| Hardware speed has been the main driving force during the spectacular rise of chess computers from duffers to Grandmasterly players within the span of the past decade. Despite frequent claims to the contrary, pushing hardware speed has proven to be a reliable way of improving chess computer performance. This dissertation will look into the past and document the hardware efforts behind ChipTest, 1987 ACM Champion, and Deep Thought, 1988 ACM Champion and the reigning World Computer Chess Champion. It will also glimpse into the future about possible hardware advances and present a new class of parallel search algorithms that promises speedup of the order of hundreds and maybe even thousands. A chess machine capable of evaluating one billion positions per second appears to be feasible with the current technology. Such a machine, projecting from the performance of existing machines, should have a good chance of defeating the human World Chess Champion.;The second half of the dissertation presents a class of new parallel ;The first half of the dissertation describes the Deep Thought chess hardware and examines some paper designs related to the final effort. Details are also given for the single chip move generator which is the core of the Deep Thought special purpose hardware and offers about three orders of magnitude improvement in speed chip-count ratio over previous designs. A surprise is that the simple move ordering provided by this single chip move generator turns out to be more efficient overall than a far more complicated move ordering proposed in earlier literature. |
