A charge-coupled device (CCD) is an integrated circuit etched onto a silicon surface forming light sensitive elements called pixels. Photons striking on this surface generate charge that can be read by electronics and turned into a digital copy of the light patterns falling on the device. CCDs come in a wide variety of sizes and types and are commonly used in high-end scientific applications.
The function of a CCD can be visualized as an array of buckets (pixels) collecting rainwater (photons). Each bucket in the array is exposed for the same amount of time to the rain. The buckets then fill up with varying amounts of water, and the CCD is then read out one bucket at a time. This process is initiated by pouring the water into the adjacent empty column. The buckets in this column transfer their water down to a final summing pixel where the electronics of the camera read-out this pixel and turn it into a number that can be recognized and stored by a computer. While this model is an oversimplification we provide an in depth explanation below.
Photons striking a silicon surface create free electrons through the photoelectric effect. A simultaneous positive charge or holes are generated as well. If nothing is done the hole and the electrons will recombine and release energy in the form of heat. Small thermal fluctuations are very difficult to measure and it is thus preferable to gather electrons in the place they were generated and count them to create an image. This is accomplished by positively biasing discrete areas to attract electrons generated while the photons strike the surface.
Simple diagram of a CCD pixel.
The substrate of a CCD is made of silicon, but photons coming from above the gate strike the epitaxial layer – essentially silicon with different elements doped into it – and generate photoelectrons. The gate is held at a positive charge in relation to the rest of the device, which attracts the electrons. Because of the insulating layer – essentially a layer of glass – the electrons can’t make it through to the gate and are held in place by the positive charge above them.
The figure to the right shows how electrons are held in place and moved to where they can be quantified. The top black line represents the potential well for the electrons that are represented by the blue color and is low, or downhill, where the potential is high since opposites attract. As the voltage adjacent to the electron’s pixel is brought high, they begin to migrate in this direction until the voltage in the preceding gate is then brought to zero, or low, thus effectively transferring all the electrons into its neighboring pixel.
Electrons are shifted in two directions on a CCD, called the parallel or serial direction. One parallel shift occurs from the right to the left (shown at left). The serial shift is performed from top to bottom and directs the electron packets to the measurement electronics.
Many CCDs are built with multiple amplifiers at each corner of the CCD and can thus be read out faster. In the example to the left, the image is split up into 2 and then 4 different sections and read-out.
The analog to digital (A/D) electronics measures the voltage created by the packet of electrons at the serial output and turns this into an electronic number that can then be digitally saved. The method of reading this voltage is called dual slope integration (DSI) and is used when the absolute lowest noise possible is required. Generally speaking, the faster a pixel is read, the more noise is introduced into the measurement. The A/D electronics output units are called analog to digital units or “ADUs”. If the gain of the measurement is known the ADU number for each pixel generated can be directly correlated to the number of electrons found in that pixel. All Spectral Instruments cameras come with a detailed test report showing the gain at a given readout speed
A/D electronics have limits on the largest number they can describe. For instance, an 8-bit A/D system cannot represent a number larger than 28 = 256. 16-bit electronics can’t describe a number larger than 216 = 65,536. Thus, a 16-bit camera can never show more than 65,535 ADUs in any given pixel. Scientific grade CCDs can generally hold anywhere from 70,000 to 500,000 electrons in any given pixel. Since this is more than the number of ADUs that the A/D electronics can express, different gains must be used for the electronics to access the entire dynamic range of the CCD. At slow read speeds, (i.e. low noise) gains of 0.25e-/ADU are common, thus reading only a maximum of 0.25*65535 = ~16.4ke- which is much lower than the dynamic range of most currently available CCDs. At higher read speeds (i.e. higher noise), gains of 5e-/ADU can be reached allowing full access to the CCDs dynamic range. This sacrifices, however, higher read noise for the extra dynamic range. All SI cameras can be read at multiple speeds and multiple gain settings to ensure access to the most critical aspects of the measurement.
Full Well Capacity is the maximum number of electrons a pixel in a CCD can hold. This number can vary widely (10ke- to 500ke-) and depends mostly on the physical dimensions of the pixel (the bigger the pixel, the more electrons it can store). When a pixel has too many electrons in it, the excess charge begins to spill into the neighboring pixels and creates imaging artifacts known as blooming. Shown at left is a picture of the Andromeda galaxy, where faint details of the dust in the spiral arms are visible but the closer stars in our own galaxy are blooming.
CCDs benefit from working at lower temperatures since thermal energy alone is enough to excite extraneous electrons into the pixels which cannot be distinguished from the actual electrons produced by photons. This noise is called dark current. For every 6-8 °C of cooling, there is about a 2X reduction in the total dark current generation rate. This does have its limits, however, since CCDs don’t function well below –120°C due to the negative effect on charge transfer efficiency. Note that especially for deep depleted CCDs, cooling below -90°C is critical to reduce the dark current to acceptable levels.
This can be observed in the video at left. At –90°C, the image looks as a deeply cooled CCD should, with only random read noise present. Note that if a bit of over scan has been included – one can tell the electronics to read more from the CCD than there actually is to help get a sense of the electronic read-noise unrelated to the electrons stored in the CCD pixels. As the temperature increases, more thermal electrons are generated. Remember that the electrons in any given pixel are moved across the CCD, so those electron packets readout last accumulate more charge as they are swept in the parallel shifting of the CCD. This creates a gradient of dark current signal increasing as the CCD is readout– a characteristic of a warm CCD. Also note the defect in this particular device, a column defect towards the bottom of the image during the CCD manufacturing process pixels can be produced that generate thermal electrons at a rate much greater than their neighboring pixels and inject charge into each electron packet swept past it.
Cooling a CCD to –110°C requires that the device be thermally isolated from its environment and therefore must be in an evacuated chamber. Today, CCDs are commonly cooled with Peltier junctions (thermoelectric coolers) and mechanical pumps (cryo-coolers). Liquid nitrogen was previously the primary method of cooling; however, today it is rarely used as LN2 impacts camera orientation and can be cumbersome to work with. Spectral Instruments offers state of the art TEC or cryo-cooled cameras.
CCDs generate photoelectrons at different rates depending on the wavelength of light. The conversion of photons into an electric signal is called quantum efficiency (QE). Anti-reflection coatings have some effect on QE, but back-thinning has the larger effect on increasing QE since light does not have to pass through the “gate material” of front-illuminated CCDs. As shown below, a normal front-illuminated device creates signal after the light has passed through the gate structures resulting in an attenuation of the incoming radiation. A back-thinned or back-illuminated CCD has the excess silicon on the bottom of the device etched away allowing unimpeded photoelectron generation to occur.
The process of back-thinning varies from company to company, which results in variation between manufacturers. Shown below is an example of some typical QE values from a CCD manufacturer (Teledyne e2v) that is commonly used in Spectral Instruments camera systems.
CCD sensors were pioneered for scientific measurement applications in the early 1980s and became the sensor of choice for nearly all imaging applications, including machine vision and consumer electronics. In the early 1990’s, CMOS image sensor development reached a point where they began to replace CCDs for some low performance applications. Over time, CMOS sensor technology has rapidly expanded while CCD technology has reached maturity. At this time, both types of sensors have their place in scientific measurement applications.
In general, CMOS image sensors are the first choice when the application calls for high frame rates, and especially low noise at high frame rates. In such applications, the optical integration time is so short that dark current and any luminescence from sensor transistors is insignificant. Readout from each column in parallel means the practical frame rate is 2 to 3 orders of magnitude higher than the typical CCD. Very small transistors in each pixel results in high sensitivity in terms of signal output per photoelectron, but it also results in limited dynamic range and noise characteristics that are less desirable than a CCD output that uses a physically large output transistor designed to have low 1/f noise. CMOS sensors designed for scientific measurement applications are currently limited in availability, but significant effort is being made to expand into the roles where CCDs have traditionally been the favored sensor.
CCDs excel in applications where the readout time is less important and readout follows a long integration time. A sufficiently cooled CCD has practically no dark current and no luminescence to mask the signal of interest. The output transistor has well-behaved low noise characteristics, and a dynamic range equal to or exceeding that of data converters. Pixels in a CCD can be “binned” to noiselessly combine charge from adjacent pixels which can be used to great advantage in low light applications (note that the term binning is also used for CMOS sensors, but it is an averaging of pixels and the improvement in signal to noise ratio is quite different). Because of the long history of development, CCDs can be optimized for best sensitivity in different wavelength ranges from near IR to x-rays by employing different silicon thickness, backside illumination (still rare in CMOS sensors), optimized backside treatment and AR coatings. Large area arrays with large pixels are routinely available. In short, if there is sufficient time for the long readout time of the CCD, there is no better sensor for very low light applications.
Scientific Charge-Coupled Devices, James R. Janesick, SPIE Press, 2001
CMOS/CCD Sensors and Camera Systems – Second Edition, Gerald C. Holst and Terrence S. Lomheim, SPIE Press, 2011
Handbook of CCD Astronomy, Steve B. Howell, Cambridge University Press, 2000
Electronic Imaging in Astronomy Detectors and Instrumentation – Second Edition, Ian S. McLean, Springer-Praxis, 2008