Abstract
This paper proposes a neural framework for power and timing prediction ofmulti-stage data path, distinguishing itself from traditional lib-basedanalytical methods dependent on driver characterization and loadsimplifications. To the best of our knowledge, this is the firstlanguage-based, netlist-aware neural network designed explicitly for standardcells. Our approach employs two pre-trained neural models of waveformprediction and delay estimation that directly infer transient waveforms andpropagation delays from SPICE netlists, conditioned on critical physicalparameters such as load capacitance, input slew, and gate size. This methodaccurately captures both intrinsic and coupling-induced delay effects withoutrequiring simplification or interpolation. For multi-stage timing prediction,we implement a recursive propagation strategy where predicted waveforms fromeach stage feed into subsequent stages, cumulatively capturing delays acrossthe logic chain. This approach ensures precise timing alignment and completewaveform visibility throughout complex signal pathways. The waveform predictionutilizes a hybrid CNN-Transformer architecture with netlist-aware node-levelencoding, addressing traditional Transformers' fixed input dimensionalityconstraints. Additionally, specialized subnetworks separately handle primarydelay estimation and crosstalk correction. Experimental results demonstrateSPICE-level accuracy, consistently achieving RMSE below 0.0098 across diverseindustrial circuits. The proposed framework provides a scalable, structurallyadaptable neural alternative to conventional power and timing engines,demonstrating high fidelity to physical circuit behaviors.