Memristor-Based Bidirectional Data Processing for Edge AI

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional processing devices cannot effectively deploy large-scale neural network algorithms in scenarios with limited volume and power resources, such as mobile or edge devices, due to high computing power and power consumption requirements.

Innovation Solution

A data processing apparatus with a bidirectional data processing module that integrates storage and computing using a memristor array, allowing for both inference and training tasks, and includes a controlling module to switch between modes, a parameter management module for weight parameter setting, and input/output modules for data processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If traditional processing devices are used for neural network algorithms, then computing power requirements can be met, but power consumption increases and computing efficiency decreases

Engineering Contradiction:

Improvecomputing powerVSAvoidpower consumption

Core Design Contradiction:

PowerVSUse of energy by moving object

Solution Approach 1:

The patent merges storage and computing functions into a single processing device. The processing device includes both a processing element with storage unit and a computing element, allowing data to be stored and processed within the same device without frequent data transfer between separate storage and computing components, thereby reducing power consumption while maintaining computing power

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The processing device is segmented into distinct functional units: a processing element with its own storage unit, a computing element, and a control element. This segmentation allows each unit to operate efficiently in its specialized function, improving overall computing efficiency while reducing the power consumption of data transfer between components

Inventive Principle:

Principle #1Segmentation

2Productivity

If traditional processing devices are used for neural network algorithms, then computing tasks can be performed, but computing efficiency is insufficient

Engineering Contradiction:

Improvecomputing efficiencyVSAvoiddevice structure

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The processing device employs dynamic reconfiguration capabilities where the control element can dynamically allocate and reconfigure computing resources based on the specific neural network algorithm being executed. This dynamic adaptation improves computing efficiency for different tasks without requiring a completely different device structure for each algorithm

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The processing device is designed with universal functionality to handle various neural network algorithms through a standardized architecture. The processing element, computing element, and storage unit can work together for different computational tasks, improving computing efficiency across multiple applications without increasing device complexity

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Reliability

If large-scale neural network algorithms are deployed, then algorithm performance is improved, but the device volume and power resources are exceeded

Engineering Contradiction:

Improvealgorithm performanceVSAvoiddevice volume

Core Design Contradiction:

ReliabilityVSVolume of stationary object

Solution Approach 1:

The processing device uses a nested structure where the storage unit is integrated within the processing element, which in turn is part of the larger processing device. This nested architecture allows efficient utilization of limited device volume by hierarchically organizing storage and computing resources, enabling deployment of large-scale neural network algorithms in resource-constrained environments while maintaining algorithm performance

Inventive Principle:

Principle #7Nested doll (Nesting)

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 efficient and low-power computing for neural network algorithms, supporting both inference and training tasks in resource-constrained environments with high computing efficiency and flexibility.

Implementation Method 1

write the weight parameter to the memristor array by changing a conductance value of each of the plurality of memristors

Methodology Applied

Scientific EffectConductance change: Electrical Resistance

Data Source

PatentUS20250005353A1Data processing apparatus and data processing method

Publication Date: 2025.01.02 TSINGHUA UNIVERSITY

AI summary

A data processing apparatus and a data processing method. The data processing apparatus includes: a bidirectional data processing module, including at least one storage and computing integration computing array; a controlling module, configured to switch a working mode of the bidirectional data processing module to an inference working mode to perform an inference computing task, and to switch the working mode of the bidirectional data processing module to a training working mode to perform a training computing task; a parameter management module, configured to set a weight parameter of the bidirectional data processing module; and an inputting and outputting module, configured to generate a computing inputting signal according to inputting data of the computing task, provide the computing inputting signal to the bidirectional data processing module, and receive a computing outputting signal from the bidirectional data processing module and generate outputting data according to the computing outputting signal.