Neuromorphic Crossbar Circuit for Negative Weights and Backpropagation

Solution Overview

Problem

Existing deep learning techniques implemented by supercomputers face high power consumption and large area requirements, necessitating a more efficient neuromorphic hardware structure that emulates neural networks to reduce hardware size and power consumption.

Innovation Solution

A neuromorphic circuit capable of controlling current direction through synapse modules and setting negative weights by controlling voltage, utilizing an input module, output module, synapse modules, and a crossing module to facilitate forward and back propagation operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If digital programming techniques using supercomputers are used for deep learning, then computing power and accuracy are improved, but power consumption and hardware area increase significantly

Engineering Contradiction:

Improvedeep learning accuracyVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by stationary object

Solution Approach 1:

The patent replaces digital computing systems with a neuromorphic analog system that emulates biological neural networks. The crossbar-array structure performs computations through physical current flows and voltage potentials, substituting the mechanical/digital processing with a biologically-inspired physical system that achieves similar computational accuracy with dramatically reduced power consumption and hardware footprint.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental operating parameters from digital binary states to continuous analog voltages and currents. By representing neural activations and weights as continuous physical quantities in the crossbar array, the system achieves high computational precision for deep learning while operating at much lower power consumption levels compared to digital supercomputers.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If digital programming techniques using supercomputers are used for deep learning, then computing power and accuracy are improved, but hardware area increases significantly

Engineering Contradiction:

Improvedeep learning accuracyVSAvoidhardware area

Core Design Contradiction:

Measurement precisionVSArea of stationary object

Solution Approach 1:

The patent replaces digital computing systems with a neuromorphic analog system that emulates biological neural networks. The crossbar-array structure performs computations through physical current flows and voltage potentials, substituting the mechanical/digital processing with a biologically-inspired physical system that achieves similar computational accuracy with dramatically reduced power consumption and hardware footprint.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent merges multiple functions into the crossbar-array structure itself. The same physical array performs both weight storage and computation operations simultaneously, eliminating the need for separate memory and processing units required in digital systems. This functional integration dramatically reduces the overall hardware area needed for deep learning implementations.

Inventive Principle:

Principle #5Merging (Combining)

3Use of energy by stationary object

If neuromorphic system operations are implemented in analog form using crossbar-array structure, then hardware size and power consumption are reduced, but implementation complexity increases

Engineering Contradiction:

Improvepower consumptionVSAvoidimplementation complexity

Core Design Contradiction:

Use of energy by stationary objectVSDevice complexity

Solution Approach 1:

The patent segments the neuromorphic system into distinct functional modules: input modules, output modules, synapse modules, and crossing modules. Each module performs a specific function in the forward propagation and back propagation processes. This segmentation simplifies the implementation by allowing each module to be designed and optimized independently while working together to achieve the overall neuromorphic computing function.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic control of current directions through the crossing modules, enabling the system to switch between forward propagation and back propagation modes. The crossing modules dynamically adjust which synapse modules receive input currents and which generate output currents, providing flexible operational control that manages implementation complexity through systematic state management.

Inventive Principle:

Principle #15Dynamics

4Ease of manufacture

If crossbar-array structure is used for neuromorphic system, then hardware integration is improved, but control of current direction and weight settings becomes more complex

Engineering Contradiction:

Improvehardware integrationVSAvoidcontrol complexity

Core Design Contradiction:

Ease of manufactureVSDevice complexity

Solution Approach 1:

The patent segments the neuromorphic system into distinct functional modules: input modules, output modules, synapse modules, and crossing modules. Each module performs a specific function in the forward propagation and back propagation processes. This segmentation simplifies the implementation by allowing each module to be designed and optimized independently while working together to achieve the overall neuromorphic computing function.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The crossing modules serve as intermediary control elements between input modules and synapse modules, and between output modules and synapse modules. These intermediary components systematically manage current direction control and weight setting operations, breaking down the complex control task into manageable sub-tasks that are easier to implement and control.

Inventive Principle:

Principle #24Intermediary (Mediator)

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 neuromorphic circuit enables efficient learning and inference processes with reduced hardware size and power consumption by accurately controlling current flow and weight settings, allowing for faster and more accurate training and identification of malfunctioning synapse modules.

Implementation Method 1

synapse modules configured to electrically connect the input module and the output module and determine a current to be transmitted to the output module

Methodology Applied

Scientific EffectElectrical Conduction: Conduction (electrical)

Implementation Method 2

a crossing module configured to control a direction of current flowing through the synapse modules

Methodology Applied

Scientific EffectElectrical Conduction: Conduction (electrical)

Data Source

PatentUS12493780B2Neuromorphic circuit capable of setting negative weights and back propagation

Publication Date: 2025.12.09 KOREA UNIV RES & BUSINESS FOUND

AI summary

Provided is a neuromorphic circuit including an input module configured to generate an input voltage, an output module configured to measure a current transmitted from the input module and generate an output voltage, synapse modules configured to electrically connect the input module and the output module and determine a current to be transmitted to the output module and including a memory element to which a weight is assigned, and a crossing module configured to control a direction of current flowing through the synapse modules.