Digital neural circuits : from ions to networks

Abstract

The biological neural computational mechanism is always fascinating to human beings since it shows several state-of-the-art characteristics: strong fault tolerance, high power efficiency and self-learning capability. These behaviours lead the developing trend of designing the next-generation digital computation platform. Thus investigating and understanding how the neurons talk with each other is the key to replicating these calculation features. In this work I emphasize using tailor-designed digital circuits for exactly implementing bio-realistic neural network behaviours, which can be considered a novel approach to cognitive neural computation. The first advance is that biological real-time computing performances allow the presented circuits to be readily adapted for real-time closed-loop in vitro or in vivo experiments, and the second one is a transistor-based circuit that can be directly translated into an impalpable chip for high-level neurologic disorder rehabilitations. In terms of the methodology, first I focus on designing a heterogeneous or multiple-layer-based architecture for reproducing the finest neuron activities both in voltage-and calcium-dependent ion channels. In particular, a digital optoelectronic neuron is developed as a case study. Second, I focus on designing a network-on-chip architecture for implementing a very large-scale neural network (e.g. more than 100,000) with human cognitive functions (e.g. timing control mechanism). Finally, I present a reliable hybrid bio-silicon closed-loop system for central pattern generator prosthetics, which can be considered as a framework for digital neural circuit-based neuro-prosthesis implications. At the end, I present the general digital neural circuit design principles and the long-term social impacts of the presented work.