physical design

This work addresses the challenging problem of spatial packaging layout and routing in three-dimensional interconnected systems with physical interactions by proposing a hybrid optimization framework. The approach employs geometric abstraction based on Maximum Disjoint Ball Decomposition (MDBD) and integrates stochastic initialization, gradient-based refinement, and an interior-point method to effectively tackle the nonlinear, non-convex, and continuous spatially coupled design problem. A newly introduced benchmark enables verifiable performance evaluation, demonstrating that the proposed method outperforms existing techniques by over 10% across multiple scenarios. The solutions achieve a relative error of only 0.6–2% compared to ground-truth values, significantly enhancing both convergence stability and solution optimality.

This study addresses the challenge of spatial layout optimization for interconnected systems within non-convex design spaces by extending the SPI2 framework. It introduces, for the first time, a geometric representation based on Maximal Disjoint Ball Decomposition (MDBD) combined with differentiable inside-outside tests, enabling component placement under arbitrary non-convex boundaries. The method integrates computations of centroid and moment of inertia and establishes an end-to-end CAD workflow that supports automatic assembly reconstruction. By simultaneously satisfying geometric constraints, routing requirements, and physical performance objectives, the approach guarantees geometric feasibility within numerical precision. The efficacy and practicality of the proposed method are demonstrated through a multi-system co-layout case study of a synthetic aircraft auxiliary unit.

This study addresses the challenge of simultaneously achieving component alignment, system coordination, solution reliability, and computational efficiency in physically interacting interconnected systems within three-dimensional space. To this end, the authors propose a decomposition-based collaborative optimization framework that, for the first time, embeds port-alignment constraints into the SPI² architecture. Treating component positions as design variables, the method employs a penalty function to enforce system-level feasibility and enables automatic generation of initial designs. By integrating gradient-based optimization for enhanced numerical stability and coupling it with NSGA-II for efficient multi-objective search, the approach achieves high-quality coordinated solutions. Demonstrated on automotive powertrain and battery-chassis integration cases, the framework significantly outperforms discrete exhaustive search, delivering superior system-level coordination while substantially reducing computational cost.

This work addresses the challenge of translating natural-language design intent into physically realizable assembly instructions. We propose a novel “Brick-Bag” physical API paradigm, using LDraw as an intermediate representation to establish an element-level, assembly-oriented universal language. Our method integrates a tool-augmented large language model (LLM), LDraw-based textual modeling, Python-based procedural generation, and structured connection-constraint reasoning—ensuring geometric validity, mechanical feasibility, and sequential constructibility. Compared with pixel-based diffusion models and conventional CAD approaches, our framework significantly improves component interoperability and capability in generating complex, multi-step assembly instructions. We validate the approach across satellite, aerospace, and architectural domains, prototyping over 3,000 components; all four original designs were successfully realized as physical assemblies. The framework delivers high-fidelity, modular, and scalable assembly outputs.

This work addresses the challenge that existing large language models struggle to simultaneously satisfy stringent geometric, routing, and electrical connectivity constraints in dense PCB layout design. To bridge this gap, we introduce OmniLayout, the first multimodal benchmark specifically tailored for PCB layout, which jointly models schematic diagrams and physical layouts. The benchmark encompasses four constraint-aware reasoning tasks designed to systematically evaluate model capabilities in geometric reasoning, routability, preservation of electrical functionality, and tool invocation. Integrating industrial-scale layout data, geometric constraint modeling, routing analysis, and circuit verification, our framework exposes critical limitations of current models—particularly their weak geometric reasoning, poor routing optimization, and insufficient functional consistency—thereby filling a crucial void in evaluating multimodal collaborative reasoning within electronic design automation.

针对VLSI物理设计中的自动宏放置挑战，提出PlaceReasoner-Beta框架，通过视觉-语言模型、几何验证器和物理验证器等组件实现闭环推理优化。

This study addresses the reliance on manual layout design for superconducting quantum chips and the limited adaptability of existing automated approaches to physical constraints. To overcome these limitations, this work establishes, for the first time, a rigorous mapping between physical properties and geometric formulations, proposing a three-stage end-to-end automated design flow. The methodology integrates geometry-aware global partitioning, integer linear programming-based port assignment, and hierarchical routing techniques, generating fabrication-compliant layouts without requiring simplifications. Experimental results demonstrate that the proposed approach reduces the design cycle to seconds or minutes, achieving a 25-fold improvement in efficiency. Furthermore, its effectiveness is validated through the open-source tool mqt-scpd.

Current large language models exhibit significant limitations in handling complex PCB routing tasks that must satisfy geometric, topological, and electrical constraints, and there is a lack of evaluation benchmarks grounded in real-world industrial scenarios. To address this gap, this work proposes OmniRouting—the first large-scale reasoning benchmark for PCB routing—comprising 1,681 industrial schematic–layout paired designs that integrate semantic and multimodal information along with joint schematic–layout constraints. The benchmark defines four progressively challenging constraint-aware tasks: geometric routing, design rule compliance, electrical functionality preservation, and tool-augmented agent reasoning, incorporating PCB geometry, component placement, netlists, stackup configurations, and EDA tool interfaces. Experiments reveal substantial deficiencies in existing models regarding path planning, rule adherence, and electrical connectivity. All data, evaluation code, and tool interfaces are publicly released to advance AI-driven PCB design research.

Existing text-to-CAD generation methods struggle to produce mechanical assemblies that are engineering-compliant, physically consistent, and reusable due to their inadequate modeling of multi-part协同 relationships and underlying engineering principles. This work proposes the first axiom-driven assembly generation framework, which translates natural language instructions into structured specifications comprising typed components, geometric ports, executable mating relations, and formal engineering axioms. By leveraging deterministic geometric solvers, the framework generates production-grade CAD assemblies that are verifiable, reusable, and explicitly grounded in engineering principles. The approach enables interpretable and generalizable parametric component synthesis by directly linking engineering intent with geometric realization, significantly outperforming current code-centric methods on the AssemBench benchmark in terms of assembly fidelity, physical validity, and cross-model generalization.

This work addresses the heavy reliance on manual intervention in macro placement for industrial physical design, where automated methods lack human-like structural awareness. The authors propose MAGE, a framework that decomposes placement into a six-stage workflow integrating structured floorplanning rules, visual inspection, and iterative refinement. Expert knowledge is injected via natural language instructions, and high-quality feedback is propagated through a multi-agent bidding mechanism. For the first time, the method introduces unsupervised, human-inspired four-dimensional layout quality metrics—gap, void, pocket, and alignment scores—enabling a fully unsupervised optimization paradigm. Evaluated on nine designs in NanGate45 and GF12nm technologies, MAGE achieves 11.1%–19.3% improvement in worst negative slack (WNS) and 70.0%–74.0% reduction in total negative slack (TNS), significantly outperforming both human experts and Hier-RTLMP, while also enhancing human-like metrics by 6%–48%, demonstrating strong generalization capability.