Very Large-Scale Neuromorphic Systems for Biological Signal Processing
Very Large-Scale Neuromorphic Systems for Biological Signal Processing
复制标题
用于生物信号处理的超大规模神经形态系统
DOI：
10.1007/978-3-319-67723-1_13
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
S. Schaafsma
中科院分区：
文献类型：
--
作者：
F. Catthoor;S. Mitra;Anup Das;S. Schaafsma
本章是一份白皮书，描述了一个将神经形态系统扩展到“人脑大小”复杂性的平台。这样的系统对于与生物数据交互的大规模搜索和分析任务是必要的。这个系统将由与成人大脑相似数量的神经元和突触组成。最大的瓶颈之一是连接数十亿神经元所产生的巨大突触复杂性。本章的目的是描述一种可行的架构，可以处理这种大规模神经形态系统所需的巨大通信带宽。提出的方法是基于这样一个假设，即只有当神经形态系统在能量消耗和大小方面与人类大脑有些相似时，我们才能理解它的效用。受SoC架构最新进展的启发，本文提出了一种新颖的可扩展集群间通信网络。一个特别有用的例子发生在全局突触通信中，连接局部突触阵列集群。所提出的解决方案的核心是CMOS后端线（BEOL）中的新型开关架构，该架构预计具有极高的功耗效率。与在所有连接的本地集群上共享的固定预定义总线相比，所建议的解决方案将允许可以动态切换的大量专用点对点连接。
This chapter is a white paper describing a platform for scaled-up neuromorphic systems to ‘human brain size’ complexity. Such a system will be necessary for massive search and analysis tasks while interacting with biological data. This system would consist of similar number of neurons and synapses as in an adult human brain. One of the largest bottlenecks is the huge synaptic complexity that would result from connecting billions of neurons. The purpose of this chapter is to describe a feasible architecture that could handle the enormous communication bandwidth necessary for such a large-scale neuromorphic system. The proposed approach is grounded in the assumption that we would only be able to appreciate the utility of a neuromorphic system when it is somewhat similar to the human brain in terms of energy consumption and size. Inspired by the recent advancements in SoC architecture, a novel scalable intercluster communication network is proposed here. A particularly useful instantiation of this occurs for the global synaptic communication, interconnecting the local clusters of synapse arrays. The core of the proposed solution is a novel switching architecture in the CMOS back end of line (BEOL) that is expected to be extremely power efficient. In contrast to a fixed predefined bus that is shared over all connected local clusters, the proposed solution will allow a multitude of dedicated point-to-point connections that can be switched dynamically.