应用于片上网络的容错机制

Fault-Tolerant Schemes for Networks-on-Chip

本文提出了一整套容错机制来解决片上网络通信中遇到的故障。对于暂时性故障,使用端对端的硬件重传机制;对于永久性故障,使用软硬件结合的方法进行故障的检测、诊断和修复。仿真显示,该容错机制能够比较全面的对故障进行处理。本论文还对提出的容错路由器和容错网络接口进行了硬件设计,在0.13μm工艺下,两个通信组件的面积仅0.12mm2和0.46mm2,相对于没有容错功能的路由器和网络接口,面积仅增加0.145mm2。

A set of fault-tolerant schemes are proposed to solve the faults in the communication of Networks-on-Chip （NoC）. When transient fault occurs, an end-to-end retransmission scheme using hardware-base approach is proposed. When permanent fault occurs, fault detection, diagnosis, recovery schemes using hardware- and software-based approaches is proposed. As shown by the simulation, these fault-tolerant schemes could deal with faults roundly. This paper also implements the hardware in the schemes, which composes of the fault-tolerant router and fault-tolerant network interface （NI）. In 0.13 μm CMOS technique, the area is only 0.12 mm^2 and 0.46 mm^2, which is only 0.145 mm^2 extra comparing with the router and NI without fault-tolerant function.