Neuristor

A neuristor is the simplest electronic element that can simulate the behavior of a simple neuron . It is a hypothetical implementation of a neuron model . A neuristor consists of a number of memristors that represent the synapses , as well as a metal-oxide-semiconductor field effect transistor (MOSFET).

technical realization

Circuit of several memristor-based neuristors according to the Hodgkin-Huxley model .

In practice Neuristoren have so far with a crossbar with memristors layer of a titanium (IV) oxide - or niobium (IV) oxide film, and an n-channel Anreichreicherungstyp- MOSFET in silicon - CMOS shown technique.

Memristors are also currently being developed, in which several memristors crossbar layers are superimposed and connected to form a three-dimensional structure. The fixed wiring in the CMOS layer is reduced and replaced by the dynamic connections of the memristor crossbar layer. This structure is more like the structure of the cortical columns in the brains of mammals, which also have three-dimensional wiring. With this structure, the neuristors are networked significantly better due to the higher number and density of memristors ("synapses"). However, this structure is also more complex to manufacture.

Schematic structure of a neuristor in CMOS technology with MOSFET and memristor crossbar

Schematic structure of a neuristor in CMOS technology with MOSFET and several memristor crossbar layers

The structure is compatible with spin FETs and can therefore also be used in spintronics , which enables considerable energy savings. However, the structure with spin-FETs has so far only been simulated in the computer , while a technical implementation is pending.

As an alternative to the use of memristors is magnetic tunnel junctions ( magnetic tunnel junction ; MTJ ) explored.

use

Neuristors can be used as a component to build artificial neural networks based on spiking neuron models. This enables biological neural circuits to be efficiently simulated as part of neuromorphing . These circuits work more efficiently than conventional logic circuits for certain tasks - such as pattern recognition and the simulation of large biological neural circuits.

Individual evidence

^ Adam Stevenson: Logic circuits that program themselves: memristors in action. In: arstechnica . January 28, 2009, accessed December 31, 2012 .
^ John Timmer: "Neuristor": Memristors used to create a neuron-like behavior. In: arstechnica . December 24, 2012, accessed December 31, 2012 .
↑ Konstantin K. Likharev: CrossNets: Neuromorphic Hybrid CMSO / Nanoelectronic Networks. In: Science of Advanced Materials (Vol. 3). American Scientific Publishers, 2011, pp. 322-331 , accessed January 11, 2014 .
↑ Mrigank 'Sharad, Charles Augustine: Proposal For neuromorphic hardware Using Spin Devices . Ed .: Purdue University . arxiv : 1206.3227 (English).
^ Adrien F. Vincent, Jérôme Larroque, et al .: Spin-Transfer Torque Magnetic Memory as a Stochastic Memristive Synapse for Neuromorphic Systems. In: IEEE Transactions on Biomedical Circuits and Systems. April 2015, accessed March 1, 2017 .

[1] Adam Stevenson: Logic circuits that program themselves: memristors in action. In: arstechnica . January 28, 2009, accessed December 31, 2012 .

[2] John Timmer: "Neuristor": Memristors used to create a neuron-like behavior. In: arstechnica . December 24, 2012, accessed December 31, 2012 .

[3] Konstantin K. Likharev: CrossNets: Neuromorphic Hybrid CMSO / Nanoelectronic Networks. In: Science of Advanced Materials (Vol. 3). American Scientific Publishers, 2011, pp. 322-331 , accessed January 11, 2014 .

[4] Mrigank 'Sharad, Charles Augustine: Proposal For neuromorphic hardware Using Spin Devices . Ed .: Purdue University . arxiv : 1206.3227 (English).

[STT-5] Adrien F. Vincent, Jérôme Larroque, et al .: Spin-Transfer Torque Magnetic Memory as a Stochastic Memristive Synapse for Neuromorphic Systems. In: IEEE Transactions on Biomedical Circuits and Systems. April 2015, accessed March 1, 2017 .