Prefetcher-based DRAM Architecture

TL;DR

This paper proposes integrating data prediction techniques into DRAM architectures like TL-DRAM and CROW to prefetch data into faster memory sections, aiming to reduce memory access latency and improve overall system performance.

Contribution

It introduces a novel approach of applying data prediction at the DRAM architecture level, specifically within TL-DRAM and CROW, to enhance data transfer efficiency.

Findings

Potential reduction in memory access latency

Improved data transfer efficiency in DRAM architectures

Enhanced overall system performance

Abstract

Advancement in Processor technology has made it easy to handle data-intensive workloads, but limiting main memory advances has created performance bottlenecks. In DRAM, there have been improvements in DRAM access latency as well as reduction in cost-per-bit with the increase in cell density. But still DRAM data transfer rate lags behind the processing speed of the current generation processors. As Memory advancements based on hardware have been progressing at a slower pace, to cope up with High-end Processors, Architectural level advancements such as Prediction techniques, Replacement policies, etc are the major subject. In the recent field of research, Data prediction is a sought out topic as correct prediction can boost performance by decreasing the amount of excess memory access by predicting data beforehand using data access trends and behaviors. Though prediction techniques have been implemented at most of the Computer Architecture, We propose implementing data prediction in DRAM level architectures like TL-DRAM and CROW. Both of these method distributes the DRAM into different parts which contain a smaller section which is faster and larger section which contains the bulk of data but is comparatively slower. We wish to use data prediction in between these sections of memory to have predicted data transferred to the faster sections to improve the overall performance by reducing the memory access time.