Leakage-aware intraprogram voltage scaling for embedded processors
Po-Kuan Huang, Soheil Ghiasi
Abstract
Po-Kuan Huang, Soheil Ghiasi
Abstract
With scaling of technology feature sizes, the share of leakage in total power consumption of digital systems continues to grow. Conventional dynamic voltage scaling (DVS) techniques fail to accurately address theimpact of scaling on system power consumption and hence, are incapable ofachieving energy efficient solutions. To overcome this problem, we utilizeadaptive body biasing (ABB) to adjust transistors' threshold voltage at runtime. We develop a leakage-aware compilation methodology that targets embedded processors with both DVS and ABBcapabilities. Our technique has the unique advantage of jointly optimizing active and leakage energy dissipation. Considering the delay and energy penalty of switching between operating modes of the processor and under deadline constraint, our compiler improves the energy consumption of the generated code by average of 13.07% and up to 30.26% at 90nm. While our technique's improvement in energy dissipation over conventional DVS is marginal (4.54%) at 130nm,the average improvement continues to grow to 7.8%, 15.94% and 29.56% for 90nm, 65nm and 45 technology nodes, respectively.
OpenAlex reports 16 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
With scaling of technology feature sizes, the share of leakage in total power consumption of digital systems continues to grow. Conventional dynamic voltage scaling (DVS) techniques fail to accurately address theimpact of scaling on system power consumption and hence, are incapable ofachieving energy efficient solutions. To overcome this problem, we utilizeadaptive body biasing (ABB) to adjust transistors' threshold voltage at runtime. We develop a leakage-aware compilation methodology that targets embedded processors with both DVS and ABBcapabilities. Our technique has the unique advantage of jointly optimizing active and leakage energy dissipation. Considering the delay and energy penalty of switching between operating modes of the processor and under deadline constraint, our compiler improves the energy consumption of the generated code by average of 13.07% and up to 30.26% at 90nm. While our technique's improvement in energy dissipation over conventional DVS is marginal (4.54%) at 130nm,the average improvement continues to grow to 7.8%, 15.94% and 29.56% for 90nm, 65nm and 45 technology nodes, respectively.
Key concepts: Dynamic voltage scaling, Computer science, Energy consumption, Dissipation, Scaling, Leakage (economics), Voltage, Transistor