Matrix power generation using an optical reduced-bandwidth systolic array

The problem of repeatedly squaring a matrix of given bandwidth on a fixed sized systolic array is considered. A general method is described for multiplying two matrices on a hexagonal systolic array with reduced input and output connections, and the concept of optical preprocessing is introduced as an efficient method for expanding host array communication bandwidths. The arrays presented increase computation by factors related only to the bandwidth of the matrices multiplied, and show the effectiveness of hybrid electro-optic devices with optics restricted to the role of flexible interconnections.