Machine Learning-Based Figure of Merit Model of SIPOS Modulated Drift Region for U-MOSFET

Area of Science:

• Power Electronics
• Semiconductor Devices
• Materials Science

Background:

• Superjunction (SJ) U-MOSFETs are critical for high-voltage applications.
• Optimizing the trade-off between breakdown voltage (BV) and specific ON-resistance (R_on) is a key challenge.
• Existing models may not fully capture the complex interactions in modulated drift regions.

Purpose of the Study:

• To develop a machine learning-based figure of merit (FOM) model for SJ U-MOSFETs with modulated drift regions.
• To optimize the breakdown voltage (BV) and specific ON-resistance (R_on) trade-off.
• To surpass the SJ silicon limit for device performance.

Main Methods:

• Utilizing semi-insulating poly-crystalline silicon (SIPOS) pillars to modulate the SJ drift region.
• Developing an analytical model considering electric field modulation, charge-coupling, and carrier accumulation.
• Employing Gaussian process regression for FOM prediction and hyperparameter optimization.
• Verifying the model against TCAD simulation results.

Main Results:

• A robust machine learning model for predicting the FOM (BV^2/R_on) of SSJ-UMOSFETs.
• Identification of optimal structural parameters for drift region, oxide thickness, and electric field modulation.
• Demonstration of a methodology to achieve a BV-R_on trade-off exceeding the SJ silicon limit.
• Comprehensive validation of the model using TCAD simulations.

Conclusions:

• The proposed machine learning model accurately predicts the performance of SJ U-MOSFETs with modulated drift regions.
• The methodology enables the design of devices with superior BV-R_on characteristics.
• This work contributes to advancing high-performance power semiconductor devices.