Simulation-Driven SoC Architecture Exploration for Edge AI Workloads
Published in International Symposium on Computer Architecture (ISCA), 2025
We present a simulation-driven design methodology that integrates Verilator-based RTL simulation with QEMU full-system emulation to support rapid architecture exploration for edge AI SoCs. The methodology allows architects to sweep across design points — cache sizes, interconnect widths, NPU configurations — while running real inference workloads compiled for the target ISA. Results demonstrate that simulation-driven exploration can identify Pareto-optimal design configurations with significantly fewer hardware iterations.
Recommended citation: Jesse. (2025). "Simulation-Driven SoC Architecture Exploration for Edge AI Workloads." International Symposium on Computer Architecture (ISCA).
Download Paper
