Abstract:The rise of machine learning technology inspires a boom of its applications in electronic design automation (EDA) and helps improve the degree of automation in chip designs. However, manually crafted machine learning models require extensive human expertise and tremendous engineering efforts. In this work, we leverage neural architecture search (NAS) to automate the development of high-quality neural architectures for routability prediction, which can help to guide cell placement toward routable solutions. Our search method supports various operations and highly flexible connections, leading to architectures significantly different from all previous human-crafted models. Experimental results on a large dataset demonstrate that our automatically generated neural architectures clearly outperform multiple representative manually crafted solutions. Compared to the best case of manually crafted models, NAS-generated models achieve 5.85% higher Kendall's $\tau$ in predicting the number of nets with DRC violations and 2.12% better area under ROC curve (ROC-AUC) in DRC hotspot detection. Moreover, compared with human-crafted models, which easily take weeks to develop, our efficient NAS approach finishes the whole automatic search process with only 0.3 days.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the automated modeling problem of chip placement and routing prediction in electronic design automation (EDA). Specifically, the author aims to use neural architecture search (NAS) technology to automatically develop high - quality neural network models to predict the routability of chip placement. This helps to guide cell placement to be closer to a routable solution. ### Problem Background 1. **Challenges of Manual Model Construction**: Traditionally, machine - learning models for predicting routability require a great deal of human expertise and engineering effort. For example, designing a neural network architecture for industrial applications may take months. 2. **Limitations of Existing Methods**: Although the existing manual designs based on CNN and other deep - learning models are effective, they have problems such as long development cycles and high costs. 3. **Automation Requirement**: In order to increase the degree of automation and reduce the need for human intervention, researchers hope to use automated machine learning (AutoML), especially neural architecture search (NAS), to automate this process. ### Research Objectives 1. **Automatically Develop High - Quality Models**: Automatically generate neural network architectures for routability prediction through NAS, thereby reducing the dependence on human experts. 2. **Improve Prediction Performance**: Ensure that the automatically generated models are superior to manually - designed models in predicting the number of networks violating DRC (design rule check) and DRC hot - spot detection. 3. **Shorten the Development Cycle**: Compared with the weeks to months required for manually - designed models, NAS can complete model search and optimization in a short time. ### Main Contributions 1. **First Application of NAS to EDA Problems**: This is the first attempt to use NAS to automatically develop machine - learning models for EDA problems. 2. **Design a Flexible Search Space**: Propose a search space that contains multiple operations and highly flexible connections to ensure the diversity of candidate models. 3. **Significantly Improve Prediction Performance**: Experimental results show that the models generated by NAS have a 5.85% improvement in Kendall’s τ in predicting the number of networks violating DRC compared to the best manually - designed model, and a 2.12% improvement in ROC - AUC in DRC hot - spot detection. 4. **Efficient Search Process**: The entire NAS search process only takes 0.3 days, while manually - designed models usually take weeks or even months. Through these improvements, this research not only improves the accuracy of routability prediction, but also greatly shortens the time of model development, providing new ideas and technical means for the automated development of EDA tools.

Automatic Routability Predictor Development Using Neural Architecture Search

Improving Routability Prediction via NAS Using a Smooth One-shot Augmented Predictor

Automating Neural Architecture Design without Search

Advances in neural architecture search

Generalized Global Ranking-Aware Neural Architecture Ranker for Efficient Image Classifier Search

Latency-aware Neural Architecture Performance Predictor with Query-to-Tier Technique

Neural Architecture Search with Representation Mutual Information

A Semi-Supervised Assessor of Neural Architectures

A Technical View on Neural Architecture Search

Accuracy Prediction with Non-neural Model for Neural Architecture Search

A Lightweight Inception Boosted U-Net Neural Network for Routability Prediction

AutoIAS

AutoSpace: Neural Architecture Search with Less Human Interference

Understanding Neural Architecture Search Techniques

NAS-CTR: Efficient Neural Architecture Search for Click-Through Rate Prediction

Automated and Agile Design of Layout Hotspot Detector via Neural Architecture Search

Automated Model Generation for Machinery Fault Diagnosis Based on Reinforcement Learning and Neural Architecture Search

Neural Architecture Search with Progressive Evaluation and Sub-Population Preservation

NASPrecision: Neural Architecture Search-Driven Multi-Stage Learning for Surface Roughness Prediction in Ultra-Precision Machining

NAS-ASDet: An Adaptive Design Method for Surface Defect Detection Network using Neural Architecture Search

Understanding Architectures Learnt by Cell-based Neural Architecture Search