Wireless synaptic / neuro devices based on interactions of local electric-fields and CDMA communication technology

摘要

In this report, we propose a novel LSI architecture that allows silicon LSIs to implement large-scale neural networks; i) neural networks consisting of a number of neuro chips (neurons) that use CDMA protocols to communicate with the other neurons, ii) neural networks on single neuro chip that consists of a number of neuro devices with a micro-antenna device. As an example of i), we propose a CDMA-Hopfield neural network, aiming at examining possible effects on retrieval properties of the network. Extensive simulation results indicated that the Hopfield network of N neurons could retrieve signal patterns from P memory patterns when P/N≈0.1. Furthermore, a wireless synaptic device based on local interactions of electrostatic fields is also introduced as an example of ii). We present simulation results of the synaptic device with basic neuron circuits, which indicates that the wireless device (chip) can be developed by using current device-manufacturing technology.