Portal del coneixement obert de la UPC

Enviaments recents

• Carregant...

  Text en actes de congrés Accés obert

  SORS: New DRAM architectures and FPGA-based frameworks to enable fundamentally better computing systems

  (Barcelona Supercomputing Center, 2025-07-31) Olgun, Ataberk

  Més...

  DRAM-based main memory is a critical component in modern computing systems. Data movement from the DRAM to the CPU incurs long latency and consumes a significant amount of energy. These costs are often exacerbated by the fact that much of the data brought into the caches is not reused by the CPU or accelerators, providing little benefit in return for the high latency and energy cost, partly due to the coarsegrained nature of DRAM data transfers and DRAM row activation. In the first part of this talk, we will describe a new, high-throughput, energy-efficient, and low-cost DRAM architecture, Sectored DRAM, that mitigates the excessive energy consumption of coarse-grained DRAM data transfers and row activation. Compared to a system with coarse-grained DRAM, Sectored DRAM reduces the DRAM energy consumption while improving performance, enabling significant system-wide energy savings. In the second part of the talk, we will introduce two new FPGA-based frameworks, PiDRAM and EasyDRAM. They can be used to develop real system prototypes with commodity DRAM chips, and showcase real, accurate system performance benefits for ideas that aim to tackle the data movement bottleneck. For example, Processing-in/Near- Memory (PiM/PnM), retention-aware intelligent refresh, and DRAM access latency reduction. Using EasyDRAM, we show that in-DRAM bulk data copy operations in real DRAM chips can improve system performance for copy-heavy workloads by 15X. We also describe the MetaSys framework that enables rapid implementation and evaluation of diverse hardware-software cooperative techniques in real hardware. Finally, we will describe the DRAM Bender infrastructure that enabled many discoveries into the robustness characteristics and the computation capability of real DRAM chips. We describe our ongoing work in BSC adapting DRAM Bender to provide support for testing DRAM chips in BSC’s MEEP cluster as part of the Severo Ochoa Mobility Program. We show preliminary data on high bandwidth memory reliability characteristics gathered using this infrastructure at BSC.

  Més...
• Carregant...

  Text en actes de congrés Accés obert

  SORS: Experimentally understanding and efficiently mitigating DRAM read disturbance

  (Barcelona Supercomputing Center, 2025-07-28) Olgun, Ataberk

  Més...

  DRAM chips are increasingly vulnerable to read disturbance phenomena (e.g., RowHammer and RowPress), where repeatedly accessing or keeping open a DRAM row causes bitflips in nearby rows, due to DRAM technology scaling. Even though many prior works develop various RowHammer solutions, these solutions incur nonnegligible and increasingly higher system performance, energy, and hardware area overheads as RowHammer vulnerability worsens. In this talk, we will present recent cutting-edge experimental studies of and solutions to read disturbance. First, we describe variable read disturbance (VRD), a phenomenon that we recently discovered in modern DRAM chips. VRD causes a DRAM row’s read disturbance threshold to change significantly and unpredictably over time, making it difficult to identify the read disturbance threshold and thus develop efficient solutions. Second, we demonstrate that real High Bandwidth Memory (HBM2) chips are largely vulnerable to read disturbance. We also describe the DRAM Bender infrastructure that enabled our discoveries as well as discoveries in other studies (e.g., the RowPress phenomenon). Third, we introduce ABACuS, a low-cost hardware-counter-based RowHammer mitigation technique that performance-, energy-, and area-efficiently scales with worsening RowHammer vulnerability. At very low RowHammer thresholds (where only 125 activations cause a bitflip), ABACuS outperforms and takes up a smaller chip area than the state-of-the-art mitigation techniques. Fourth, we describe Self- Managing DRAM (SMD), a new, low-cost DRAM architecture that enables efficient and autonomous in-DRAM maintenance operations (e.g., periodic and RowHammer-preventive refresh) through a single, simple modification to the DRAM interface. This single modification enables implementing new maintenance mechanisms (or modifying existing ones) with no further changes in the DRAM interface or other system components, thereby enabling innovative ideas in DRAM architecture to rapidly come to fruition. SMD provides substantial performance and energy benefits while also improving system robustness across a variety of workloads.

  Més...
• Carregant...

  Text en actes de congrés Accés obert

  LOCA series: Development of advanced soot models for non-volatile particulate matter produced by aircraft engines using advanced computational models

  (Barcelona Supercomputing Center, 2025-07-22) Sriram, Niranjana

  Més...

  The study of soot has become important with thegrowing concerns for environment. Soot is one of the major byproducts of hydrocarbon combustion contributing to air pollution. Also, soot has harmful effects on the health of humans. In addition, soot affects the performance of combustors as well. Numerous models have been developed to study soot in various conditions. The existing soot models will be studied and improved to increase the accuracy of prediction of soot formation. The main focus will be on the univariate and the bivariate sectional soot model. Different models for soot processes like multi PAH nucleation, reversible nucleation, extend HACA mechanism will also be implemented.The bivariate soot model will be included in ALYA and comparisons between the univariate and bivariate soot model will be carried out. The goal of the work will be to develop a robust soot model that can be used for various laminar and turbulent flame applications.

  Més...
• Carregant...

  Text en actes de congrés Accés obert

  SORS: The AI frontier: transformative role of foundation models across scientific disciplines

  (Barcelona Supercomputing Center, 2025-07-16) Trisovic, Ana

  Més...

  This study provides an extensive analysis of artificial intelligence (AI) integration within academic sciences, focusing on the adoption and use of AI foundation models. By manually collecting data on almost 2,000 foundation models—including details such as model size, institution of origin, openness, training data, and software availability—we build a dataset that captures the landscape of AI resources available to researchers. Combined with a corpus of nearly half a million openaccess academic papers from Semantic Scholar that cite these models, our analysis explores how AI is engaged in scholarly work. Using large language models (GPT-4.1), we categorize this engagement into three main applications: developing novel AI technologies, customizing existing models, and employing AI as a routine tool in scientific research. Our findings reveal transformative trends in computational science, including a rapid increase in model complexity and the growing expertise and resources required to use these technologies effectively. We also identify a shift toward industrial dominance in AI development, which could affect the independence of academic research due to industry’s control over talent and resources. Finally, we observe a preference for open-source models among researchers addressing socially significant issues, underscoring the importance of open AI in advancing both scientific and societal goals.

  Més...
• Carregant...

  Text en actes de congrés Accés obert

  SORS/WomenInBSC: epitope engineering and natural variation in antibody-antigen recognition

  (Barcelona Supercomputing Center, 2025-07-10) Lepore, Alba

  Més...

  Antibody–antigen interactions are central to modern immunotherapies. While efforts have focused on enhancing antibody binding to improve targeting, an emerging strategy involves engineering cell surface antigens to evade recognition. This 'cell shielding' approach enables selective targeting by protecting healthy cells from therapeutic antibodies, opening new possibilities for immune engineering and disease treatement. In this talk, I will present an application of epitope engineering in acute myeloid leukemia. Our work demonstrates that engineering a CD45 epitope enables transplanted hematopoietic stem cells to resist the effects of a potent antibody–drug conjugate (ADC), allowing for the targeted elimination of malignant cells while preserving normal hematopoiesis. This work prompts a broader question: how resilient are antibody therapies to natural variation in their target epitopes? To address this, we conducted a systematic analysis of natural single nucleotide variants (SNV) and their impact on antigen recognition across a broad panel of therapeutic antibodies, both approved and in clinical development. For every antibody analysed, we identified multiple missense SNVs within or near the antibody-antigen interface, a subset of which were predicted and experimentally confirmed to abolish antibody binding entirely. Although many of these variants are globally rare, some are enriched in specific populations, with important implications for patient stratification, therapy design and clinical decision-making.

  Més...