Unveiling Resistance Switching Mechanisms in Undoped HfO_x Ferroelectric Tunnel Junction Using Low-Frequency Noise Spectroscopy

Proposition of Adaptive Read Bias: A Solution to Overcome Power and Scaling Limitations in Ferroelectric‐Based Neuromorphic System

“…[10,14,18] In contrast, ferroelectric tunnel junctions (FTJs), which have the advantages of fast operation speed, low current density, high reliability, process simplicity, and high integration (4F 2 , F: Minimum feature size), are considered the strongest candidates for implementing synapses in neuromorphic computing systems. [20][21][22][23][24][25] This movement gained significant traction following the breakthrough discovery of ferroelectricity in hafnium oxide (HfO 2 ) thin films in 2011, which was fully compatible with conventional CMOS processes. Subsequently, numerous studies have reported the possibility of implementing neuromorphic systems using HfO 2 -FTJs.…”

“…For instance, resistive random‐access memory (ReRAM) stores information as a resistance, [ 7 , 8 , 9 , 10 ] phase‐change memory (PCM) stores it as a phase change, [ 11 , 12 , 13 , 14 ] magnetic tunnel junction (MTJ) stores it magnetically, [ 15 , 16 , 17 , 18 ] and ferroelectric (FE)‐based memory stores it in a polarization state. [ 19 , 20 , 21 , 22 , 23 ] To realize highly integrated, low‐power, high‐performance neuromorphic systems, synaptic devices require high integration, low current density, high reliability, and high operating speed. However, PCM and MTJ suffer from high power consumption owing to their high current density, and ReRAMs suffer from reliability issues when learning frequently.…”