Advanced Imaging - August 1999
Photon Vision Systems
Creates CCD Performance with CMOS Value
By Sheldon Liebman
When decisions are being made about what type of imager to use for specific applications, one of the points to consider has always been the quality of CCD imaging products vs. the lower cost of CMOS devices. Photon Vision Systems, a two-year old company in Cortland, NY, may have come up with a way for customers to have both.
Over the past few years, PVS has been working on a new, patent pending technology called Active Column Sensor (ACS), which is designed to increase the quality of CMOS imagers while decreasing their complexity and lowering their costs. Over the next few months, the company will be delivering its first products created with this new technology.
FPN, The Problem with CMOS Imagers
Most CMOS imagers today use Active Pixel Sensor technology (APS), which utilizes an amplifier for each pixel. Each amplifier requires at least 3 Field Effect Transistors (FET) to implement. Due to process variations during the manufacture of these amplifiers, the actual gain and offset of each amplifier is slightly different. In fact, PVS suggests that the average output gain is approximately 85%. As a result, APS imagers suffer from high Fixed Pattern Noise (FPN) problems. The resulting video appears as if you are looking through a dirty window.
APS systems typically counteract the gain and offset issues by creating tables of multiplier and offset values to correct the incoming video. As long as the temperature remains relatively constant, these tables can be used with little or no modification. However, the process does add to the complexity of the system.
Improving the Process with Unity
Since the biggest cause of FPN is the variation in gain from one amplifier to another, an obvious solution to the problem is to implement a Unity Gain Amplifier (UGA) for every pixel. Unfortunately, each UGA requires the use of at least 6 FETs, increasing the complexity of the product significantly. In addition, each FET decreases the active area of its pixel. With 6 or more FETs per pixel, the active area would be greatly reduced. In addition, the added complexity would raise the cost of the product, eliminating one of the major advantages that CMOS technology provides.
The challenge for the people at Photon Vision Systems was to find a way to get the benefits of a UGA without having to use 6+ FETs per pixel. Jeff Zarnowski and Matthew Pace, two of the founders of the company, looked at the problem from a variety of angles, and ultimately discovered a pattern - all but one of the FETs of each UGA for a column of pixels was redundant. The input FET changed for the UGA of each pixel, but the rest were the same. If they could be shared among all the pixels in a column, the benefits of UGA could be realized without all of the complexity and cost. The successful implementation of this concept is called Active Column Sensor and, according to PVS, it has allowed them to create the next generation of CMOS imagers.
Higher Quality Without the Cost
The best way to see the implication of ACS technology is by using the example of a 1000 x 1000 pixel array. Using APS technology, this array requires the use of 3,000,000 FETs. To implement this design using a UGA per pixel would require over 6,000,000 transistors. However, ACS technology can be used to implement this design with only 1,004,000 transistors (1,000,000 input FETs and 4 shared FETs for each of the 1000 columns). The actual numbers may be slightly different, but the concept is the same.
This design completely eliminates the gain issue that is the largest contributor to FPN while decreasing the number of transistors by over 60 percent. Higher quality is therefore achieved from a CMOS imager that is less complex and less costly to manufacture.
The second biggest contributor to FPN is variations in the offset between amplifiers. These variations are typically handled with correlated double sampling. Before each frame is captured, the initial value of each pixel is stored in a table indicating the offset compared to true black. After the frame is captured, each pixel's value is adjusted up or down based on this initial value.
The use of correlated double sampling is standard in the CMOS imager industry, but companies can choose to perform this operation either on-chip or off-chip. Since most of the APS systems already perform the gain adjustments off-chip, offset is usually handled in the same manner. Again, this increases the complexity of the system.
Photon Vision Systems has implemented their correlated double sampling on the chip. By adjusting for the variations in offset on the chip and eliminating the variations in gain, the company is able to virtually eliminate FPN before the image leaves the chip. Other factors, such as temperature, can add to the level of FPN, but gain and offset are the two biggest factors.
From Custom to Standard
When Photon Vision Systems first opened, the company offered custom design services only. Earlier this year, they began shipping their LIS family of Linear Imaging Sensors. The area sensors will be marketed under the ACS-I name and are currently in alpha testing at a number of companies. PVS plans to offer sensors up to 2K x 2K resolution with a number of features incorporated into the product. Some of these features include multiple regions of interest, an electronically controlled shutter, progressive scan and 10-bit A/D conversion with programmable gain and offset.
As part of their shift from custom to standard products, PVS is speaking with a number of large manufacturers that may be interested in utilizing CMOS imagers instead of CCD devices. Price is an obvious feature for this market, but other factors include fewer components, lower power consumption and a smaller package size. Unfortunately, PVS can't tell us yet who they're speaking with, but more information may be available before the end of this year.
Lower cost and higher quality indicate that this technology should be attractive to virtually every area of imaging, but there are a few that truly stand out. Machine vision applications, which depend on pattern recognition, can obviously benefit from higher quality without longer processing times. In fact, PVS claims the system can handle up to 20,000 frames per second in real-time.
Space-critical applications are also attracted to this technology, as much of the "support" hardware is eliminated. Compared to CCD systems, which utilize multiple clocks and more than one voltage, the ability of the ACS-based imager to operate using a single clock and one voltage is another strong incentive to explore this technology.
Lower cost, simpler design, smaller packaging, and high quality. It doesn't seem possible that one product could provide all of this, but Active Column Sensor technology from Photon Vision Systems seems close to delivering it. It will be interesting to look at this after they begin shipping finished products to see if the reality lives up to the promise.
Back to Recent L&S Marketing Articles