Special-purpose processors in theoretical physics

Reviewer: John Keith Lowell

As the development of specialized-purpose computer processors continues, no discipline has so reaped the benefits of these systems as computational physics. In an area where multitudes of variables undergo numerous perturbations, specialized processors have revolutionized and streamlined our ability to perform calculations which years ago would have required decades to complete. To put this development into perspective, the authors address two main issues. They review the kinds of physical calculations which are ideal for a hardware implementation. Two ideal candidates, the Monte Carlo simulation of the three-dimensional Ising model, and the SU(3) lattice gauge theory, are chosen to illustrate what makes a particular algorithm an ideal candidate for a specialized processor. The remaining portion serves to exemplify how some particular processor architectures have been tailored to solving these and other problems in theoretical physics. In conclusion, the authors provide some thoughts on future directions. While the treatment of the subject is commendable, the prospective reader should be aware that the authors are speaking from a physics perspective, not one based in traditional computer science. Anyone looking for detailed expositions on microsystem architectures will not find it here. Nor will the nonphysicist easily comprehend and appreciate the computational advancement achieved in elementary particle theory and statistical mechanics. What results is a descriptive, but mostly nontechnical survey of special-purpose processor applications in physics only. This approach works well when addressing scientific problems but can be misleading when relating to relevant computing issues. To illustrate, the authors refer to computer-aided design techniques in speculating that VLSI technology may someday permit actual hardware realizations of conceptual algorithms through a process known as a “silicon compiler.” Upon reading further, the impression emerges that this process is more of a dream than a reality. While “silicon compilers” have yet to be perfected, current practice is that many custom and semicustom companies not only have tools today, but are producing working chips from customers' “high level” designs. The best part of the paper is its account of how the development of microprocessor systems has significantly advanced computer physics. This is a noteworthy achievement due to equal efforts by both computer scientists and theoretical physicists. Not surprisingly, hardware specialists interested in applications, and scientists looking for unique computer solutions to numerical problems, will find this article of particular interest. While the evolution of special-purpose processors is by no means complete, the authors give evidence that the future of special processors in theoretical science looks bright.