Architecture and design of a pseudo two-port VLSI snoopy cache memory

A CMOS VLSI cache memory subsystem that includes a 72K-bit cache memory, an 11K-bit tag memory, a 1.3K-bit state array, two special buffers and cache control logic, has been designed and integrated on a microprocessor chip. The architecture, design and analysis of the cache design are presented. The design achieves higher system performance by reducing the cache reload penalty through a pseudo-two-port architecture which utilizes a reload buffer and a store-back buffer. It also maintains cache data coherency and supports multiprocessing by bus snooping. A single-port tag is used for concurrent snooping and CPU access with an enhanced write-once protocol. A cost-effective 'locked replacement' scheme was incorporated to maintain data coherency in these two special buffers. Cache modeling and analysis were carried out to derive the proper design point.