Simulator acceleration and inverse design of fin field-effect transistors using machine learning

Area of Science:

• Semiconductor device physics
• Computational materials science
• Artificial intelligence in engineering

Background:

• Traditional simulation and design of electronic devices like transistors are time-consuming and costly.
• Existing methods rely on intuitive design and model equations, limiting efficiency.
• Advancements in machine learning offer potential for optimizing these semiconductor industry processes.

Purpose of the Study:

• To implement and evaluate machine learning models for accelerating fin field-effect transistor (FinFET) simulations.
• To develop and assess machine learning for the inverse-design of FinFETs, enabling targeted specification achievement.
• To demonstrate the efficacy of AI in addressing simulation and design challenges in electronic device engineering.

Main Methods:

• Utilized various machine learning approaches, including neural networks, for simulation acceleration.
• Developed an inverse-design model trained on extensive simulation data relating device design to performance specifications.
• Compared the performance of the machine learning models against traditional simulation techniques.

Main Results:

• The proposed neural network model achieved simulation results with high accuracy (R² = 0.99), comparable to traditional simulators.
• The machine learning simulation was over 122,000 times faster than conventional methods.
• The inverse-design model successfully generated accurate design parameters meeting target specifications (R² = 0.96).

Conclusions:

• Machine learning models provide highly efficient and accurate solutions for electronic device simulation and inverse-design.
• The developed approach significantly reduces the time and cost associated with transistor design and simulation.
• This AI-driven methodology can be extended to more complex devices and critical semiconductor manufacturing processes.