Autonomic fault-handling and refurbishment using throughput-driven assessment

A new paradigm for online EH regeneration using Genetic Algorithms (GAs) called Competitive Runtime Reconfiguration (CRR) is developed where performance is assessed based upon a broad consensus of the population instead of an individual-centric fitness function. Outliers in a population of alternatives are identified using detection techniques that leverage temporal fitness information while requiring neither exhaustive testing nor resource-intensive evaluation. Relative fitness measures support graceful degradation even in the presence of unpredictable changes in the operational environment, inputs, or the FPGA application. CRR adapts using a Sliding Evaluation Window in conjunction with a periodically updated outlier cut-off that aims toward fault identification with at least 99.5% coverage assuming equiprobable inputs, without the extensive downtime associated with exhaustive testing. CRR achieves regeneration without additional test vectors while maintaining a significant percenta...

• Type: Journal
• Year: 2011
• Where: ASC
• Authors: Ronald F. DeMara, Kening Zhang, Carthik A. Sharma
• Keywords: Artificial Intelligence, ASC 2011, Competitive Runtime Reconfiguration, Exhaustive Testing, Online Eh Regeneration