Does Your CCD Camera Need Cooling?
The answer may lie in how much noise is generated by dark current.
David W. Gardner
To those in the light-starved world of astronomy, the result of cooling cameras to ultralow
temperatures is obvious: Truly amazing images can be produced in situations where
only a few photons are available. The need for cooling may not be so obvious in
applications such as web inspection, laser profiling and medical imaging. Yet even
here, cooling the camera (albeit not to cryogenic levels) can often provide a dramatic
boost to image quality.
In most cases, the benefits of cooling come from
its ability to reduce the dark current generated by the camera. This pesky current
represents the collection of all unwanted free electrons that thermal noise generates
in the CCD. In applications such as astronomy and spectroscopy, it can mean the
difference between a good image and no image at all. In more conventional industrial
applications, the amount of dark current correlates with a lack of image clarity.
Decoding dark current
The term “dark current” comes from
the fact that the current has nothing to do with the incident light and is generated
equally well in complete darkness. Depending on where the unwanted electrons are
generated in the silicon, some of the charge will collect in the individual CCD
pixels and contaminate signal electrons (related to the image). At the CCD output,
dark-current-generated electrons appear identical to signal-generated electrons,
so dark current appears as noise in the image.
At a given temperature, the average
dark current in a given pixel is relatively constant, which results in a fixed offset
that is added to each pixel value. But the generation rate for that current typically
varies spatially across the array. This process produces a fixed background pattern
that will be superimposed on top of the desired image. By capturing an image in
complete darkness, a pixel-by-pixel representation of the average dark current can
be obtained for that temperature and exposure period. Subtracting this dark reference
image from all subsequent real images partially eliminates the effects of the spatially
varying dark-current offset. Another name for the subtraction process is dark-field
Dark-field correction can dramatically reduce the contamination effect that dark noise has on an image.
Unfortunately, there is a second component
of dark-current generation — a randomly varying component related to shot
noise, which scientists approximate as the square root of the dark current collected
in a given pixel. For example, a CCD that generates 4000 such electrons in a pixel
during a one-second exposure will have an offset of 4000 electrons that can be subtracted,
plus a random noise equal to roughly 63 electrons rms that cannot.
Consider an application involving a
12-bit camera where each pixel can hold a maximum of 40,000 electrons (the full-well
capacity). If properly calibrated, each analog-to-digital count of the camera output
would represent 40,000 electrons divided by 212, or a total of 10 electrons. An
increase of 10 photoelectrons captured by a given pixel would increase the camera’s
digital output by one count.
If this camera has a dark current of
4000 electrons per second, and a one-second exposure is taken, the average pixel
would have an offset of 4000 electrons that could be subtracted and, for the most
part, eliminated. Keep in mind, however, that the dark current has consumed 4000
of the 40,000 maximum, so the signal full well has been reduced to 36,000 electrons.
Furthermore, the process of subtracting
images takes time and computational power. Added to this 4000-electron offset is
random noise from the dark current of 63 electrons rms. With the camera biased so
that one analog-to-digital count represents 10 electrons, this means that your camera
has about 6.3 counts of rms noise and that any signal lower than this will never
be seen. As a result, the 12-bit camera is providing a real dynamic range of only
(40,000 — 4000)/63 = 571:1, or about 9 bits.
Just as dark current grows with increasing
temperature, it decreases if the sensor is cooled. This allows some camera manufacturers
to reduce dark current to extremely low levels. One field that requires ultracooled
cameras is astronomy. Because of low light levels, researchers may wish to cool
to as low as 2100 °C and take an exposure that is 30 minutes to an hour long.
Ultracooling is tricky
Cooling a camera to this level involves significantly
more than simply pouring liquid nitrogen over the sensor while the camera is running.
Cameras of this type can be quite complex and often use liquid nitrogen or multistage
thermoelectric coolers combined with vacuum enclosures and closed-loop feedback
to lower sensor temperatures.
An extensive knowledge of heat transfer,
materials science and semiconductor physics is key to successfully cooling a camera
in applications where dark current must be virtually eliminated. Only a few highly
skilled companies, such as Spectral Instruments in Tucson, Ariz., and Roper Scientific
in Trenton, N.J., offer products that are geared toward the ultracooled market.
One might ask whether high frame rates
make dark current insignificant and therefore eliminate the need for cooling. After
all, because the amount of dark charge captured is directly proportional to the
exposure time, a shorter exposure (higher frame rate) will reduce its effect at
a given temperature. However, the user should not assume that running at high frame
rates alone will provide immunity to dark current and other thermal issues.
CCD and camera manufacturers typically
specify dark current for a certain temperature, often around 25 °C. Higher-frame-rate
cameras have significant self-heating within the sensor because of higher amplifier
bias currents and higher-speed clocking of the gate capacitance. Therefore, it is
not unusual to find a commercial camera running at 30 frames with a sensor temperature
of 45 °C. A rough approximation is that dark current doubles for every 6 °C
rise. Using this rule of thumb, the sensor at 45 °C would produce dark current
roughly 10 times greater than the specification shows. The dark current not only
is significant on this high-frame-rate camera, but also changes by an order of magnitude
between the time the sensor is turned on (cool) and the time it reaches thermal
equilibrium (perhaps 30 minutes later).
To further complicate things, the dark
current will also change in an uncooled camera when there is a change in operating
modes, such as binning, frame rate, time delay and integration shift rate, and region
of interest. Each time the dark current changes, the noise level and “black
level” also change. To calibrate out these variations, the camera system would
have to store dark-field calibration frames for every possible operating mode, and
these would not be valid until the camera had reached thermal equilibrium.
With the current push for ever-increasing speeds,
many camera manufacturers have moved toward multiple-output sensors. This is particularly
true for time delay and integration and area-scan sensors used in high-speed inspection
where composite data rates may exceed 1 billion pixels per second. For these multichannel
cameras, varying sensor temperature adds a whole new set of challenges.
In sensors with multiple outputs, the
gain and offset for each individual video channel must be precisely matched to all
the other channels. Moreover, these gains and offsets must not change over time
and temperature. If they do, the resultant image will show “tiling”
effects, where the image section from a given output varies in contrast and brightness
with respect to adjacent image areas.
Both the gain and offset for these
areas are affected by the operating temperature. This, in turn, is influenced by
an uncooled camera’s operating mode, ambient temperature, frame rate and other
factors. Significant problems related to maintaining a good gain and offset balance
can result. Attempts to correct this mismatch can be both time-consuming and expensive
for the customer — and, in the end, a futile effort because the next change
in the operating mode or temperature will again create imbalance.
The cost of cooling
Clearly, low dark current and extreme thermal
stability offer significant advantages in light-starved applications such as spectroscopy,
crystallography and astronomy. In higher-frame-rate applications, the performance
enhancement from a cooled camera also can be substantial. Reduced dark current noise,
coupled with a dramatic increase in stability and repeatability, directly benefits
applications ranging from semiconductor and web inspection to laser profiling and
medical imaging. The problem is that such inspection applications often don’t
warrant the cost and complexity of cryogenically cooled sensors.
Fortunately, in the majority of these
applications, most of the performance improvement can be accomplished simply by
lowering the sensor temperature to a few degrees below ambient and stabilizing it.
Decreasing to just 20 °C reduces dark current in most inspection applications
to an insignificant level. The thermal stabilization ensures that performance characteristics
remain constant regardless of frame rate or operating mode. For higher-speed cameras,
this low-level cooling and thermal stabilization provide significant gains in imaging
performance with little or no impact on camera cost.
Meet the author
David W. Gardner is president of Summit Imaging
Inc. in Colorado Springs, Colo., which specializes in high-performance area-scan
camera design. He was the founder and former president of Silicon Mountain Design
MORE FROM PHOTONICS MEDIA