Cross-Bar ReRAM On-Chip Processing for Low-Power Image Compression

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image compression methods are inefficient in terms of power consumption and circuit space, particularly in mobile devices, and do not effectively utilize cross-bar non-volatile memory devices for super-sparse image compression.

Innovation Solution

A cross-bar non-volatile memory (NVM) array is trained using an unstructured Hebbian training procedure to generate and store a dictionary of image elements, allowing for super-sparse image compression by replacing image patches with corresponding dictionary elements, which are identified through cross-bar column indices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If conventional image compression methods are used, then image compression is achieved, but power consumption and circuit space are excessive

Engineering Contradiction:

Improvepower consumptionVSAvoidimage compression efficiency

Core Design Contradiction:

Loss of energyVSProductivity

Solution Approach 1:

The image is divided into multiple patches that are processed independently through the cross-bar NVM array. Each patch can be compressed separately, allowing parallel processing and reducing the computational burden on any single processing element, thereby lowering overall power consumption while maintaining compression efficiency

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent replaces conventional software-based image compression algorithms with a hardware-based cross-bar NVM array system. This substitution leverages the inherent parallelism and non-volatile storage capabilities of the NVM array to perform compression operations directly in hardware, significantly reducing power consumption and circuit space requirements compared to traditional approaches

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Area of stationary object

If conventional image compression methods are used, then image compression is achieved, but circuit space is excessive

Engineering Contradiction:

Improvecircuit spaceVSAvoidimage compression efficiency

Core Design Contradiction:

Area of stationary objectVSProductivity

Solution Approach 1:

The cross-bar NVM array serves multiple functions simultaneously: it stores the dictionary of image elements, performs the compression operation through parallel processing of image patches, and maintains the compressed data. This multi-functionality eliminates the need for separate storage and processing circuits, dramatically reducing overall circuit space while preserving compression efficiency

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent merges the dictionary storage function and the compression processing function into a single cross-bar NVM array structure. By combining these previously separate functions into one integrated system, the circuit space required for image compression is significantly reduced while maintaining or improving compression performance

Inventive Principle:

Principle #5Merging (Combining)

3Loss of energy

If cross-bar NVM array is used for super-sparse image compression, then power consumption is minimized, but implementation complexity increases

Engineering Contradiction:

Improvepower consumptionVSAvoidimplementation complexity

Core Design Contradiction:

Loss of energyVSDevice complexity

Solution Approach 1:

The cross-bar NVM array performs self-configuration through training procedures where it automatically learns and stores the optimal dictionary of image elements specific to the application domain. This self-service capability eliminates the need for external programming or manual configuration, reducing implementation complexity despite the advanced hardware architecture

Inventive Principle:

Principle #25Self-service

4Loss of information

If cross-bar NVM array is used for super-sparse image compression, then compression ratio is enhanced, but training time is extended

Engineering Contradiction:

Improvecompression ratioVSAvoidtraining time

Core Design Contradiction:

Loss of informationVSLoss of time

Solution Approach 1:

The dictionary training is performed in advance during a preliminary phase, separating the time-consuming training operation from the actual compression operations. Once trained, the NVM array can perform rapid compression without requiring repeated training, thus achieving high compression ratios with minimal time loss during actual use

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12394195B2Non-volatile cross-bar memory with on-chip processing circuitry

Publication Date: 2025.08.19 WESTERN DIGITAL TECHNOLOGIES INC

AI summary

Exemplary methods and apparatus are disclosed that implement super-sparse image/video compression by storing image dictionary elements within a cross-bar resistive random access memory (ReRAM) array (or other suitable cross-bar NVM array). In illustrative examples, each column of the cross-bar ReRAM array stores the values for one dictionary element (such as one 4×4 dictionary element). Methods and apparatus are described for training (configuring) the cross-bar ReRAM array to generate and store the dictionary elements by sequentially applying patches from training images to the array using an unstructured Hebbian training procedure. Additionally, methods and apparatus are described for compressing an input image by applying patches from the input image to the ReRAM array to read out cross-bar column indices identifying the columns storing the various dictionary elements that best fit the image. This may be done in parallel using a set of ReRAM arrays.