MESO Recurrent Neuron Logic for Low-Energy Signal Propagation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Recurrent neural networks (RNNs) using traditional CMOS technology face limitations in component count, energy efficiency, and signal propagation distance due to reliance on spin-polarized currents, which restrict their scalability and performance in applications like natural language processing and image/video captioning.

Innovation Solution

Implementing RNNs with magneto-electric spin orbit (MESO) logic, which reduces component count, operates with low energy consumption (10 attojoules per switching), and enables unlimited input-output distance by using electric charge currents, allowing for faster switching times (<1 nanosecond) and improved scalability through MESO devices in a feedback loop configuration.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If CMOS technology is used for RNN implementation, then device compatibility is maintained, but component count is high and energy efficiency is poor

Engineering Contradiction:

Improveenergy consumptionVSAvoidcomponent count

Core Design Contradiction:

Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The patent combines multiple functions (thresholding, weighting, feedback, and memory storage) into a single MESO device. The MESO device integrates the magnetization port for thresholding, input ports for weighting factors, and feedback loop capability, eliminating the need for separate CMOS components for each function and reducing overall component count while improving energy efficiency

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The MESO device serves multiple purposes simultaneously: it acts as a thresholding unit, a weighted summing unit, a feedback element, and a memory storage device. This multi-functionality allows a single MESO device to replace what would traditionally require multiple separate CMOS components, thereby reducing component count and improving energy efficiency

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Length of moving object

If spin-polarized currents are used for signal propagation, then magnetic state control is achieved, but signal propagation distance is limited

Engineering Contradiction:

Improvesignal propagation distanceVSAvoidsignal propagation reliability

Core Design Contradiction:

Length of moving objectVSReliability

Solution Approach 1:

The patent introduces electric charge currents as an intermediary to transfer information between MESO devices. Instead of relying on spin-polarized currents that decay rapidly over distance, the charge currents serve as a robust mediator that can propagate signals over unlimited distances, while the MESO devices themselves maintain magnetic state control through their magnetization ports

Inventive Principle:

Principle #24Intermediary (Mediator)

3Speed

If traditional CMOS components are used, then manufacturing maturity is maintained, but switching speed is slow and scalability is limited

Engineering Contradiction:

Improveswitching speedVSAvoidscalability

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The patent changes the fundamental operating parameters by using magnetization switching in MESO devices instead of voltage switching in CMOS transistors. This parameter change enables switching times of less than 1 nanosecond and allows for scalable integration of multiple MESO devices in feedback loop configurations, overcoming the speed and scalability limitations of traditional CMOS

Inventive Principle:

Principle #35Parameter changes

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

The MESO-based RNNs achieve improved performance and energy efficiency compared to CMOS-based RNNs, enabling efficient signal propagation and scalability, suitable for various applications including natural language processing, image/video captioning, and sentiment classification.

Implementation Method 1

an output port coupled to a spin orbit effect stack of the SC node

Methodology Applied

Scientific EffectSpin orbit effect:

Data Source

PatentUS11410021B2Recurrent neuron implementation based on magneto-electric spin orbit logic

Publication Date: 2022.08.09 INTEL CORP

AI summary

Techniques are provided for implementing a recurrent neuron (RN) using magneto-electric spin orbit (MESO) logic. An RN implementing the techniques according to an embodiment includes a first MESO device to apply a threshold function to an input signal provided at a magnetization port of the MESO device, and scale the result by a first weighting factor supplied at an input port of the MESO device to generate an RN output signal. The RN further includes a second MESO device to receive the RN output signal at a magnetization port of the second MESO device and generate a scaled previous RN state value. The scaled previous state value is a scaled and time delayed version of the RN output signal based on a second weighting factor. The RN input signal is a summation of the scaled previous state value of the RN with weighted synaptic input signals provided to the RN.