A mixed analog/digital chaotic neuro-computer prototype: a 200-neuron dynamical associative memory

摘要

We have proposed a system architecture for a large-scale chaotic neuro-computer. The system has a hierarchy including "board," "unit," and "system" levels In this paper, we implement these hierarchy onto a mixed analog/digital circuit system. The board-level consists of a neuron-board and a control-board. The unit contains some neuron-boards (max. of 10) and one control-board. A connection of some units (max. of 100) through a specially designed ring bus completes the systems. We build a two-unit prototype system to confirm the proposed architecture. The two-unit system is sufficient to verity the whole system composed of 10,000 neurons because the system has a hierarchical structure and at the same time, the hardware for each hierarchy has a module structure. We observe a dynamical auto-association from the two-unit prototype using 169 neurons.