Amorphous slack methodology for autonomous fault-handling in reconfigurable devices

Abstract

Amorphous Slack fault handling methodology utilizes adaptive runtime redundancy to improve survivability of FPGA based designs. Unlike conventional static redundancy based methods to achieve fault resilience, the proposed system operates in uniplex arrangement under non-contingent conditions. The proposed fault isolation algorithm is invoked upon fault detection which employs a health metric of the application operating over reconfigurable platform. This assertion applies if a signal-to-noise metric is known, as well as applications that do not possess a readily correlated metric to identify anomalous behavior. In particular, readily available processor cores allow dynamic fault identification by executing a software specification of the signal processing algorithm which is used to periodically validate critical outputs of the high-speed hardware circuit within tolerances. The results from H.263 video encoder and Canny edge detector implemented over Xilinx Virtex-4 device demonstrate autonomous recovery from permanent stuck-at faults while maintaining the throughput during fault-handling operations. The fault-detection and isolation applications are executed on on-chip PowerPC processor while the Circuit-Under-Test (CUT) is realized in hardware fabric. The proposed architecture allows on-chip processor based functional monitoring of the contained hardware resources subjected to the actual inputs of the circuit.