Research On Adaptive Evolvable Hardware And Its Fault-Tolerance Technology

The motivation of developing electronic and computer engineering is some requirements in special context, nowadays, the human beings explore the unkown world is more active than ever before, as a consequence, the micromation and intelligent emerging as a new requirement of systems, which exhibits the future development of electronics. An obstacle in front of traditional EDA technologies is the inconsistency between micromation and reliability, intelligent and modularization, Evovlable Hardware (EHW) is the most idea technology to overcome these problems so far. EHW introduce evolutionary mechanism to implement in-situ-self reconfigurable system, it includes various systems which capable of adapting to environment and improving the perfomance during its execution. The evolutionary mechanism brings self-organic, self-adaptive and self-recovery to EHW, the self-recovery be derived from its capability of self-adaptive to environments, structural evolution with EHW is the most significant approach of constructing self-recovery systems. Howerver, many adaptive systems at present are based on parameter adjustment, few investigations are about structural evolution. Hence, This dissertation concentrates on researching structural evolution based Adaptive Evolvable Hardware (AEH), the ultimate intention is to construct different self-recovery systems by each adaptive behavior of EHW.Real-time is crucial to self-adaptive and self-recovery EHW, it depends on evolution efficiency, this dissertation discusses this problem from multiaspect, such as representation parameters, evolutionary algorithm, evolution granularity, the complexity of EHW, etc. The reprsentation is the key issue of EHW technology, a good representation is propitious to improve the evolution efficiency. In order to avoid some shortcomings of Cartesian Genetic Programming (CGP), this dissertation proposes an Extend Cartesian Genetic Programming (ECGP) method, it places the dedicated output nodes behind array nodes, all of the nodes in the array are potential usable, so it improves the efficiency of genetic operators. Furthermore, the link codes and function codes in ECGP are two different parts of the chromosome, both of them are arranged sequentially, so it is able to avoid illegal individuals and take separate genetic operators to different parts. The experimental results show that the ECGP takes prominent improvement in efficiency campare with CGP. Moreover, due to the redundancy of ECGP, adopt the innovational uniform crossover operators is helpful for promoting the evolution efficiency.This dissertation defines "searching space" to generalize the complexity of EHW, it is exponential with the scale of the target circuits, this is the essential reason of low evolution efficiency. Searching space appropriately interpret the scalability problem of EHW, it incorporates as "fitness stalling effect". Divide-and-Conquer is one of the possible method to takle the scalability problem of EHW, This dissertation proposes a Parallel and Recursive Decomposition (PRD), it regards the targets before and after Shannon decomposition as "equivalent circuits", meanwhile, introduces "elitist individual" preserval and "genotype melioration" mechanisms into output decomposition, it transforms the evolution of complex circuits into parallel evolution of the sub-circuit of it or its equivalent circuit. The experimental results indicate that the generations of PRD to evolve an available configuration is only linearity increased with the scale of the target circuitry, as well as it can skip the "fitness stalling effect" in time, and consequently overcome the scalability problems.Most EHW-based fault-tolerant systems at present are isomerous redundancy fault tolerant (IRFT), these systems are not structural evolution, they only utilize EHW technology to design some fault tolerance, defect tolerance or noise robust circuits, and then combine with traditional redundancy technology to realize system-level fault tolerance. There are no evolution mechanism existed in IRFT systems, their structurals can not be adjusted with environment, they are still confined to traditional traditional redundancy technology, so they are unable to achieve self-recovery in situation. This dissertation attempts to introduce the structural evolution of EHW into system-level fault-tolerance, it constructs a self-adaptive EHW system base on the structural adjustment of combinational logic circuits, then simulates the self-adaptive behaviors from three scenarios of improve structure, undeterminated task and phenotypic plasticity, these self-adaptive behaviors can achieve different fault-tolerant models. The adaptive of EHW improve structure conduces self-reconfigurable fault tolerance (SRFT), it takes EHW as the detected system as well as reconfigurable target, when the system failed, EHW reconfigure itself to find a new structural to "bypass" the fault components, the system consequently restores itself. The advantage of SRFT systems is able to implement self-recovery with very little redundancy, but it also has two drawbacks:First, the recovery process needs a relative long time, it is difficult to meet the real-time requirements of self-recovery; Second, during EHW is reconfiguring,the system is always in a chaotic state, which could damage the system structure.In order to overcome the deficiencies of SRFT, this dissertation presents a compensation fault tolerance model (CFT). CFT regards EHW as an auxiliary repair module of the detected system, it separates the failed system and reconfigurable module to avoid the risk of destructing the original structural which exists in self-reconfigurable fault tolerance. The experiments show that the compensation fault tolerant system is not only save much recovery time, but also can greatly promote the reliability with small amount of redundancy. On this basis, an evolvable self-recovery system is further constructed, it establishes a pre-configured database by failure modes and effects testing (FMET) to recover the anticipated faults, which consequently avoid the unnecessary evolution and reduce the recovery time. Simultaneously, the evolvable self-recovery system utilizes the self-adaptive characteristic of EHW to recover unanticipated faults, consequently improve the reliability in different environments, the experimental results illustrate this evolvable self-recovery system ensures a high reliability even in rigorous fault scenarios.