Authors: Antonino Tumeo (Pacific Northwest National Laboratory (PNNL)), Johannes Gebert (High Performance Computing Center (HLRS), Stuttgart; University of Stuttgart), Suma George Cardwell (Sandia National Laboratories), Sara Achour (Stanford University, Unconventional AI), Josef Weidendorfer (Technical University of Dresden), Talya Vaknin (Lightsolver), Michael Förtsch (QANT GMBH), Connah Johnson (Pacific Northwest National Laboratory (PNNL))
Abstract: Analog computing is re-emerging as a broad approach to scientific computation, offering opportunities beyond energy efficiency, including low-latency physical dynamics, reduced data movement, computation integrated with sensing, and new methods for simulation, optimization, and artificial intelligence. Yet progress remains fragmented across physical substrates, applications, architectures, mathematics, and software. This highly interactive BoF will bring the SC community together to identify challenges, promising workloads, shared infrastructure, and opportunities for collaboration. Building on the DOE ASCR report Basic Research Needs for Analog Computing for Science, participants will prioritize research directions and help shape a community agenda for analog and hybrid analog-digital computing.
Long Description: Analog computing is re-emerging as a broad approach to scientific computation as conventional digital scaling faces growing limits in power, data movement, and time-to-solution. By exploiting the continuous dynamics of physical systems—including electrical, photonic, mechanical, chemical, biological, and other substrates—analog and hybrid analog-digital systems may offer benefits extending beyond energy efficiency. These include low-latency computation, direct solution of continuous models, reduced data movement, computation integrated with sensing and transduction, and new methods for simulation, optimization, sampling, linear algebra, and artificial intelligence.
Despite this potential, the field remains fragmented across physical technologies, application domains, mathematical models, architectures, and software ecosystems. Different communities often use incompatible terminology, precision and error models, benchmarks, performance metrics, and evaluation methodologies. Progress therefore requires a coordinated effort spanning applications, algorithms, mathematics, devices, system architecture, programming models, compilers, interfaces, validation, and benchmarking.
SC is the right venue because it convenes the full scientific-computing ecosystem needed to evaluate analog computing at system level. Application scientists can identify high-value workloads and accuracy requirements; applied mathematicians and algorithm developers can address stability, uncertainty, and continuous computation; architects and systems researchers can assess scalability and integration; software and compiler researchers can develop usable abstractions; facilities experts can evaluate operational requirements; and industry can clarify technology maturity and deployment pathways. This breadth is essential for moving beyond isolated component demonstrations toward credible, reproducible, end-to-end scientific workflows.
The discussion will use the six Priority Research Directions identified by the DOE ASCR report Basic Research Needs for Analog Computing for Science (https://www.osti.gov/servlets/purl/2476965) as a starting framework: mathematical foundations; high-performance analog systems; novel device primitives; analog computing integrated with sensing and the edge; natural and unconventional computing substrates; and full-stack co-design, software, and benchmarking. Participants will be invited to challenge, extend, and prioritize this agenda for the broader international scientific-computing community.
The BoF will explicitly examine both opportunities and limitations, including noise and variability, precision and uncertainty, analog-digital interfaces, scalability, manufacturability, programmability, software portability, application mapping, and the possibility that conversion, calibration, or integration costs may outweigh component-level gains. Rather than promoting analog computing as a replacement for digital systems, the session will focus on defensible system-level advantages and practical integration into heterogeneous scientific workflows.
The BoF will be organized as a facilitated community working session. Brief provocations from experts across academia, national laboratories, HPC centers, and industry will frame contrasting perspectives, while most of the time will be devoted to live polling, moderated discussion, prioritization, and small-group exchanges. Participants will identify promising workloads, credible evaluation criteria, benchmark and software needs, and mechanisms for sustained collaboration.
The session is well positioned for success. An analog-computing BoF at ISC 2025 attracted approximately 80 participants, demonstrating substantial interest, and the DOE workshop report identified continued community convening as important for progress. The intended outcomes are community-ranked research priorities, candidate workload and benchmark categories, gaps in software and evaluation infrastructure, and concrete options for follow-on working groups or recurring meetings. A public summary will document areas of agreement, disagreement, and next steps.
Website: https://future-computing.eu/gatherings/analog-computing-post-moore/