Synthetic PCB Autorouting Dataset for Two-Layer Zero-Obstacle Pathfinding

1 million synthetic samples for a fundamental printed circuit board autorouting subproblem. The dataset, created by tscircuit in collaboration with seveibar and makeshifted, focuses on two-layer routing with no obstacles or impedance constraints. It frames autorouting as a multi-agent pathfinding problem for vision model integration.

Use Cases

• Training vision models to predict optimal trace paths between components on a two-layer PCB grid.
• Benchmarking multi-agent reinforcement learning algorithms on the pathfinding subproblem of PCB autorouting.
• Developing models that learn routing heuristics from the 1 million synthetic board state and solution pairs.
• Simulating and analyzing the density limits of trace routing in a simplified, obstacle-free environment.

Strengths

• Contains 1 million synthetic samples, providing a substantial volume for model training.
• Focuses on a well-defined, fundamental subproblem (two-layer, zero-obstacle routing) for clear benchmarking.

Limitations

• Synthetic data may not capture the full complexity and noise of real-world PCB design constraints.
• Limited to a specific, simplified routing scenario without obstacles, pads, keepouts, or impedance requirements.

Provenance

Source

tscircuit, in collaboration with seveibar and makeshifted.

Collection Method

Synthetically generated autorouting dataset.

Freshness

Last updated on 2026-02-26.