Chip simulates retina

Tokyo—Since the invention of the camera, engineers have sought to perfect image capturing techniques. Microelectronics have expanded the search in fascinating new directions, including solid-state imaging, lasers, and digital cameras. Now, with the release of Mitsubishi Electric's Artificial Retina Chip, technology comes ever closer to simulating the most sophisticated of image capturing devices—the human eye.

With the ability to capture and process the image by extracting essential features directly from the projected images, the chip functions much like a retina. Its architecture consists of a 256×256 array of 35×26µm Variable Sensitivity Photodector Cells (VSPD's) embedded on a Complimentary Metal Oxide Semiconductor (CMOS) where each cell corresponds to one pixel of imaging.

"The key technology" says Project Team Leader Kenichi Tanaka, "is that we are able to change the gain, the positive and negative sensitivity, for each cell. Output voltage, therefore, is a function of the light intensity, allowing image processing directly on the chip." The VSPD array also realizes programmable filtering by employing between-pixel current mode calculations.

Because total imaging area measures 0.35×0.26 inch and total chip size is only 0.46×0.46 inch, the Artificial Retina Chip can be developed into a system that is one-half the size of the competing technology, charged couple devices (CCD). A bulky DSP or ASIC image processor is not needed; instead, a microchip located directly behind the Artificial Retina chip processes the image.

Operating at 3-5V on a single power supply, the Artificial Retina Chip uses 1/5 of the power required by a CCD camera, saving money further. It can also be manufactured using conventional semiconductor fabrication. "One problem with CCD technology is that it needs a special fabrication system," says Tanaka. "CMOS chips," however, "can be manufactured on the same line as DRAM, SRAM, and microprocessors, saving both time and money."

Based on a proprietary image-processing algorithm, the Artificial Retina module comprises a lens unit, signal processing unit, and communications unit.

One of the most distinct advantages of the technology is the ability to recognize images using a programmable scanning feature. Since each cell has a specific voltage output, the scanner can sum up the light intensity to obtain a distinct value for each image. For example, the Artificial Retina can recognize a distinct letter in the alphabet since each will have its own voltage signature.

Another useful feature is edge extraction, which can be used to identify an image such as a face or fingerprint. Varying the sensitivity repeatedly along the cells in the array will allow the retina to distinguish sudden changes in light intensity along the surface of an image. If, for example, one cell has a negative output voltage corresponding to the dark eyebrow on a face, and the adjacent cell has a positive output voltage corresponding to the light reflecting from the skin, an edge will be drawn representing the outline of the eyebrow. Since every face is unique each will have its own edge-extraction signature.

Security systems are among the very first applications. More than 500 mini artificial retina cameras have been mounted in a garage in West Ginza allowing a person in a control room to monitor 500 vehicles simultaneously. If a car leaves the lot, a green light will flash indicating a space is available. In the future, the chip will be able to detect an intruder using a motion algorithm that can calculate the "optical flow" showing the person's location and direction of escape.