EDA² | ISEDA 2025 | Technical Session

Advanced Simulation and Optimization in Semiconductor Processes

Abstract: This paper proposes a novel global modeling framework based on a Mixture of Experts (MoE) architecture to address the challenge that conventional compact global models face in capturing the size-dependent operational characteristics of gate-all-around field-effect transistor (GAAFET). The proposed framework dynamically activates specialized expert models through an adaptive gating network, effectively integrating device geometric parameters with bias voltage conditions to enable precise prediction of device current characteristics. Comprehensive experimental results demonstrate that the proposed MoE-based model can significantly improve prediction accuracy while ensuring computational efficiency, especially when dealing with high-dimensional and complex current voltage (IV) characteristics, exhibiting strong adaptability and generalization ability. Compared to conventional global models with equivalent parameter counts, the proposed method reduces root mean square error (RMSE) by 20.97% while accelerating inference speed by 53.57%. When benchmarked against models with comparable inference speeds, it achieves superior performance with 54.26% lower RMSE and 48.88% reduced mean absolute percentage error (MAPE). These results highlight the framework's strong adaptability to device scaling variations and enhanced generalization capabilities across diverse operating regimes.

Abstract: A wide-band ANN-based inductor model up to 110GHz for 0.1um GaAs pHEMT technology is presented. Different with the traditional equivalent circuit structure, the proposed ANN model takes high-order parasitic effect into account, such as skin effect and coupling effect, which is significant in high-frequency applications. The proposed ANN model has the input layer of geometric dimensions of the inductor and the frequency, with S-parameters as the output layer. The relationship between inputs and outputs is described through nonlinear expressions training of the model. The model results demonstrate high accuracy in comparison with the EM simulation results across the 110GHz frequency range. By integrating into the PDK, the efficiency and accuracy of MMIC design can be significantly enhanced.

Abstract: Aging delay prediction is one of the key tasks for modern circuit design during reliability assessment. In traditional methods, static timing analysis is utilized to estimate accurate and effective guard bands. However, it requires intensive Monte-Carlo simulations which could bring high computational overheads. In modern IC design, deployment of neural networks assists to shorten time consumption and enhance prediction efficiency. Due to the black-box training procedure, the prediction results usually suffer from the uncertainty in networks and prediction accuracy could be unsatisfying in some risk-sensitive applications. In this work, we proposed a simple method to improve the aging delay predicted by machine-learning-based method. By implementing uncertainty estimates based on hybrid graphic neural distribution with Monte Carlo network and Dropout, prediction accuracy can be enhanced without further training while enabling the network to maintain high efficiency. Then, statistical analysis is developed based on predictive distribution and an innovative scheme for guard bands design is provided. Experiment results demonstrate that our method shows higher accuracy compared with benchmark networks with uncertainty unconsidered.

Abstract: In this work, a Recurrent Neural Network-based Open Model Interface (RNN-OMI) is proposed to enable AI driven compact model (CM) to characterize the typical aging behavior of MOSFETs, specifically Negative Bias Temperature Instability (NBTI). The long-term sequence learning ability of RNN assists in fine-tuning the sensitive parameters within AI-CM. The enhanced AI-CM predicts the drift of SRAM butterfly curves, enabling the temporal prediction in circuit aging evaluation.