Fault-tolerant evolvable hardware using field-programmable transistor arrays

Abstract

Abstract-The paper presents an evolutionary approach to the design of fault-tolerant VLSI (very large scale integrated) circuits using EHW (evolvable hardware). The EHW research area comprises a set of applications where GA (genetic algorithm) are used for the automatic synthesis and adaptation of electronic circuits. EHW is particularly suitable for applications requiring changes in task requirements and in the environment or faults, through its ability to reconfigure the hardware structure dynamically and autonomously. This capacity for adaptation is achieved via the use of GA search techniques. In our experiments, a fine-grained CMOS (complementary metal-oxide silicon) FPTA (field-programmable transistor array) architecture is used to synthesize electronic circuits. The FPTA is a reconfigurable architecture, programmable at the transistor level and specifically designed for EHW applications.The paper demonstrates the power of EA to design analog and digital fault-tolerant circuit. It compares two methods to achieve fault-tolerant design, one based on fitness definition and the other based on population.The fitness approach defines, explicitly, the faults that the component can encounter during its life, and evaluates the average behavior of the individuals. The population approach, on the other hand, uses the implicit information of the population statistics accumulated by the GA over many generations.The paper presents experiment results obtained using both approaches for the synthesis of a fault-tolerant digital circuit (XNOR) and a fault-tolerant analog circuit (multiplier). The experiments show that the EA (evolutionary algorithm) can synthesize fault-tolerant designs for both the analog and digital functions circuits that can recover for functionality when lost due to a-priori unknown faults by finding new circuit configurations that circumvent the faults.The paper shows that although the classic fault-tolerant design approach is able to create a reliable circuit design by evaluating the behavior of the circuit when well known faults are injected during the evolutionary process, better circuit performance, in less computation time, for a same fault-tolerant degree is achieved by allowing the evolutionary design process to be free of all faults constraints.

Autonomic fault-handling and refurbishment using throughput-driven assessment

“…CRR, by virtue of depending on a population of higher-fit alternatives that are evaluated temporally over many iterations, precludes the need for rollback of configurations and ensures higher populational fault tolerance capability. Significantly, as opposed to the work of Keymeulen in populational fault tolerance [Keymeulen00], CRR achieves device refurbishment at runtime, while ensuring sustainable levels of throughput with graceful degradation. As compared to the Roving STARs approach [Abramovici01], CRR minimizes detection latency, as faults are evident immediately upon a discrepancy at the outputs.…”

“…Hence, recent research has focused on employing the capability for reconfiguration inherent in field programmable devices to increase reliability and autonomy [Keymeulen00] [Lohn03] [DeMara05a] [Moore05] [Zhang05]. In these experiments, fault-tolerance is evolved at design-time, or achieved at repair-time using evolution after taking a detected failed unit offline.…”

Addressing the Metric Challenge: Evolved versus Traditional Fault Tolerant Circuits

“…Stefatos and Arslan [12]proposed a twolayered fault-tolerant VLSI architecture where the second layer provides for detection and correction of the first layer. At NASA, field-programmable transistor arrays (FPTA) are evolved to be fault-tolerant [7] and fault-recovering [17]. Genetic representations are also investigated with regards to fault recovery on FPGAs [9].…”

Dynamic Voting Schemes to Enhance Evolutionary Repair in Reconfigurable Logic Devices

“…The evolutionary process, which utilizes techniques such as crossover and mutation, continues till a perfect repair is found, or the pre-determined number of generations is evolved. Refer to [6] for a more detailed description of evolution and GAs, and to [2] for a description of the main concepts in Evolvable Hardware (EHW).…”