论文部分内容阅读
离散 Ｈｏｐｆｉｅｌｄ 联想存储器（ Ｄ Ｈ Ａ Ｍ）在联想记忆方面有很大的优点，但是在硬件实现过程中存在着各种故障。对于软硬故障同时存在情况下 Ｄ Ｈ Ａ Ｍ 的性能进行分析是很有必要的。该文对 Ｄ Ｈ Ａ Ｍ 建立了同时包含权值误差、权值断路、神经元输出固值等各种软、硬故障的故障模型，定义了概率意义下 Ｄ Ｈ Ａ Ｍ 的鲁棒性。运用概论统计的方法推导出了运行于同步方式下 Ｄ Ｈ Ａ Ｍ 的鲁棒性，推导中对结果进行了近似处理。对 Ｄ Ｈ Ａ Ｍ 系统进行了仿真实验，与利用这种鲁棒性分析方法计算进行了对比，结果相差非常小，证实了此方法的正确性，可以用于 Ｄ Ｈ Ａ Ｍ 的可靠性计算。
Discrete Hopfield associative memory (DMA) has great advantages in associative memory, but there are a variety of failures in the hardware implementation. It is necessary to analyze the performance of D H A M in the presence of hard and soft faults. In this paper, we establish a fault model which includes all kinds of soft and hard faults, such as weight error, weighting circuit breaking, output value fixing of neuron and so on, and defines the robustness of D H A M in the sense of probability. The robustness of DHA M running in synchronous mode is deduced by using the method of general statistics. The results are approximated by the derivation. The simulation experiment of DHA M system is carried out. Comparing with the calculation using this robust analysis method, the difference is very small. The correctness of this method is proved and it can be used in the reliability calculation of DHA M system.