Deep-Learning-Based Optimization for a Low-Frequency Piezoelectric MEMS Energy Harvester

High operational frequency is one of the major limitations of the conventional MEMS vibration energy harvesters. In this work, we present a piezoelectric MEMS energy harvester with the capability of operating at a low resonant frequency (i.e., less than 200 Hz). The proposed harvester has a symmetric serpentine structure with a doubly clamped configuration comprising several proof masses at the junctions. In order to facilitate the design process and determine the optimum physical dimensions, an artificial neural network is used to model the design. In the first step, a dataset with 108 samples is generated by finite element modeling (FEM) to train a deep neural network. The validation results indicate that the trained deep neural network model can achieve around 90% estimation accuracy of device features, such as resonant frequency and harvested voltage. Next, this trained model is integrated with genetic algorithm as a performance estimator to optimize the geometry of the harvester to lower the resonant frequency and improve the harvested voltage. An optimized harvester with a total area of 8.7 mm ² has been fabricated through a standard micromachining process. Our measurement results confirm that the proposed deep-learning-based method can help reach the balanced summit of both higher power density and lower resonant frequency among the published works. Our prototype device features the first resonant frequency of 121.7 Hz and the harvested power of $0.73~\mu \text{W}$ under 0.1g input acceleration.