This paper describes a very fast Systolic Array for Matrix Product. Systolic Sistems usually achieve high performance by allowing computations to be pipelined over a large array of processing elements. To achieve even higher performance the systolic array proposed in this paper utilizes a column broadcasting technique plus a second level of pipelining by allowing the processing elements themselves to be pipelined. This systolic array shows not only high performace, but also optimality in terms of area and time complexity by applying recursively the fitting rows method to reduce the 1/0 bandwidth and speedup the matrix product algorithm.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados