AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies

Authors: Jian Gao, Weidong Cao, Junyi Yang, Xuan Zhang

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show the remarkable generation performance of Analog Genie in broadening the variety of analog ICs, increasing the number of devices within a single design, and discovering unseen circuit topologies far beyond any prior arts.
Researcher Affiliation Academia 1 Northeastern University, 2 The George Washington University EMAIL {weidong.cao}@gwu.edu
Pseudocode No The paper describes methods in prose and includes theoretical proofs (Theorem 3.2.1) but does not provide structured pseudocode or algorithm blocks.
Open Source Code Yes Our source code is available at https://github.com/xz-group/Analog Genie.
Open Datasets No For our open-sourced circuit dataset, we provide its statistics in Appendix A.1. We will make our code and dataset public on Github in the future.
Dataset Splits Yes During training, we first split the topology data set into train and validation sets with a 9 to 1 ratio.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions using a 'GPT model' and 'Ngspice simulation infrastructure' but does not specify any version numbers for software dependencies.
Experiment Setup Yes Our Analog Genie model is a decoder-only transformer consisting of 6 hidden layers and 6 attention heads with 11.825 million parameters in total. The vocab size is 1029. The maximum sequence length is 1024.