Publication: Automated soc static low-power isolation convergence with machine learning
Loading...
Date
2024-10
Authors
Teow, Wei Seng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
With the rise of wireless devices, minimizing power consumption is crucial, prompting engineers to focus on low-power SoC designs and use the Unified Power
Format (UPF) to manage power intent early in the design process. Traditionally, generating set_isolation commands manually is time-consuming and error-prone. The project "Automated SoC Static Low-Power Isolation Convergence with Machine Learning" addresses power leakage in SoC design by using machine learning to automatically generate UPF set_isolation commands. This research employs the Scikit-Learn library and Python to develop an MLPClassifier model, optimized
through GridSearchCV, and evaluates its performance with various training-testing data splits. The 80_MLP.joblib model, trained on 80% of the data, achieved the highest accuracy of 99.9509% with reasonable training time, underscoring the importance of sufficient training data. The methodology involves cloning a previous project model, extracting and modifying data from "soc_upf_config.txt" to simulate real-world challenges, and developing scripts to process error reports and generate set_isolation commands. The findings demonstrate the potential of machine learning to automate and optimize SoC design verification tasks, reducing manual effort and errors, paving the way for more efficient UPF generation techniques and suggesting future work to enhance dataset diversity, explore advanced algorithms, and conduct real-world evaluations.