Deep learning-driven performance prediction and design of high-DoF MEMS resonators

Area of Science:

• Microelectromechanical Systems (MEMS)
• Computational Physics
• Artificial Intelligence in Engineering

Background:

• Microelectromechanical systems (MEMS) resonator design is computationally intensive.
• Existing methods face limitations in structural degrees of freedom and speed.

Purpose of the Study:

• Introduce a novel structural design framework for MEMS resonators.
• Enable rapid physical performance prediction and efficient optimization.
• Overcome computational cost and design freedom limitations.

Main Methods:

• Developed a residual network-enabled solver (ResNES) for performance prediction.
• Employed a stochastic topology generation strategy for diverse designs.
• Integrated ResNES with particle swarm optimization for multi-objective design.

Main Results:

• ResNES achieves a ~1000x speedup over FEM with <3% prediction error.
• Experimental validation shows <5% discrepancy.
• Optimized designs identified in minutes, reducing design cycle by 70%.

Conclusions:

• ResNES provides a scalable paradigm for intelligent MEMS structural design.
• The framework enables seamless integration of analysis and fabrication.
• Accelerated design process leads to improved high-performance MEMS devices.