[an error occurred while processing this directive]

Skip to Content

Computer Science Colloquia

Friday, March 30, 2012
Liang Wang
Advisor: Kevin Skadron
Attending Faculty: Sudhanva Gurumuthi, Chair; Kim Hazelwood and Westley Weimer

12:00 PM, Rice Hall, Rm. 204

Ph.D. Quals Exam Presentation
Mitigating Dark Silicon with Near-Threshold Computing


Due to limited scaling of supply voltage, power density is expected to grow in future technology nodes. This increasing power density potentially limits the number of transistors switching at full-speed in future chips, leaving others inactive, a phenomenon which some now refer to as dark silicon. This project proposes the idea of improving throughput with higher utilization of die area by exploiting near-threshold computing (dim silicon) instead of dark silicon. To test this hypothesis, an analytical model is developed to quantify the performance of many-core systems operating at near-threshold voltage. The model augments Amdahls Law with detailed scaling of frequency and power, calibrated by circuit-level simulations using a modified Predictive Technology Model (PTM) and factoring in effects of process variations. Our analysis shows the effectiveness of near-threshold computing in improving the chip utilization, as well as higher throughput, of up to 2x over the dark silicon. Our results also suggest that near-threshold computing cores favor the high-performance process with a lower threshold over the low-power process with a higher threshold. Results also indicate near-threshold computing in future technology nodes is more vulnerable to process variations. In the worst case, variations eliminate all benefits of near-threshold computing for both the throughput and chip utilization. Finally, our results show that dim operation is preferable to dark operation, but that near-threshold is not viable without mechanisms to tolerate process variation.