A fast and efficient technique to apply Selective TMR through optimization

Fault tolerance is an important factor for circuits in critical applications, especially those working in harsh environments. There are many techniques to increase reliability of circuits, being those based on redundancy very popular. In this way, Triple Modular Redundancy (TMR) is frequently used, but it usually incurs high area costs. That is why other alternative techniques, as Selective TMR, are used in order to reduce this cost. In this technique, only a subset of registers is tripled, those that are more sensitive and produce a higher error rate in the circuit. However, the problem of these methodologies is the complexity of finding the optimal set of registers to triple, what usually leads to very high computation times. In this paper, a novel solution that improves Selective TMR is presented, based on the automatic and fast calculation of an initial partition prior to the optimization process. The solution has been tested on a real communication circuit, a Feed-Forward Equalizer.

► We have proposed a new technique to apply Selective TMR.
► We have defined a set of topological criteria to model circuits statically.
► We have used an optimization algorithm to automate the whole process.
► The model has been tested on a large number of examples and a real case study (FFE).