Research & Papers

Context-aware Simopt-Power cuts FPGA dynamic power by 6.8%

Yosys-based framework recycles discarded simulation data for smarter optimizations.

Deep Dive

Context-aware Simopt-Power, an open-source simulation-guided FPGA design optimization framework, combines switching-activity metadata from pre-implementation simulations with structural features like logic depth and fan-out to target high-impact netlist regions. On the Koios deep-learning benchmarks, it achieved an average 6.8% dynamic-power reduction with 11.2% LUT overhead, using a Yosys/ABC flow.

Key Points
  • Pre-implementation simulation metadata is typically discarded; Simopt-Power reuses it for power optimization.
  • Achieves average 6.8% dynamic-power reduction on Koios deep-learning accelerator benchmarks.
  • Open-source implementation using Yosys/ABC flow; replaces fixed thresholds with architecture-aware parameters.

Why It Matters

Enables FPGA designers to cut power consumption with minimal area overhead using widely available simulation data.