tilos-ai-institute / multi-source-prim-dijkstra

Multi-Source Prim-Dijkstra

UPDATE #3
We thank you very much for participating in the MSPD contest.
Results have been compiled, and winners have been notified of their placement.

Here is the final leaderboard:
1st Place: Team DIAG, Score: 0.003106342769
2nd Place: Team FU-Berlin, Score: 0.1761129309
3rd Place: Team Moto, Score: 0.337300873

We are very impressed by the quality and strength of your models, and hope that you are proud of your team's work. We also hope that you enjoyed this contest, and gained some new insight or skill by participating.

Thank you once again,
Shreyas Thumathy and Mingyu Woo
Contest Organizers

UPDATE #2
Registration deadline: ${\color{red}06/15/23}$
Please register using the following form.

UPDATE #1
Our MSPD Machine Learning Contest is now open for registration !!!
Thanks to The OpenROAD Project, the contest has over $7500 in available prizes!

---

Welcome to the Multi-Source Prim-Dijkstra repository on GitHub. Multi-Source Prim-Dijkstra (MSPD) is a new and efficient heuristic rectilinear Steiner tree construction that more effectively trades off between the competing objectives of minimum tree cost and minimum tree skew. The method was reported in our ISQED-2023 paper. This repository contains code and data, and hosts a machine learning contest that runs from April 2023 to November 2023.

This repository contains our work for the ISQED-2023 paper:
Andrew B. Kahng, Shreyas Thumathy and Mingyu Woo, "An Effective Cost-Skew Tradeoff Heuristic for VLSI Global Routing", Proc. Intl. Symp. on Quality Electronic Design, April 2023
[PDF], [PPTX], [MP4].

• src/ contains the implementations of MSS and MSPD for underlying PDRev (PD+HVW+DAS) and STT (stt+DAS) Steiner constructions.

• contest/ contains the contents and descriptions for our machine learning contest, which launches on April 5, 2023 and aims to inspire an ML-based solution for predicting good combinations of sources to use in MSPD with the STT construction. More than $7500 in prizes are available!

• LUTs/ contains a python script (LUT.py) that gives experimentally-determined normalized cost, normalized skew, variance in normalized cost, and variance in normalized skew, for given $N$, $AR$, $p$, and $\alpha$ parameters. This corresponds to Section 3 in our ISQED-2023 paper.

When referring to either our paper or contest, please cite as shown with the following bibtex:

@inproceedings{ISQED2023,  
  author    = {Kahng, Andrew B. and Thumathy, Shreyas and Woo, Mingyu},  
  title     = {An Effective Cost-Skew Tradeoff Heuristic for VLSI Global   Routing},  
  booktitle = {2023 International Symposium on Quality Electronic Design (ISQED)},
  year      = {2023},  
  pages     = {1-8}. 
}