Lorentzian Switching Dynamics in HZO-Based FeMEMS Synapses for Neuromorphic Weight Storage

Neuromorphic computing demands synaptic elements that store and update analogue weights with high precision while minimizing read disturbance. Conventional ferroelectric synapses typically encode weights in remanent polarization states and infer them from polarization-dependent electrical characteristics. Here, we demonstrate a ferroelectric MEMS (FeMEMS) synapse in which analog weights are stored in the effective piezoelectric coefficient d_31,eff of a released unimorph beam with a 20 nm HZO layer. Partial domain switching modulates d_31,eff, and a low-amplitude AC drive under subcoercive conditions converts the programmed state into beam displacement proportional to d_31,effV_ac, realizing single-device analogue multiplication during readout. The switching-threshold distribution follows a Lorentzian form, and the median threshold obeys a Merz-type field-time law. Using this framework, we demonstrate ∼200 electromechanical weight levels. We further show representative retention and endurance, establishing a compact FeMEMS synaptic weight element for calibration-aware electromechanical multiplication in neuromorphic hardware.