Deep Learning Based Multiresponse Optimization Methodology for Dual-Axis MEMS Accelerometer.

Area of Science:

• Engineering
• Materials Science
• Computer Science

Background:

• Microelectromechanical systems (MEMS) capacitive accelerometers are crucial for inertial sensing.
• Optimizing MEMS accelerometer design involves complex trade-offs between geometric parameters and operating conditions.
• Existing optimization methods can be computationally intensive and may not efficiently handle multi-response objectives.

Purpose of the Study:

• To develop a novel deep neural network (DNN) based methodology for the design optimization of dual-axis MEMS capacitive accelerometers.
• To create a unified model that analyzes the impact of individual design parameters and operating conditions on sensor performance.
• To enable efficient simultaneous optimization of multiple output responses for MEMS accelerometers.

Main Methods:

• A deep neural network (DNN) model was designed and trained using geometric design parameters and operating conditions as inputs.
• The DNN model was used to predict the output responses of the MEMS accelerometer.
• The performance of the DNN-based optimization was compared against the Design of Computer Experiments (DACE) based multiresponse optimization methodology.

Main Results:

• The DNN-based model effectively analyzes the influence of individual design parameters on MEMS accelerometer output responses.
• The proposed DNN methodology achieved simultaneous optimization of multiple output responses efficiently.
• The DNN approach demonstrated superior performance compared to DACE, showing lower Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).

Conclusions:

• Deep neural networks offer an efficient and effective approach for the design optimization of dual-axis MEMS capacitive accelerometers.
• The DNN-based methodology provides a powerful tool for analyzing parameter sensitivities and optimizing complex MEMS devices.
• This work establishes a new benchmark for MEMS accelerometer optimization, highlighting the potential of DNNs in micro-device design.