OPTIMIZING ENERGY-EFFICIENT COMPUTER ARCHITECTURE: A MEMORY-CENTRIC APPROACH WITH PROCESSING-IN-MEMORY (PIM) AND NON-VOLATILE MEMORY (NVM)

Abstract

Modern computing architectures move towards memory-centric systems because processor speed now surpasses memory performance thus requiring optimized power efficiency and computational throughput. The research investigates how PIM combined with NMP and IMP alongside NVM systems helps overcome memory wall challenges. The evaluations show NVM-based structures deliver a 30% energy efficiency increase together with 25% increased computations and 40% reduced latency than conventional compute-centric structures demonstrate. The combination of PIM and IMP systems decreases energy usage by 20%–25% yet NMP methods shorten execution duration by 15%. Memory-centric design architectures reduce latency when processing AI workloads which leads to performance enhancements of 2x above CPU-GPU implementations thus making these designs well-suited for deep learning applications. The challenges of NVM durability and hardware-software conglomerates need more research about adaptive memory management systems for their effective implementation. The examined hybrid memory design proves adaptable to multiple computing settings which establishes advanced energy-efficient computing solutions for AI technology along with IoT and edge computing applications.