Polarization current-based reservoir computing utilizing an anti-ferroelectric-like HfZrO2 capacitor
Shin-Yi Min, Eishin Nako, Ryosho Nakane, Mitsuru Takenaka, Kasidit Toprasertpong, Shinichi Takagi · APL Machine Learning · 2025
We have experimentally demonstrated the physical reservoir computing by employing the polarization switching current dynamics of an Hf1−xZrxO2 (HZO)-based metal/ferroelectric/metal capacitor with Zr content x = 0, 0.5, and 0.75. The spatial distribution of the crystalline phase of an HZO film reveals that the tetragonal phase is a dominant crystal structure in the HZO film with [Zr] = 75%, resulting in anti-ferroelectric (AFE)-like double polarization switching. Analyses using t-distributed stochastic neighbor embedding (t-SNE) find that the AFE-HZO capacitor effectively transforms the 3-bit time-series input into eight different reservoir output states. In reservoir computing tasks, the AFE-HZO capacitor with [Zr] = 75% achieves improved computational capacities compared with the other MFM capacitors with [Zr] = 0% and 50%. The AFE-HZO capacitor can effectively diversify time-series input signals through dynamic double polarization switching, leading to a more sub-divided and dispersive weight distribution across the adjustable weights in the readout part of our RC system.