Uploaded September 2025 | Updated November 2026, 16 hours ago
Speaker: Dr. Nezam Rohbani, Phd in Computer Engineering, Barcelona Supercomputing Center (BSC)
In-memory processing has emerged as a promising paradigm to overcome the memory bottleneck in modern computing systems by reducing data movement between memory and processors. DRAM, as the dominant main memory technology, offers unique architectural and physical features that can be exploited to perform computation within the memory itself. In this talk, we explore how specific characteristics of DRAM—such as differential cell access using paired bitlines, the structure and behavior of sense amplifiers, and the inherent dynamic nature of charge leakage in DRAM cells—can be leveraged to enable efficient bitwise operations directly inside DRAM arrays. We highlight the fundamental challenges of implementing logic operations in commodity DRAM devices and present some of our previous works that demonstrate innovative techniques for in-DRAM computation. These approaches show the potential of DRAM not only as a storage medium but also as an active computational substrate, paving the way for more energy-efficient and data-centric computing architectures.
Further information here: bsc.es/research-and-development/research-seminars/sors-leveraging-dram%E2%80%99s-intrinsic-features-memory-processing
Speaker: Dr. Nezam Rohbani, Phd in Computer Engineering, Barcelona Supercomputing Center (BSC)
In-memory processing has emerged as a promising paradigm to overcome the memory bottleneck in modern computing systems by reducing data movement between memory and processors. DRAM, as the dominant main memory technology, offers unique architectural and physical features that can be exploited to perform computation within the memory itself. In this talk, we explore how specific characteristics of DRAM—such as differential cell access using paired bitlines, the structure and behavior of sense amplifiers, and the inherent dynamic nature of charge leakage in DRAM cells—can be leveraged to enable efficient bitwise operations directly inside DRAM arrays. We highlight the fundamental challenges of implementing logic operations in commodity DRAM devices and present some of our previous works that demonstrate innovative techniques for in-DRAM computation. These approaches show the potential of DRAM not only as a storage medium but also as an active computational substrate, paving the way for more energy-efficient and data-centric computing architectures.
Further information here: bsc.es/research-and-development/research-seminars/sors-leveraging-dram%E2%80%99s-intrinsic-features-memory-processing










