Linear Pseudosystolic Array for Matrix Algorithms

In Chapter 5 and 6 we have shown that the partitioned realization of matrix algorithms on systolic-type linear arrays (with a given number of cells) has several advantages over two-dimensional architectures. For large problems, linear structures are simpler to build and simpler to program, might have lower bandwidth requirements and better utilization, and are more suitable to incorporate fault-tolerant features than two-dimensional arrays. Based on these facts, it is of interest to look into the architecture of a class-specific linear computing structure.