Design and Implementation of an AlScN-Based FeMEMS Multiplier for In-Memory Computing Applications

This paper reports on the design, fabrication, and experimental validation of an aluminum scandium nitride (AlScN) based Ferroelectric Micro-Electro-Mechanical Systems (FeMEMS) Multiplier—a core component for multiply-accumulate (MAC) operations in next-generation in-memory computing applications. The FeMEMS multiplier leverages ferroelectric polarization switching in AlScN to change the piezoelectric coefficient (d
31
), facilitating non-volatile, analog memory storage for weights in a neural network. The piezoelectric parameters of the films are then used to change a capacitive gap for readout. The ferroelectric thin films could be partially polarized and reached a peak remnant polarization of 216µC/cm
²
at a voltage of 100V V
P
(5MV/cm). Experimental results on optically measured displacements confirmed the AlScN unimorph multiplier’s operation. The maximum resonance mode displacement was linearly dependent on the polarization and input voltages. This work provides foundational insights into utilizing AlScN in in-memory computing, opening new avenues for high-speed, low-power, and high-accuracy computing applications.