The role of charge coupled devices in low light level imaging


A technical guide from PixCellent Limited


Summary

This booklet is intended to provide a simplified guide to the principles and practice of low light level imaging systems. In particular it gives an introduction to the properties of charge-coupled devices (CCDs), their evaluation and suitable applications. Finally, details of typical performance figures are given for PixCellent cooled CCD imaging systems.


Low light level imaging


Introduction

Electronic imaging light detectors such as television cameras have been in existence for over sixty years. All imaging systems are characterized by a few simple parameters. They include the size of the sensitive area, the number of picture elements (pixels) and the range of light levels that the detector can work with. There is always a high light level above which the detector is saturated and a low light level below which a useful image cannot be obtained. Domestic television cameras usually have their sensitivity defined in terms of the lux. For quantitative work the lux is a very unsatisfactory unit since it is defined with a broad spectrum light source rather like a domestic tungsten lamp which has a significant proportion of its output in the near infrared region of the spectrum. This means that detectors with a lot of far red sensitivity will be quoted as having a good sensitivity in terms of lux. Very roughly, low cost monochrome TV cameras that use CCDs are often able to work down to 0.1 lux or better which corresponds to photon arrival rates (photons are the elementary light packets or quanta that make up any image) of about 10^10 photons per second per square millimetre. Colour CCD cameras have sensitivity limits in the region of 10 lux. In the context of this guide these are very high light levels indeed. We shall next look at the different imaging technologies to compare their strengths, weaknesses and sensitivity limits as illustrated in Figure 1.

Figure 1: Sensitivity limits of light detectors


Unintensified imaging detectors


Unintensified vidicons

These are a form of TV camera tube where the image that falls on the target progressively discharges the target. A scanned electron beam recharges the target each time the image is ‘read out’. Some targets have high sensitivity (the efficiency with which light falling on the detector is converted to output signal) and because the target is scanned it is possible to read out a vidicon with a very large number of pixels. This is not the same as high resolution, and it is important never to confuse pixel size and numbers with image resolution. The resolution of vidicons is fairly generally poor. In addition, the target recharging mechanism is highly non-linear so it is difficult to use vidicons for accurate photometric work. The low light level sensitivity limit for vidicons is set by internally generated noise sources which occur even in the absence of an input signal (dark current and read-out noise). This limits vidicon sensitivity to around 10-3 lux at best.


Unintensified charge coupled device TV cameras

We shall look in detail at the internal structure of CCDs later in this guide. They consist of a slice of silicon that absorbs photons and directly generates a charge (one electron per absorbed photon). The generated charges are held in place by an array of electrodes and the signal is read out by transferring the charge in each pixel to the device output, one pixel at a time. At low signal level the performance of a CCD TV camera is limited by the internally generated dark current and by the noise generated in the first stage signal amplifier that is built as part of the CCD chip itself. The best monochrome CCD TV cameras will work down to about 10-2 lux. The CCD response is highly linear, although CCD cameras are often designed with an intentional electronic non-linearity that gives a cosmetically more appealing picture on a TV monitor. Their resolution is largely set by the pixel size and their sensitivity is very high. By virtue of being solid state integrated circuits they are rugged and highly resistant to damage from light overload.


Unintensified photodiode arrays

These are included here because they are used quite widely for spectroscopic applications and are being increasingly displaced by CCD systems. They are generally one-dimensional devices, often with long, thin pixels (2.5 mm by 25 microns is common). They have similar advantages to CCDs in that they are solid state devices and are therefore rugged. The large area of their pixels gives a high dark current at room temperatures and even if the device is cooled and in the absence of signal or dark current, the readout structure generates a rather high readout noise.


Intensified imaging detectors

In trying to achieve low light level performance with any of the above systems we find we are limited by sources of noise that are generated internally by the detector itself. Clearly if we can amplify the signal we wish to detect and couple that intensified signal to the detector then we should be able to work at lower light levels. This can be achieved by an image intensifier which is placed in front of the detector and consists of a photocathode which emits many photons or electrons for each incident photon.

Image intensifiers produce an improvement in sensitivity which can be as high as a factor of 1000, sometimes more. However, the intensifiers have a number of consequences on the overall performance of the system.

Firstly, the light detection efficiency of the photocathode is often poorer than that of the detector to which it is coupled. Second, the overall luminous gain of the intensifier plus the image coupling to the detector itself has to be carefully selected because too high a gain will cause the detector stage to saturate on a relatively low input signal consisting of too few photons to produce an image of acceptable quality. Third, an intensifier reduces the dynamic range of the detector (the ratio of the strongest detectable signal to the weakest detectable signal in a single image) although with care this effect can be minimised. Fourth, with all intensifiers there is a lot of signal-induced background which limits the dynamic range to a level that can be as low as 100:1. Clearly these effects reduce the overall imaging quality of the system and imperfections in the intensifier itself (photocathode non-uniformities, geometric distortions and resolution) will be added to those inherent in the unintensified detector. However, intensifiers are generally capable of making much lower light levels accessible to the scientist, albeit at a considerable increase in overall system cost.


Intensified vidicons

The most commonly used intensified vidicon is the Silicon Intensified Target (SIT) vidicon which encapsulates a single stage intensifier inside the vacuum envelope of a standard vidicon. This gives a signal gain of about 2000 and allows work down to 10-4 lux. As with all vidicons the linearity is poor. The resolution is reduced although it is possible to purchase units with fairly high resolution. Further improvements in sensitivity are possible by gating off the readout for several frames while letting the stored image integrate up before reading it out. This cannot be used too much as the intensifier photocathode itself suffers from dark current which limits the sensitivity that may be achieved.


Intensified CCD TV cameras

As with the SIT camera, adding an intensifier to a standard CCD TV camera gives a great improvement in low light level sensitivity and levels of 10-5 lux may be achieved with careful component selection. As with the SIT camera the intensifier has lower detection efficiency (photocathode rather than silicon) and both resolution and dynamic range are degraded.


Intensified photodiode array

Use of a high gain image intensifier allows the high readout noise of the unintensified photodiode array (PDA) to be overcome. Intensified PDAs are widely used for spectroscopic applications where the importance of the gain in sensitivity more than offsets the lower detection efficiency and poorer resolution of the system.


Video signal handling

Although not strictly a low light level issue it is important to think a little about the way the image output signal is to be used for the application in mind. Both vidicons and CCD TV cameras (intensified or not) produce standard video output signals. These consist of 525 or 625 lines of picture information every 30 or 40 milliseconds. They have a big advantage in that image changes such as motion may be displayed immediately on a standard TV monitor. Computer cards known as frame grabbers allow a single TV frame or a sequence of frames to be digitized and passed to the computer software analysis package. This has several consequences:

· The image can never have more pixels than the roughly 700 x 500 format of the image.

· The high pixel rate (up to 10 MHz) means that the digitizing is usually done to no more than 8 bits (256 levels) even though the video signal may contain higher quality information (it usually does not, however).

· The fast readout is often a source of noise limiting the useful range of the data.

One method of improving the signal to noise that can be achieved from a single video frame is to use a frame grabber that allows a series of consecutive frames to be co-added and averaged. This can significantly improve the signal to noise ratio in proportion to the square root of the number of frames averaged. At first sight this can seem an attractive option. However, most TV cameras are manufactured so that for single frame operation the camera performs well. As soon as many frames are averaged the summed image can show other fixed pattern noise that cannot be suppressed by averaging. Frame averaging will only work if truly random noise limits the performance of the system. Video-rate cameras usually suffer from many other noise sources and these are not improved by frame-averaging.

Remarkable claims are often made for video-rate systems with frame averaging. The potential customer is strongly encouraged to ask for a clear demonstration from a supplier that their system has the sensitivity and dynamic range required by their application.


Cooled CCD systems

The CCD is clearly attractive as an imaging detector because it is rugged, compact, has good resolution, excellent linearity and high detection efficiency. By cooling the CCD, by slowing down the read-out of the CCD and by breaking free from the restrictions of TV output format (analogue video signals), a much better noise performance can be achieved, higher resolution images can be obtained and dramatic sensitivity improvements are possible, albeit at a longer cycle time between images. This approach to achieving very low light level sensitivity without using an image intensifier was originally developed for astronomy.

At room temperatures, standard CCD TV cameras generate a dark current that is very high - often hundreds or thousands of electrons per pixel per second is typical. If the CCD is cooled, the dark current reduces roughly by a factor of 10 for every 20oC. Typical figures are 10 electrons per pixel per second at -40oC, and less than one electron per pixel per hour at -140oC.

The next barrier to achieving very low light level performance is the noise that is generated when the CCD is read out at video rates (i.e. pixel rates of several megahertz). If the readout rate is reduced it becomes possible to use special electronic signal processing procedures (called double correlated sampling) to give a read-out noise of only a few electrons. This can to be compared with a read-out noise often several hundred times higher at TV read-out rates. The net effect of using cooled CCDs and slower read-out rates is that it is possible to achieve limiting sensitivities of 10-10 lux and below at full image resolution.

 


Principles of CCD operation


Physical characteristics

A charge-coupled device (CCD) is an ultra-sensitive light detector made primarily of silicon and manufactured using similar technologies to those used in making computer silicon chips.

One of the basic properties of silicon is its very high sensitivity to light - it can very easily absorb photons of light that are incident on its surface. When a photon of light is absorbed, a single electron is released that is free to move around in the silicon crystal lattice structure. However, CCDs are specially designed to ‘store’ these generated electrons and prevent them from wandering around the lattice. In this way a pattern of electrons is built up in the CCD that directly corresponds to the pattern of the photons of light received.

The resultant electron charge pattern can then be read out from the CCD using analog electronics and digitized to give a very accurate digital representation of the light image that has fallen on the CCD.

The CCDs used by PixCellent are very similar to those used for domestic TV applications such as home video cameras. However, PixCellent has developed technologies that run the CCDs in very different ways to produce the most sensitive imaging systems available anywhere, for almost any application.

The CCD chip basically consists of a thin slice of silicon substrate covered by a two-dimensional array of polysilicon electrodes which are separated by oxide insulation layers. A cross-section of a typical CCD is shown in Figure 2:

Figure 2: Cross-section of a CCD

 

 

 

 

 


Charge generation and collection

The electrodes are held at different potentials and when the CCD is exposed to light, photons which fall on the surface of the CCD pass through the electrodes and cause electrons to be generated in a doped depletion layer on the silicon substrate. These generated electrons are held in position by the applied voltages on the electrodes and are stored in potential wells.

After the CCD has been exposed to light, the accumulated electric charge in each potential well is then transferred (or coupled) to adjacent electrodes by altering their relative potentials. In this way the charge pattern, corresponding to the intensity of incident photons of light, can be moved along the CCD and into an output register and amplifier at the edge of the CCD for digitization.

Many of the CCDs used in PixCellent’s imaging systems are known as three-phase CCDs. In these types of CCD the silicon is covered with three sets of electrode strips and each set of electrodes is isolated from the silicon substrate and from one another. One of the three electrodes is biassed more positively than the other two, and it is under this one that the electrons generated by the incident light are accumulated.

The electrons are restrained from moving along the length of the electrode by channel stops, which are narrow regions of heavily doped material. Their negative charge repels electrons and therefore prevents movement across the stops. This defines the pixel extent in that direction, as illustrated in Figure 3.

 

Figure 3: Layout of a three-phase CCD

 

 

 

Figure 4 shows the basic layout of a three-phase, two-dimensional CCD. The sequence 1, 2, 3 on each set of parallel and serial electrodes indicates the normal direction of charge transfer.

 

Figure 4: Basic layout of a three-phase two-dimensional CCD


Charge transfer

At any time the charge accumulated under one electrode on the CCD can be transferred to the area below an adjacent electrode if the adjacent electrode potential is raised while the first is lowered, as illustrated in Figure 5.

 

Figure 5: Charge transfer in a three-phase CCD

 

 

 

 

 

 

 

 

The transfer of charge may be in either direction depending only on the order in which the electrode voltages are raised and lowered. All the charges stored over the entire two-dimensional imaging area are moved simultaneously in the relevant direction. This process may be repeated in order to transfer the accumulated two-dimensional charge over many pixels.

Figure 6: Repeated charge transfer in a three-phase CCD

 

 

 

 

 

The drive pulses which must be generated by the electronics system in order to transfer the charge in this manner are shown in Figure 7.

Figure 7: Drive pulses during CCD charge transfer

 

 

 

Most CCDs are manufactured for use at domestic TV frame rates, and their internal organization reflects this. TV applications require that light falls continuously onto the sensitive surface (no shutter), and thus the device organization must allow the accumulated charge image to be moved out quickly (to minimize image blurring) then read out slowly to give the continuous data stream that constitutes the TV image.

Although all CCDs have similar physical characteristics and use the same principles of operation for charge transfer and readout, there are many different designs of CCDs available that have been customized for use in specific imaging applications. The CCDs used in PixCellent's imaging systems are known as frame-transfer CCDs and full-frame CCDs, each of which are extremely useful for different scientific imaging applications.

Figure 8 shows the full layout of a typical frame-transfer CCD. The main light sensitive areas of the CCD are called the image and store sections of the CCD. Most PixCellent CCD systems treat these together as one continuous area by wiring the pins that are connected to these areas in parallel. This gives the three clock lines (f1, f2 and f3) needed to transfer the image across the area of the CCD chip. When the driver electronics pulses these three clock lines, one entire row of charge is transferred across the CCD, with all pixels in that row moving in parallel. These lines are known as the parallel clock lines.

If the electronics drives the parallel clock lines high in the order f1, f2 and f3, the charge distribution in the CCD is moved towards the serial output register. Driving them in the reverse order f3, f2 and f1 moves the charge distribution in the opposite direction. Normally it is the former sequence that is used and the row of charge is moved towards the serial output register.

Once the charge from one row is in the serial output register it may be transferred to an output amplifier, located at one end of the serial output register, by pulsing the serial clock lines Rf1, Rf2 and Rf3. The sense of pulsing Rf1, Rf2 and Rf3 transfers the charge towards the output amplifier, while the reverse order transfers charge away. The serial clock lines are normally pulsed so as to transfer charge towards the output amplifier.

The charges in the output register are transferred one at a time to the output amplifier in exactly the same way that the charges are transferred across the parallel part of the CCD.


Frame-transfer operation

The light falling on a frame-transfer CCD with image and store electrodes connected is controlled by a mechanical shutter. For exposures of longer than a few tens of milliseconds this is satisfactory, but for shorter exposures mechanical shutters are unreliable. An alternative approach is to cover the store section with an opaque screen so that light falls on the image section only. After the required exposure time, both image and store clocks are operated rapidly to move the accumulated image from the image section into the store section. If this is done fast enough then the blur that results is acceptable and the store section may be read out at whatever speed is desired.

This technique is known as frame-transfer operation. Half the number of CCD pixels are used for imaging (and the other half to store the image) so a smaller image results. However it allows shorter exposure times and, while the store section is being read out, the image section is accumulating the next image. In this way frame-transfer operation allows light to be gathered nearly 100 percent of the time, while in full-frame operation light gathering is interrupted by image read-out and can result in inefficient duty cycles (e.g. 20 millisecond exposure followed by a few seconds’ read-out).

 

Figure 8: Layout of a frame-transfer CCD

 

 


On-chip charge binning

The previous section described the normal procedure for reading out the electric charge pattern stored in the CCD, pixel by pixel. In this procedure, a single parallel transfer places the accumulated charges from one row of pixels into the serial output register. The output register is then emptied by transferring the charge from each element of the output register to the output amplifier and reading the amount of charge in each.

However, if one parallel transfer is immediately followed by others (without reading out the output register first) then the charges in the second or subsequent row are added to the charges already in the output register, as shown in Figure 9.

 

Figure 9: Adding charges in the parallel registers of a CCD

 

 

 

 

 

 

 

 

 

 

 

 

 

In exactly the same way it is possible to add charges in the serial output register, as shown in Figure 10.

 

Figure 10: Adding charges in the serial register of a CCD

 

 

 

 

 

 

 

 

 

 

The addition of charges in this way is known as charge binning or just ‘binning’ for short. Binning the parallel transfer electrodes is known as parallel binning and binning the serial transfer electrodes is known as serial binning. Sometimes Parallel and Serial binning are referred to as vertical and horizontal binning, reflecting the way that CCDs are oriented to produce a TV picture.

In many CCDs the charge capacity of the CCD output register is significantly larger than that of the imaging pixels. This allows binning into the output register to reach higher peak signals without saturation.

Any parallel binning factor may be combined with any serial binning factor. For example, the previous two diagrams illustrate 2x2 charge binning. This means a sequence of two parallel transfers followed by a serial readout where pairs of pixels are transferred to the CCD output node to be measured together.

The binning of the charge pattern is a totally noiseless process and leads to higher signal levels in the output register and also significantly reduces the readout time, which may be important in some applications and is especially important for initial alignment and approximate focusing of the camera head.

With a low light level CCD system the best performance is achieved with fairly slow readout rates (25,000 to 100,000 pixels per second). The time taken is to allow the proper analog signal processing to occur on each pixel read out but the time taken to transfer and add charges is tiny by comparison and may be done rapidly as the process is totally noiseless. Performing a 2x2 charge binning of a CCD will produce only a quarter of the number of output pixels and therefore the entire electric charge pattern in the CCD can be read out in roughly a quarter of the time.


Charge binning strategies

The readout noise level which provides the lower limit to the sensitivity of a CCD camera system is generated by the noise currents in the CCD output amplifier. The advanced signal processing procedures used in all PixCellent CCD imaging systems ensure that this readout noise is essentially a constant noise added to the signal from each read out element. It is important to distinguish these read out elements from CCD pixels.

The CCD chip is a two-dimensional array of pixels designed to allow the charges in the pixels to be transferred to the output amplifier for measurement. The output amplifier measures whatever signal is passed to it. The parallel clocks are capable of adding many rows of charge into the serial output register and the serial clocks are capable of adding the charges in many elements of the output register into the output amplifier. The fixed readout noise is added once for each output amplifier read operation and is unaffected by where the charge it is measuring came from.

In this way the CCD may be operated to allow n rows to be added into the serial output register and for m serial output register elements (corresponding to the CCD columns) to be added into the output amplifier. This is described as binning n x m pixels on-chip, and the readout of this signal (which will be n x m times the signal in a single pixel for uniform illumination) will only have one unit of readout noise added by the output amplifier.

There are advantages and disadvantages in binning the CCD image on-chip by n x m. The advantages are:

A typical timing budget for a high precision read-out operation is:

Parallel transfer 40 microseconds
Serial transfer 2 microseconds
Pixel read time (including serial transfer) 32 microseconds

 

To read out an EEV CCD06-02 of 578 x 385 pixels at 16-bit accuracy would then take microseconds or 7.1 seconds.

If a CCD is binned 4 times in the parallel direction and 4 times in the serial direction these numbers become: or 0.55 seconds, a factor of nearly 14 times faster (compared with a factor of 16 times fewer pixels to be read out).

The disadvantage of on-chip binning is that the image resolution is clearly degraded by n x m and the trade-off requires the user to decide which is preferable.


Examples of charge binning

As an extreme example of the usefulness of charge binning, some users (such as spectroscopists) are often only interested in the variation in signal in one direction. The variation in the other direction is of no importance but they are keen to use all the light they can to maximize sensitivity.

An EEV CCD02-06 has 385 columns of 578 pixels each. If the CCD is run in a cycle of:

this generates a single line image of 385 pixels where each contains the sum of all the charges in the corresponding column.

Alternatively, the following procedure could have been used:

this generates a single line image of 578 pixels where each contains the sum of all the charge in the corresponding row.

The first procedure is faster since it needs 578 parallel and 385 serial transfers only, whereas the second needs 578 parallel and 385 x 578 serial transfers (they may be fairly fast but they do still take time). However, the second procedure gives a longer (more pixels) line image or spectrum.

The large variety of charge binning capabilities enable the speed disadvantages inherent in any slow-scan CCD system to be partially overcome. For example, when initially setting up the camera head for correct alignment, 3x3 binning can be used with a short exposure time (the average signal will be 9 times larger). This will give a readout time roughly one ninth of the unbinned image (probably a fraction of a second). For initial focussing, charge binning can also be used to speed the cycle time between images, but for accurate focussing, where it is important to retain maximum image resolution, another technique called sub-array read out, can be used.


Sub-array read out

We have seen how the charge in each pixel may be transferred quickly across the CCD. With slow-scan CCD systems it is the business of accurately reading the signal in each pixel that takes most of the time. This can be used to advantage when only a relatively small part of the image is of interest. In these cases the unwanted parts of the image may be flushed away rapidly with the slower read out limited to the area of interest. This arises, for example, when the camera is to be focussed or when a small area is required to be read out very rapidly or many times in quick succession.

The read-out time is mainly set by the size of the area of interest - if it is one tenth of the area of the CCD, the read-out time will be nearly ten times faster than the full-frame read-out time. All PixCellent CCD imaging systems have these facilities as standard.


Other CCD structures

The description of how a CCD works is based around the three-phase devices such as those made by EEV and SITe. Other manufacturers use different structures. For example, Thompson generally use four-phase devices where the sequence of operation is exactly as for a three-phase device but extended to a fourth phase.

Kodak, however, make two-phase devices. The direction of charge flow is fixed by the doping in the silicon under each electrode during manufacture. This structure does not allow the reverse clocking that is needed for a few specialist applications and which is possible in principle with three-phase and four-phase designs. However, some new dark current suppression features being used by manufacturers such as EEV forbid reverse clocking despite a three-phase electrode structure.


CCD performance specification

The previous section explained how CCDs record an image when used in any general purpose optical instrument. This section moves on from there and describes the performance characteristics of CCDs and how, when used in an PixCellent system, the CCD comes very close to being a theoretically perfect light detector.

PixCellent’s CCD systems use CCDs manufactured by EEV, SITe and Kodak, although it also supports CCDs manufactured by other CCD suppliers. The performance characteristics described here are those obtained typically by PixCellent with EEV, SITe or Kodak CCDs incorporated into its camera systems. Nevertheless, the principles described here apply to all types of CCDs supported by PixCellent. These include the EEV (UK), SITe (US) and Loral (US) three-phase CCDs, the Thompson (France) four-phase CCDs, the Kodak (US) two-phase CCDs and the Texas Instruments (US) single-phase CCDs. Others are being added regularly.

These different makes of CCD differ in relatively minor details but the broad principles of operation are the same. Some CCDs are available only in full-frame format (no separate image and store regions). The four-phase and three-phase CCDs have their charge-transfer direction determined by the order of clocking the electrodes as described in the previous section. With two-phase CCDs the charge-transfer direction is fixed by doping implants in the silicon. This means that it is not possible to reverse clock the CCD as may be done with three-phase and four-phase CCDs in specialist applications, but for most applications this restriction is unimportant.


Operational mode

Although PixCellent uses the same CCDs as those that are used in domestic TV cameras, the CCD is operated in a very different fashion. The two main differences are:

The benefits provided by these two techniques are described in the next two sections.


Cooling the CCD

The thermal energy of the electrons in the silicon allows some of the electrons to break away from the lattice and become free to move through the silicon in just the same way as electrons excited by incident photons. These electrons constitute the dark current and are seen as a signal which is present even when there is no light falling on the CCD. This signal is generated at all times - between exposures, during an exposure and during readout, although not necessarily at the same rate.

To reduce the amount of dark current present, the CCD is cooled. This reduces the dark current by a factor of about 10 for every 20oC temperature reduction. Typical values for EEV CCDs operated in standard mode are given in Table 1.

Temperature (oC)

Dark current (e-/pixel/sec)

Typical area

+20

10,000(std) 100 (MPP)

Room temperature

-40

10 (std) 0.1 (MPP)

Air cooled camera head

-60

1 (std) 0.01 (MPP)

Water cooled camera head

-140

< 1 e-/pixel/hour (std)

Liquid nitrogen cooled camera head

Table 1: Dark current at different temperatures

 

The dark current can also be reduced during long exposure times by using special types of CCDs called MPP CCDs. These CCDs have a Multi-Phase Pinned (MPP) architecture which is capable of reducing the dark current by a factor of 25 to 1000.

In all MPP CCDs, as the dark current is reduced, the capacity of the pixels of the CCD to store charge is reduced. This capacity is known as the full well capacity and at the lowest dark current levels, it is typically half to two thirds of standard mode operation full well capacity. However, in all precision PixCellent systems fitted with an MPP CCD (provided the CCD in question permits such operation), the operating mode of the CCD is software selectable, so it is possible to make a trade-off between the dark current level and the full well capacity depending upon the particular requirements of an application.

Using PixCellent’s camera control software, certain MPP CCDs from EEV can be operated in three different modes:

The typical full well capacity and dark current figures for an PixCellent high precision system using a thermoelectrically cooled (Peltier stack) camera head fitted with an EEV CCD02-06 are given in Table 2. The table shows the figures obtained for the CCD operating in the three different software-selectable modes:

Mode

Dark current (e-/pixel/sec)

Full well capacity (e-)

Super-MPP

0.02

40,000 - 50,000

MPP

0.2

90,000 - 120,000

Standard

20

250,000 - 500,000

Table 2: Full well capacity and dark current for a high precision cooled CCD system

 

The benefits provided by MPP CCDs are substantial, especially for applications where the exposure times are much longer than the readout times. In such applications, Peltier cooled camera heads fitted with MPP CCDs can now be used where the low dark current performance of a liquid nitrogen cooled camera head is required along with the versatility of the smaller Peltier cooled camera heads.

In order to achieve the lowest possible dark current in MPP, Super-MPP and standard mode, care must be taken to avoid exposing the CCD to excessive light levels as this will increase the dark current significantly for several hours.

A more recent development by EEV is an advanced form of MPP operation that gives the exceptionally low dark current of Super-MPP operation yet retains the full well capacity of standard mode operation. EEV use the term inverted mode operation (IMO) for their MPP devices and the new structure is called AIMO (advanced inverted mode operation). This is remarkable in that the full dark suppression of the best Super-MPP CCDs from EEV is achieved but with negligible reduction in full well capacity. This is best seen in the EEV CCD30-11 where the dark current is very low yet the full well capacity is typically 500,000 electrons.

All the devices from Kodak are intrinsically of MPP design and their dark current is extremely low. This is partly because they generally have very small pixels (many of their devices are in the 6.8 to 9 micron square range of pixel sizes). Another consequence of the small pixel size is that the full well capacity is also smaller.

CCDs made by SITe are available with MPP structures. Generally they have chosen a less extreme level of dark current supression (factors of 30 to 100) in order to retain a higher full well capacity.


Reading out the charge in the CCD slowly

The charge that is read out from the CCD is passed via a buffer amplifier transistor or transistors that is part of the CCD itself. This amplifier has an internal noise level that is always added to the signal whenever a pixel is read out. The analog signal processing used by PixCellent is principally designed to minimize this noise. The more slowly the CCD is read out, the lower the noise, up to a cut-off point varies between 40 microseconds per pixel with the lowest noise EEV CCDs and 2 microseconds per pixel with the Kodak CCDs.

 

Table 3 illustrates the benefits that are obtained by reading out the CCD more slowly. The table gives some typical readout noise levels at different pixel rates for an PixCellent Antares, cooled CCD imaging system fitted with an EEV CCD02-06 (587x385 pixels) or a Kodak KAF-0400 CCD (768x512 pixels).

Model

Number of ADCs

Pixel rate (kHz)

Full-frame read time (seconds)

Noise (e- RMS) (typical)
     

EEV CCD06-02

Kodak KAF0400

EEV CCD06-02

Kodak KAF0400

Antares

4202

2

167 (max)

25

2.0

11.0

3.5

16.5

12

5 (lowest)

9

7 (lowest)

Table 3: Readout noise for an PixCellent Antares cooled CCD imaging system

 

The readout noise is added even in the absence of any signal. On-chip binning has no effect on the readout noise since it is only added to the charge in each binned pixel.

Reading out the CCD very slowly is acceptable for some applications. However, other applications performing real-time imaging require the CCD to be read out at video-rates (several million pixels per second). Electronic constraints make it much harder to suppress many of the CCD noise sources when operating at such high rates. However, it is possible to build fast readout rate CCD camera systems that can then be slowed down to give improved readout noise.

Table 4 gives some typical readout noise levels at different pixel rates for an PixCellent Capella cooled CCD imaging system fitted with various Kodak CCDs and the EEV CCD57-10 frame-transfer CCD.

Pixel rate

Full-frame read time (seconds)

Noise (e- RMS, typical)

(MHz)

Kodak KAF0400

Kodak KAF1600

EEV CCD57-10

Kodak KAF0400

Kodak KAF1600

EEV CCD57-10

8

0.06

0.22

n/a

set-up

set-up

n/a

5.5

0.08

0.32

0.05

40

40

50

4

0.12

0.48

0.07

25

25

30

2

0.22

0.88

0.15

15

15

18

1

0.45

1.8

0.30

10

10

12

0.5

0.88

3.5

0.60

8

8

8

Table 4: Readout noise for an PixCellent Capella cooled CCD imaging system

 


Sensitivity and quantum efficiency

The sensitivity of CCDs to light is normally given in terms of quantum efficiency where quantum efficiency is defined as the effectiveness of a CCD in generating electrons from the incident light falling on the CCD, as a function of wavelength. Figures 11, 12 and 13 show the typical of responses of EEV, Kodak and SITe CCDs respectively to different wavelengths of light.

Figure 11: Spectral response curves for EEV CCDs

Figure 12: Spectral response curves for Kodak CCDs

 

Figure 13: Spectral response curves for SITe CCDs

 

 

The standard, front-illuminated CCD has little response at wavelengths shorter than 400 nm. This is because the polysilicon electrodes which cover the device act as a yellow filter by blocking the bluer light. In the far red the sensitivity extends beyond one micron but here the silicon is progressively more transparent and a rapidly decreasing fraction of the incident light is absorbed in the layer of the CCD over which charge is collected by the electrodes.

The blue sensitivity of the CCD may be enhanced in two ways:

 

Figure 14: Cross-section of standard and thinned CCDs

 

 

 

 

 


Linearity

Photons are absorbed in the CCD and generate electrons that are held in place by the voltages applied to the electrodes of the device. Once an electron has been generated there is nowhere for it to go or to be lost. The absorption of other photons is unaffected by the charge already present, so the CCD is clearly a device that is intrinsically linear. Since the PixCellent readout electronics is also designed to be linear, the overall system is highly linear.

Linearity here means that as the exposure time on a fixed scene is increased, so the output signal goes up in exactly the same proportion as the exposure. Two parts of the image with a certain brightness ratio will be detected with that same ratio over a wide range of exposure times. Linearity will eventually break down once the CCD reaches saturation.

In practice CCD linearity is limited by the performance of the on-chip amplifier that is part of the CCD itself. If the signal is large enough to shift the working point of this transistor enough to affect its gain, then a small non-linearity may be observed. If this occurs, it can be corrected using image processing software.

In testing a CCD camera system for linearity it is important to remember that real mechanical shutters have a finite response time. This might mean that a 100 second exposure might turn out to be 100.01 seconds long - a negligible error, but a 0.01 second exposure might turn out to be 0.02 seconds long, giving a result that could easily be misinterpreted as demonstrating a gross non-linearity in the system that was not real. In addition, real shutters are often close to the CCD chip so that the centre of the CCD is exposed for slightly longer or at a different time from the outer parts of the device. For long exposures these effects are utterly negligible but for short exposures the effects can be very important.


Dynamic range

The dynamic range of a detector is defined as the ratio of the largest signal which the detector can handle to the readout noise in a single exposure. Typical values for an EEV CCD02-06 are 500,000 and 5 electrons respectively giving a dynamic range of 100,000:1. This wide dynamic range is achieved because CCDs are designed for use at much higher light levels and with higher dark currents. Cooling the CCD and reading it out slowly dramatically reduces the minimum noise level but has no effect on the maximum signal that the CCD can store.

In practice the wide dynamic range figure has to be looked at rather cautiously. Firstly, the electronic system includes an analog to digital converter that provides a limit to the dynamic range. A 12-bit converter has 4096 grey levels, a 14-bit converter has 16384 levels and a 16-bit converter (standard in all PixCellent high precision systems) has 65536 grey levels. Clearly even with a 16-bit converter 100,000:1 dynamic range per pixel is impossible to achieve. However it is important to realize that real objects actually extend over several pixels. So an object with an area of 9 pixels can be up to 9 times the signal of one pixel but the noise of 9 pixels added together is only 3 times greater (square root of 9, see below) so a dynamic range of nearly 200,000:1 can be obtained in this case.

When certain specific older design EEV CCDs are run in MPP mode to dramatically reduce the dark current it is found that the maximum signal that can be stored in each pixel is reduced to as low as 40,000 to 50,000 electrons per pixel thus reducing the dynamic range. However, it is easy to change from MPP mode (low dark current, lower dynamic range) to standard mode (higher dark current, wider dyanmic range) using PixCellent's camera control software.

It is also important to appreciate that real optical systems are often incapable of creating an image onto the detector with this sort of dynamic range because of defects (internal reflections, scattering of light etc.) in the optical system.


Saturation characteristics

Sometimes it is necessary to use a CCD to look for faint features near to much brighter ones. It is acceptable to expose the CCD so that the bright feature saturates the CCD locally. What happens within the CCD is that the light eventually spills over into adjacent pixels giving an image that is smeared in the parallel transfer direction. Other parts of the CCD will work normally. This effect is also known as blooming.

This saturation effect is much more pronounced in intensified imaging detectors than in CCDs. A CCD is much more resistant to high light overloads (remember they are used in daylight at TV readout rates) although the blue sensitive phosphor coating can have its efficiency degraded by exposure to excessive levels of ultra violet light for long periods of time.

If the CCD becomes saturated because of very substantial light overload, or because the system was powered off when the CCD was cold, then the device will have to have any excess charge flushed out. Even after several flushes, liquid nitrogen cooled systems will exhibit higher than normal dark current (a factor of a few higher).

Therefore, if the ultimate in dark current performance is needed it is essential to avoid saturating the device or powering the system off while cold. Normal dark current is achieved after the device has been warmed up and then re-cooled. No long-term damage occurs due to switching off the system when the device is still cold.

There is a similar characteristic at any temperature with MPP CCDs. Exposure to high light levels will significantly increase the dark current in the CCD temporarily, even if the CCD is warm and/or powered off.

Certain CCDs are available with a modification to their electrode structure that stops saturation from occurring. Anti-blooming drains are placed next to each electrode of the imaging area to remove charge in a pixel above a threshold which is below the saturation point at which blooming or saturation occurs. Anti-blooming drains are extremely effective and can permit a thousand-fold over exposure to light without adverse effects. They do, however, occupy part of the area of the silicon and therefore reduce the overall quantum efficiency of the CCD by 15 to 30 percent. Many (though not all) Kodak CCDs are available optionally with anti-blooming drains.


Charge transfer efficiency

The image quality which is obtained with a CCD system is greatly affected by the completeness of the transfer of charge from one pixel or row to the next during readout. If any charge is left behind then the image will look smeared and finer detail (resolution) of the image will be lost. Most CCDs work well in this respect at high signal levels but many manufacturers have systems that show poorer charge transfer at the lowest signal levels. This is simply because it requires the most careful electronic design and optimization for the best transfer efficiency to be achieved.

The requirements in some applications are very demanding indeed. For example, in some spectroscopic applications requiring full serial or parallel binning an output signal level of 50 electrons per binned output pixel might imply a mean charge level before binning below 0.1 electrons per pixel. Clearly this actually implies 1 electron per 10 pixels in practice but good charge transfer efficiency (CTE) requires that these individual electrons are transferred across one centimeter. of silicon or more without being left behind, a remarkable achievement for the designers of CCDs.

Better CTE is achieved at higher light levels. Several manufacturers help the CTE of their system by pre-flashing the CCD before readout with a weak uniform light source to give a signal pedestal to improve the CTE. This works but of course it adds unnecessary and unwanted photon shot noise to the image. Typical levels are 100 to 300 electrons, adding 10 to 17 electrons of noise for unbinned images but more than twenty times this for a fully binned application. It rather defeats efforts to make a low-noise, low dark current CCD system. Clearly, systems that avoid the use of a pre-flash are greatly preferable to those that do not. No PixCellent CCD system has ever needed to resort to pre-flashing.


Cosmic ray events

Energetic atomic particles are shot out from the Sun and strike the upper atmosphere of the Earth. These cosmic rays generate secondary particles called muons which are detected by CCDs (and all other types of light detectors). Typical rates are about 2 events per square centimeter of detector active area per minute. The rates vary slightly with solar activity level. Each detected event is very compact, covering only 1 to 3 pixels typically and having a total signal of 2000 electrons typically, though there is a great spread in these characteristics.

Should they cause any ambiguity in the images taken then it is possible to remove them (using image processing software) or to split the exposure to give two shorter exposure images which allow real objects (present on both) to be distinguished from cosmic ray events (present only on one or the other).


Signal to noise considerations

In order to judge how to get the best from your CCD system it is important to try to estimate the signal levels you expect to work at and then the signal to noise ratio you expect to achieve. Always remember that the CCD does not distinguish between signal and noise. All sources of charge within the CCD are read out as signal. It is up to you to decide which of these charges you want (the signal) and those that make your signal more difficult to see (the noise).

There are several sources of the electronic charge which is read out of the CCD at the end of the exposure, some of which have been discussed already:

The presence of light leakage in an experimental set-up may be traced easily and safely with a flashlight. Unlike many low light detector systems, CCDs are undamaged by light-levels tolerated by the human eye.

Although we might think of some of these sources as being noise, as far as the CCD is concerned they are a signal to be measured as precisely as possible. However if we have an output signal of q electrons which is to be measured there is a fundamental physical limit which says we must expect that if we identically repeat the experiment many times we will find that the uncertainty of the value q is only defined to within plus or minus . This uncertainty is called the shot noise or photon noise in the measured signal.

The CCD chip will accumulate and transfer charge without loss or added noise but the amplifier attached to the output of the CCD has an internal source of noise which is expressed in electrons. If the readout noise of your system is s, then the effect it has is exactly the same as if you had a uniform signal of exactly s2 electrons linearly added to each and every pixel read out. It is a noise that is present irrespective of whether there is any signal being read out. The readout noise is added in quadrature with the photon shot noise. For a readout noise of s electrons (root mean square or RMS) and a signal level of q electrons, the total noise (NT) is then:

and the signal to noise ratio (SNR) of the measurement of the charge in that pixel is:

At high signal levels, q>>s, and the signal to noise ratio is essentially independent of the readout noise. It is nearly equal to , the level it would have in the absence of any readout noise. However at lower signal levels , the readout noise significantly degrades the signal to noise ratio achieved.

It is necessary to estimate as well as you can what sort of light level you expect to encounter and what sort of interfering background signals you have to cope with. PixCellent can help you to make these estimates and can advise from their experience about problems users in similar applications have experienced.


Finding out more about CCDs

PixCellent has available a wide range of selection guides, data sheets and application guides. We do not expect users to be expert in CCD technology - few of our users are and many do not want to be. Our role is to help you select the best CCD system for your particular application. We will usually have experience of at least one that is closely similar to yours and we will be delighted to discuss your application in as much detail as necessary.

If you would like more information from PixCellent   please contact us by e-mail at sales@pixcellent.com

Web site: http://www.pixcellent.com/ or http://www.pixcellent.co.uk/


Glossary


Anti-blooming drain

A structure implanted into the silicon to allow local light overload conditions to be handled without affecting adjacent pixels of the CCD.

Luimogen

A special coating that is applied to the surface of a CCD to increase the quantum efficiency of a CCD at blue and ultra-violet wavelengths.

Advanced inverted mode operation

A special design of MPP CCD that combined exceptionally low dark current with high full-well capacity.

Binning

A technique for reading an electron charge pattern out of a CCD. The technique involves adding the charge stored in either the serial or parallel electrodes or both, before it is read into the output amplifier.

CCD

See Charge-coupled device.

CTE

See Charge transfer efficiency.

Channel stop

Narrow regions of heavily doped material in a CCD that lie between the parallel transfer electrode strips. The channel stops are negatively charged and repel electrons generated in the CCD, thus preventing electrons from wandering to another pixel on the CCD.

Charge binning

See Binning.

Charge transfer efficiency (CTE)

The efficiency with which a CCD transfers the electronic charge pattern generated by incident light across its surface and into its output register for read-out.

Charge-coupled device (CCD)

An analog integrated circuit that uses a one- or two-dimensional array of electrodes to convert incident photons of light into a proportional electric charge.

Cosmic rays

Energetic atomic particles that are shot out of the Sun and strike the upper atmosphere of the earth generating muons which are detected by CCDs as a source of noise.

Dark current

An omnipresent signal in a CCD that is caused by the thermal energy of the electrons being able to break away from the CCD structure and become free to move through the silicon in just the same way as electrons generated in the CCD by incident light photons.

Dynamic range

The ratio of the largest signal which a CCD can handle to the read-out noise in a single exposure.

Four-phase CCDs

A type of electrode structure on the surface of a CCD comprising four sets of electrode strips.

Frame-transfer CCD

A type of CCD that is has a separate image and store section to allow fast operation without a mechanical shutter. This type of CCD is used in many of PixCellent's fast imaging systems.

Full Frame CCD

A type of CCD that is mostly used in scientific imaging equipment. It requires a mechanical shutter or other control to stop image smearing during read out. This type of CCD is used in many of PixCellent's fast imaging systems.

Full-well capacity

The maximum amount of charge that can be stored by a pixel on a CCD.

Inverted mode operation (IMO)

See Multi-phase pinned.

Linearity

When used in relation to CCD technology it refers to the relationship between exposure time and output signal. Up to certain light levels, as the exposure increases the output signal generated by the CCD should increase in a linear fashion. Above such levels linearity is limited by saturation effects.

MPP

See Multi-phase pinned.

Multi-phase pinned (MPP)

A type of CCD architecture that is designed to reduce the dark current by a factor of 25 to1000.

Output amplifier

The component on a CCD that receives the electron charge pattern from the serial output registers and amplifies the signal so that it can subsequently be read by the Digital to Analog convertors in the digital driver electronics unit.

Parallel binning

Adding the electron charge pattern collected in two or more sets of parallel transfer electrodes into the serial transfer output electrodes.

Parallel transfer electrodes

Electrodes that lie on the surface of a CCD. They are held at different potentials and are used to transfer the electron charge generated underneath them across the CCD and into the serial output register electrodes.

Photon noise

The minimum noise that is always associated with any light signal. It is equal to the square root of the total number of photons detected and is sometimes referred to as shot noise.

Potential wells

Individual sites on a CCD that can store the ‘mobile’ electrons generated in the CCD. The electric charge generated in each well can be transferred to adjacent wells by altering the relative potentials of the electrodes covering each well.

Quantum efficiency

The effectiveness of a CCD in generating electrons from the incident light falling on the CCD.

Serial binning

Adding the electron charge pattern collected in two or more of the serial output register electrodes into the output amplifier.

Serial output register electrodes

Electrodes that lie on the output edge of a CCD. They are held at different potentials and are used to transfer the electron charge into the output amplifier.

Shot noise

See Photon noise.

Single-phase CCDs

A type of electrode structure on the surface of a CCD comprising one set of electrode strips.

Super-MPP

A software-selectable mode of operation of MPP CCDs that reduces the dark current by a factor of 1000 better than standard CCDs for many low-light level applications.

Three-phase CCDs

A type of electrode structure on the surface of a CCD comprising three sets of electrode strips.

Two-phase CCDs

A type of electrode structure on the surface of a CCD comprising two sets of electrode strips.

Virtual-phase CCDs

See Single-phase CCDs.


Home:


The pages on this Web Site are Copyright of PixCellent Imaging Ltd., Cambridge, England. Reproduction in part is permitted with acknowledgement to PixCellent Imaging Ltd., and of this Web Site Address (http://www.pixcellent.com/).
Comments, please, to webmaster@pixcellent.com
Site last updated: 1 November 2006.