A novel Markov model for the reliability prediction of fault tolerant non-homogenous multipipelines

A novel Markov model for the reliability prediction of fault-tolerant non-homogenous VLSI and WSI multipipeline arrays is presented. The PEs (processing elements) of the array are assumed to fail independently (with a constant failure rate) at different moments and the transition rate between two different error states is constant. A total system failure is reached when the number of working pipelines becomes less than a predetermined number S/sub m/. Thus the reliability of the multipipeline array is defined as the probability of having S(t) greater than or equal to S, where S(t) is the number of survived pipelines at time t, and S/sub m/ is the minimum number of survived pipelines that is needed for the multipipeline to be considered in a working condition. In addition to predicting the reliability, the Markov model can be used in design optimization to determine the best possible design among multiple alternatives. Several experiments are conducted that demonstrate the ability of the proposed Markov model to predict the reliability and to evaluate various design alternatives.