Abstract

Write-through caches potentially have higher reliability than write-back caches. However, write-back caches are more energy efficient. This paper provides a comparison between the write-back and write-through policies based on the combination of reliability and energy consumption criteria. In the experiments, SIMPLESCALAR tool and CACTI model are used to evaluate the characteristics of the caches. The results show that a write-through cache with one parity bit per word is as reliable as a write-back cache with SEC-DED code per word. Furthermore, the results show that the energy saving of the write-through cache over the write-back cache increases if any of the following changes happens: i) a decrease in the feature size, ii) a decrease in the L2 cache size, and iii) an increase in the L1 cache size. The results also show that when feature size is bigger than 32nm, the write-back cache is usually more energy efficient. However, for 32nm and smaller feature sizes the write-through cache can be more energy efficient.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call