AI-Driven Learning and Regeneration of Analog Circuit Designs From Academic Papers

Abstract

This paper presents an artificial intelligence (AI)-based framework designed for learning and regenerating analog circuits from academic papers. The framework comprises four distinct modules: circuit extractor, table extractor, text extractor, and simulation executor. The circuit extractor module utilizes deep learning object detection to identify devices and their associated textual descriptions while extracting interconnections between devices. The table extractor module handles textual and image-based tables, extracting device parameters, and simulation data. The text extractor module leverages optical character recognition (OCR) and AI models to extract supplementary information. The simulation executor employs this information to conduct simulations and optimize circuit performance. In our experiments, our method effectively extracts multimodal circuit design information, achieving an average accuracy of up to 97% in target detection within the circuit extractor module. The improved performance during the simulation process further validates the effectiveness of our framework.