The Only New Research Program in NSF's $1.5 Billion Package Is About Computing With Waves. Concept Outlines Are Due November 19.

Eleven of the twelve funding opportunities NSF released on August 17, 2026 were consolidations. Existing programs merged, existing solicitations retired, existing money redistributed under fewer documents.

The twelfth funds something that did not have a program before.

NSF 26-524 — Emerging Frontiers in Research and Innovation: Wave-Based Computing (EFRI-WBC) commits $30 million, at $15 million per fiscal year across FY2027 and FY2028, to a proposition that sounds like a category error until you look at the physics: that computation can be performed by waves interacting with each other, rather than by transistors switching.

Don Millard, head of NSF Engineering, framed the payoff plainly — this investment "will help realize wave-based computing and its seamless integration with traditional digital computing," with significant impact on artificial intelligence, communications, and high-performance computing. The program was identified as a priority research opportunity by the Engineering Research Visioning Alliance, and it is being run by Engineering in collaboration with the Mathematical and Physical Sciences and Technology, Innovation and Partnerships directorates.

The first gate is not the proposal. It is a two-page email, and it is due November 19, 2026.

Why waves, and why now

The case for wave-based computing is a case about physics running out of road.

Digital computing moves charge. Every logic operation charges and discharges capacitance, every bit of that is dissipated as heat, and the industry has spent two decades managing the consequences — clock speeds plateaued around 2005, and the response has been parallelism, specialization and increasingly heroic cooling. AI workloads have made the arithmetic brutal: the dominant operation in a neural network is matrix multiplication, and doing it in digital silicon means shuttling enormous quantities of data between memory and compute, burning power at every hop.

Waves do some of that work for free. When waves superpose, they sum. When they pass through a structured medium, they transform. Interference performs, as a matter of physics rather than instruction execution, operations that a digital processor has to sequence. An optical system can propagate a full matrix multiplication at the speed of light through a passive element, with no clock and comparatively little dissipation. NSF's own framing points at exactly these properties: "the high speed, massive parallelism, low latency, low power usage, and high efficiency" of the approach.

The reason it is a research program and not a product line is that nobody has solved the surrounding engineering. Analog wave systems accumulate noise. They are hard to program, hard to cascade, and hard to reconcile with the digital infrastructure everything else runs on. Which is why Millard's phrase — seamless integration with traditional digital computing — is the actual technical target, not a throwaway.

Two tracks, and what each one demands

The solicitation splits into tracks with sharply different scale and expectations.

Track 1 — WBC Exploratory Research (WBC-ER) funds $100,000 to $400,000 for up to two years, awarded through the EAGER mechanism. A single PI can hold it. It must address at least one of the first two research thrusts, and NSF states it "should also include a testbed unless the proposed research is purely theoretical."

Track 2 — WBC Multidisciplinary Research (WBC-MR) funds up to $2,000,000 over four years, and it requires a minimum of one PI and two co-PIs — three investigators, no exceptions. Track 2 proposals must address all three thrusts.

Across both tracks NSF anticipates 15 to 30 awards, with up to fifteen per fiscal year.

The three thrusts are the structural spine of the whole solicitation:

Thrust 1 — Physical Principles. The wave physics itself: electromagnetic and acoustic waves, linear and nonlinear processes, emerging optical materials, artificially structured media, and chip-scale integration.

Thrust 2 — Information Processing. The computational layer: analog, digital, mixed-signal and neuromorphic architectures, quantum information processing, and learning algorithms designed for wave substrates.

Thrust 3 — Proof-of-Concept Testbeds. NSF is explicit that this means "development of testbeds that translate the early-stage research and novel concepts...into an actual platform that validates these concepts," demonstrating "novel designs of device and system components."

Thrust 3 is where most Track 2 proposals will be won or lost. A team of physicists and a team of computer architects can each write a compelling contribution to Thrusts 1 and 2 without ever building anything. NSF has made the hardware demonstration a mandatory component of the largest award, which means the proposal has to name the platform, the fabrication pathway, and who on the team has actually built one before.

The eligibility rules that will disqualify people

This solicitation carries more entry restrictions than anything else in the August package, and several of them are easy to miss.

The lead PI must be tenured or tenure-track. For proposals from institutions of higher education, NSF requires that "the lead Principal Investigator (PI) must be full-time, tenured or tenure-track faculty." Research faculty, research scientists, and non-tenure-track teaching faculty cannot lead. In a field this instrument-heavy, where much of the relevant fabrication expertise sits in research-scientist appointments, that is a meaningful constraint on team structure.

Engineering must be represented in the leadership. "Either the PI or one of the co-PIs must have a full-time, tenured or tenure-track faculty appointment within a College/Department of Engineering." A pure physics or applied-math team does not qualify regardless of technical merit. This is an ENG-led program and NSF is enforcing that at the org-chart level.

One proposal per person per year. "An individual may participate as PI or co-PI on a maximum of one proposal submitted to this solicitation in a single fiscal year." You cannot hedge with a Track 1 and a Track 2. You pick.

Spin-wave computing is explicitly out of scope. The solicitation states that "proposals focusing specifically on spin-wave computing are not considered within scope." Magnonics is one of the most active communities in this general space, and it has been carved out. Anyone whose research program centers on spin waves needs to read that line before writing anything.

Collaborative proposals must use subawards. "Separately submitted collaborative proposals are not permitted." A multi-institution Track 2 team submits as one proposal from one lead institution with subawards — a materially different budgeting and negotiation exercise, and one that takes weeks longer than most teams assume.

Voluntary committed cost sharing is prohibited, and institutions are limited in practice: "the EFRI Office will not normally award more than one proposal from any one lead institution in this competition in a single cycle." There is no formal cap on submissions per organization, but a large university sending four Track 2 proposals should understand they are largely competing with each other. Budgets must also include travel for the mandatory annual EFRI grantees' meeting.

The timeline is longer than it looks

Two cycles are on the table:

• FY2027: concept outlines due November 19, 2026 (5 p.m. submitter's local time), full proposals due February 11, 2027
• FY2028: concept outlines due November 16, 2027, full proposals due February 10, 2028

The concept outline is a two-page research concept outline submitted by email to EFRI2027-2028@nsf.gov — required of every prospective PI who intends to submit. A webinar for prospective PIs is scheduled for October 2026.

Unlike the rest of the August package, this program kept its deadlines. While BIO, GEO and the MPS divisions moved to rolling submission, EFRI-WBC runs a conventional two-stage calendar — which for a program funding genuinely speculative work is the right call. Batched review against a known cohort is how you calibrate what counts as an acceptable risk.

The real scheduling problem is Track 2's team requirement. Three investigators spanning wave physics, computational architecture and hardware demonstration — with at least one holding a tenured or tenure-track appointment in engineering, and all under a single institution's submission with subawards for the rest — is not a team you assemble in the eight weeks before a concept outline. Groups that will win in February 2027 are having those conversations now, and groups that miss November have a full year to wait.

NSF's review criteria make the ambition explicit. Beyond intellectual merit and broader impacts, reviewers are asked whether the work "represents an opportunity for a significant leap or paradigm shift in foundational engineering knowledge" and whether there is "potential for making significant progress on a strategic technology priority and/or national S&T mission." Track 1 is judged additionally on its mentoring plan, Track 2 on its management plan — a signal that NSF expects coordination failure, not scientific failure, to be the main risk in a four-year three-PI hardware project.

In a package otherwise devoted to reorganizing how NSF funds work it already funded, $30 million for computing with light and sound is the one line that buys a new question — and the first thing it costs is two pages and a November email. Sorting which of the twelve new solicitations your project actually belongs in, before you commit a semester to writing, is the kind of work Granted can compress into an afternoon.