Three-dimensional computational wavefronts for matrix product

Abstract A three-dimensional wavefront array for matrix product with minimal block pipelining period of 1 is introduced and compared to existing systolic array architectures for matrix product. An optimal processor-time product of n 3 with cycles defined computationally by two operations is obtained when successive problem instances are considered. The 3-D architecture; is extensible and scalable, is cycle invariant (all respects), is node invariant (all respects), has minimal node complexity of one multiply and one addition per cycle, has unidirectional and local data forwarding in three dimensions, has 100% utilization of processing elements, and has a cycle-invariant one-to-one correspondence between input/output ports and input/output matrix elements.