Processing-in-Memory Decompression for Bandwidth Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory systems face inefficiencies in processing performance and power consumption due to the need to transfer data from memory arrays to processing resources, which can be external and not aligned with memory cell pitch, leading to increased memory bandwidth requirements and costs.
Innovation Solution
The implementation of processing-in-memory (PIM) devices that form sensing circuitry and compute components on the same pitch as memory cells, allowing for on-pitch data processing and reducing the need for external data transfer by performing operations directly on the memory array, such as decompression and logical operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is transferred from memory arrays to external processing resources, then processing operations can be performed, but memory bandwidth requirements increase and power consumption increases
Solution Approach 1:
The patent combines processing resources directly with the memory array by forming sensing circuitry and compute components on the same pitch as memory cells. This integration eliminates the need for external data transfer, allowing processing operations to be performed directly on data stored in the memory array, thereby reducing power consumption while maintaining processing performance.
Solution Approach 2:
The patent introduces an intermediary processing layer between the memory array and external processing resources. This intermediary layer performs processing operations locally within the memory device, reducing the amount of data that needs to be transferred externally and thereby reducing power consumption and memory bandwidth requirements.
2Productivity
If data is transferred from memory arrays to external processing resources, then processing operations can be performed, but memory bandwidth requirements increase
Solution Approach 1:
The patent merges processing resources with the memory array structure, forming sensing circuitry and compute components on the same pitch as memory cells. This integration enables processing operations to be performed directly on data within the memory array, eliminating the need for external data transfer and thereby reducing memory bandwidth requirements while maintaining processing performance.
3Productivity
If processing resources are implemented externally to memory arrays, then processing operations can be performed, but time is consumed for data transfer
Solution Approach 1:
The patent combines processing resources with the memory array by forming sensing circuitry and compute components on the same pitch as memory cells. This integration eliminates external data transfer steps, allowing processing operations to be performed directly on data stored in the memory array, thereby reducing processing time while maintaining processing performance.
4Loss of energy
If sensing circuitry and compute components are formed on the same pitch as memory cells, then data transfer is reduced, but device complexity increases
Solution Approach 1:
The patent segments the memory array into multiple banks, with each bank containing sensing circuitry and compute components formed on the same pitch as memory cells. This segmentation allows processing operations to be performed locally within each bank, reducing power consumption while managing device complexity through modular organization.
Solution Approach 2:
The patent implements local processing capabilities by forming sensing circuitry and compute components on the same pitch as memory cells within specific banks. This local quality approach enables processing operations to be performed close to where data is stored, reducing power consumption for data transfer while maintaining manageable device complexity through localized functionality.
Data Source
AI summary
The present disclosure includes apparatuses and methods for operations using compressed and decompressed data. An example method includes receiving compressed data to a processing in memory (PIM) device and decompressing the compressed data on the PIM device.