Scalable design of a programmable NMR voter with inputs' state descriptor and self-checking capability

Abstract

One of the most frequently used techniques for increasing systems' fault-tolerance is N-Modular Redundancy (NMR). The voter in an NMR system besides outputting the voting result can also indicate the situation on its input side, for instance, how many modules agree to the result of voting and which of the modules do not agree. Having this information, elaborated systems may initiate various actions e.g., interrupt or reset a processing unit or increment an error counter. Furthermore, since voters are a single point of failure in an NMR system, additional measures to increase its dependability are needed. Here, we present a design method for an NMR voter which along with the voting result, outputs the state of its inputs. It also makes self-checks of the consistency of its operation and signals errors. At last, the voter allows for each of its inputs to be defined whether the input takes part in voting or not i.e., the voter is programmable. The design method is based on a binary matrix (built according to the voter inputs) which has specific mathematical properties that enable scaling the design, as well as building the mentioned capabilities.