Quantum Kernel Learning for Small Dataset Modeling in Semiconductor Fabrication: Application to Ohmic Contact

Area of Science:

• Semiconductor device fabrication
• Quantum machine learning
• Materials science

Background:

• Modeling complex semiconductor fabrication processes, like Ohmic contact formation, is challenging due to high-dimensional parameters and limited data.
• Classical machine learning (CML) struggles with nonlinear scenarios and small datasets, common in advanced materials research.

Purpose of the Study:

• To investigate quantum machine learning (QML) as a viable alternative for modeling semiconductor fabrication processes with limited experimental data.
• To develop and evaluate a quantum kernel-aligned regressor (QKAR) for predicting Ohmic contact formation in Gallium Nitride High Electron Mobility Transistors (GaN HEMTs).

Main Methods:

• Development of a quantum kernel-aligned regressor (QKAR) using a shallow Pauli-Z feature map and a trainable quantum kernel alignment (QKA) layer.
• Utilizing a dataset of 159 experimental GaN HEMT samples for training and validation.
• Comparative analysis against seven baseline CML regressors using a unified PCA-based preprocessing pipeline.

Main Results:

• The QKAR model consistently outperformed all classical baseline models across multiple evaluation metrics (MAE, MSE, RMSE).
• Achieved a mean absolute error (MAE) of 0.338 Ω·mm on experimental data, demonstrating high predictive accuracy.
• Demonstrated noise robustness and generalization capabilities through cross-validation and new device fabrication assessments.

Conclusions:

• Carefully constructed QML models offer significant predictive advantages in data-constrained semiconductor modeling.
• QML presents a promising complementary approach to CML for complex process modeling tasks, with potential for near-term quantum hardware deployment.
• This study validates the potential of QML in addressing challenges in semiconductor fabrication modeling.