Realistic yield-evaluation of fault-tolerant programmable logic arrays

When analytic yield-evaluation methods for fault tolerant systems are being considered, the question of their capacity to represent and conform to reality soon becomes apparent. The goodness of analytic yield-evaluation methods depends on their ability to account for the relationship between basic and faulty components and the reconfiguration strategy (R-S). With regards to redundant programmable logic arrays (RPLA), two R-S have been proposed in the literature: static R-S: the diagnosis and the reconfiguration phases are performed independently, dynamic R-S: the diagnosis and the reconfiguration phases, for some kind of faults, are performed simultaneously. This paper highlights the necessity to model the: relationship between faulty and basic components, and adopted R-S, in order to achieve a realistic yield evaluation. We show that the yield evaluation method used in the literature for two R-S for fault tolerant RPLA is unrealistic; we propose to use two analytic yield-evaluation methods, each of which is adopted for a specific R-S. These two methods are based on fault pattern statistics, and are: Markov based method (MBM) fault pattern & reconfiguration method (FP&RM). They model the steps implemented in the R-S. Extensive Monte Carlo based simulation experiments validate the analytic approach. We present two comparisons: qualitative: between realistic yield evaluation methods and the yield evaluation method of RPLA proposed in the literature; and quantitative: between the static R-S and dynamic R-S.