Formal methods based computer-Aided Synthesis of STochastic inExact in-memoRy computing systems
Abstract
The need for speed for computer systems is out-pacing the systems abilities to process the information. The state-of-the-art in computing has grown from teraflops (10^12 FLOPS - floating-point operations per second) and terabytes of storage to petaflops (10^15 FLOPS) of computation and petabytes of storage. Dennard scaling is a scaling theory which approximately states that as transistors decrease in size, the transistor s power density stays constant, so that the power use is proportionate to area; however, this scaling theory has not held up because as the transistors become smaller, more are added, and the heat of the chip becomes too great and parts of the chip must be shut off (resulting in "dark silicon") which is of no use. This proposed research will investigate formal methods based algorithmic synthesis of approximate stochastic computing systems that exploit emerging ultra-dense crossbar memories to perform in-memory execution of software systems. This project will transform two known deficiencies of emerging memories, stochasticity and sneak-paths, into solutions to the dark silicon problem besides leveraging their well-known benefits: speed, size, energy and the ability to store multiple bits. The results of this research work can have a substantial impact on execution within software systems by advancing state of the art within memory execution by providing a solution to the "dark silicon" problem and will improve the warfighters systems dramatically.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 15, 2016
- Source ID
- FA95501610255
Entities
People
- Sumit Kumar Jha
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Central Florida