A New Chip Design May Lead to New Sight

March 29 -- For decades, scientists and eye doctors have been trying to develop artificial eyes that would return the sense of sight to blind and visually impaired people. And the thought of the "bionic eye" may not be too far fetched.

Many companies, such as Optobionics in Wheaton, Ill., have taken the first steps with tiny microchips that can mimic certain parts and function of the human eye — such as the rods and cones, sensors that convert light into electrical impulses at the retina located at the back of the eye.

But scientists at the Office of Naval Research in Arlington, Va., believe they are on the path to a chip that could truly mimic the entire nerve system of the retina back of the human eye.

At the heart of their potential artificial eye is a well-known chip design called a cellular nonlinear network, or CNN. In the chip, individual computer circuits are connected to each other in a checkerboard array. Each connection can be given a mathematical "weight" that "describes" the relationship of each circuit to each other.

When the chip is exposed to image data, each pixel or point of light in the picture is sent to a specific cell in the chip. Mathematical algorithms can then manipulate each connection's weight to produce different resulting images. One set of algorithms could help find the edges of an object in the image. Another set of algorithm could then find corners, while another set define contours.

Unique Advantages

Larry Cooper, the program manager at Office of Naval Research who specializes in nanoelectronics, says the CNN chip has multiple advantages that make it ideal for use in an artificial retina.

For one, the connections between each circuit are parallel, or "non-linear." That means, the calculation for each circuit is happening almost simultaneously and allows for very rapid image processing. "The time it takes a chip to [process a function] is about a microsecond," says Cooper.

Another advantage: The chip is an analog processor. Common microprocessors, such as those used in desktop PC are digital — dealing with values of "1" and "0." But the CNN chip can perform its calculations using image values that aren't as exacting — which is the same way our brain processes information.