Overview of emerging semiconductor device model methodologies: From device physics to machine learning engines.

Area of Science:

• Semiconductor device physics and modeling
• Machine learning applications in engineering
• Materials science and nanotechnology

Background:

• Semiconductor industry advancements introduce novel materials and device structures, posing complex physics challenges for circuit-level characterization.
• Accurate modeling of emerging devices is essential for physics-driven TCAD-to-SPICE flows and design technology co-optimization (DTCO).
• Quantum effects in ultra-scaled devices necessitate empirical parameters, disconnecting models from manufacturing processes.

Purpose of the Study:

• To provide a comprehensive overview of emerging device modeling methodologies.
• To analyze and structure current research in machine learning-assisted compact modeling (MLCM).
• To demonstrate how MLCM can address limitations of traditional compact modeling.

Main Methods:

• Review and synthesis of traditional 'white-box' and emerging 'black-box' modeling approaches.
• Focus on machine learning-assisted compact modeling (MLCM) using neural networks.
• Training MLCM on experimental and simulated data for accurate input-output mapping.

Main Results:

• MLCM provides a general-purpose modeling approach for complex physics and mathematics.
• Neural networks trained on data generate accurate closed-form mappings for device characteristics.
• MLCM effectively bridges the gap between device physics and manufacturing processes.

Conclusions:

• MLCM overcomes limitations of traditional compact modeling, offering a powerful alternative.
• MLCM significantly contributes to effective design technology co-optimization (DTCO).
• This approach is vital for advancing future semiconductor technologies.