Improving direct-mapped cache performance by the addition of a small fully-associative cache and prefetch buffers

Projections of computer technology forecast processors with peak performance of 1,000 MIPS in the relatively near future. These processors could easily lose half or more of their performance in the memory hierarchy if the hierarchy design is based on conventional caching techniques. This paper presents hardware techniques to improve the performance of caches.