Protecting QDI interconnects from transient faults using delay-insensitive redundant check codes

Abstract

Asynchronous circuit design is a promising technology for large-scale multi-core systems. As a family of asynchronous circuits, Quasi-delay-insensitive (QDI) circuits have been widely used to build chip-level long interconnects due to their tolerance to delay variations. However, QDI interconnects are vulnerable to faults. Traditional fault-tolerant techniques for synchronous circuits cannot be easily used to protect QDI interconnects. This paper focuses on protecting QDI interconnects from transient faults. The first contribution of this paper is a fault-tolerant delay-insensitive redundant check code named DIRC. Using DIRC in 4-phase 1-of-n QDI pipelines, all 1-bit and some multi-bit transient faults are tolerated. The DIRC and basic pipeline stages are mutually exchangeable. Arbitrary basic stages can be replaced by DIRC ones to strengthen fault-tolerance. This feature permits designers to use DIRC flexibly according to the practical design requirement. The second contribution is a redundant technique protecting the acknowledge wires (RPA). Experimental results indicate that DIRC pipelines have moderate area and speed overheads. Compared with unprotected basic pipelines, the average speed decrease of DIRC pipelines is less than 50%, with the 128-bit 1-of-2 DIRC pipeline only 28% slower. In severe environments with multi-bit transient faults, the fault-tolerance capability of DIRC pipelines increases thousands-fold.