Machine Learning Enabled FBAR Digital Twin for Rapid Optimization

In this paper we discuss a machine learning-based method to obtain a digital twin of a Thin Film Bulk Acoustic Wave Resonator (TFBAR) that can be used as a surrogate for simulations to estimate resonance frequencies of devices. Normalized root mean square error values better than 0.04% and 0.1% were achieved for 1D and 2D models, respectively. Training times for neural networks were ~20 s for ~2000 epochs and hundreds of datasets.