A semi-empirical approach to calibrate simulation models for semiconductor devices

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Semiconductor device optimization traditionally involves time-consuming fabrication and testing cycles.
• Accurate simulation models are crucial for efficient prototyping but require precise calibration of material parameters.
• Calibration often demands extensive characterization data and expert knowledge, posing a significant bottleneck.

Purpose of the Study:

• To develop a hybrid method for calibrating multiple semiconductor device simulation models.
• To reduce the reliance on extensive characterization data and expert tuning during model calibration.
• To demonstrate the method's effectiveness using a silicon solar cell as a case study.

Main Methods:

• A hybrid approach combining minimal characterization data with machine learning-based prediction models was proposed.
• Optical and electrical simulation models for an industrially manufactured silicon solar cell were calibrated.
• Simulated device performance was compared against measurement data from the physical device.

Main Results:

• The hybrid method successfully calibrated simulation models for a photovoltaic device.
• The calibrated models accurately predicted the performance of the silicon solar cell.
• The approach demonstrated time and resource efficiency compared to conventional methods.

Conclusions:

• The proposed hybrid method offers an efficient strategy for calibrating semiconductor device simulation models.
• This technique can significantly accelerate computer-based prototyping in semiconductor research and development.
• Minimal characterization data combined with machine learning can effectively bridge the gap in model calibration.