Compressing for Compression With the XOR Cache
IEEE Micro · 2026-09 · 46 (5) · 12-19
DOI: 10.1109/mm.2026.3701038 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Micro, 2026-09. Verfasst von Zhewen Pan und 1 weiteren Autor:innen. Die Arbeit nennt 12 Literaturangaben.
Abstract (Originalsprache)
Modern processors dedicate a large share of their resources to caching, especially to the last-level cache (LLC). We observe that today’s memory hierarchies contain untapped compression potential in the redundancy created by private caching and inclusion. To exploit it, we introduce XOR Cache, a new LLC architecture based on XOR compression. Instead of storing cache lines as-is, XOR Cache stores the bitwise XOR of line pairs, halving the number of stored lines through a form of inter-line compression. When paired with existing compression schemes, XOR Cache also boosts their intra-line compression ratios: XORing similar lines lowers the entropy of the data before it is compressed. Our evaluation shows that XOR Cache reduces LLC area by 1.93× and power by 1.92×, with only a 2.06% performance overhead compared to a larger uncompressed cache.
Zitieren
APA
Pan, Z., & Miguel, J. S. (2026). Compressing for Compression With the XOR Cache. IEEE Micro, 46(5), 12-19. https://doi.org/10.1109/mm.2026.3701038
Vancouver
Pan Z, Miguel JS. Compressing for Compression With the XOR Cache. IEEE Micro. 2026;46(5):12-19. doi:10.1109/mm.2026.3701038
BibTeX
@article{Pan2026,
author = {Pan, Zhewen and Miguel, Joshua San},
title = {Compressing for Compression With the XOR Cache},
journal = {IEEE Micro},
year = {2026},
volume = {46},
number = {5},
pages = {12--19},
doi = {10.1109/mm.2026.3701038}
}
EUVIMED