A University of Southern California professor is claiming he has developed the first analogue-to-digital converter suitable for integration into a neural network, reports Microbytes Daily. The significance of Dr Bing Sheu’s converter chip is its speed: Sheu claims it can be used to interpret analogue signals from a robotics or vision system based on a neural network, and process them in a digital computer in a matter of minutes, against the six hours or so required by conventional analogue-to-digital circuitry. Since the chip can handle the parallel processes of a neural network, the size and complexity of the conversion equipment in the robot or vision system is greatly reduced, Sheu claims. He reckons that the size reduction will, among other things, lead to production of smaller robots, so saving space in installations such as factory assembly lines, and enable the design of more mobile and portable robots, making them more appropriate for use in space exploration – Carver Mead’s Silicon Retina was cited as a system that could particularly benefit from the new circuit. The 4-bit analogue-to-digital converter circuit, which is based on the Hopfield Circuit designed by biophysicist John Hopfield, has been built and tested but has not yet been integrated into a neural system. Dr Sheu is also working on an 8-bit version of the chip.
