神经形态多核处理器的架构模型研究

项目名称： 神经形态多核处理器的架构模型研究

项目编号： No.61504059

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 肖昊

作者单位： 合肥工业大学

项目金额： 18万元

中文摘要： 神经形态芯片通过模拟生物神经系统的信息处理方式和结构特征实现低功耗的实时智能计算，对人工智能和非冯诺依曼结构的新型处理器研究有极其重要的意义。本申请项目拟将神经形态系统的理论模型与多核处理器的架构设计相结合，从数字神经元、神经元互连和突触存储三个研究点入手，研究神经形态多核处理器的架构模型。项目拟采用系统级的片上系统设计方法，通过对非线性神经元专用处理单元、基于事件驱动的分布式神经元间通信策略以及基于内存中计算的突触可塑性模型等三个关键科学问题的研究，探索神经形态多核处理器的架构模型和设计方法，为实现低功耗的实时人工智能和探索非冯诺依曼结构的新型处理器提供思路和科学依据。

中文关键词： 神经形态芯片；多核处理器；并行处理；片上互连网络；神经网络

英文摘要： Neuromorphic Chip aims at the realization of real-time and energy-efficient intelligent computation by emulating the structure and processing method of biological nervous systems. This technique is considered to be of great significance for the research of real-time artificial intelligent and non-von neumann processor architectures. Inspired by the idea of using the theory of neuromorphic systems to design the architecture of multi-core processors, this proposal aims to explore the architectural model of neuromorphic multi-core processor from the aspect of digital neuron, neuron interconnection and synapse storage. We will use the system-level design methodology and focus on three critical issues - the processing element design of nonlinear neuron models, the event-driven distributed communication protocol and the synaptic plastic model based on in-memory computing. The proposal’s objectives are two-fold: 1) to provide a massively parallel processing platform for achieving low-power and real-time artificial intelligent; and 2) as an aid in the investigation of new computer architectures, which break the rules of conventional von neumann architecture.

英文关键词： Neuromorphic Chip;Multi-core Processor;Parallel Processing;Network-on-Chip;Neural Networks