Performance recovery in direct-mapped faulty caches via the use of a very small fully associative spare cache

Single chip VLSI processors use on-chip cache memories to satisfy the memory bandwidth demands of CPU. By tolerating cache defects without a noticeable performance degradation, the yield of VLSI processors can be enhanced considerably. In this paper we investigate how much of the lost hit ratio due to faulty block disabling in direct-mapped caches can be recovered by the incorporation of a very small fairly associative spare cache. The recovery percentage that can be achieved as a function of the primary cache's parameters (cache size, block size), the number of faulty blocks and the size of the spare cache is derived by trace driven simulation. The results show that when the number of the faulty blocks is small the use of a spare cache with only one block offers a hit ratio recovery of more than 70%, which increases further with cache size. A spare cache with two blocks is justified only in the case of a large number of faulty blocks.<
>