Implementing neural networks in silicon

In spite of dramatic increases in the capacity and throughput of automated systems, there remain a number of descriptively simple yet highly desirable tasks that have remained elusive. These tasks are associated with the process known as pattern recognition. If machines were able to identify patterns in electrical, visual, mechanical, acoustic, or chemical signals as quickly and reliably as living systems, our world would be a very different place. A number of tedious operations could be performed tirelessly and accurately. We would no longer have the need for locks on our automobiles and homes or keyboards on our computers. For many years, engineers and mathematicians have worked to perform computer-based pattern recognition using geometric and statistical methods, but levels of accuracy commensurate with those of human operators have been difficult to obtain. To address the overwhelming utility to perform these tasks, engineers have begun to take cues from biological systems, the simplest of which are able to perform pattern recognition with relative ease and high reliability, as a matter of their very survival.