Introduction

According to their website, “The International Workshop on Logic and Synthesis is the premier forum for research in synthesis, optimization, and verification of integrated circuits and systems.”. I would like to give an overview of the 2020, 2021 and 2022 challenges by mostly providing links (the challenge content is all over the place and I found it hard to find stuff).

2020 Challenge

Given were several, freshly constructed benchmarks or in other words, training sets. The training sets had features of $\mathbb{B}^n$, where $\mathbb{B}$ denotes the set {0, 1} and labels $\mathbb{B}$, so a binary classification task. The challenge was to come up with an AIG with a maximum of 5000 nodes for each training set that represents a good fit.

Contest description is here.

They made a paper [1] out of the contest results.

2021 Challenge

This time, the task was to come up with AIG classifiers that fit CIFAR-10 well. Input were 32x32 8-bit 3-channel color images, so 24576 in total and there are 10 output classes. There were 3 “leagues”; a maximum of 10,000, 100,000 and 1,000,000 nodes.

Call for contest is here.

They only made slides out of the contest results, no paper.

The winning team used XGBoost which they converted into circuit, the 2nd team started out with a CNN and the 3rd team used random forests for 10,000 and 100,000 nodes and a CNN for 1,000,000 nodes.

The graphic below is taken from the contest website

2022 Challenge

The 2022 contest was about exact logic gate synthesis, so given a truth table, construct an AIG. There was no node limit, but fewer nodes gave more points.

Contest announcement is here.

Again no paper, but some slides.

The contest GitHub contains the benchmarks and circuit submissions.

Video from the winners, the 2nd team and 3rd team.

References

1. [1]S. Rai et al., “Logic synthesis meets machine learning: Trading exactness for generalization,” in 2021 Design, Automation & Test in Europe Conference & Exhibition (DATE), 2021, pp. 1026–1031.