Neuromorphic applications of silicon oxide memristors

Lead Research Organisation: University College London
Department Name: Electronic and Electrical Engineering

Abstract

The project will investigate the potential for silicon oxide resistance switches (memristors) to be used in neuromorphic systems for non-Von Neumann computation. Through a series of experimental and modelling studies it will develop the elements of hardware neural networks and establish to what degree existing CMOS systems can be replaced by memristive systems. This has direct relevance to the EPSRC ICT theme research area of microelectronics device technology

Publications

10 25 50
publication icon
Berg AI (2019) Synaptic and neuromorphic functions: general discussion. in Faraday discussions

publication icon
Mannion D (2020) Memristor-Based Edge Detection for Spike Encoded Pixels in Frontiers in Neuroscience

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/N509577/1 01/10/2016 30/09/2021
1917855 Studentship EP/N509577/1 01/10/2017 30/09/2021 Daniel John Mannion
 
Description I have shown that changes in an oxides conductance can be used to detect the edges within an image. The sensors are on the nanoscale and can therefore can be placed in between neighbouring pixels to detect the difference in their intensities. If this measured difference is large, then we assume there may be an edge lying across the neighbouring pixels. This finding demonstrates it may be possible to detect edges within images at a reduced power consumption which would be ideal in autonomous robotics and computer vision applications, for example, drones with tight power budgets. This finding was published in a journal article - DOI: 10.5522/04/9741722

In addition to this applied research, there have been findings relating to the fundamental mechanisms occurring within these oxide devices. I have investigated what may be causing the changes in oxide conductance. In doing so, I will potentially challenge the existing models addressing the case of current transients observed in oxide capacitors. However, it is too early to say with confidence whether this is the case. Hence, this is work which I shall carry on developing throughout the remainder of this award.
Exploitation Route The findings on edge detection could potentially be taken up by industry looking to develop low power vision sensors. The system architecture and device requirements have all been published in a journal with open access rights.
Sectors Digital/Communication/Information Technologies (including Software),Electronics

URL https://www.frontiersin.org/articles/10.3389/fnins.2019.01386/full
 
Title Memristor-Based Edge Detection Dataset 
Description This is the data produced for the paper titled: Memristor-Based Edge Detection for Spike Encoded Pixels. In this study, we investigate the use of silicon dioxde memristors in edge detection. Devices exhibit analogue and volatile behaviours and are connected in potential divider arrangements. Encoding image pixels as spike trains and applying these to the memristors allow us to detect sharp changes in pixel intensity and in turn predict edges within an image. The dataset contains data from: device characterisation, interpolated simulation models, output images generated in simulation and device variance data. 
Type Of Material Database/Collection of data 
Year Produced 2019 
Provided To Others? Yes  
Impact This dataset was crucial in developing simulations that went on to produce a journal paper. 
URL https://rdr.ucl.ac.uk/articles/Memristor-Based_Edge_Detection_Dataset/9741722
 
Description A Talk to Entrepreneurs on Neuromorphic Engineering and its Applications to Machine Learning 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Local
Primary Audience Industry/Business
Results and Impact I presented a talk on the topic of neuromorphic engineering to a group of entrepreneurs at an incubator based in London. The evening was centred around machine learning and so I discussed the limitations of today's hardware and presented the alternative - neuromorphic hardware.
Year(s) Of Engagement Activity 2018
URL https://www.youtube.com/watch?v=j50wbVyf0Dw&t=235s