Synaptic and neuronal functionalities on a single oxide film
Abstract
Several Exabytes of data are produced worldwide every day. This number is predicted to grow dramatically, posing a new and very challenging problem: how do we handle so much data? New ways of processing, classifying and discarding information are needed to keep up with the accelerating pace of today’s information society. Nature might offer a solution to this problem since biological brains are very good precisely at this task: processing large amounts of data and extracting patterns from them. Neuromorphic computing is anew computation paradigm which attempts to mimic biology by creating hardware that emulates the behavior of neural networks. For this purpose, it is fundamental to find new materials or devices that emulate the fundamental properties of neurons and synapses.The most promising way to achieve this is by using an effect known as resistive switching, in which the resistance of a material can be modified by applying a voltage to it. There are two different types of resistive switching: volatile and non-volatile, which have been shown ideal for replicating the basic functionalities of neurons and synapses, respectively. However, these two types of switchinghave different origins and are observed in different types of materials, a big obstacle towards practical implementation. Non-volatile switching is observed in most oxides, and is caused by the drift of oxygen vacancies under strong electric fields. But volatile switching is only observed in a very particular type systems that feature an insulator-to-metal transition (IMT), which can be triggered by applying voltage.We propose to bridge these differences by implementing both volatile and non-volatile switching in a singlesystem, in a single thin film. We will use oxides with an IMT, such as VO2 or SmNiO3, and engineer the interface between these materials and different metallic electrodes. By switching between an ohmic or a Schottky-like contact, we can tailor where the voltage drops in the system. This will in turn result either in a drift of oxygen vacancies (non-volatile switching) or a voltage-driven IMT (volatile), depending on the oxide/metal contact, ultimately allowing us to choose between neuron-like or synapse-like characteristics. If successful, this technique would allow for patterning neurons, synapses and the whole neural network in the same material andsubstrate, paving the way for upscaling neuromorphic hardware.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 09, 2021
- Source ID
- N629092112028
Entities
People
- Stefano Gariglio
Organizations
- Office of Naval Research
- United States Navy
- University of Geneva