Research On Real-time Fault-tolerant Mechanisms Based On Evolvable Hardware

With the development of electronic information technology, The electronic systems are becoming intelligence and miniaturization. It is a urgent problem nowadays to design an innovative and reliable electronic system. As a new technology, evolvable hardware combines the evolutional capacity of the evolutionary algorithm with the reconfigurable characteristics of the programmable devices. It has the characteristics of self-organizing, self-adaptive and self-repairing, well to make up for the shortcomings of traditional EDA design method and favored by the electronic system designers. This paper mainly research on the real-time fault-tolerant system, including circuit coding method and real-time fault-tolerant mechanism.When using the evolvable hardware to design circuit, an appropriate coding scheme should be selected firstly. Efficient coding scheme can not only improve the success rate of evolution, but also to reduce evolution time. In this paper, a dynamic adaptive coding scheme based on the traditional CGP is proposed. It mainly improved from the output selection and fitness evaluation. A dynamic random sampling model is used to select the output nodes, and the sampling probability dynamically adjusted with the number of iterations and the distribution of population fitness. A fast fitness evaluation model is built, which uses the parent individuals'fitness to calculate the offspring's fitness. Experiments show that our method can reduce the fitness evaluation time and accelerate the convergence speed.Fault-tolerant system is essentially a real-time systems. However, the common fault-tolerant technologies are difficult to meet two requirements of fault tolerance and real-time. In order to take into account both real-time and fault-tolerance. In this paper, we study the real-time fault-tolerant mechanism based on evolvable hardware, and design a real-time fault-tolerant system framework with the mechanism. To accelerate the fault repair time, the faults are divided into anticipated and unanticipated faults. For the anticipated faults, a static configuration library is applied and for the unanticipated faults, an evolutionary algorithms based on the similarity is used. When the configuration library overflows, in order to ensure long-term stable operation of the system, the system is reconfigured by using the compensation repair mode to ensure the reliability of the system.