Teledyne Photometrics KINETIX, Manual, Page: 20 / 48

11
58-723-004 Rev A01
Chapter 3 – Theory of Operation
Introduction
Backside-illuminated scientific CMOS (BSI Scientific CMOS) sensors are a recent but rapidly
maturing development in image sensor technology. Boasting near-perfect 95% quantum
efficiency (QE), they are able to provide the highest levels of sensitivity. New cameras can
now leverage these high-QE sensors to offer a combination of low read noise, high full well
capacity, and fast readout rates that is ideal for modern low-light imaging.
CMOS Image
Sensor
Structure
A major difference between traditional CCD sensors and CMOS sensors is the location where
charge-to-voltage conversion of accumulated photoelectrons takes place.
CCD sensors transfer the pixels’ accumulated signal in charge packets in “bucket brigade”
fashion across the sensor to a common output node where charge is converted to a voltage.
The voltage is then sampled using off-chip analog-to-digital converters (ADCs) and
transferred to the PC as digital grey values.
While providing excellent quantitative photometry and very high image quality, the large
number of transfers and sequential digitization of pixels results in low frame rates. This
speed penalty increases with the number of pixels to be digitized.
CMOS sensors leverage many of the same analog signal concepts used in CCDs but place
the output node circuitry inside each pixel. This eliminates the charge transfer process.
To read the signal from a given row, the accumulated charge is converted to a voltage
inside the pixel, then each pixel in the row is connected to the appropriate column
voltage bus, where the on-chip ADCs convert the voltages to a grey value.
The parallel digitization of all pixels in a row provides CMOS devices with a tremendous
speed advantage. Imagine a CCD with 3200 x 3200 pixels – and each pixel’s voltage is
measured in 1 µsec. To read a single row, 3200 voltage measurements are performed in
serial fashion taking slightly longer than 3.2 msec, and when repeated for 3200 rows, the
entire image takes more than 10 seconds to be digitized.
On a CMOS device, the entire 3200 voltage conversions needed to digitize a row happen in
parallel. If the time to digitize a pixel remains at 1 µsec, the time to read the entire frame is
now 3.2 msec. In practice, the time saving is split between faster frame rates and slowing
the rate of pixel measurement to reduce electronic noise. For example, if the time to
measure a pixel was increased to 3.749 µsec to reduce noise, then the image sensor can
still be read in 12 msec (for a maximum 83 fps).