Sensors Support Low-Light Imaging
Sandor L. Barna
One key to customer acceptance of CMOS imagers in the mobile market space is low-light performance. This is often specified in terms of the lowest light level at which the sensor produces an image with a minimally "acceptable" signal-to-noise ratio. This value is highly customer- and application-dependent, with high-end digital still cameras requiring better performance than low-end video cameras.
In enhancing the signal-to-noise ratio, the designer rarely has the opportunity to improve the input signal level. The signal level at each pixel will be proportional to the scene lighting level (L), scene reflectivity (R), lens transmission (T) and pixel size (A), and inversely proportional to the square of the lens f number, which is related to the lens aperture:
S ≈ L*R*T*A/f2
Usually, the designer cannot influence these parameters in mobile applications where the user determines the light level and scene characteristics -- which may be improved with a flash -- and cost considerations drive the lens f number and pixel size. For instance, in wireless applications such as the mobile-phone camera, the f number is typically fixed at 2.8, and pressure from the phone manufacturer imposes constraints on the camera module size, which reduces the acceptable pixel size.
Given this, the only option is to reduce noise as much as possible. The absolute limit for noise reduction is given by the Poisson statistics (shot noise) of the incoming light. This is a physical limit that states that the maximum signal-to-noise ratio that can be achieved is the square root of the number of photons detected by the pixel.
The primary sources of additional noise in the image come from pixel performance itself. These include dark current, photoresponse nonuniformity and pixel-reset noise. Dark current is the accumulation of electrical charge in the photodiode from electron-hole pairs that are generated independent of the photodetection process. A primary source of this is the presence of impurities or lattice defects in the silicon substrate. Because these defects are localized, the dark current level is different for each pixel, leading to a fixed pattern noise in the image resembling a starry sky at long integration times in a dark image.
This magnitude of the noise source is proportional to the integration time and exponentially increases with temperature. Photoresponse nonuniformity results from pixel-to-pixel variation in the capacitance of the sense node and appears as a gain error. This can be seen as a noise source that is proportional to the signal level and that will provide the maximum limit to the signal-to-noise ratio in bright lighting conditions. The pixel-reset noise is determined by the thermal noise of the pixel photodiode in a standard three-transistor CMOS photodiode pixel. More advanced pixel structures are needed to reduce pixel-reset noise.
Active-pixel CMOS architectures use intrapixel amplification and temporal and fixed-pattern noise suppression circuitry to image with good dynamic range(~75 dB) and low noise (~15 e rms noise floor). Courtesy of Micron.
The Mobile Sensor Group of Micron Technology Inc. in Pasadena, Calif., has significantly reduced the effects of each of these noise sources through advances in process technology. Image sensor dark current is closely associated with the processes that cause reduced refresh times in DRAM, an issue central to our core business. Photoresponse nonuniformity is addressed through precision process control and device matching.
Finally, very low pixel-reset noise is accomplished through an advanced pixel that separates the photodiode region from the sense node. Specialized process steps enable complete charge transfer, resulting in a true correlated double sampling of the accumulated photodiode charge. This eliminates the thermal noise.
Zoomed fragments within this GretagMacBeth ColorChecker Chart illustrate the signal-to-noise ratio produced with low-light CMOS sensor technology. The image was taken at 5 lux and 5 fps with a fluorescent light source and f/2.8 lens.
This pixel technology places new constraints on the circuit designers, who now have to design readout circuitry that preserves the high-pixel signal-to-noise ratio. With the elimination of the pixel-reset noise, the noise of the readout circuitry must be reduced to very low levels. A critical requirement is good rejection of substrate noise, especially in more complex camera-on-a-chip applications in which full camera functionality is implemented on chip in digital logic. Micron has designed signal chains with noise floors of only a few electrons (referred to the pixel photodiode), even in the presence of hundreds of thousands of gates of digital logic running at the pixel rate on the same substrate.
These process and design improvements have resulted in a family of image sensors having extremely good low-light performance. The sensors are competitive in performance with several comparable CCDs on the market today, with considerably more functionality at a much lower overall system cost than traditional CCD products.