Periodically Self Restoring Redundant Systems for VLSI Based Highly Reliable Design

Abstract : Error masking by majority voting remains an important fault tolerance technique for realizing highly reliable computer systems for critical control applications. However, VLSI technology has imposed a relatively high cost on hardware voter circuits because of their high interconnect complexity. In this paper we present and analyze a new design for dedundant microcomputer systems which appears well suited for implementation from VLSI modules. In the proposed design, redundant computing units (CU's) making up the system communicate with each other periodically to restore units that may have been disabled by transient faults. The system is also protected against permanent failures. To evaluate the proposed approach a reliabilty mode of triple redundant periodically self restoring systems is developed in this paper. The model accounts for both permanent and transient faults during system operation, as well as the possibility of undetected failures in the redundant units at the start of the mission. (Author)