Where we were back in 1997.
In January of 1996, the common wisdom shared by many digital camera companies was that Complementary Metal Oxide Semiconductor (CMOS) technology just couldn’t compete with Charge Coupled Device (CCD) technology because of certain shortcomings in image quality and sensitivity that were inherent to CMOS. Many of these companies had actually built prototype CMOS cameras and based their opinions on their practical experience with these designs. However, as is often the case in high technology, what is true today, may not be true tomorrow.
CMOS and CCD technology share many common attributes. They are, after all, similar in structure, material, and basic operation. The principal advantage of CMOS is that it can be fabricated at facilities that have higher volume, more modern equipment, higher yield and therefore lower cost than CCD manufacturers. In addition, CMOS requires less sophisticated driving and A/D circuitry which also affects the cost of implementation within the camera. These are reflected in the elimination of specialized clock drivers and correlated double sampling A/D converters that are required for CCD implementations. CMOS also operates at lower voltages, which means that less power is required, resulting in longer battery life.
CMOS does, however, have two principal disadvantages. First, the stationary noise pattern is higher than found in a CCD; and second, in that each pixel contains not just a photo diode, but in addition, amplifiers and selection circuitry, the actual area of the pixel that gathers photons is smaller. This is reflected as a reduction in sensitivity in comparison to a similar sized CCD. This is often referred to by describing the “fill factor” of the pixel, or the percentage of the pixel that gathers photons.
In January of 1997, Sound Vision introduced its first in a series of CMOS digital cameras, the SVmini. Eventually, this camera became available in the retail market under the brand names: the”ViviCam-3000″ by Vivitar; and the “Sharpset-8000” by UMAX Technologies. The SVmini is based upon a custom 1000 by 800 CMOS sensor manufactured for us by VLSI Vision Ltd., an Edinburgh based company which was spun off from the University of Edinburgh. This camera demonstrates clearly, both the advantages and disadvantages of CMOS.
For example, the retail cost of the camera is approximately $399, significantly less expensive than CCD based cameras with comparable resolution and features. If one were to disassemble the SVmini and, for example, an Olympus D-200 or D-300, the difference in cost would be more apparent. You would find essentially a single circuit board within the SVmini in comparison to 6 circuit boards in the Olympus cameras. This significant reduction in complexity and component count directly translates into a lower selling price. As such, the SVmini offers tremendous value at its market price.
However, this first CMOS camera also demonstrates some of the weaknesses expected from CMOS. For example, the reduction in sensitivity makes indoor photography without a flash virtually impossible. While the SVmini excels in the capture of outdoor pictures on sunny days, it suffers in low light conditions. Second, the processing time between pictures, which can be in excess of 20 seconds, is a direct result of the need to perform substantial image processing on each image, prior to compression, in order to extract every last bit of quality from the high level of stationary noise. I often muse at the thought that the DSP within the SVmini, which executes a multiply and add every 36 nanoseconds, must execute 600,000,000 instructions per picture. Most of Sound Vision’s added value to the CMOS technology is within this software.
Where it is going.
Based upon the results that we have achieved with the SVmini, it is now clear to most in the industry that CMOS has arrived. Clearly, it represents a cost/performance tradeoff in comparison to CCD technology, that cannot be ignored, and will have an important place in, at least, the consumer portion of the digital camera business. However, my own beliefs still differ from the “current wisdom.” Based on what we have learned at Sound Vision, I am convinced that CMOS will eventually dominate the consumer market and may even possibly dominate the niche markets as well. For this to happen, four improvements must occur:
The low fill factor of current CMOS detectors can be improved in either of two ways. First, micro-lenses can be added to the detectors. This innovation is not new to CMOS; it has been done for several years on other technologies. Each miniature lens is only the size of a single pixel and it is deposited on the sensor using photo etching techniques while the sensor is still part of a larger wafer. The purpose of each lens is to gather light from the insensitive portions of the 10.8 micron square pixel and “focus” it down to the photo diode. Current sample tests have demonstrated that this technique clearly works and it increases the sensitivity of our current sensor by almost a factor of two.
The second technique available to increase the sensitivity is to use a finer structure in the creation of the portions of each pixel that are not light sensitive. Again, this is not new technology. Our current sensor is fabricated with a 0.8 micron structure while many fabricators are commonly producing 0.35 micron devices. However, the principal advantage of staying with the larger structure is that yields can be higher (since you are more tolerant of imperfections). Therefore, it may only be necessary or desirable to use micro-lenses to achieve comparable sensitivity to CCDs.
As with all new semiconductor devices, analysis of early designs indicate changes that are desirable for increasing the tolerance of the device to certain types of imperfections during fabrication as well as further controlling the sources of noise. In addition, process controls can be determined that also increase the yield. This is especially true of these types of devices that have several fabrication steps, namely: CMOS detector fabrication, Color deposition, Lens deposition and sensor encapsulation. In time, the skill acquired in controlling these steps in combination with subtle design changes will greatly increase yield and further drive costs down.
3. Color Deposition
Possibly, the single largest area for improvement lies in the field of color deposition. Further improvements in the spectral characteristics of dyes will greatly enhance the color quality and indirectly, the sensitivity, noise and spatial resolution of the camera. Again, as in the case of micro-lenses, we have already tested prototype sensors using vastly improved dye sets that can markedly increase the performance of our camera.
4. Better Processing
The final area for improvement lies with advancements in image processing. As we gain a better understanding of the performance of the CMOS sensor it becomes possible to improve both the quality and the speed of images. Our latest research is quite encouraging about the ability to increase the resolution and color quality from even our existing sensor.
CMOS has already proven its viability as a competitive technology in digital photography. In the near future, though, it will, through several well understood advancements, move to become the dominant technology through its clear cost/performance advantages. Sound Vision is committed to continuing its effort to pioneer this technology and, through it, to continue to force higher quality in combination with lower cost for digital cameras. We are at the threshold of a major change in the photographic industry and CMOS is the vehicle to carry us through that change.
Where we are in 1999.
Two years have passed since I first wrote this article and it seemed to me that it deserved to be updated. The past two years have seen several important improvements to CMOS sensor technology as well as a confirmation of the opportunity that CMOS represented. Today, there are many CMOS foundries that have announced or are already in production of CMOS sensors. Devices range from QCIF to Megapixel resolution and already, CMOS is replacing the CCD in high volume applications where cost is a driving factor.
In fact, for 1999 and 2000, CMOS sensors will probably overtake CCDs for the VGA and lower resolution portions of the digital camera market. Prices for VGA CMOS sensors are already below $10 and continue to fall. As well, the dark current and resulting stationary noise patterns of these devices have fallen dramatically as the foundries have learned to slightly modify their processes to improve device qualities. As a result, CMOS devices of today offer no serious drawbacks in comparison to CCDs for these low-resolution applications.
At the lower price category, the additional cost difference represented by the integration of the A/D and timing circuits required by CCDs represents a more important advantage than at the higher cost camera segment of the market. This further draws CMOS into these markets.
The future, with respect to digital cameras, is clear. However, the more important opportunity for CMOS sensors still rests with its ability to mix other functions with the sensor in a single die for exceedingly low-cost applications. These “smart sensors” which will include some processing and “seeing” in a single package will revolutionize the sensor market and open extraordinarily high volume opportunities in industrial and consumer imaging. In most of these applications, a person will never actually see the acquired image, but rather, a processed result will be offered outside the device. That result might be an indication of motion, identity or position or an endless list of other image attributes.
We are at the edge of a major transition in the inclusion of imaging in the machines around us and the advent of the CMOS sensor was a critical step to the growth and success of this market.