Sabina Spiga: Resistance switching memories for spiking neural networks

Memristive device technologies have been receiving an increasing interest for a wide range of applications, such as storage class memory, non-volatile logic switch, in-memory and neuromorphic computing. Among the proposed devices, oxide-based resistance switching memories (RRAM) are based on redox reactions and electrochemical phenomena in oxides and are very promising because of low power consumption, fast switching times, scalability down to nm scale and CMOS compatibility. Therefore, they can be used as new building blocks for brain-inspired computing technologies towards, for instance, spiking neural networks (SNN) for real time and low power computation systems. In this seminar, after a brief introduction of the current emerging technologies and devices for neuro-inspired architectures, I will present our work which span from materials and devices to system level simulation of SNN, including RRAM as key elements of synaptic nodes. In particular, it is possible to tune the RRAM conductance either in a digital or in an analog fashion by using various programming schemes, and the update of the conductance values can be achieved by a spike timing and rate dependent plasticity mechanism, which is demonstrated also at hardware level. System level simulation of SNN is implemented by using experimental RRAM characteristics for the synaptic nodes and neural equations derived from hardware CMOS implementation [1-3].