Information Technology
SuperTag Cache for Energy-Optimized Compression
WARF: P130076US01
Inventors: David Wood, Somayeh Sardashti
The Wisconsin Alumni Research Foundation (WARF) is seeking commercial partners interested in developing methods that improve compression effectiveness and reduce system energy by exploiting spatial locality of cache blocks.
Overview
Cache memories play a critical role in reducing a computer system’s energy. Generally, they are fast access memories that store data reflecting locations in a corresponding main memory. Typically, the data stored in caches are organized into ‘blocks.’ Each block has a corresponding set of tags, which also are stored in cache. Such tags can include address tags identifying the area of main memory that maps to the corresponding block.
Although caches consume lots of power, they also can save system power by reducing costly off-chip accesses to main memory. Clearly, cutting energy usage requires making caches as effective as possible. Cache compression is one technique for increasing cache capacity and reducing misses. However, previous designs have focused on performance rather than energy efficiency.
A new compression approach could be more effective by exploiting the spatial locality of data blocks. This would recognize that several contiguous blocks often coexist in memory, having a similar compression ratio that typically is higher in larger block sizes.
Although caches consume lots of power, they also can save system power by reducing costly off-chip accesses to main memory. Clearly, cutting energy usage requires making caches as effective as possible. Cache compression is one technique for increasing cache capacity and reducing misses. However, previous designs have focused on performance rather than energy efficiency.
A new compression approach could be more effective by exploiting the spatial locality of data blocks. This would recognize that several contiguous blocks often coexist in memory, having a similar compression ratio that typically is higher in larger block sizes.
The Invention
UW–Madison researchers have developed a compressed cache, called SuperTag, which exploits spatial locality to optimize compression effectiveness and energy use.
SuperTag cache manages cache at three granularities: ‘super blocks,’ single blocks and fractional data segments. Since contiguous blocks have the same tag address, SuperTag increases per-block tag space by tracking super blocks (for example, a group of four aligned contiguous blocks of 64 bytes each). It also breaks each cache block into smaller data segments for storage.
To improve compression ratio, the technique uses a variable-packing scheme allowing variable-size compression blocks without costly compaction. It also co-compresses contiguous blocks, including within the same super block, thereby producing data segments for storage.
SuperTag cache manages cache at three granularities: ‘super blocks,’ single blocks and fractional data segments. Since contiguous blocks have the same tag address, SuperTag increases per-block tag space by tracking super blocks (for example, a group of four aligned contiguous blocks of 64 bytes each). It also breaks each cache block into smaller data segments for storage.
To improve compression ratio, the technique uses a variable-packing scheme allowing variable-size compression blocks without costly compaction. It also co-compresses contiguous blocks, including within the same super block, thereby producing data segments for storage.
Applications
- Memory system development
Key Benefits
- Improves energy and performance for memory-intensive applications
- Improves compression ratio
- Higher compressibility without high area overheads
Additional Information
Related Technologies
Tech Fields
For current licensing status, please contact Emily Bauer at [javascript protected email address] or 608-960-9842