Scaled, Ferroelectric Memristive Synapse for Back‐End‐of‐Line Integration with Neuromorphic Hardware

• Published in: Advanced electronic materials Vol. 8; no. 6
• Main Authors: Bégon‐Lours, Laura, Halter, Mattia, Puglisi, Francesco Maria, Benatti, Lorenzo, Falcone, Donato Francesco, Popoff, Youri, Dávila Pineda, Diana, Sousa, Marilyne, Offrein, Bert Jan
• Format: Journal Article
• Language: English
• Published: 01.06.2022
• Subjects: back‐end‐of‐line; ferroelectrics; resistive switching; synaptic weight; transport
• ISSN: 2199-160X; 2199-160X
• DOI: 10.1002/aelm.202101395

Ohmic, memristive synaptic weights are fabricated with a back‐end‐of‐line compatible process, based on a 3.5 nm HfZrO4 thin film crystallized in the ferroelectric phase at only 400 °C. The current density is increased by three orders of magnitude compared to the state‐of‐the‐art. The use of a metallic oxide interlayer, WOx, allows excellent retention (only 6% decay after 106 s) and endurance (1010 full switching cycles). The On/Off of 7 and the small device‐to‐device variability (<5%) make them promising candidates for neural networks inference. The synaptic functionality for online learning is also demonstrated: using pulses of increasing (resp. constant) amplitude and constant (resp. increasing) duration, emulating spike‐timing (resp. spike‐rate) dependent plasticity. Writing with 20 ns pulses only dissipate femtojoules. The cycle‐to‐cycle variation is below 2%. The training accuracy (MNIST) of a neural network is estimated to reach 92% after 36 epochs. Temperature‐dependent experiments reveal the presence of allowed states for charge carriers within the bandgap of hafnium zirconate. Upon polarization switching, the screening of the polarization by mobile charges (that can be associated with oxygen vacancies and/or ions) within the ferroelectric layer modifies the energy profile of the conduction band and the bulk transport properties. A back‐end‐of‐line, ferroelectric synapse shows excellent cycle‐to‐cycle and device‐to‐device variation (2%), retention and endurance (1010 cycles). A 10% error is measured in a 3 × 3 crossbar (92% accuracy predicted on the MNIST data set). Ohmic conduction at low bias is ideal for multiply and accumulate operation. Temperature measurements reveal the analog resistive switching mechanisms in semiconducting and ferroelectric HfZrO4.