基于故障检测与流量分析的低功耗片上网络可靠性最优路径容错路由算法研究

Network-On-Chip(NOC) is one of highly concerned research topics and one of advancing front questions in the field of microelectronic and communication technology today. Fault tolerant mechanism has been the hotpots in research of NoC. However, there are still a lots of problems to be perfected and more optimization on the fields of concrete fault model, fault tolerant mechanism for specific fault, fault detection mechanism, port selection, and the problem of considering the service quality in the fault-tolerant routing algorithm.This project aimed at three important topics as follows:1) Power optimization in NoC system which is the key module easily to be interfered;2) Route architecture design built-in error detection unit;3) Design of the flow detection unit which perceives flow status;4) Design of the fault-tolerant routing algorithm of the best route based on transient and permanent fault-tolerant mechanism.Following questions is to be solved:1) design the LDO architecture which is suitable for NoC system power supply which is on-chip integration without output, capacitance type, quick response and low power consumption;2)model a functional fault of a transient fault-tolerant mechanism of Noc; 3) key techniques such as fault-tolerant routing algorithm of node unit. The novelty of research includes:1) optimization in NoC system which is the key module easily to be interfered; 2)design of route architecture design built-in error and flow detection units. This project is full of significance to prompt research of NoC.

片上网络（NoC）是当前微电子与通信技术的热点、前沿问题。容错机制是NoC领域研究的热点，但在具体的故障模型、针对特定故障的容错机制、故障检测机制、端口选择、考虑服务质量的容错路由算法等方面仍有待完善和进一步优化。本项目针对新形势下NoC的热点问题拟重点研究：1）NoC易受干扰关键模块的电源优化；2）内置错误检测单元的路由架构设计；3）感知流量状况的流量检测单元的设计；4）基于瞬时故障及永久性故障容错机制的最优路径容错路由算法设计等四方面问题。突破1）适合NoC系统供电的片上集成无输出电容型快速响应超低功耗LDO架构设计；2）瞬时故障的容错机制NoC功能故障模型的建模；3）以节点为单位的容错路由算法研究等关键技术。在1）优化 NoC系统中易受干扰关键模块的电源；2）加入故障检测单元、流量检测模型的路由架构的设计这两方面取得创新性成果。本项目的实施对促进国内NoC研究具有积极意义。

本课题针对片上网络具体的故障模型、故障检测机制、故障容错机制等问题进行深入的研究和优化。（1）在共享缓冲区的路由架构设计上，对改进的奇偶转弯路由架构设计，无虚通道的改进型XY容错路由架构设计以及基于虚通道的容错路由架构设计进行了研究，并取得相应成果。（2）设计了三款LDO：基于辅助推挽输出级的片上集成LDO、基于自适应输出级的片上集成LDO、基于自适应偏置电路的片上集成LDO。（3）提出了一种针对片上跨时钟域互连线的全速测试方法，并采用映射算法来实现流量的检测。（4）研究了瞬时故障容错机制和永久故障容错机制，并提出了针对这两种机制的细粒度故障模型和容错路由算法，包括无虚通道的改进型XY容错路由算法、单节点故障下最短路径路由的容错路由算法、基于虚通道的多故障节点的容错路由算法以及一种基于全局网络状况的容错路由算法。（5）在完成了项目相关任务后，又进一步针对NoC的映射过程进行研究，提出了基于链路的测试感知动态映射算法和延长NoC和Core寿命的LBC+LBL的生命周期感知映射算法。本课题对NoC系统全面分析故障发生的类型，研究对应的容错路由算法，对今后 NoC 在故障容错方面的进一步研究有重要的意义。