FLASH In-Memory Computing for Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image compression methods, such as JPEG and JPEG2000, face issues like increased quantization step size, decreased bit per pixel, blocking effects, and noise in decoded images when increasing compression ratio, which are not effectively addressed in existing semiconductor and integrated circuit technologies.
Innovation Solution
A system and method utilizing a FLASH in-memory computing array that includes convolutional neural networks for encoding and decoding, along with a quantization module, to process images efficiently by performing matrix-vector multiplication operations in parallel, reducing energy and hardware resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional image compression methods (JPEG, JPEG2000) increase compression ratio, then storage efficiency improves, but quantization step size increases causing blocking effects and noise in decoded images
Solution Approach 1:
The patent replaces traditional mechanical/image-processing compression methods with in-memory computing based on FLASH memory devices. The FLASH memory cells perform parallel computing operations to execute image compression algorithms, substituting conventional sequential processing with parallel in-memory computation, thereby achieving high compression ratios while preserving image quality through efficient neural network operations
Solution Approach 2:
The patent utilizes the physical parameters of FLASH memory devices, specifically the threshold voltage characteristics of memory cells, to implement computing operations. By programming different threshold voltage levels in FLASH cells, the system performs analog computing for image compression, enabling continuous parameter adjustment that maintains image quality across varying compression ratios
2Quantity of substance
If traditional image compression methods increase compression ratio, then bit per pixel decreases, but this results in loss of image detail and increased noise
Solution Approach 1:
The patent substitutes traditional sequential bit-processing methods with parallel in-memory computing using FLASH devices. The system performs matrix multiplication and neural network operations directly in memory, enabling efficient processing that preserves image details even at low bit per pixel rates through parallel computation of multiple pixel values simultaneously
3Productivity
If conventional image compression is implemented, then encoding and decoding operations consume significant time and computational resources, but speed improvement is limited
Solution Approach 1:
The patent merges the storage function of FLASH memory with the computing function by performing image compression operations directly within the memory array. This combination of storage and computation in the same physical substrate eliminates data transfer between separate memory and processing units, significantly improving compression speed while reducing overall energy consumption
Solution Approach 2:
The patent replaces conventional CPU/GPU-based sequential processing with parallel in-memory computing using FLASH devices. The system utilizes the inherent parallelism of the memory array structure to perform multiple compression operations simultaneously, achieving substantial speedup and reduced energy consumption compared to traditional processing architectures
Data Source
AI summary
A system and a method for compressing an image based on a FLASH in-memory computing array are provided. The system includes: a convolutional neural network for encoding of the FLASH in-memory computing array, a convolutional neural network for decoding based on the FLASH in-memory computing array, and a quantization module; the convolutional neural network for encoding based on the FLASH in-memory computing array is configured to encode an original image to obtain a feature image; the quantization module is configured to quantize the feature image to obtain a quantized image; the convolutional neural network for decoding based on the FLASH in-memory computing array is configured to decode the quantized image to obtain a compressed image.