Complementary Memristive Synapse for Bidirectional Conductance Control

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing artificial neural networks using memristive devices of PCM type cannot implement the STDP learning rule effectively due to the inability of these devices to progressively decrease conductance, which is essential for unsupervised learning methods like STDP, as they only vary conductance in one direction.

Innovation Solution

A neural network architecture utilizing two memristive devices in opposition to form an artificial synapse, allowing for both increasing and decreasing conductance based on the state of connected neurons, with specific pulse configurations and thresholds to achieve alternating conductance variations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If unipolar PCM memristive devices are used to implement artificial synapses, then high integration density and fast state changes are achieved, but the ability to progressively decrease conductance is lost

Engineering Contradiction:

Improveintegration densityVSAvoidconductance variation capability

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The synapse is segmented into two separate unipolar PCM devices (first and second memristive devices) connected in opposition. Each device maintains its unipolar characteristics for high integration density, while the opposing connection enables bidirectional conductance control through differential operation

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The two unipolar PCM devices are connected in opposition to each other, where one device's conductance increase counterbalances the other's conductance decrease. This anti-weight configuration allows the synapse to achieve progressive conductance decrease capability while each individual device maintains its unipolar nature for high integration density

Inventive Principle:

Principle #8Anti-weight (Counterweight)

2Device complexity

If unipolar PCM devices are used for STDP implementation, then device complexity is reduced, but the learning rule cannot be properly implemented due to unidirectional conductance variation

Engineering Contradiction:

Improvedevice structureVSAvoidSTDP rule implementation

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The synapse is divided into two unipolar PCM devices connected in opposition, allowing each device to maintain simple unipolar structure while the combined system achieves the bidirectional conductance variation needed for reliable STDP implementation

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of requiring each device to independently perform bidirectional conductance variation, the invention inverts the approach by using two unipolar devices in opposition where one device's unidirectional change is counterbalanced by the other, achieving the desired bidirectional synapse behavior through inverted operational logic

Inventive Principle:

Principle #13The other way round (Inversion)

3Ease of operation

If complex programming signals are used to implement STDP with unipolar devices, then conductance control is attempted, but the method fails to account for the absence of progressive conductance decrease

Engineering Contradiction:

Improveprogramming signal complexityVSAvoidlearning rule implementation

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The invention inverts the conventional STDP approach by using two unipolar devices in opposition rather than attempting to make a single device perform bidirectional variation. This inversion allows standard unipolar programming signals to work effectively while the opposing connection provides the necessary conductance decrease capability

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The output neuron's activation state provides feedback that determines the polarity of pulses applied to the two memristive devices. This feedback mechanism dynamically switches between potentiation and depression modes, enabling proper STDP implementation without complex pre-programmed signals

Inventive Principle:

Principle #23Feedback

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Enables the implementation of unsupervised learning methods like STDP by allowing progressive variation in synapse conductance in both increasing and decreasing directions, overcoming the limitations of unipolar PCM devices.

Implementation Method 1

resistive components whose conductance varies as a function of the voltage applied across their terminals

Methodology Applied

Scientific EffectMemristive effect: Electrical Resistance

Implementation Method 2

said output neuron integrates the difference between the currents originating from the first and second devices

Methodology Applied

Scientific EffectElectrical conduction: Conduction (electrical)

Data Source

PatentUS9396431B2Network of artificial neurons based on complementary memristive devices

Publication Date: 2016.07.19 COMMISSARIAT A LENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES

AI summary

A neural network comprises a plurality of artificial neurons and a plurality of artificial synapses each input neuron being connected to each output neuron by way of an artificial synapse, the network being characterized in that each synapse consists of a first memristive device connected to a first input of an output neuron and of a second memristive device, mounted in opposition to said first device and connected to a second, complemented, input of said output neuron so that said output neuron integrates the difference between the currents originating from the first and second devices.