Architecture and Applications of DADO: A Large-Scale Parallel Computer for Artificial Intelligence

摘要

As part of our research on very high performance parallel architectures, we have been investigating; machine architectures specially adapted to the highly efficient implementation of artificial intelligence (AI) software. In the course of our research we designed DADO, a highly parallel, VLSI-based, tree-structured machine, and implemented a high-speed algorithm for production systems on a simulator for DADO. Subsequent research has convinced us that DADO can support many other AI applications, including the very rapid execution of PROLOG programs, and a large share of the symbolic processing typical of contemporary knowledge-based systems. In this brief report, we outline the hardware design of a moderate size DADO prototype, comprising 1023 processing elements, which is currently under construction at Columbia University. We then sketch the software base being implemented on a small 15 processing element prototype system including several applications written in PPL/M, a high-level language designed for specifying parallel computations on DADO.