AI/ML Based Design Optimization & Fabrication of ALN Piezo-Cantilever Array for Bio-Mimetic MEMS Cochlea

We propose a biomimetic, fully implantable MEMS-based cochlear implant utilizing an AlN piezo-cantilever array. Gaussian Process Regression (GPR) integrated with Genetic Algorithm was used to optimize cantilever geometry for maximum charge output at a desired frequency within 0.25 mm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> area. Multiple machine learning models were trained with COMSOL data, and GPR gave best prediction with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}=0.9999$</tex> for resonant frequency and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}=0.98$</tex> for charge. Optimized designs generate 0.7-8 mV, sufficient for direct auditory neuron stimulation. Fabrication employed a new two-step method to grow thick, high quality AlN layer. We also report initial measurement of resonant frequencies and sensitivity of the fabricated array.