Oxide based resistive memories for low power embedded applications and neuromorphic systems

E. Vianello (1) , D. Garbin (1) , N. Jovanovic (1) , O. Bichler (2) , O. Thomas (1) , B. de Salvo (1) , L. Perniola (1)

1 CEA-LETI - Commissariat à l'énergie atomique et aux énergies alternatives - Laboratoire d'Electronique et de Technologie de l'Information
2 DACLE-LIST - Département d'Architectures, Conception et Logiciels Embarqués-LIST

Oxide based resistive memories (OxRAMs) is one of the potential candidates for non-volatile logic circuits and neuromorphic circuits in the applications of wearable devices, internet of things (IoT), cloud computing, and big-data processing. One of the main OxRAMs issue is the noise behavior of the high resistance state (HRS). In this work, we will demonstrate a hybrid (CMOS logic plus ReRAM devices) Non Volatile Flip Flop designed to face OxRAM variability. Concerning neuromorphic circuits, we will focus on the impact of resistance variability on the performance of Convolutional Neural Network (CNN) systems for visual pattern recognition applications.

Keywords: Big data; Computation theory; Data handling; Embedded systems; Flip flop circuits; Networks (circuits); Neural networks; Pattern recognition; Pattern recognition systems; Convolutional neural network; Embedded application; High-resistance state; Internet of Things (IOT); Neuromorphic circuits; Neuromorphic systems; Non-volatile flip-flops; Visual pattern recognition; Logic circuits

Domains: Engineering Sciences [physics]

E. Vianello, D. Garbin, N. Jovanovic, O. Bichler, O. Thomas, et al.. Oxide based resistive memories for low power embedded applications and neuromorphic systems. ECS Transactions, 2015, 69 (3), pp.3-10. ⟨10.1149/06903.0003ecst⟩. ⟨cea-01837009⟩