MSS: message-based systolic structures for real-time signal processing applications

The authors present initial results of research aimed at developing a class of VLSI structures better suited for the parallel implementation of dense and structured sparse matrix-matrix and triple matrix product algorithms. They propose a structure for the efficient evaluation of such algorithms that consists of systolic nodes of fixed dimension connected via a global system network. The nodes have their own local memory and can access a common memory. Communication among the nodes is performed by message passing. This class of arrays is called message-passing systolic structures (MSS). In addition, a nonconventional method is developed that analyzes the evolution of results in the MSS array. >