Harvesting through array partitioning: a solution to achieve defect tolerance

The problem of "harvesting" aims at obtaining the largest possible working array from an original array in which a (possibly high) number of faults are present, typically at the end of production. While complete spares efficiency has been proved to require channel width and interconnection length proportional to the dimensions of the array, techniques previously presented in literature achieved fixed-dimensions for channels and links by accepting low spares efficiency. In the present paper an approach previously adopted for survival to faults (i.e., to achieve arrays of fixed dimensions with a pre-determined distribution of spares) is extended to the harvesting problem by proving that partitioning the original array into subarrays (within a limited set of constraints) lends to effective reconfiguration. Spares efficiency is seen to be quite satisfactory.