Three-dimensional multiprocessor system-on-chip thermal optimization

  • Authors:
  • Chong Sun;Li Shang;Robert P. Dick

  • Affiliations:
  • Queen's University, Kingston, ON, Canada;Queen's University, Kingston, ON, Canada;Northwestern University, Evanston, IL

  • Venue:
  • CODES+ISSS '07 Proceedings of the 5th IEEE/ACM international conference on Hardware/software codesign and system synthesis
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

3D stacked wafer integration has the potential to improve multiprocessor system-on-chip (MPSoC) integration density, performance, and power efficiency. However, the power density of 3D MPSoCs increases with the number of active layers, resulting in high chip temperatures. This can reduce system reliability, reduce performance, and increase cooling cost. Thermal optimization for 3D MPSoCs imposes numerous challenges. It is difficult to manage assignment and scheduling of heterogeneous workloads to maintain thermal safety. In addition, the thermal characteristics of 3D MPSoCs differ from those of 2D MPSoCs because each stacked layer has a different thermal resistance to the ambient and vertically-adjacent processors have strong temperature correlation. We propose a 3D MPSoC thermal optimization algorithm that conducts task assignment, scheduling, and voltage scaling. A power balancing algorithm is initially used to distribute tasks among cores and active layers. Detailed thermal analysis is used to guide a hotspot mitigation algorithm that incrementally reduces the peak MPSoC temperature by appropriately adjusting task execution times and voltage levels. The proposed algorithm considers leakage power consumption and adapts to inter-layer thermal heterogeneity. Performance evaluation on a set of multiprogrammed and multithreaded benchmarks indicates that the proposed techniques can optimize 3DMPSoC power consumption, power profile, and chip peak temperature.