The dream of building machines that mimic the behavior of animals and people dates to the dawn of technology. With the advent of cheap and powerful digital computers, that dream is becoming reality. Computers, though, struggle hard to process the type of visual input people can instantly absorb, as when we identify a familiar face, read a scrawl, or hit a hurtling tennis ball. One reason for the contrast is that visual images contain so much information: a 1-second-long, uncompressed NTSC video segment amounts to about 22 MB of data. The need to process, store, and ship such vast data streams is what hampers machine vision. Yet miracles of real-time visual behavior are performed by the common house fly, whose brain is the size of a grain of rice. Clearly there is much to learn from the computational strategies of the nervous system.

For a decade now, research at many university laboratories has sought to understand the biological circuits and principles that underlie vision and vision-based behavior in flies, frogs, cats, monkeys, and human beings. At the same time, the complexity of the circuitry a single silicon chip can support has reached new heights. Capitalizing on these gains, a few researchers have built electronic chips that mimic neurobiological circuits related to visual processing: so-called neuromorphic integrated circuits, a term coined by Carver Mead at the California Institute of Technology (Caltech), in Pasadena. A neuromorphic imaging sensor consists of arrays of photoreceptors combined with analog circuitry at each picture element (pixel) in such a way as to emulate the vertebrate retina. More specifically, the sensor, like the retina, can adapt locally to vast changes in brightness, can detect edges, can signal temporal changes, and can detect motion [Fig. 1].

Until recently, vision chips of this kind were laboratory curiosities. Now, they are powerful enough for use in a variety of products. In the long term, the principles of neuromorphic design will enable machines to interact with the environment and with persons, not through keyboards or magnetic-strip cards, but with the help of robust, cheap, small, and real-time sensory systems of the type that have been ubiquitous for the past 400 million years. Appliances will become life-like: smart doors will let us pass once they have seen our faces; cars will navigate by themselves; and roach-like cleaners will scurry along floors to remove dust and dirt. Last but not least, because of their similarity to biological nervous systems, neuromorphic systems can provide a "natural" substitute for damaged parts of the human nervous system, such as the retina or parts of the cerebral cortex.