¡¡Chinese Journal of Computers   Full Text
  TitleInjecting Redundant Instructions Behind Branches for a Fault-Tolerant Architecture
  AuthorsZHANG Shi-Jian HU Wei-Wu
  Address(Key Laboratory of Computer System and Architecture,Institute of Computing Technology,Chinese Academy of Sciences, Beijing 100080)
(Graduate School of Chinese Academy of Sciences, Beijing 100039)
  Year2007
  IssueNo.10(1674¡ª1680)
  Abstract &
  Background 		Abstract Since deep submicron manufacturing process is widely used in microprocessors, transient faults have become the main source of chip faults. A new fault-tolerant technique is proposed that inject redundant instructions behind primary branch instructions. It utilizes the wasted processing bandwidth during branch misprediction for redundant execution, hence the performance overhead is mitigated. The experiment results show that performance penalty resulted from the solution is ranging from 6% to 31%, with an average of 21%, which is much lower than that of MBI technique and almost equal to that of DIE technique. The proposal can detect and recovery faults occurring in the entire pipeline, provides short fault detection latency and requires modest hardware cost. It is well suited to realize a fault-tolerant embedded microprocessor which has a simple branch predictor.

keywords transient fault; temporal redundancy; fault-tolerant; branch prediction; embedded microprocessor

background With the widespread adoption of embedded microprocessor-based systems for safety critical application, fault-tolerant mechanisms have to be built into microprocessors. In this paper, a fault-tolerance microarchitecture is proposed that inject redundant instructions behind primary branch instructions. This work is an important component of the research on low power and high reliability for the godson1 microprocessor. The project is supported by the National Basic Research Program of China(973 Program) under grant No.2005CB321600 and the National Natural Science Foundation of China under grant No.60325205 and No.60603049 and Beijing Natural Science Foundation under grant No.4072024.