A Novel Methodology for Evaluating the Energy Consumption of IP Blocks in System-Level Designs

System designers often face a serious challenge when estimating the energy requirements of their system. This information is essential for the design process. Especially for IP blocks, this is hard to estimate, since the influence of the surrounding system on the IP block needs to be considered. This creates need for a multi-level simulation: The system level must be simulated in conjunction with the IP core which is at gate level. Common approaches to this problem are too low-level and slow for system design. This paper shows another method: Using transpilation from netlists to a system design language, one can generate high-level modules which simulate the logic of the IP block and collect statistics about the energy consumption at runtime, using a statistically trained energy model. This has the advantage that no external tooling or complicated setups for crossing the boundary from system-level simulation to gate-level simulation are necessary. Benchmarks show that this approach, while rather simple to implement, yields energy values much closer to the ones we obtained with a professional gate-level toolchain than with simple average case figures, when evaluated with real world stimuli.