AnalogRetriever: Tri-Modal Framework Revolutionizes Analog Circuit Search

The introduction of AnalogRetriever marks a significant breakthrough in the field of Electronic Design Automation (EDA), specifically for analog circuit design. Historically, searching and reusing existing analog intellectual property (IP) has been hampered by the heterogeneous nature of design representations, including SPICE netlists, schematics, and functional descriptions. Existing methods are largely confined to exact matching within a single modality, failing to capture the crucial semantic relationships across these diverse formats. AnalogRetriever addresses this by providing a unified tri-modal retrieval framework, enabling engineers to semantically search and retrieve analog circuits regardless of their representation. This capability is poised to dramatically accelerate design cycles and enhance the efficiency of a domain critical to nearly all modern electronics.

AnalogRetriever's technical architecture is built upon a robust foundation. It leverages a two-stage repair pipeline to create a high-quality dataset from Masala-CHAI, achieving a 100% netlist compile rate from an initial 22%. This cleaned dataset is crucial for effective training. The framework then employs a vision-language model to encode schematics and descriptions, while a port-aware relational graph convolutional network handles netlists. The key innovation lies in mapping all three modalities into a shared embedding space through curriculum contrastive learning. This shared space allows for seamless cross-modal querying. Experimental results are compelling, demonstrating an average Recall@1 of 75.2% across all six cross-modal retrieval directions, significantly outperforming previous baselines. Furthermore, when integrated into the AnalogCoder agentic framework, AnalogRetriever consistently improves functional pass rates and enables the completion of previously unsolved design tasks, highlighting its practical utility.

The forward-looking implications for analog circuit design are profound. By enabling efficient semantic search and retrieval across diverse design artifacts, AnalogRetriever can drastically reduce the time and effort required for IP reuse, a cornerstone of modern chip design. This will not only accelerate product development cycles but also lower design costs and potentially lead to more innovative and complex analog circuits. The release of the code and dataset will further democratize access to this technology, fostering broader adoption and research. This paradigm shift could empower engineers to focus more on novel design challenges rather than tedious search tasks, ultimately driving advancements in critical sectors such as automotive, aerospace, medical devices, and the Internet of Things, all of which heavily rely on sophisticated analog components.