A content-addressable systolic array for sparse matrix computation

Abstract A systolic array is proposed which is specifically designed to solve a system of sparse linear equations. The array consists of a number of processing elements connected in a ring. Each processing element has its own content-addressable memory where the nonzero elements of the sparse matrix are stored. Matrix elements to which elementary operations are applied are extracted from the memory by content addressing. The system of equations is solved in a systolic fashion and the solution is obtained in NZ + 5n − 2 steps, where NZ is the number of nonzero elements along and below the diagonal and n is the number of equations.