Stochastic behavior of random telegraph noise in ferroelectric devices: Impact of downscaling and mitigation strategies for neuromorphic applications

Abstract

This study investigates the stochastic behavior of random telegraph noise (RTN) in ferroelectric tunnel junctions (FTJs) considering the downscaling effect and its implications for neuromorphic systems. Through low-frequency noise spectroscopy and DC current fluctuation measurements of fabricated FTJs with varying top electrode areas, we quantified the stochasticity of the tunneling current as a function of applied voltage and device area. Our results indicate a significant increase in RTN-related stochasticity with decreasing FTJ area, resulting in higher RTN amplitude and a greater number of devices exhibiting RTN. Analysis of the capture and emission time constants of RTN shows that RTN arises from the interaction between the metal top electrode and a dominant trap site, located 4 nm deep from the top electrode, with a trap energy 1.8 eV below the conduction band of the HZO layer. To assess the impact on neuromorphic systems, we performed system-level simulations incorporating the measured device non-idealities (nonlinearity, limited dynamic range) and stochasticity (1/f noise and RTN), and demonstrated that RTN can severely degrade system accuracy as device size decreases. To mitigate this problem, we proposed a limited dynamic range scheme that confines device operation to RTN-safe conductance levels, effectively minimizing accuracy degradation. This study clarifies the origin of the stochastic behavior of RTN in FTJs and also provides system-level solutions for high-density neuromorphic hardware systems affected by RTN.