A new hybrid neural system interfacing neurons and silicon hardware for fast signal recognition

Abstract

Built on the biological neural network (BNN) theories, artificial neural network (ANN) has exhibited many significant advantages as of now. But yet, the high complexity of live beings' nervous system leads to quite limited knowledge on the working principles of learning, thinking and cognition at molecular level today, i.e. in a sense, the development of ANN has to be confined by the understanding of BNN. On the other hand, the huge memory space in an ANN chip for storing all connection weights is also a serious problem. In this paper, a novel mixed neural system interfacing biological neurons and semiconductor chip on a shared silicon wafer substrate for fast signal recognition is proposed, where three blocks are designed and interconnected. Recorded simulations with a 5 /spl times/ 5 microelectrode-array covered by a 100 /spl times/ 100 BNN show that combining the individual advantages of large-scale integrated circuits and BNN, this system has faster and more intelligent capabilities for fuzzy control, speech or pattern recognition as compared with common ways. At the same time, it can resolve the problems of huge memory space in ANN chips and the high complexity for algorithms, with an average 90.3% degree reduced efficiently between 5 trials.