Low Voltage SRAM Cache Compression and ECC Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low voltage operation of SRAM-based caches leads to persistent bit errors, increasing power consumption and reducing reliability, as the number of faults grows exponentially with decreasing voltage and operating frequency.
Innovation Solution
Implementing data compression and an error correction code cache to store error-recovery metadata, allowing the cache to operate at lower voltages while maintaining reliability by prioritizing the placement of compressed data in fault-free subportions of cache lines and dynamically managing error protection resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If supply voltage is lowered to reduce power consumption, then power consumption decreases, but bit error rate increases exponentially
Solution Approach 1:
The cache is divided into multiple subportions (e.g., half-cache or quarter-cache) within each cache line. Error correction code (ECC) metadata is stored in dedicated ECC subportions, while data is stored in data subportions. This segmentation allows the system to operate at low voltage with ECC protection for only the critical metadata portions, rather than protecting entire cache lines, thus reducing power consumption while maintaining reliability.
Solution Approach 2:
Different parts of the cache structure receive different levels of error protection. ECC metadata in ECC subportions is protected with full ECC coverage, while data in data subportions uses compressed or reduced ECC coverage. This local differentiation optimizes power consumption by applying error protection only where most critical, enabling low-voltage operation without sacrificing overall reliability.
2Reliability
If error correction code cache is used to protect against low voltage faults, then reliability improves, but device complexity increases
Solution Approach 1:
Each cache line is segmented into ECC subportions and data subportions. ECC subportions store compressed ECC metadata and are fully protected, while data subportions store application data with reduced protection. This segmentation reduces the total amount of ECC metadata needed compared to protecting entire cache lines, thereby reducing complexity while maintaining reliability.
Solution Approach 2:
The ECC coverage parameter is dynamically adjusted based on the subportion type. ECC subportions use full ECC coverage for maximum reliability, while data subportions use compressed or reduced ECC coverage. This parameter change allows the system to achieve acceptable reliability with less ECC overhead, reducing device complexity.
3Productivity
If compressed data is placed in faulted subportions of cache lines, then cache utilization improves, but performance loss increases due to decompression overhead
Solution Approach 1:
Cache lines are segmented into faulted subportions and non-faulted subportions. Compressed data is placed only in non-faulted subportions to avoid decompression of corrupted data. This selective placement improves cache utilization by keeping faulted lines enabled while minimizing decompression overhead by only decompressing data from reliable subportions.
Solution Approach 2:
The patent extracts and separates ECC metadata from data, storing them in different subportions. Compressed data is extracted and placed only in non-faulted subportions, leaving faulted subportions for ECC metadata or unused. This extraction improves cache utilization by enabling faulted lines while reducing unnecessary decompression operations.
Data Source
AI summary
A method of operating a cache in a computing device includes, in response to receiving a memory access request at the cache, determining compressibility of data specified by the request, selecting in the cache a destination portion for storing the data based on the compressibility of the data and a persistent fault history of the destination portion, and storing a compressed copy of the data in a non-faulted subportion of the destination portion, wherein the persistent fault history indicates that the non-faulted subportion excludes any persistent faults.