VHE: GPU-Accelerated Gate-Level Simulation at Zero License Cost

Source: Dev.to

How we built a GPU simulator to verify a 6.7 M‑gate NPU when Verilator failed

Our NPU design hit 1.4 M gates. Verilator started a convolution test:

• Runtime: 139 billion cycles
• VCD trace: 56 GB
• Status: Killed after 3 days

Commercial emulators cost a lot, and as a startup in India that wasn’t feasible.

VHE (Virtual Hardware Emulator) – a GPU‑accelerated gate‑level simulator.

┌─────────────┐    ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│   Yosys     │───▶│   Parser    │───▶│  Levelizer  │───▶│    CUDA     │
│  JSON Net   │    │  (Python)   │    │  (DAG sort) │    │   Kernel    │
└─────────────┘    └─────────────┘    └─────────────┘    └─────────────┘
                                                               │
                                                               ▼
                                                        ┌─────────────┐
                                                        │  Simulation │
                                                        │   Output    │
                                                        └─────────────┘

Design

Gates

Gate‑level netlists are represented as a directed acyclic graph (DAG). We topologically sort the gates into levels so that gates at level N depend only on gates at levels …

• WIOWIZ website:
• VHE project page: