Architectures and Performance group

Evaluating emerging computing architectures for scientific discoveries.

At Oak Ridge National Laboratory, the Architectures and Performance group in the Computing and Computational Science Directorate is transforming how scientists harness the power of supercomputers and advanced computing systems by leveraging new computing paradigms such as neuromorphic, analog, quantum, and heterogeneous computing. Their work addresses a crucial challenge in modern scientific research: making increasingly complex computing systems more accessible and efficient for researchers across different fields. The group’s mission is to create a collaborative microelectronics excellence center dedicated to energy- and performance-efficient codesign, heterogeneous computing, and the alignment of emerging computing paradigms with the Department of Energy’s scientific objectives. Leveraging its strategic position, the team advances research across novel computing paradigms, runtime systems, high-level programming models, and high-performance computing and machine-learning applications—optimizing both performance and energy use.

The group's flagship innovation is towards IRIS (Intelligent Runtime System), a R&D 100 2024 award-winning sophisticated heterogeneous runtime framework that enables scientific applications to run seamlessly across various computing systems, from smartphones to supercomputers. They've enhanced IRIS with cutting-edge machine learning capabilities, specifically developing heterogenous memory handler, automatic data orchestration, automatic data flow optimizations, and a mechanism to use Graph Neural Networks to intelligently optimize how computational tasks are scheduled and executed. These advancements help scientists focus on their research with a runtime supporting high programmability, productivity, and performance rather than wrestling with low-level and accelerator-specific technical computing challenges

The group has provided outstanding architectures for ultra-low-latency digital-twin machine-learning solutions on FPGAs—deployed at the Spallation Neutron Source (SNS) and Center for Nanophase Materials Sciences (CNMS) user facilities—and has advanced neuromorphic research through the Neuro-Spark architecture with ultra-low latency (~50 nanoseconds) prediction solution for high-PT and low-PT classification of charge particles for Large Hadron Collider and is working on ASIC chip tapeout.

Additionally, this group is actively conducting co-design research on emerging memory technologies such as ReRAM and ECRAM and also into advance processing in memory strategies for HPC and machine learning applications.

Their practical impact is demonstrated through several key projects, including:

· Sub-microsecond Spiking Neural network solution (Neuro-spark architectures)

· ReSpike: Codesign framework for evaluating SNNs on ReRAM

· MatRIS: A multi-level truly heterogeneous math library addressing the challenges of portability, performance and heterogeneity

· Abisko: Deep codesign of an architecture for spiking neural networks using novel neuromorphic materials

· Improving the performance of the Spallation Neutron Source facility through errant beam detection using machine learning-based optimization, directly contributing to more efficient clean energy research.

What makes their approach particularly valuable is the combination of practical solutions with forward-thinking innovation. Rather than creating isolated tools, they're building an integrated ecosystem that bridges the gap between complex computing hardware and the scientists who need to use it. Their work spans from improving basic programming tools to developing sophisticated AI-driven systems that automatically optimize performance.

Additionally, the research group also offers the Experimental Computing Laboratory (ExCL) to support the CCSD Mission to conduct research in a safe and secure environment, enabling researchers to focus on research instead of focusing on system maintenance, and enable the accessibility of research equipment at the CCSD/lab-wide/global levels. The Experimental Computing Lab (ExCL) is a laboratory designed for computer science research. At a time when heterogeneity defines the path forward, this system offers heterogeneous resources that researchers can use in their work. The computational resources provided by ExCL comprise diverse technologies in terms of chips, memories, and storage. ExCL will also adapt to the ever-changing computing ecosystem and will incorporate the latest technology and make it available to its users.

Looking ahead, the group is positioned to play a crucial role in democratizing access to advanced computing resources. As supercomputers become more powerful and complex, their tools and systems will be essential in ensuring that these resources can be effectively used by researchers working on society's most pressing challenges, from climate modeling to drug discovery. Their vision of creating more intelligent and accessible computing systems promises to accelerate scientific discovery across disciplines, ultimately contributing to breakthroughs that benefit humanity.