies to as high as 75 percent, and extending response into the near-UV. This technology represents the best price-vs.-performance tradeoff for the types discussed here. Advantages are low cost, though not so low as standard front-illuminated, and improved quantum efficiency. A disadvantage is that selection is limited. Back illuminated: By building the CCD as in the standard process, but turning the device over and allowing the incident radiation to enter from the rear side, the losses incurred in a front-illuminated CCD are greatly reduced. The advantage is high quantum efficiency -- some manufacturers report it as high as 90 percent. The disadvantages are higher cost and poorer selection. Some manufacturers offer special coatings to enhance spectral response of CCDs. The two common coating types are: Antireflection coatings: AR coatings reduce losses at surfaces and in thin layers, thereby increasing quantum efficiency. These coatings can be "tuned" or optimized for certain wavelength ranges, enhancing the visible or NUV response as desired. For example, see the curves for the SITe CCDs in Figure 2. Fluorescence: These coatings, commonly known as lumogen and used only with front-illuminated devices, enhance response only in the short-wavelength part of spectrum. They do this by taking radiation at a wavelength that the CCD is not sensitive to and converting it to a longer wavelength that the CCD is sensitive to. The response is extended into the shorter wavelengths, but the improvement amounts only to about 10-15 percent. For example, see the curves for the Kodak CCDs in Figure 2. Some CCDs are made with features that reduce blooming. This is the reaction of a CCD to very bright illumination, where vertical spikes of white "bloom" from bright points in the image. Blooming occurs when a pixel or pixels are overfilled, beyond their full-well capacity, usually by several times. Antibloom reduces this effect considerably, sometimes to many thousands of times full-well signal, but with a price: loss of linearity in response, or a reduction in active area in a pixel, known as fill factor. There are several different methods of controlling the duration of exposure or integration time. These are: Interline transfer: CCDs designed for video applications use this technology almost exclusively. The good thing about these devices is that they can be "shuttered" very quickly, in the microsecond time regime or faster. This makes them extremely powerful sensors for comparative measurements where two events are separated by only a tiny slice of time, or where very short exposures are required. The drawbacks to this technology are that it has less than 100-percent fill factor, reducing quantum efficiency and introducing potentially negative optical effects. Fill factors are generally about 70 percent, and QE can be down around 15-20 percent. Kodak and Sony supply several devices in this category. Frame transfer: This type of CCD is a cross between an interline device and a full-frame device. It has the interline CCD's ability to electronically shutter, but has the full-frame's fill factor and QE. The electronic shuttering is not as fast as the interline device's, since it requires a complete vertical clock cycle for every line the imager has in the Y direction. This means that a 512-x-512-pixel device needs 512 vertical clocks to completely shutter for a full image exposure. Since the shuttering is accomplished by transporting an image into a storage area, there is a possibility that significant signal will fall on the image while it is moving into the storage area, resulting in smear. Because the image and storage areas are controlled separately, some interesting control methods can be used to achieve high frame rates along with low noise. Pipelining subarray images through the storage area can result in rates of 200 per second or higher, while holding the noise floor down to 10 e- RMS, making possible applications in adaptive optics or similar areas. Manufacturers that supply these types of devices are Texas Instruments, Thomson CSF, and Philips, among others. Full-frame: This technology has no intrinsic shuttering features, which means that unless some means are used to block the incident flux, photons will fall on the image area during the entire readout period. This usually results in unacceptable levels of smear, and so necessitates the use of an electromechanical shutter. Since no features of the architecture are used for exposure control, these devices have high QE and fill factor. They are also less expensive than their frame-transfer equivalent, since there is no doubling of the number of pixels for the storage array. Many manufacturers offer full-frame CCDs, among them Kodak, Texas Instruments, Thomson CSF, Philips, EEV, EG&G, and others. Among optical considerations is fill factor, the ratio of the actual collecting area of a pixel to its overall geometric area. Though at first it might seem that these would be equal, some technologies such as the interline transfer CCD reduce this ratio from unity. Typically, fill factors of 70 percent are about the best currently available with interline CCDs. Besides the loss of efficiency, reduced fill factor causes other deleterious effects: aliasing of spatial frequencies and reduction of contrast at high spatial frequencies are typical problems. Antibloom features can also reduce fill factor. Typically, CCDs with antibloom will use a drain electrode that must be positioned next to each pixel to sweep off the excess charge created by high illumination levels. This drain blocks or reflects light, and hence lessens fill factor. Some specialized processes, such as that used by Texas Instruments, eliminate or greatly reduce these losses through the use of transparent materials for the overflow drain, so that not all CCDs with antibloom will suffer from reduced fill factor. The answer to the question of what pixel size is right depends on several things unique to an application. The primary consideration is usually spatial resolution, but to select a pixel size several things need to be nailed down. In determining the required spatial resolution, the user should be mindful of the deleterious effects of too large a pixel (spatial resolution too low), such as aliasing, reduced modulation transfer function (MTF), etc. It is better to err on the small side. If the focal length of the optics to be used can be adjusted as required, then selecting based on pixel size is easier. A 12-mm pixel used with a 100-mm focal length achieves the same spatial resolution as a 24-mm pixel and a 200-mm focal length, but at a lower cost, since bigger pixels usually implies higher prices. If the optical configuration is fixed, however, then the pixel size must be adjusted to achieve the target resolution. This may or may not result in a practical solution. If, when all is considered, a 3-mm pixel is needed, there will be a very small list of potential devices to choose from. Conversely, if a 50-mm pixel is needed, there are few devices so big. But in this case, a technique known as binning can synthesize big pixels from small. The camera selected must support binned operation for this to be possible, so the camera specs should be checked. One the required spatial resolution has been determined, the array size can be calculated. This is pretty straightforward: it is simply the ratio of the linear dimension of the image area and the pixel size. For example, if the pixel size is 10 mm, and an area of 5 mm x 5 mm must be imaged, then a 512-x-512-pixel array would be fine. Budget may determine the choice of array size, since big pixels can have big price tags. The biggest CCDs can be as large as 4000 x 4000 pixels. SpectraSource Instruments (SSI) has supplied thousands of camera to researchers, scientists, and engineers for diverse application environments around the world. Below are listed some general application areas, along with recommended camera and CCD configurations. Radiography: Medical and industrial radiographic applications will have differing requirements that affect the selection process. In this digital form of x-ray imaging, for some systems lens-coupling the camera to the phosphor output will be adequate, but in other applications fiber-optic coupling is needed to insure the highest coupling efficiency. SSI supplies either using many different CCDs on either the Teleris or Orbis platforms. Biomedical: Frequently the application in biomedical imaging involves low flux conditions, as in green fluorescent protein or fluorescent in-situ hybridization fluorescence imaging, so low noise and low dark current are of importance. Also, these applications are often microscope-based imaging, so in these areas a smaller pixel size can be an advantage. Spectroscopy: Often, as in Raman applications, high quantum efficiency and low dark current are of primary concern, since flux rates can be extremely low. In this case, back-illuminated CCDs or those with newer enhanced front-illuminated technologies like Kodak's Blue Plus and Texas Instruments' Virtual Phase CCDs would be the sensors of choice, because of their high quantum efficiencies. Special CCDs optimized for spectroscopy, such as rectangular arrays, are available in back-illuminated configurations. Supercooled dewar systems using LN2 can be configured for the CCD of choice, insuring extremely low noise levels even in the long-duration integrations commonly required in this field. Astronomy: This can be a challenging arena, both in the selection process as well as in the performance of the camera system. For deep-sky work, LN2 systems with low noise and high quantum efficiency are the systems of choice, while in planetary imaging a simpler, less expensive Peltier thermoelectrically cooled system could be used. For more information contact Stephen B. McArthur, president of SpectraSource Instruments, 31324 Via Colinas, Ste. 114, Westlake Village, CA 91362; (818) 707-2655; fax: (818) 707-9035; E-mail: spectrasrc@earthlink.com; www.spectrasource.com.
|
