Skip to content

Implemented a memory subsystem consisting of an L1D cache and a main memory. Whenever there is a miss in L1D cache, an error tolerance information is sent along with the data request that aims to reduce power consumption. This project is implemented using the HDL language - SystemVerilog.

Notifications You must be signed in to change notification settings

prateek-0328/Memory-Subsystem-with-Approximate-Data-Transfer

Repository files navigation

Memory-Subsystem-with-Approximate-Data-Transfer

Implemented a memory subsystem consisting of an L1D cache and a main memory. Whenever there is a miss in L1D cache, an error tolerance information is sent along with the data request that aims to reduce power consumption. This project is implemented using the HDL language - SystemVerilog.

Project Description :

In this project, we have a 2 KB, 4 way set associative L1D cache. Each cache line is of 32 bytes each and processor word size is of 4 bytes. Note that this size excludes the tag information present in each cache block. Just as in a typical cache, just passing the address should give a word size value. During the evaluation, a 32 bit address will be given and the output should be a word (4 byte) value. Whenever there is a cache miss in L1D cache, the memory request is sent to a 8 KB main memory, with each location of 4 bytes each. Note that in case of cache miss, the cache line is fetched and not just the word missed. As each cache line is of 32 bytes and processor size is of 4 bytes, the transfer happens as a stream of eight 4-byte numbers. Make sure that the first data is multiple of 8, else wrong data may be passed as output. As the main memory may not be available, a simple protocol consisting of four signals- CLK, VALID, READY and DATA. A transfer takes place when both VALID and READY are asserted.

The data transfer in this case happens as a sequence of eight-four byte numbers. This causes a lot of area consumption. In order to reduce the data transfer, we compress the data before sending it across the protocol. In this project, you will implement the AxDeduplication compression scheme discussed in AxBA. Note that you just need to implement the AxDeduplication scheme and not the software aspects of AxBA.

About

Implemented a memory subsystem consisting of an L1D cache and a main memory. Whenever there is a miss in L1D cache, an error tolerance information is sent along with the data request that aims to reduce power consumption. This project is implemented using the HDL language - SystemVerilog.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published