A Vertical Systematic Generalization Towards Real IP Analog-Mixed-Signal AMS Schematics Structure Recognition

AMS circuits are crucial components in modern electronic systems. However, AMS circuits are variable and complex, making their design and verification difficult. Design and verification of AMS circuits demand accurate analog circuit structure recognition for layout generation, performance optimization, and hierarchical breakdown. In context with this, the current work introduces a robust, systematic, and exhaustive approach to AMS schematic recognition to enhance the capabilities of existing approaches within the AMS circuits hierarchy. Our solution applies graph attention networks (GATs) and machine learning (ML) techniques to analyze the structure and connectivity of circuits. To cope with the complexity of AMS circuits, we use graph-based learning methods. Graph attention techniques enable the model to focus on the circuit's most intriguing sections, thereby enhancing the accuracy of structure recognition. We evaluated the proposed approach on a large dataset of AMS circuits with various levels of complexity. Our method is superior for AMS schematic structure recognition, particularly for large-scale and highly dynamic circuits. Our strategy of vertical generalization enhances performance. This essential characteristic enables our method to scale successfully, precisely, and efficiently despite circuit complexity. Our data augmentation approach showed outstanding outcomes compared to others. This enhances learning by providing diverse circuit examples and makes the model more robust against the diversity and complexity of AMS circuits. These enhancements increase the efficiency and dependability of analog circuit design and verification, thereby resolving current electronic system issues.