CMOS Coming Out Party?
CMOS Image Sensors: Imaging Systems On a ChipTop: Progression path for on-chip logic integration, present system boundaries (shown by dotted lines) will eventually merge to produce single chip imaging solutions.
The Imaging Market and Key Technologies
The market for image sensors has undergone a dramatic shift in the last five years. Up to the mid 90's the imaging market was dominated by CCDs, however, advances in CMOS fabrication technology, the need for low power portable devices and price pressures have been responsible for enabling CMOS imaging technology to become a serious contender to the mainstream CCD technology. CMOS sensors accounted for less than 10% of the 90 million imagers shipped in 1999, however, their sales are expected to grow to over 50% of the 300 million imagers expected to be sold by the year 2004. This explosive growth will be partially fuelled by existing applications such as fax machines, scanners, security cameras, and camcorders. It will be, however, the other rapidly growing applications in the consumer imaging industry such as digital still cameras, toy and PC cameras, cameras for cell phones and PDAs, biometrics, and automobiles which will result in the exponential growth in both the imaging market and consequently the market share of CMOS imagers.
Growth in markets for both CMOS and CCDs is expected to remain strong for the next five years. Both technologies will establish their respective niches based on their strengths and weaknesses. While sales for CCDs are expected to remain strong in the high resolution applications and in captive markets such as camcorders, the usage of CMOS sensors will experience steady growth in the lower end, but high volume, consumer applications that are driven by power consumption and cost.
The first CCD was invented at the Bell Labs; it was not long before that CMOS image sensors were announced. The limitations of the available CMOS process at the time made the performance of these sensors unacceptable for most applications and the technology lay dormant till the late 1980's. Starting in the 1990's with the availability of sub-micron CMOS processes that offered considerably reduced leakage currents and noise, the interest amongst the technical community in CMOS sensors started to grow once again. The early pioneering work came from researchers at the Edinburgh University in Scotland and the Jet Propulsion Laboratory (JPL) in California. Since then, a number of key semiconductor suppliers whom already have access to leading edge CMOS processes have started to assimilate imagers into their product lines.
Sensors at Work
Solid state imagers, including CCDs and CMOS imagers, use a particular physical property of silicon called the photoelectric effect. When photons strike silicon, they excite electrons from the valence band to the conduction band. The number of electrons thus "freed" are proportional to the photon flux or simply the number of photons striking silicon and their wavelength. Thus in effect the number of electrons collected vary with the intensity and the color of the illumination. If these electrons can be collected, metered, and measured — an electronic representation of the incident scene can be created and stored. For a typical imaging application, the scene is focussed on the pixel array with the help of an optical lens for spatial sampling. The size and number of pixels ultimately determines the minimum size features that the sensor is able to resolve. In this regard the larger the number of pixels, the finer is the sensor's resolution. The pixel is the eye of the sensor, and its conversion efficiency of photons to electrons determines its sensitivity. This is typically a function of the size and aperture of the pixel. The charge to voltage conversion ratio is also important since ultimately the electron charge packet has to be converted to a signal that can be measured and manipulated for signal processing.
As the pixels are typically laid on an orthogonal grid (for area arrays), a mechanism has to exist to readout all the rows and columns of individual pixels that comprise the sensor. Up to this point, CMOS and CCD sensors are identical in their functionality. It is only in the implementation where these two differ.
The fabrication technology for CCDs has not been subject to much evolution since its invention. CCDs have to be fabricated on NMOS processes. The pixel sizes are rarely constrained by the minimum feature size that the process allows. As a result, the process technology for CCDs has not seen a significant change in the last 30 years. Perhaps the most significant shift has been in the area of yields; improved yields now allow manufacturers to produce multi-megapixel resolution CCD sensors in volume. Slow equipment evolution has also forced CCD manufacturers to continue using 4"-6" wafer fabrication lines where the rest of the industry is on the 8"-12" curve. The mismatch in processing technology for CCDs precludes the integration of additional on-chip functionality beyond light sensing and charge-to-voltage conversion. The result — all signal processing has to be done off-chip with CMOS ICs. The inherent advantages of CMOS processes form the basis of the advantages of CMOS imagers. First, CMOS sensors can take full advantage of the economies of scale offered by this technology, hence keep costs low. Second, the products can be run on most commercially available sub micron CMOS processes with minimal changes thus allowing fabless design houses to just as easily manufacture at foundries. Third, since CMOS sensors are manufactured on the same processes as those for other analog and digital signal processing circuits, virtually any functionality can be integrated on chip, thereby reducing the number of components and achieving lower system cost. Fourth, CMOS ICs run on 3.3V power supply and are rapidly moving to 2.5V. CCDs, on the other hand, use a 15V power supply. The power requirements for CCDs makes it difficult to integrate them into portable, low power applications such as cell phones, PDAs and other hand held scanning devices.
in the Bayer pattern.
CMOS Coming On
With that said, CMOS sensors have their own set of problems. They face the daunting task of unseating a proven and mature technology that already has several captive markets. The image quality of CMOS sensors can vary significantly between different manufacturers and in most cases is not at par with the CCDs. The main issue at hand with CMOS sensors is their noise performance. These sensors use active pixels, hence their alternate name Active Pixel Sensors (APS). Each pixel has an amplifier that converts the collected charge packet to a voltage. Mismatches in transistors can result in mismatches in amplifiers thus causing a mismatch in the
A representative architecture of a CMOS imager (designed by Motorola).
response of pixels. The other major source of noise is the dark
leakage current which results in collection of thermally generated
charge that contaminates the photo electrically generated charge
that is collected from the incident scene. Since each pixel has an
amplifier, the entire pixel area cannot be used for photon
detection. This results in the aperture or fill factor being less
than 100% causing a drop in sensitivity and the output signal. As
the signal drops and the noise floor rises, the net signal-to-noise
ratio (SNR) deteriorates resulting in a lower system dynamic range.
Some of this loss is offset by the fact that all signal processing
is done on chip, close to the signal source. Improved mixed signal
design methodologies have also resulted in substantially reducing
these noise sources. Advances have thus resulted in image quality
of CMOS sensors that was barely acceptable a decade ago, being
comparable to that of CCDs today.
The future for CMOS technology looks bright (as does the image quality). The continued development of this technology should have a dramatic effect on the digital camera market — perhaps much sooner than originally anticipated.