How Deep Learning Can Drive Physical Synthesis Towards More Predictable Legalization

Abstract

Machine learning has been used to improve the predictability of different physical design problems, such as timing, clock tree synthesis and routing, but not for legalization. Predicting the outcome of legalization can be helpful to guide incremental placement and circuit partitioning, speeding up those algorithms. In this work we extract histograms of features and snapshots of the circuit from several regions in a way that the model can be trained independently from region size. Then, we evaluate how traditional and convolutional deep learning models use this set of features to predict the quality of a legalization algorithm without having to executing it. When evaluating the models with holdout cross validation, the best model achieves an accuracy of 80% and an F-score of at least 0.7. Finally, we used the best model to prune partitions with large displacement in a circuit partitioning strategy. Experimental results in circuits (with up to millions of cells) showed that the pruning strategy improved the maximum displacement of the legalized solution by 5% to 94%. In addition, using the machine learning model avoided from 22% to 99% of the calls to the legalization algorithm, which speeds up the pruning process by up to 3×.

Algorithm Selection Framework for Legalization Using Deep Convolutional Neural Networks and Transfer Learning

“…The only related work that focuses on predicting the quality of legalization algorithms is our previous work [28]. In that work, we trained an ML model to identify circuit regions that would result in large displacement after legalization.…”

“…This section presents a comparison with a related non-convolutional model from the state of the art [28]. This comparison is performed to ensure that our claim that CNNs provide better results than non-convolutional models is correct.…”

“…This comparison is performed to ensure that our claim that CNNs provide better results than non-convolutional models is correct. In order to perform this comparison, we trained artificial neural networks (ANNs) for legalization outcome prediction using the same features from our previous work [28]. For fairness, we used as input data the same placements generated for the CNN results and the same training and validation sets, but used the feature extraction strategy from [28] instead of using the placement images.…”

“…In order to perform this comparison, we trained artificial neural networks (ANNs) for legalization outcome prediction using the same features from our previous work [28]. For fairness, we used as input data the same placements generated for the CNN results and the same training and validation sets, but used the feature extraction strategy from [28] instead of using the placement images. In addition, we trained the ANNs for 15 epochs (the same total number of epochs as the CNNs).…”