Compact MEOL OxRAM with 14 Conductance Levels for Dense Embedded Inference Computing

Download PDF
Open Webpage

Joel Minguet Lopez, Sylvain Barraud, Manon Dampfhoffer, Aurelie Souhaité, Theophile Dubreuil, Jean-Michel Pedini, Corinne Comboroure, Ahmed Gharbi, François Boulard, Clément Castan, Amélie Lambert, François Andrieu

Published: 2025, Last Modified: 16 Apr 2026IRPS 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: A compact 1T1R memory cell with 14 conductance levels is demonstrated in 28nm node technology. The HfO2-based OxRAM is integrated in the 40nm×40nm drain contact of thin gate oxide Fully Depleted Silicon-On-Insulator (FDSOI) transistors, opening the way towards memory arrays with 0.0357μm2 bitcell area. The OxRAM stack and programming protocols are optimized, resulting in 106 endurance, 2h data retention at 150°C and 109 read disturb cycles, while keeping low SET currents (30μA) and 1.2V RESET voltages. Furthermore, the pertinence of this OxRAM for Quantized Spiking Neural Network (QSNN) synaptic weight hardware implementation for Automatic Lip-Reading (ALR) inference computing on-chip is elucidated, reaching 74.3% near state-of-art accuracy.