An optical data-flow computer

For many applications, such as signal and image processing, computer vision, and artificial intelligence, the current achievable performance is much lower than that needed. Von Neumann models cannot achieve computational rates equivalent to billions of operations per second that will be required for these applications. Therefore, parallel models of computation must be used. One attractive concept for future computer architecture is the data-flow model of computation. Theoretically, maximal concurrency can be exploited using such a model. However, the lack of efficient implementation has prevented its wide use. This paper examines how optical systems can be used to break the bottlenecks that conventional electronic implementation imposes and presents a high-performance and scalable optical data-flow architecture. The architecture exploits the high degree of connectivity and inherent parallelism in optics for implementing a highly parallel instruction-level data-flow multiprocessing system. Both architectural and implementation issues are considered. Copyright .