Block-Based Data Compression for Off-Chip Memory Bandwidth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hardware accelerators and convolutional neural networks (CNNs) on processor chips face memory capacity limitations, leading to bandwidth bottlenecks when transferring data from off-chip memory due to incompatible compression techniques with their operating frequencies.
Innovation Solution
An off-chip data compression technique that splits data into blocks and applies lossless compression methods like run-length encoding (RLE) or partial Huffman encoding based on block characteristics, generating headers for decoding, and enabling parallel processing to optimize bandwidth and reduce memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data is compressed using traditional lossless compression techniques, then data size is reduced, but compatibility with hardware accelerator operating frequencies is lost
Solution Approach 1:
The patent divides data into fixed-size blocks that can be independently compressed and processed. Each block is handled separately through the memory interface, allowing parallel processing that matches the hardware accelerator's operating frequency requirements while achieving overall data compression through selective application of compression algorithms to individual blocks.
2Speed
If data is transferred from off-chip memory without compression, then processing speed is maintained, but bandwidth bottlenecks occur
Solution Approach 1:
The patent applies compression algorithms to data blocks before they are transferred from off-chip memory through the memory interface to the hardware accelerator. This preliminary compression reduces the total volume of data that needs to be transferred, eliminating bandwidth bottlenecks while the hardware accelerator processes the compressed data at its native operating frequency without speed degradation.
3Quantity of substance
If compression algorithms are applied to all data blocks, then data size is reduced, but processing complexity increases
Solution Approach 1:
The patent evaluates each data block individually and applies compression algorithms selectively based on the specific characteristics of that block. Some blocks may be compressed while others are left uncompressed, optimizing the balance between data size reduction and processing complexity by adapting the compression approach to local data properties rather than applying a uniform compression strategy to all data.
Data Source
AI summary
Methods, apparatus, systems and articles of manufacture to compress data are disclosed. An example apparatus includes an off-chip memory to store data; a data slicer to split a dataset into a plurality of blocks of data; a data processor to select a first compression technique for a first block of the plurality of blocks of data based on first characteristics of the first block; and select a second compression technique for a second block of the plurality of blocks of data based on second characteristics of the second block; a first compressor to compress the first block using the first compression technique to generate a first compressed block of data; a second compressor to compress the second block using the second compression technique to generate a second compressed block of data; a header generator to generate a first header identifying the first compression technique and a second header identifying the second compression technique; and an interface to transmit the first compressed block of data with the first header and the second compressed block of data with the second header to be stored in the off chip memory.