LiDAR 2.0: Hierarchical Curvy Waveguide Detailed Routing for Large-Scale Photonic Integrated Circuits

Hongjian Zhou, Haoyu Yang, Ziang Yin, Nicholas Gangi, Zhaoran Huang, Haoxing Ren, Joaquin Matres, Jiaqi Gu

TL;DR

This paper introduces LiDAR 2.0, an automated detailed router for large-scale photonic integrated circuits (PICs). It tackles the growing complexity of PICs by introducing a hierarchical routing approach, adaptive crossing insertion, and optimizations for curvy waveguides, significantly improving routing efficiency and reducing insertion loss compared to previous methods.

What Problem Does This Paper Address?

Photonic Integrated Circuits (PICs) are becoming increasingly complex, with thousands of components per chip, driven by applications like optical neural networks and high-bandwidth communication. Traditional manual design processes are becoming unmanageable. Existing automated routing tools, often adapted from electronic design, struggle with PIC-specific challenges like curvy waveguides, precise port alignment, and the need for adaptive crossing insertion. These tools often fail to produce "design-rule-violation-free" (DRV-free) layouts and cannot scale effectively. This gap necessitates specialized Electronic-Photonic Design Automation (EPDA) tools to streamline the physical design of PICs.

Methods

LiDAR 2.0 builds upon its predecessor, LiDAR, with several key enhancements:

  • Hierarchical Routing: The framework now leverages the inherent hierarchical structure of PIC netlists. It processes modules and submodules independently, treating them as private routing domains, and then integrates them. This modular approach significantly improves scalability.
  • Reusability of Routing Results: For identical modules within a hierarchical design, LiDAR 2.0 reuses previously generated routing solutions (after translation, rotation, or flipping), drastically reducing computation time.
  • Conflict-Resilient Routing:
    • A* Offset Neighbor: Introduces "offset neighbors" in the A* search algorithm to correct small port misalignments without introducing unnecessary bends, improving routing quality and reducing runtime.
  • Preserved Crossing Space: A "crossing space penalty" is incorporated into the A* cost function. This encourages the router to leave sufficient space around potential crossing locations, preventing conflicts and enabling more crossings when needed.
  • Routing Order Optimization: The routing order for nets within a group is determined by the Manhattan distance to the average source port location, resolving conflicts caused by improper ordering, especially in multi-port components.
  • Rip-up and Reroute Enhancements: The rip-up and reroute strategy is improved to handle more complex conflict scenarios, including adjusting routing order and dynamically extending port regions to prevent subsequent blocking.
  • Curvy-Aware A* Engine: Continues to use a grid-based A* search that supports non-Manhattan, curvy waveguides, with adaptive neighbor generation based on bend radius and grid size.
  • Adaptive Crossing Insertion: The router can now intelligently insert crossings when necessary, rather than relying on manual placement, to overcome obstacles or reduce detours.

Datasets

The paper introduces and utilizes a new suite of PIC benchmarks generated using a YAML-based Photonic Intermediate Representation (PIC IR). These benchmarks are designed to be scalable and cover diverse routing complexities, including:

  • Spacious Benchmarks: (e.g., Clements, ADEPT, TeMPO, GWOR) These are designed for evaluating routing under less constrained conditions.
  • Compact Benchmarks: (e.g., Clements_C, ADEPT_C, TeMPO_C, GWOR_C, Benes) These feature higher component density, smaller die sizes, and more crossings, posing greater routing challenges and reflecting real-world constraints.

The benchmarks are categorized into "photonic tensor core" and "optical switch" types, with specific examples like Clements, ADEPT, TeMPO, Light, GWOR, and Benes.

Key Insights and Ideas

  • Hierarchy is Key for Scale: The paper strongly demonstrates that leveraging the hierarchical nature of PIC designs is crucial for achieving scalability in automated routing. Treating modules as independent routing domains significantly reduces complexity.
  • Reusing Routing is a Game Changer: For designs with repetitive modules, reusing previously computed routing solutions offers massive speedups, making complex designs feasible within practical timeframes.
  • Proactive Conflict Prevention Beats Reactive Resolution: LiDAR 2.0's emphasis on preserving crossing space and optimizing routing order before conflicts become severe is more effective than solely relying on rip-up and reroute.
  • Curvy Waveguides Require Specialized Routing: The paper highlights that standard Manhattan routing is insufficient. The adaptive A* search with parametric neighbors is essential for handling the geometric nuances of curvy PIC waveguides.
  • "Design Rule Free" is Achievable: The results show that LiDAR 2.0 can consistently produce layouts that are nearly or completely free of design rule violations, a significant improvement over prior methods.

Key Results

LiDAR 2.0 demonstrates significant improvements over previous methods:

  • Spacious Benchmarks: Achieves up to 16% lower insertion loss (IL) and 7.69x speedup compared to PROTON.
  • Compact Benchmarks: Achieves 9% lower IL and 6.95x speedup compared to LiDAR 1.0.
  • DRV-Free Layouts: Consistently produces nearly DRV-free layouts across various benchmarks, even under challenging compact conditions.
  • Robustness: The tool handles varying crossing sizes and bend radii effectively, demonstrating reliability for different PIC technologies. For example, it successfully routes GWOR_32x32 with bend radii up to 80 µm.
  • Runtime Efficiency: The hierarchical approach and result reuse lead to substantial runtime reductions, especially for complex and repetitive designs.

Limitations and Future Work

  • Limitations:
    • While LiDAR 2.0 achieves nearly DRV-free layouts, some minor DRVs were observed in ADEPT cases on compact benchmarks.
  • For some extremely compact benchmarks (e.g., TEMPO and Benes), no valid routing solution could be found due to insufficient spacing, indicating the limits of even advanced routing in extremely dense scenarios.
  • The paper mentions that the "group congestion penalty" parameter needs to be adjusted to balance routing complexity and wirelength.
  • Future Work:
    • The paper implicitly suggests that further research into optimizing the balance between crossing avoidance and path length for insertion loss reduction is valuable.
  • The development of more sophisticated benchmark suites and the exploration of new routing strategies for even more challenging, ultra-compact PIC designs are implied.
  • The paper's introduction of a YAML-based PIC IR and open-source benchmarks lays the groundwork for future research and development in the EPDA community.

Summary updated

Abstract

Driven by innovations in photonic computing and interconnects, photonic integrated circuit (PIC) designs advance and grow in complexity. Traditional manual physical design processes have become increasingly cumbersome. Available PIC layout tools are mostly schematic-driven, which has not alleviated the burden of manual waveguide planning and layout drawing. Previous research in PIC automated routing is largely adapted from electronic design, focusing on high-level planning and overlooking photonic-specific constraints such as curvy waveguides, bending, and port alignment. As a result, they fail to scale and cannot generate DRV-free layouts, highlighting the need for dedicated electronic-photonic design automation tools to streamline PIC physical design. In this work, we present LiDAR, the first automated PIC detailed router for large-scale designs. It features a grid-based, curvy-aware A* engine with adaptive crossing insertion, congestion-aware net ordering, and insertion-loss optimization. To enable routing in more compact and complex designs, we further extend our router to hierarchical routing as LiDAR 2.0. It introduces redundant-bend elimination, crossing space preservation, and routing order refinement for improved conflict resilience. We also develop and open-source a YAML-based PIC intermediate representation and diverse benchmarks, including TeMPO, GWOR, and Bennes, which feature hierarchical structures and high crossing densities. Evaluations across various benchmarks show that LiDAR 2.0 produces nearly DRV-free layouts, achieving up to 16% lower insertion loss and 7.69× speedup over prior methods on spacious cases, and 9% lower insertion loss with 6.95× speedup over LiDAR 1.0 on compact cases. Our codes are open-sourced at link.
BibTeX
@misc{zhou2025lidar,
  author = {Hongjian Zhou and Haoyu Yang and Ziang Yin and Nicholas Gangi and Zhaoran Huang and Haoxing Ren and Joaquin Matres and Jiaqi Gu},
  title = {{LiDAR 2.0: Hierarchical Curvy Waveguide Detailed Routing for Large-Scale Photonic Integrated Circuits}},
  year = {2025},
  eprint = {2505.17239},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2505.17239},
}

Loading related content...

Figure