Dagstuhl Seminar 25151
Disruptive Memory Technologies
( Apr 06 – Apr 11, 2025 )
(Click in the middle of the image to enlarge)
Permalink
Please use the following short url to reference this page:
https://www.dagstuhl.de/25151
Organizers
- Haibo Chen (Shanghai Jiao Tong University, CN)
- Ada Gavrilovska (Georgia Institute of Technology - Atlanta, US)
- Jana Giceva (TU München - Garching, DE)
- Frank Hady (Intel Corporation - Portland, US)
- Olaf Spinczyk (Universität Osnabrück, DE)
Contact
- Marsha Kleinbauer (for scientific matters)
- Simone Schilke (for administrative matters)
Shared Documents
- Dagstuhl Materials Page (Use personal credentials as created in DOOR to log in)
Schedule
The continued increase in demand for data and data-intensive applications is exposing scaling limitations in the capacity and performance of traditional DRAM-based memory system designs. In response, new memory system designs are emerging, based on disaggregation, new heterogeneous memory components, and near-/in-memory processing. However, to fully leverage the benefits of the hardware innovation requires redesign of the software stack, from the low-level operating system and virtualization primitives managing hardware access, to programming and compiler support and application runtimes. This seminar gathered researchers from industry and academia working across the entire stack, to discuss the pressing challenges in this new memory landscape, and to identify the most promising paths forward. The five-day seminar was structured to guide the discussion across the different layers:
Day 1 set the stage by first discussing the driving challenges from the database and datacenter communities, and on surveying the state-of-the-art of the hardware technologies for processing-near memory, processing-in memory and for disaggregation (with emphasis on CXL).
Day 2 included a deep-dive in diverse use cases, including bioinformatics and drug discovery, database processing, visual analytics, and AI/ML, followed by presentations on lessons-learned from practical adoptions of memory disruptors such as persistent memory.
Day 3 focused on identifying the limitations of state-of-the-art software technologies in leveraging new memory capabilities, complemented by a panel discussion addressing challenges of cross-stack co-design to optimize their potentials, considering both general-purpose and domain-specific needs.
Day 4 explored vision talks outlining future research directions for Processing-in-Memory (PIM) , Processing-near-Memory (PNM) , hyper-heterogeneous computing, and software-hardware co-design, etc.
Day 5 concluded with collaborative proposal discussions, synthesizing insights to define a strategic roadmap for advancing these technologies.
The breakout sessions, visionary talks, and panel discussions proved highly effective in identifying the challenges and opportunities posed by disruptive memory technologies. These discussions were particularly impactful due to the diverse mix of participants, including experts from hardware, computer architecture, operating systems, databases, and parallel software domains, representing both academia and industry. For instance, industry leaders and academic researchers approached emerging memory technologies from complementary perspectives, and the ensuing vigorous debates helped broaden understanding across both sectors. Insights shared by practitioners provided researchers with valuable context to refine and prioritize critical research questions. Our report offers a comprehensive summary of the seminar’s key discussions and outcomes.
Creative Commons BY 4.0
Haibo Chen, Ada Gavrilovska, Jana Giceva, Frank Hady, and Olaf Spinczyk
Memory is a central component in every computer system. The technological evolution has led to greater capacities and higher speeds, but essential properties of the interface between hardware and software have been unchanged for decades: Main memories were usually passive, largely homogeneous, and volatile. These properties are now so firmly anchored in the expectations of software developers that they manifest in their products.
However, a wave of innovations is currently shattering these assumptions. In this sense, several new memory technologies are disruptive for the entire software industry. For example, new servers combine "high-bandwidth memory" with classic memory modules and CXL enables even more hybrid architectures (non-homogeneous). The "in-/near-memory" computing approaches abandon the traditional Von Neumann architecture and promise enormous performance improvements by allowing a vast number of parallel operations on data objects in or close to the memory (non-passive). Finally, "persistent memory" is available for server systems and embedded systems (non-volatile), which can be used for persistent in-memory data structures or even fully persistent processes.
As always, these innovations arrive with high expectations. And the memory demands of AI make these expectations even higher. But as with all technologies, these innovations encounter system realities. To be useful these innovations must shine through within a full system of hardware and software. But existing software and algorithms are finely polished for existing systems. Breaking the system inertia requires innovations of such impact that they move the architecture with them to achieve much better energy consumption, processing speed, reliability, or cost. Which technologies can deliver at this systems level? How must systems change to enable these advantages? Where is co-optimization across the silicon technology, hardware subsystems, layers of software, and even algorithms warranted? What are the new architecture models we expect and what is the migration path that will lead us there?
This Dagstuhl Seminar with about 40 leading experts from industry and academia will tackle these difficult questions in a holistic fashion. Dagstuhl Seminars are highly interactive. We plan a mix of presentations, where expert knowledge is shared, open brain-storming sessions, and group discussions. Towards the end of the week, we hope to have ideas for shaping future research in this area and groups of participants who follow up on the developed ideas in collaborative research efforts.
Haibo Chen, Ada Gavrilovska, Jana Giceva, Frank Hady, and Olaf Spinczyk
Especially for participants from the systems community outside Europe, the Dagstuhl Seminar is scheduled for the week after the EuroSys and ASPLOS 2025 conferences, which will be held in Rotterdam.
Please log in to DOOR to see more details.
- Gustavo Alonso (ETH Zürich, CH) [dblp]
- Oana Balmau (McGill University - Montréal, CA) [dblp]
- Antonio Barbalace (University of Edinburgh, GB) [dblp]
- Frank Bellosa (KIT - Karlsruher Institut für Technologie, DE) [dblp]
- Daniel Berger (Microsoft - Redmond, US) [dblp]
- Jerónimo Castrillón-Mazo (TU Dresden, DE) [dblp]
- Haibo Chen (Shanghai Jiao Tong University, CN) [dblp]
- Jian-Jia Chen (TU Dortmund, DE) [dblp]
- David Cohen (Solidigm - Santa Fe, US) [dblp]
- Christian Dietrich (TU Braunschweig, DE) [dblp]
- Thaleia Dimitra Doudali (IMDEA Software Institute - Madrid, ES) [dblp]
- Michal Friedman (ETH Zürich, CH) [dblp]
- Birte Kristina Friesel (Universität Osnabrück, DE) [dblp]
- Ada Gavrilovska (Georgia Institute of Technology - Atlanta, US) [dblp]
- Jana Giceva (TU München - Garching, DE) [dblp]
- Boris Grot (University of Edinburgh, GB) [dblp]
- Timo Hönig (Ruhr-Universität Bochum, DE) [dblp]
- Yu Hua (Huazhong University of Science & Technology, CN) [dblp]
- Sudarsun Kannan (Rutgers University - Piscataway, US) [dblp]
- Sanidhya Kashyap (EPFL - Lausanne, CH) [dblp]
- Kimberly Keeton (Google LLC - South San Francisco, US) [dblp]
- Hoshik Kim (SK hynix - San Jose, US) [dblp]
- Wolfgang Lehner (TU Dresden, DE) [dblp]
- Alberto Lerner (University of Fribourg, CH) [dblp]
- Till Miemietz (Barkhausen Institut - Dresden, DE) [dblp]
- Onur Mutlu (ETH Zürich, CH) [dblp]
- Ilia Petrov (Hochschule Reutlingen, DE) [dblp]
- Tilmann Rabl (Hasso-Plattner-Institut, Universität Potsdam, DE) [dblp]
- Tajana Simunic Rosing (University of California - San Diego, US) [dblp]
- Kai-Uwe Sattler (TU Ilmenau, DE) [dblp]
- Wolfgang Schröder-Preikschat (Universität Erlangen-Nürnberg, DE) [dblp]
- Kevin Skadron (University of Virginia - Charlottesville, US) [dblp]
- Olaf Spinczyk (Universität Osnabrück, DE) [dblp]
- Michael Swift (University of Wisconsin-Madison, US) [dblp]
- Jürgen Teich (Universität Erlangen-Nürnberg, DE) [dblp]
- Nandita Vijaykumar (University of Toronto, CA) [dblp]
- Tianzheng Wang (Simon Fraser University - Burnaby, CA) [dblp]
- Zeke Wang (Zhejiang University - Hangzhou, CN) [dblp]
- Thomas Willhalm (Intel Deutschland GmbH - Feldkirchen, DE) [dblp]
- Youjip Won (KAIST - Daejeon, KR) [dblp]
- Chun Jason Xue (MBZUAI - Abu Dhabi, AE) [dblp]
- Huanchen Zhang (Tsinghua University - Beijing, CN) [dblp]
Classification
- Databases
- Hardware Architecture
- Operating Systems
Keywords
- Processing in Memory (PIM)
- Persistent Memory (PMem)
- Disaggregated Memory
- Data-centric Computing
- System Software Stack
