PCBRouteAgent: How Close Are LLM Agents to Becoming PCB Routing Engineers?
AuthorsYue Zhang, Jilei Sun, Liqiang Jing, Yijie Wei, Ruohan Yang
AffiliationsThe University of Texas at Dallas · Texas Instruments Inc · Iowa State University
PCBRouteAgent pairs an LLM’s board-level routing plans with KiCad’s routing engine to tackle complex PCB layouts.
Key results
Total synthetic and real-world PCB boards in the benchmark.
Claude Opus 5.5 result on small real-world boards.
Claude Opus 5.5 result on medium real-world boards.
Claude Opus 5.5 result on large real-world boards.
Percentage-point increase in clean-pass rate.
Reduction in average violations versus Freerouting.
What the paper found
PCBRouteAgent tackles PCB routing by separating board-level strategy from trace-by-trace execution. Its multimodal planner uses board images and structured electrical and geometric data to build minimum-spanning-tree connection plans, prioritize nets, and define routing corridors and layer transitions. A detailed executor then uses KiCad’s native push-and-shove router, bounded local recovery, and A* detours; when a route still fails, structured feedback prompts targeted plan revisions while preserving accepted connections. Evaluated with Anthropic’s Claude Opus 5.5 on 247 boards, the system achieved clean-pass rates of 94.9% on D3-A, 100% on D3-B, and 40% on the largest-board split, D3-C. Those results beat the strongest evaluated baselines by 13.9 percentage points on D3-A, 36 points on D3-B, and 10 points on D3-C. On D3-C, it also cut average design-rule violations from 32.0 to 6.3, an 80.3% reduction, despite leaving six of ten boards incomplete. The study also compares against interactive agents using OpenAI’s GPT-5.4 and Anthropic’s Claude Haiku 5.5. Its central caveat is that connected, rule-clean traces are not necessarily electrically validated: in one example, the agent routed USB D+ and D- independently rather than as a differential pair, while crowded-board routing order remains a challenge.
Original abstract
Automating printed circuit board (PCB) routing requires coordinating board-level connectivity decisions with geometric execution under strict design rules. Although recent agents can interact with native routing tools, reliable routing on complex boards remains challenging. We present PCBRouteAgent, a multimodal PCB routing agent inspired by how human engineers coordinate global strategy with local execution. A Global Routing Planner combines board images and structured electrical and geometric information to specify connection topology, routing order, corridors, and layer transitions. A Detailed Routing Executor realizes these guides through KiCad's native routing engine and performs bounded local recovery. Unresolved failures are returned as structured feedback, enabling targeted guide revisions while preserving accepted routes. We evaluate PCBRouteAgent on 247 synthetic and real-world boards from four PCBWorld-Bench test splits. It achieves clean-pass rates of 100%, 94.9%, 100%, and 40% on D2, D3-A, D3-B, and D3-C, respectively. On the three real-world subsets, PCBRouteAgent outperforms the best evaluated baseline in clean-pass rate by 13.9, 36, and 10 percentage points, respectively.
Read the original paperMore in AI Hardware
Browse all 40 papers →Halley: A Predictive Control Plane for Optical AI Data Center Networks
Manav Mishra, Krishna M. Sivalingam
Halley anticipates failures in optical data-center networks and coordinates lasers, switches, and thermal controls to keep more AI training capacity online.
QFabric: Guarded Cross-Domain Task Recovery and Its Limits on a Linux-RTOS Board
Ayushman Bhattacharya, Anwesha Chakraborty
QFabric tests whether tasks can escape timing trouble by moving between Linux and an RTOS, and finds that careful guards mostly prevent moves that would make performance worse.
A Hybrid GPU and Processing-Near-Memory Architecture with Shared DRAM for Efficient Prefill-Decode Disaggregation
Sanghyeok Han, Inseong Choi, Byungkuk Yoon, Gyeonghwan Park, Jae-Joon Kim
DuoPD pairs GPUs with memory-side processors to run LLM prefill and decoding together and improve inference throughput.