Memristive synapses

Members: Andreas Grübl, Aurel A. Lazar, Bernabe Linares-Barranco, Charles Clercq, Christoph Posch, Eero Lehtonen, Giacomo Indiveri, Jennifer Hasler, Eric Ryu, Jayawan Wijekoon, Jonathan Tapson, Kaijun Yi, Alexander Kazeka, Love Cederstroem, Marc-Olivier Schwartz, Martin Salinga, Mika Laiho, Radoslav Prahov, Robert Nawrocki, Yulia Sandamirskaya, Simon Friedmann, Shih-Chii Liu, Sim Bamford, Sebastian Millner, Tobi Delbruck, serrano, Vasilis Thanasoulis, André van Schaik

Discussions

This group will concentrate on discussing various aspects of a memristor, it's operation and modeling, as well as the possibility of utilizing this two-terminal device as a memristive synapse. Inorganic based memristor will be compared to an organic memristor. An organic bistable device (OBD) will be discussed, including its physical and electrical properties, its relation to a memristor, as well as the possibility of using OBD for the function of a binary and non-binary artificial synapse.

Various implementations of memristive, spiking synapses will be introduced, including NOMFET or nanoparticle organic memory field effect transistor (Alibart, 2010) and memristive cross-bars (Jo, 2010).

Meetings

Day 1

18 participants

An introduction to a memristor was given: what is a memristor, when the behavior is observed, their IV characteristics, response to voltage in time, how to construct such devices. An introduction to Organic Bistable Devices (OBDs) was presented, their bistable nature, materials used, their ability to operate as a binary synapse, electrical characteristics, and possible explanation(s) of resistive switching. A circuit that approximates a behavior of non-spiking neuron (Synthetic Neural Network, or SNN) utilizing OBDs as synthetic synapses was shown. Additionally two alternative proposals of constructing devices demonstrating memristive behavior (NOMFET - Nanoparticle Organic Memory Field Effect Transistor) and utilizing memristive synapses via cross-bar architecture was visited.

Day 2

15 participants

The discussion followed the topics introduced during the first meeting. Martin briefly introduced the concept of phase-change materials and their relation to approximating the behavior of an artificial synapse. Aurel introduced a question of "what is a memory in a neuromorphic system." Subsequently the discussion continued with the aim of understanding how memory is encoded in networks of spiking neurons including a brief comparison with networks of non-spiking neurons.

Day 3

16 participants

The meeting started with Eero introducing modeling of memristive behavior alternative to the one proposed by Biolek (2009). Subsequently Eero discussed alternative uses for memristors including implication logic. Barnabe introduced a model of for constructing network of spiking neurons, based on the use of flux-driven memristors utilized to act as synapses that use a spiking input signal. The meeting ended with a discussion of a question proposed by Eero, namely "are binary synapses enough to encode complex information."

Projects

A possible project will include an introduction to Emergent (introduced next) for simple, yet biologically realistic neural simulations.

Links/Software?

Emergent is a neural network simulator that allows for the creation and analysis of complex, sophisticated models of the brain in the world.  http://grey.colorado.edu/emergent/index.php/Main_Page

Micro-Cap is a SPICE-based electronic circuit simulator. As of the latest version, it features memristors, memcapacitors, and meminductors, as defined by Chua (2009) and Biolek (2009).  http://www.spectrum-soft.com/index.shtm

Multisim is SPICE-based electronic circuit simulator by National Instruments (the maker of LabView? ( http://www.ni.com/academic/purchasing.htm)). It is extremely easy to use, as all of the analysis are done using GUI, which emulates the work in a lab (it's actually called Electronic Workbench).  https://lumen.ni.com/nicif/us/academicevalmultisim/content.xhtml or if that link does not work, try  http://www.ni.com/academic/multisimse.htm