TOKYO Canon Corp. has developed a 3-million-pixel CMOS sensor that challenges the widely held idea that CMOS modules can't beat the resolution of charge-coupled-device (CCD) sensors in digital cameras.
Canon will use the sensor first in the digital version of its EOS single-lens reflex camera, dubbed EOS-D30, to be marketed in September.
"Canon introduced a digital still camera in the early '80s, but it was too early and was not successful," said Fujio Mitarai, president of Canon. "Since then, we have been concentrating on analog cameras and have let our competitors get ahead in the digital still camera market. But while we were behind, we used the time to develop key components, and now our digital still camera strategy is in good shape," he said.
"We will never sell the CMOS sensor alone to other companies," Mitarai said. "We'll use the sensor for our strategic products. I expect that the sensor will pull up our share in digital still cameras to around 20 percent."
The CMOS sensor has a total of 3.25 million pixels. Its effective sensor size is 22.7 x 15.1 mm, or APS-Classic size, with 3.11 million effective pixels.
CMOS sensors in general have speed and power consumption advantages over CCD sensors because CMOS parts run on about 5 volts, while CCDs need about 20 V. But CMOS sensors fall behind when it comes to picture quality, especially in a dim light, because of the noise in each pixel.
Canon engineers developed three new technologies for their CMOS sensor: on-chip noise reduction; electronic charge transfer; and on-chip programmable gain amplification. All were key breakthroughs, said Shigeyuki Matsumoto, senior general manager at Canon's device development center.
The new sensor has an amplifier in each pixel to convert optical signals to electronic signals. The unevenness of those amplifiers causes noise that has a fixed pattern. To eliminate the noise, Canon engineers developed on-chip noise-reduction circuitry that reads out the mixture of noise and optical signals in 10 milliseconds and then pure noise in another 10 ms. When the pure noise is subtracted from the mixture, a pure optical signal remains.
Random noise is another problem. The noise is due to the unevenness of the initial level of optical signals at each pixel. Canon engineers improved the pixel structure by transferring electronic charges completely to reset the initial levels of optical signals and noise in each pixel.
The programmable gain amplifier is the first block integrated on the sensor chip to take advantage of the CMOS process. The amplifier eliminates noise and reads out signals at a speed of 12 MHz.
Moving forward, Canon engineers have two major goals in mind for the next step in CMOS sensor development. One is to develop a larger sensor with a higher resolution. Another goal is to develop system-on-chip ICs that integrate the sensor with various blocks on a single chip.
The sensor is fabricated on a 0.35-micron process. Compared with CCD sensors with the same resolution, the CMOS sensor requires about 15 percent fewer masks, which, ultimately, will give them a cost advantage, said Matsumoto.