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

By Hongjian Zhou, Keren Zhu and Jiaqi Gu^†.

This repo is the official implementation of "LiDAR:Automated Curvy Waveguide Detailed Routing for Large-Scale Photonic Integrated Circuits", ISPD 2025.

Table of Contents

• Introduction
• Folder Structure
• Installation

Introduction

LiDAR is developed for automatically generating large-scale real-world PIC routing solutions while considering specific photonic design rules. LiDAR features a grid-based curvy-aware A* engine with adaptive crossing insertion, congestion-aware net ordering and objective, and crossing-waveguide optimization scheme, all tailored to the unique property of PIC.

• Animation

16x16 Photonic Tensor Core (ADEPT)	WRONoC_north

To efficiently enable curvy-aware A* search, we propose parametric curvy-aware methods to generate neighbor candidates based on parametric bending geometry and perform a comprehensive DRC check to select legal neighbors for exploration. Each routing node is defined by its spatial location and orientation (x, y, orientation), and we have 45° neighbors to enable 45° routing.

To ensure that only feasible neighbors are considered for exploration, a GridMap-based legality check is necessary. Each routed grids are assigned with the waveguide orientation for further crossing insertion.

If a neighboring candidate hits a previously routed waveguide (marked as an obstacle), we need to check whether it is feasible to insert a waveguide crossing to pass through it.

To further enhance overall port accessibility, several port access assignment techniques are proposed.

• Port Propagation: Propagates the ports that are inside the component's bounding box.
• Bending-Aware Port Access Region Reservation: To prevent other waveguides from blocking port regions, grids in front of each port are reserved for the corresponding net, ensuring they cannot be crossed by other nets.
• Congested Port Spreading: Spreads the access ports with a predefned extension length and spacing for ports in the same grid.
• Channel Planning via Staggered Access Point Offsets: For densely placed ports, we progressively extend the access region length and use staggered access ports to facilitate the placement of consecutive crossings.

Folder Structure

File	Description
benchmarks/	Netlists and its corresponding scripts
config/	Parser and router config
database/	Script for netlist loading and LiDAR database construction
drc/	Scripts for design rule checking
main/	Main function
queue/	Queue data structure for A star search
routing/	Implementation of routing algorithm
scripts/	Script for batch execution of benchmarks
utils/	Utilities for post-processing and logging

Installation

Prerequisites

• Python >=3.11
• GDSFactory ==8.26.1
• Other required Python packages (listed in requirements.txt)
• klayout and klive (Optional)

Get the LiDAR Source

git clone https://github.com/ScopeX-ASU/PICRoute.git

Usage

1. How to get benchmarks

cd src/picroute
python benchmarks/clements.py
python benchmarks/MMIports.py

Benchmark currently provides photonic computing circuits. Users can generate different sizes and configurations of benchmarks based on these scripts.

2. How to run

cd src/picroute
python scripts/lidar/run_route.py

The output layout gdsii files: result/LiDAR/main_results
The ouput log files: log/LiDAR/main_results

or

cd src/picroute
python main/picroute.py --benchmark path-to-benchmark --config path to router config

The input to LiDAR is formatted as a netlist, and the tool's configuration is as yaml, similar to LEF/DEF.

LiDAR's Configurations

YAML Parameter	Description
maxIteration	maximum routing iteration
net_order	net routing order
group	enable group routing order or not
enable_45_neighbor	enable diagonal neighbors or not
historyCost	history cost of routing grid
ripup_times	maximum ripup and reroute times
grid_resolution	resolution of routing grid >=2
bend_radius	radius of bend
net_default_bound	minimum routing boundary (um)
net_bound_scale_factor	scale factor of the routing boundary
loss	loss for the propagation, bending, and crossing
show_temp	show the intermediate routing result or not (needs klayout and klive)

Citing LiDAR

@inproceedings{hzhou2025lidar,
  title={LiDAR: Automated Curvy Waveguide Detailed Routing for Large-Scale Photonic Integrated Circuits},
  author={Hongjian Zhou and Keren Zhu and Jiaqi Gu},
  booktitile={International Symposium on Physical Design (ISPD)},
  year={2025}
}

Hongjian Zhou, Keren Zhu and Jiaqi Gu, "LiDAR: Automated Curvy Waveguide Detailed Routing for Large-Scale Photonic Integrated Circuits," International Symposium on Physical Design (ISPD), 2025.