On-Line Global Energy Optimization in Multi-Core Systems Using Principles of Analog Computation

This work presents the design and the silicon implementation of an on-line energy optimizer unit based on novel analog computation approaches, which is capable of dynamically adjusting power supply voltages and operating frequencies of multiple processing elements on-chip. The optimized voltage/frequency assignments are tailored to the instantaneous workload information on multiple tasks and fully adaptive to variations in process and temperature. The optimizer unit has a response time of less than 50 us, occupies a silicon area of 0.021 sqmm and dissipates 2 mW/task.