Applying perceptrons to computer architecture

Speculation plays an ever-increasing role in optimising the execution of programs in computer architecture. Speculative decision-makers are typically required to have high speed and small size, thus limiting their complexity and capability. Because of these restrictions, decision-makers often consider only a fixed subset of the available data in making decisions, and consequently do not realize their potential accuracy. Perceptrons, or simple neural networks, could be highly useful in speculation as a means of examining a larger quantity of available data, and identifying which data lead to accurate results. Recent research has demonstrated that perceptrons can operate successfully within the narrow restrictions of speculation in computer architecture. In this paper, we examine two successful applications of perceptrons to speculation problems: branch prediction and confidence estimation. We analyse the limitations of perceptrons in these two applications, and use them to discuss how perceptrons could be applied in future computer architecture applications. Finally, we present three promising new applications of perceptrons in computer architecture.