The parallel nature of optics and free-space propagation, together with its freedom from communication interference, makes it ideal for designing massively paralle l computers. Our architecture is highly amenable to optical implementations and aims at data-parallel applications. �Pt ics, due to its inherent p arallelism and no ninterfering communications, is under serious consideration for de signs of massively parallel processing systems of the future. To contribute to this un dertaking, designers at the University of Arizona's Department of Electrical and Computing Engi neering explored a three-dimensional oplical computing architecture under a grant from the US National Science Foundation. 11lis model-a single-instruction, multiple-data system, or SIMD-e xploits spatial parallelism and processes 2D binary images as fundamental computational entities bas ed on symbolic substi tution logic. A better alternative than electronic mesh computers, this system effectively imple ments highly structured data-parallel algorithms, such as signal and image processing. partial dif terential equations. multidimensional numerical second) and input dat a rates appr oaching 1 Gbyte/s. A common factor of these applications is a high degree of data parallelism in which simple arith metic and logic operations must simultaneously take place across all data points. 1 Computing these applications with high-throughput rates requires massively parallel processing; however, traditional electronic technology faces major limitations in achieving massive parallelism. A key feature of this type of processing is the l arge amount of communication required among the processing elements (PEs). While the design of high-perfor mance PEs has pr ogressed significantly, the progress in designing high-perfoffilancc intercon nection networks has not been satisfactory. The major bottlenecks in today's massively parallel processing systems include the limited communi cation bandwidth and the lack of cost-effective means of achieving parallel I/O.'"