L2 Cache


The L2 cache is a 256KB, 4-bank, 4-way set associative shared L2 cache. The latency of L2 cache is 4 cycles at hit. The L2 cache RAM reading and writing processes are pipelined into 4 stages for less RAM access and higher frequency. The L2 cache is designed as a non-blocking cache which can handle hit-under-miss and miss-under-miss using the Missing Status Holding Registers (MSHRs). With non-blocking L2 cache design, memory system can execute out-of-order and more latency can be hidden.


Fig.1 L2 cache bank connection


The parameter of L1 data cache is as follows:

Cache capacity

Cache line numbers

Cache line capacity

Mapping method

256 KBytes


32 Bytes

4-way set associative

L2 cache pipeline

The L2 is designed as 4-stage-pipeline for low power and high frequency. In the first 3 stages, valid, tag, lru, dirty and data RAMs are serially checked, which means some of the RAMs are not needed to be accessed if the information got from previous stages tells the control logic not to.

The Missing Status Holding Registers lie in the stage 4, which has the ability to hold multiple cache missed request to the next level memory, without blocking the whole pipeline. This is a key feature for Out-of-Order memory system.


Fig.2 L2 cache pipeline overview