3DIC SoC Test Benchmarks (Power Traces)
3DSIC SOC Test Benchmarks (Power Traces)
The format for power traces files for each die is based on the power trace format used in the Hotspot thermal simulator. Each power trace file corresponds to a single test of the benchmark, as symbolized by the name. More specifically, the format for each file name is as follows...
(MODULE_NAME)t(TEST_NUMBER).ptrace
Power Trace Format
The first line of each floor plan lists each module in the benchmark, including modules in other dies if a multiple-die stack. Each subsequent line represents the average dynamic power consumed by the test across 400 clock cycles in each module. For most tests, only one module will consume power, but modules that have embedded modules will also have their embedded modules consuming power as well.Scaling
The way power values are generated may not be appropriate for all studies. However, there are ways to scale power values to comply with a study.The power density of circuits used in these benchmarks ranged between 0.9 and 2.0 W/mm^2. Although these values are reasonable given the power density of modern integrated circuits and the power during test is substantially higher than during normal operation, these values may still be out of bounds for a particular study. Scaling these values can be easily done by multiplying/dividing each value in each power trace by a given constant.
Power traces were generated presuming modern integrated circuit construction (<45 nm feature sizes) with a test clock of 120 MHz. This may not be optimal for several studies, dynamic power values can be scaled to fit the needs of the study. In general, larger feature sizes imply less power consumption, and faster clock speeds imply larger power consumption. Since dynamic power consumption is linked the number of switching transistors, and the number of switching transistors does not change with different feature sizes and clock speeds, power values can be scaled to meet a project's technology needs.
Static Power
Although these benchmarks do offer dynamic (switching) power values, they do not offer static (leakage) power values. This is mainly due to the need of temperature values in determining leakage power, and temperature greatly depends on the environment in which benchmarks are used. However, researchers are encouraged to use novel approaches to include steady state power, such as assigning constant values based on module areas.home
