Array architectures for iterative matrix calculations

The paper considers the important problem of fast computation of iterative matrix equations. Three architectures for fast implementation of Faddeev´s matrix inversion algorithm are described. The recursive nature of Faddeev´s algorithm is indicative of many other algorithms, based on the Cayley-Hamilton theorem. All such algorithms seem well suited to the class of orbital architectures reported by Porter and specialised herein. The architectures presented here are essentially 3-dimensional structures. They clearly demonstrate several advantages over comparable 2-dimensional structures particularly with respect to speed and I/O requirements. Besides, the presented architectures are chosen to highlight various 3-dimensional solutions and associated tradeoffs that are representative of the subject class of problems.