Time Moore: Exploiting Moore's Law From The Perspective of Time

Abstract

Moore's law has served as a goal for the semiconductor industry for more than 50 years. After decades of relentlessly racing forward, convincing new evidence now shows that the end of conventional Moore's law scaling is really near. Besides the More More Than and Beyond Moore approaches, which concentrate most of their efforts on squeezing processing power from the perspective of space, it is useful to inspect Moore's law from the time perspective since matter exists in both space and time. In microelectronics, time is reflected as clock frequency. Clocking is another direction to exploit Moore?s law in greater depth: we will play with frequency, which is a forgotten opportunity that remains open for a potentially significant profit. This article investigates a more sophisticated strategy to manipulate frequency for clocking an electronic system. The aim is to obtain higher compute energy efficiency from given silicon real estate, not by packing more transistors into it, not by blindly increasing clock speed, but by adjusting clock frequency dynamically wherever and whenever possible. This approach is termed Clock Moore. Furthermore, we propose an idea of using the rate of switching as a computational variable to encode information, which is termed Rate Moore. Together, Clock Moore and Rate Moore make up Time Moore, which enhances transistors' collective computing capability by using time more efficiently.