Multiple binary OxRAMs as synapses for convolutional neural networks

E. Vianello (1) , D. Garbin (1) , O. Bichler (2) , G. Piccolboni (1) , G. Molas (1) , B. De Salvo , L. Perniola (1)

1 DACLE-LETI - Département d'Architectures, Conception et Logiciels Embarqués-LETI
2 DACLE-LIST - Département d'Architectures, Conception et Logiciels Embarqués-LIST

Oxide-based resistive memory (OxRAM) devices find applications in memory, logic, and neuromorphic computing systems. Among the different dielectrics proposed in OxRAM stacks, hafnium oxide, HfO$_2$, attracted growing interest because of its compatibility with typical BEOL advanced CMOS processing and promising performances in terms of endurance (higher than Flash) and switching speed (few tens of ns). This chapter describes an artificial synapse composed of multiple binary HfO$_2$-based OxRAM cells connected in parallel, thereby providing synaptic analog behavior. The VRRAM technology is presented as a possible solution to gain area with respect to planar approaches by realizing one VRRAM pillar per synapse. The HfO$_2$-based OxRAM synapse has been proposed for hardware implementation of power efficient Convolutional Neural Networks for visual pattern recognition applications. Finally, the synaptic weight resolution and the robustness to device variability of the network have been investigated. Statistical evaluation of device variability is obtained on a 16 kbit OxRAM memory array integrated into advanced 28nm CMOS technology.

Domains: Computer Science [cs]

E. Vianello, D. Garbin, O. Bichler, G. Piccolboni, G. Molas, et al.. Multiple binary OxRAMs as synapses for convolutional neural networks. Advances in Neuromorphic Hardware Exploiting Emerging Nanoscale Devices. Cognitive Systems Monographs, 31, Springer Verlag, pp.109-127, 2017, ⟨10.1007/978-81-322-3703-7_6⟩. ⟨cea-01818392⟩