Reliability Analysis of Biased Majority-Vote Systems

Majority-vote systems (n-modular redundancy) are commonly used in the synthesis of reliable digital systems from unreliable components. This note presents a generalized analysis of the reliability of majority voting in terms of the conditional reliabilities of voters. The model is able to handle the case of identically distributed statistically-biased binary voters which favor one type of decision over another. The main conclusion is that for arbitrarily large n¿1. If both voter conditional reliabilities exceed 1/2, the majority-vote system is perfectly reliable. 2. If only one of the two conditional reliabilities exceeds ¿, even with average voter reliability greater than ¿, there exists an optimum finite number of modules at which the majority-vote system has peak reliability.