FN Synapse Memory Consolidation With Differential Floating Gates

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing synaptic models for artificial neural networks struggle to achieve optimal memory consolidation and plasticity-stability trade-offs, making it difficult to scale and implement in-silico, and the physical realization of synaptic devices lacks tunable consolidation properties.

Innovation Solution

The use of Fowler-Nordheim (FN) synapses, which operate using quantum-mechanical tunneling, to store synaptic weights differentially and implement synaptic memory consolidation, allowing for tunable plasticity-stability trade-offs through a reservoir model and modulation of synaptic weights.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If standard synaptic storage elements are used in artificial neural networks, then the network structure is simple, but the memory consolidation capability and plasticity-stability trade-off are insufficient

Engineering Contradiction:

Improvememory consolidation capabilityVSAvoidsynaptic device structure

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The synaptic device is segmented into multiple functional regions within a single memory cell: a first region storing synaptic weight information and a second region storing consolidation information. This segmentation enables both memory consolidation capability and plasticity-stability trade-off without requiring complex coupling of multiple devices, thus improving reliability while maintaining relatively simple device structure.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements nesting by embedding consolidation information storage within the same synaptic device structure as weight storage. The consolidation information is stored in a separate region within the same memory cell, allowing the device to perform both weight storage and consolidation functions hierarchically, enhancing memory consolidation capability without proportionally increasing device complexity.

Inventive Principle:

Principle #7Nested doll (Nesting)

2Reliability

If complex coupling of dynamical states and diffusion dynamics is used to achieve optimal memory consolidation, then the consolidation characteristic is optimal, but the implementation is difficult to scale in-silico

Engineering Contradiction:

Improvememory consolidation characteristicVSAvoidscalability in-silico

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The patent extracts the essential consolidation mechanism from complex diffusion dynamics and implements it through controlled charge transfer between distinct regions within the memory cell. By taking out only the necessary consolidation function and implementing it through electrostatic charge distribution rather than complex diffusion processes, the system achieves optimal consolidation characteristics while improving ease of simulation and scaling.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent replaces complex diffusion dynamics (mechanical/physical process) with electrostatic charge transfer and control (electrical process). This substitution allows the same consolidation function to be achieved through electrical fields and charge distribution, which are more easily simulated and scaled in silico compared to diffusion-based mechanisms.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If algorithmic synaptic consolidation models are used, then the plasticity-stability trade-off can be tuned, but it is not clear if optimal consolidation characteristics can be naturally implemented on the synaptic device

Engineering Contradiction:

Improveplasticity-stability trade-off tuningVSAvoidnatural implementation on device

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements dynamic plasticity-stability trade-off tuning through controllable charge transfer mechanisms. The consolidation information storage region allows dynamic adjustment of synaptic characteristics by controlling charge distribution between regions, enabling the device to naturally exhibit tunable plasticity-stability trade-offs that are reliably implemented through physical charge control rather than purely algorithmic approaches.

Inventive Principle:

Principle #15Dynamics

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

FN synapses achieve near-optimal memory consolidation characteristics, outperforming elastic weight consolidation networks in continual learning tasks with energy efficiency and scalable implementation, providing femtojoules per synaptic update.

Implementation Method 1

Fowler-Nordheim (FN) synapses, which operate using quantum-mechanical tunneling, to store synaptic weights differentially

Methodology Applied

Scientific EffectFowler-Nordheim tunneling:

Data Source

PatentUS20250371330A1Fowler-nordheim devices and methods and systems for continual learning and memory consolidation using fowler-nordheim devices

Publication Date: 2025.12.04 WASHINGTON UNIV IN SAINT LOUIS

AI summary

A synaptic array includes a plurality of Fowler-Nordheim (FN) synapses. Each FN synapse connected to at least one other FN synapse of the plurality of FN synapses to form a network. Each FN synapse includes a pair of FN tunneling devices each including a floating gate. Each FN synapse is operable to store a synaptic weight as a differential voltage across the floating gates of its FN tunneling devices and to implement synaptic memory consolidation.