Abstract
The automation of logic circuit design enhances chip performance, energyefficiency, and reliability, and is widely applied in the field of ElectronicDesign Automation (EDA).And-Inverter Graphs (AIGs) efficiently represent,optimize, and verify the functional characteristics of digital circuits,enhancing the efficiency of EDA development.Due to the complex structure andlarge scale of nodes in real-world AIGs, accurate modeling is challenging,leading to existing work lacking the ability to jointly model functional andstructural characteristics, as well as insufficient dynamic informationpropagation capability.To address the aforementioned challenges, we proposeAIGer.Specifically, AIGer consists of two components: 1) Node logic featureinitialization embedding component and 2) AIGs feature learning networkcomponent.The node logic feature initialization embedding component projectslogic nodes, such as AND and NOT, into independent semantic spaces, to enableeffective node embedding for subsequent processing.Building upon this, the AIGsfeature learning network component employs a heterogeneous graph convolutionalnetwork, designing dynamic relationship weight matrices and differentiatedinformation aggregation approaches to better represent the original structureand information of AIGs.The combination of these two components enhancesAIGer's ability to jointly model functional and structural characteristics andimproves its message passing capability. Experimental results indicate thatAIGer outperforms the current best models in the Signal Probability Prediction(SSP) task, improving MAE and MSE by 18.95\% and 44.44\%, respectively. In theTruth Table Distance Prediction (TTDP) task, AIGer achieves improvements of33.57\% and 14.79\% in MAE and MSE, respectively, compared to thebest-performing models.