FPGA realization of the LIRA neural classifier
Optical Memory and Neural Networks
| Hi-index | 0.00 |
An analog hardware neuron model having biological neuron characteristics is proposed and its basic characteristics are reported. The proposed circuit model is constructed from a membrane capacitance part, negative resistance circuit part, and reactance circuit part, and realizes the membrane excitability observed in biological neurons by voltage-controlled negative resistance characteristics. This model exhibits an action potential generation mechanism similar to that of biological neurons. HSPICE simulation reveals that this model can reproduce both the well-known responses to excitatory inputs and the postinhibitory rebound (PIR) firing which is a neuronal activity occurring by the release from an inhibitory input discharge, and that the characteristics of the input pulse amplitude-latency and entrained response to the periodic input are similar to those of biological neurons. In addition, this model is developed by considering compatibility with integrated circuit processes, and design results of LSI implementation of the model are also presented in this article. PIR firing cannot be reproduced by the simple neuronal model used in conventional neurochips, but it plays an important role in information processing in the brain. The functionality of neurochips is expected to be improved by simulating biological neurons by such a biologically inspired hardware neuron model. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(6): 84–91, 2005; Published online in Wiley InterScience (). DOI 10.1002/scj.10676