Quilter

Quilter is a physics-driven AI platform for electronics design that automates PCB placement and routing. Its engine uses reinforcement learning trained entirely on synthetic data generated by simulation — the site compares it to AlphaGo learning Go — rather than on human-designed boards, so it does not imitate human layouts or train on customer IP. For each job, Quilter generates dozens or hundreds of candidate board layouts in parallel and ranks them against a five-part scoring system: routing completion percentage, manufacturability (no DRC violations, trace/spacing/drill compliance), physics rule checks (ground plane overlap, overheated trace length, differential length mismatch, thermal and signal performance), fabrication fit against real fabricator stack-ups, and resource efficiency (layer count, via count, trace length, component density). Engineers can filter candidates by layer count, trace width, and vendor rules. The product works with native ECAD files: users upload Altium, Cadence Allegro, Siemens Xpedition, or KiCad projects, define the board outline, pre-place connectors, and set the floorplan, and Quilter returns files in the same format for DRC, polishing, and fab file generation — no proprietary formats or lock-in. It surfaces physics-aware considerations for review, such as bypass capacitors, impedance-controlled nets, and differential pairs, and gives transparent feedback on which design aspects are done and which need more work. Current capability scope: 100 to 1,000 components per board, roughly 2,000–5,000 pins, pin density under 20%, and through-hole vias, with blind/buried vias, length matching, and RF support in development and BGA fanout generation in beta. In a published comparison, layout work that professionals quoted at 428 hours took 38.5 hours of human cleanup with Quilter. Deployment options include Quilter Cloud on AWS (SOC 2 Type II) and self-hosted Kubernetes deployments via Helm chart with NVIDIA GPU resources, including fully air-gapped operation. A free tier covers academic, personal, and eligible professional use, while paid access is priced per project by unrouted pin count rather than per seat.