HZO-based FerroNEMS MAC for In-Memory Computing

TL;DR

This paper introduces a hafnium zirconium oxide (HZO)-based ferroelectric NEMS unimorph device that enables low-energy in-memory multiply-accumulate operations through analog control of its piezoelectric coefficient, promising high throughput and CMOS compatibility.

Contribution

The work presents a novel ferroelectric NEMS unimorph device with programmable piezoelectric properties for in-memory computing, demonstrating its potential for scalable, low-energy MAC operations.

Findings

Achieved analog control of the piezoelectric coefficient via partial ferroelectric switching.

Demonstrated programmable displacement response for multiply-accumulate operations.

Indicated CMOS compatibility and high computational throughput potential.

Abstract

This paper demonstrates a hafnium zirconium oxide (HZO)-based ferroelectric NEMS unimorph as the fundamental building block for very low-energy capacitive readout in-memory computing. The reported device consists of a 250 $\mu$m $\times$ 30 $\mu$m unimorph cantilever with 20 nm thick ferroelectric HZO on 1 $\mu$m $SiO_2$.Partial ferroelectric switching in HZO achieves analog programmable control of the piezoelectric coefficient ($d_{31}$) which serves as the computational weight for multiply-accumulate (MAC) operations. The displacement of the piezoelectric unimorph was recorded by actuating the device with different input voltages $V_{in}$. The resulting displacement was measured as a function of the ferroelectric programming/poling voltage $V_p$. The slopes of central beam displacement ($\delta_{max}$) vs $V_{in}$ were measured to be between 182.9nm/V (for -8 $V_p$) and -90.5nm/V (for 8 $V_p$), demonstrating that $V_p$ can be used to change the direction of motion of the beam. The resultant ($\delta_{max}$) from AC actuation is in the range of -18 to 36 nm and is a scaled product of the input voltage and programmed $d_{31}$ (governed by the $V_p$). The multiplication function serves as the fundamental unit for MAC operations with the ferroelectric NEMS unimorph. The displacement from many such beams can be added by summing the capacitance changes, providing a pathway to implement a multi-input and multi-weight neuron. A scaling and fabrication analysis suggests that this device can be CMOS compatible, achieving high in-memory computational throughput.