Cache Line Encoding with Transformer Ensembles for Zero-Value Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data compression techniques for cache memories do not optimize the removal of zero-value bytes efficiently, leading to suboptimal storage capacity utilization in cache lines.
Innovation Solution
An ensemble of data transformers applies different transform operations to cache line chunks, selecting the transformed chunk with the minimum number of non-zero bits, followed by zero-value compression to maximize zero-bit removal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional zero-value compression is used without data transformation, then the compression process is simple and fast, but the compression ratio is suboptimal because zero-values are not maximized
Solution Approach 1:
The patent applies data transformation operations (such as XOR, rotation, or other encoding schemes) to the input data before performing zero-value compression. This preliminary action increases the number of zero-values in the data, thereby improving the compression ratio. The transformation is performed in advance to prepare the data for more effective compression without requiring complex real-time processing during the compression phase.
Solution Approach 2:
The patent introduces an intermediate transformation step between the input data and the zero-value compression process. This intermediary transformation layer converts the original data into a form that has more zero-values, making it more suitable for compression. The intermediary step acts as a bridge that enhances the effectiveness of the subsequent compression operation without directly modifying the compression algorithm itself.
2Quantity of substance
If data transformation is applied before compression, then the compression ratio improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent divides the data processing into distinct segments: first applying data transformation operations to generate transformed data with more zero-values, then performing zero-value compression on the transformed data. This segmentation allows each step to be optimized independently, with the transformation step preparing the data and the compression step efficiently removing zero-values. The segmented approach enables better compression ratios while managing processing time through modular operation.
Data Source
AI summary
Provided are a compression unit, a cache memory, and method for using an ensemble of data transformers to encode data before zero-value compression. A plurality of data transformers that each perform a transform on an input chunk to produce a transformed chunk having more zeroes than the input chunk. A selection unit selects a transformed chunk having a minimum number of non-zero bits. A zero-value compressor processes the selected transformed chunk to output a compressed chunk having all non-zeroes and fewer bits than the input chunk. The compressed chunk is stored in the cache memory cell array.