Peak power estimation using activity measured on emulator

Peak power analysis is a critical requirement for the design of robust System-on-Chips (SoCs). Typically power estimation relies either on “rule of thumb” activity rates or on limited simulation runs that do not guarantee a real silicon worst case estimate. Rather, we have developed and implemented a method based on extracting activity data from an emulator box running a typical real-life application with suitable instrumentation. The search-and-refine method focuses on the time window exhibiting the highest activity. The time window is in a second step analyzed using Back-End power estimation tools. Results on a typical block of a SoC are presented and discussed.