7

Current VLSI technologies provide for the fabrication of chips with several million transistors. With these technologies a single chip may contain one powerful digital processor, a huge memory containing millions of very simple units placed in a regular structure, and other complex functions. A powerful combination of a simple logic processor placed in a regular structure is the cellular automaton invented by John von Neumann. The cellular automaton is a highly parallel computer architecture. Although many living neural circuits resemble this architecture, the neurons do not function in a simple logical mode: they are analog “devices.” The cellular neural network architecture, invented by Leon O. Chua and his graduate student Lin Yang [1] has both properties: the cell units are nonlinear continuous time dynamic elements placed in a cellular array. Of course, the resulting nonlinear dynamics in space could be extremely complex. The inventors, however, showed that these networks can be designed and used for a variety of engineering purposes, while maintaining stability and keeping the dynamic range within well-designed limits. Subsequent developments have uncovered the many inherent capabilities of this architecture (IEEE conferences: CNNA-90, CNNA-92, CNNA-94, 96, 98, 00, 02; Special issues: Int. J. Circuit Theory and Applications, 1993, 1996, 1998, 2002 and IEEE Transactions on Circuits and Systems, I and II, 1993, 1999, etc.). In the circuit implementation, unlike analog computers or general neural networks, the CNN cells are not the ubiquitous high-gain operational amplifiers. In most practical cases, they are either simple unity gain amplifiers or simple second-or third-order simple dynamic circuits with one to two simple nonlinear components. Tractability in the design and the possibility for exploiting the complex nonlinear dynamic phenomena in space, as well as the trillion operations per second computing speed in a single chip are but some of the many attractive properties of cellular neural networks. The trade-off is in the accuracy; however, in many cases, the accuracy achieved with current technologies is enough to solve a lot of real-life problems.