Machine learning-based inductance extractors for interconnects
Over the past few decades, there has been a continuous reduction in the feature size of integrated circuits (ICs) accompanied by an increase in their performance, as dictated by Moore's law. When aiming for miniaturization and performance optimization in the realm of IC design, the importance o...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/174026 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Over the past few decades, there has been a continuous reduction in the feature size of integrated circuits (ICs) accompanied by an increase in their performance, as dictated by Moore's law. When aiming for miniaturization and performance optimization in the realm of IC design, the importance of an inductance extractor cannot be overstated. While some notable inductance extraction tools have already been developed and can achieve good accuracy and less computational time, it is still desirable to further accelerate the inductance extraction process to speed up the iterative design exploration.
This dissertation proposes machine learning algorithms for rapid extraction of inductances of interconnects. Three machine learning methods are examined to generate surrogate models and estimate the inductances of interconnects. These algorithms include Back Propagation (BP) network, Support Vector Machine (SVM), and Extreme Learning Machine (ELM). By specifically designing network structures, adjusting loss functions, and training and testing the networks, the algorithms can accurately estimate the inductances due to varying geometrical parameters. To generate the data for training and testing, a popular physics-based inductance extractor, called VoxHenry, is utilized to extract reference inductance values and construct the dataset used in training and testing the machine learning-based inductance extractors.
Numerical results show that machine learning-based inductance extractors require less computational time than physics-based inductance extractors while providing sufficiently accurate results. Among the machine learning algorithms, the ELM-based inductance extractor outperforms the others. |
|---|