Constrained binding and scheduling of triplicated algorithm for fault tolerant datapath synthesis

The combination of triple algorithm redundancy and vote-writeback (TAR/VW) is a promising approach for designing an application specific fault-tolerant datapath circuit. The cone partitioning of an input application algorithm in TAR/VW framework increases the opportunity of resource sharing while keeping the fault tolerance ability. However TAR/VW combined with cone partitioning requires a specialized high level synthesis which can treat the complicated resource sharing conditions for fault tolerance. This paper treats high level synthesis for this purpose, and proposes a novel heuristic approach based on a two-phase binding. The resource binding in the first phase utilizes three reservation tables to assign resources to each cone properly while keeping the resource sharing constraint for fault tolerance. The second phase is the detailed resource binding and scheduling in each cone. Through experiments, our heuristic method is shown to produce comparable solutions to ILP-based exact solutions with much faster than ILP solvers.