A high level implementation and performance evaluation of level-I asynchronous cache on FPGA

To bridge the ever-increasing performance gap between the processor and the main memory in a cost-effective manner, novel cache designs and implementations are indispensable. Cache is responsible for a major part of energy consumption (approx. 50%) of processors. This paper presents a high level implementation of a micropipelined asynchronous architecture of L1 cache. Due to the fact that each cache memory implementation is time consuming and error-prone process, a synthesizable and a configurable model proves out to be of immense help as it aids in generating a range of caches in a reproducible and quick fashion. The micropipelined cache, implemented using C-Elements acts as a distributed message-passing system. The RTL cache model implemented in this paper, comprising of data and instruction caches has a wide array of configurable parameters. In addition to timing robustness our implementation has high average cache throughput and low latency. The implemented architecture comprises of two direct-mapped, write-through caches for data and instruction. The architecture is implemented in a Field Programmable Gate Array (FPGA) chip using Very High Speed Integrated Circuit Hardware Description Language (VHSIC HDL) along with advanced synthesis and place-and-route tools.