Year: 2025
Type: article
Abstract: In high-performance computing (HPC) architectures, optimizing memory hierarchy is crucial for enhancing system performance and efficiency. The memory hierarchy consists of various levels of storage, each with distinct characteristics in terms of speed, cost, and capacity. As the gap between processor speeds and memory access times widens, effective memory management becomes essential to minimize latency and maximize throughput. This paper explores several strategies for optimizing memory hierarchy, including dynamic reconfiguration of cache systems, integration of emerging memory technologies, and the implementation of behavior-aware cache hierarchies. Dynamic memory management techniques enable the adaptive configuration of cache and translation lookaside buffer (TLB) sizes based on workload demands, significantly improving performance by reducing miss penalties. Emerging memory technologies such as ReRAM, PCM, and MRAM offer non-volatile options that can bridge the speed and capacity gaps inherent in traditional DRAM and NAND flash systems. Additionally, behavior-aware cache hierarchies allow for optimal allocation of multi-level cache resources tailored to application-specific access patterns, resulting in reduced energy consumption and enhanced data throughput. This comprehensive review highlights the importance of memory hierarchy optimization in HPC environments and presents a framework for future research aimed at developing more efficient memory architectures that can support increasingly complex computational tasks
Author: Muthukumaran Vaithianathan
Institution: Samsung Semiconductor (United States)
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4411575770