Multi-Objective Design and Optimization of Complementary FETs Using Holographic Counterpart Predictive Model

With the relentless scaling of consumer devices, Complementary Field-Effect Transistors (CFETs) have emerged as a promising candidate to overcome the limitations of FinFETs. Nonetheless, precisely modeling and optimizing their intricate design parameters and electrical characteristics still remains a significant challenge. This work integrates a state-of-the-art holographic counterpart predictive model for designing and optimizing CFETs through an Attention-Based Convolution Skip Bidirectional Long Short-Term Memory Network (Att-CB-LSTM). The model generates synthetic data based on five key design variables: n and p-channel FETs (NFET/PFET) separation distance (D_n/p), nanosheet separation (D_nsht), channel width (W_nsht), channel thickness (T_nsht), and oxide thickness (T_oxi). To provide stable input scaling and avoid training bias, Distribution Scaling Normalization (DSN) is used. The Att-CB-LSTM model precisely predicts the important electrical parameters, Noise Margin Low (NM_L), Noise Margin High (NM_H), Gain (G), Power-Delay Product (PDP), and Frequency (Freq) by identifying intricate nonlinear relationships. Hyperparameters are tuned using the Walrus Optimization Algorithm (WaOA) to improve prediction accuracy and generalization. Experimental testing reveals considerable enhancements over conventional models, with NMlow increased by 7.5%, Gain enhanced by 22.6%, and PDP decreased by 20.8%, demonstrating the effectiveness and stability of the proposed method.