Affiliations: Aihara Complexity Modelling Project, ERATO, JST.,
Shibuya-ku, Tokyo, 151-0064, Japan | Institute of Industrial Science, The University of
Tokyo, Meguro-ku, Tokyo, 153-8505, Japan
Abstract: The two major principles in silicon neuron implementations are
phenomenological and conductance-based. The former reproduces some properties
perceived by the designers and does not claim mechanisms are consistent. The
latter reproduces the dynamics of the ion channels on the nerve membranes.
Although it makes the silicon neurons more similar to biological ones, the
implementations tend to be complicated because it attempts to replicate the
detailed dynamics of the biological components. In previous
work [1], we proposed a simple and biologically
realistic MOSFET-based Class 2 silicon nerve membrane. It reproduced basic
mathematical structures that produce resting potential and threshold in the
Hodgkin-Huxley equations [2,3]. In this paper, we
focus on a method of designing such a silicon nerve membrane, which is based on
mathematical analyses that have been applied to biological neuron models. The
concept of the method is to reproduce the topological structure in phase
portraits of biological nerve membrane models utilizing the characteristic
curves of basic MOSFET circuitries as the elements of these for silicon nerve
membranes. The design method also revealed how to tune their parameters up.
