FoMR: Post-Silicon Microarchitecture

North Carolina State University, Raleigh NC

Investigators

Abstract

High-performance microprocessors are the computational horsepower behind all of society's infrastructure. The seemingly unending march of their progress, known to the public as "Moore's Law," has all but slowed owing to semiconductor technology and manufacturing limits. Continued progress requires breakthroughs in microarchitecture: the way the microprocessor executes applications, in particular, the way it unlocks hidden parallelism within them. Researchers have proposed many exotic microarchitecture enhancements in recent decades, but their commercialization has been stymied because they are too specialized. Strategies that speed-up any application were exhausted long ago. All that remains are strategies that pinpoint specific application behaviors, leading to complex microarchitecture enhancements that perform exceedingly well on some, but not all, tasks. This project fundamentally changes the value proposition of any microarchitecture enhancement. The impacts of this project are: scalable microprocessor performance in the post-Moore era, working closely with industry partners to ensure meaningful results and speed technology transfer, and training students of diverse backgrounds for the workforce. The key idea of the project is to tightly couple the fixed hardware components of a state-of-art microprocessor with reconfigurable (not fixed) hardware components. The fixed hardware still fetches and executes instructions in basically the same way. Enhancements programmed into the reconfigurable hardware intervene to unlock more parallelism in the instruction stream. Different enhancements can be synthesized into the reconfigurable hardware as the task at hand changes. Thus, microarchitecture enhancements are instantly commercializable/deployable. Moreover, taken altogether, the suite of available and future microarchitecture enhancements covers a wide range of application behaviors for acceleration. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.