Topological Semimetals Beyond Copper: The Case of Niobium Phosphide
In January 2025, a Stanford-led team published a result in Science 1 that briefly took over the tech press: A thin film of niobium phosphide (NbP), a material almost no one outside condensed-matter physics had heard of, had been measured with roughly twice the electrical conductivity of copper at the same thickness. Better still, the film was grown at 400 °C, a temperature compatible with how chips are actually manufactured. The headlines wrote themselves: copper has been beaten at the nanoscale.
A year is long enough to ask what happened next. Did the result hold up? Did anyone else reproduce it? Did the chip industry move on it? And, the question that brings me to go deeper here in Raw Science: what does it mean for the energy footprint of the data centres now consuming a growing share of the world’s electricity?
Spoiler: the result has held up beautifully, in my opinion. The story has also grown more interesting, in a direction the original headlines did not anticipate. There are now half a dozen materials in the race, not only one. Two of the biggest tech companies in the world are working on “topological” chips, but on completely different problems, and the press routinely confuses them. And the part of the AI-energy story this material actually addresses turns out to be one of the most overlooked.
This is a tour through the year, written for readers who want the full picture of NbP and topological semimetals.
What was actually claimed in the article
A 1.5-nanometer-thick film of niobium phosphide (NbP), a topological semimetal, showed a resistivity of about 34 µΩ·cm at room-temperature. For comparison, copper at the same thickness is around 100 µΩ·cm. I want to point out that while copper exhibits excellent bulk conductivity, its performance deteriorates at the nanometre scale because electrons begin scattering more from the wire’s surface than through its interior.
Two additional properties of NbP were also identified that could prove highly relevant from an industrial perspective:
- The films were not crystalline. NbP films have only short-range nanocrystalline order in an otherwise amorphous matrix. This was the real surprise. Earlier work on topological semimetals had always assumed you needed clean, single-crystal samples to see the effect, yet Khan et al. showed the property survives even in disordered films. That matters industrially, because real fab processes do not produce perfect crystals.
- The deposition temperature was 400 °C. The back-end-of-line of a modern chip, i.e., the metal-wiring layers added on top of the transistors, has a strict thermal budget. Anything that needs more than around 450 °C cannot be used without redesigning the rest of the process. NbP at 400 °C fits inside that envelope.
What happened in 2025–2026
The mechanism of NbP films was independently confirmed
In May 2025, a group at IBM Research Europe – Zurich, working with the Max Planck Institute for Chemical Physics of Solids in Dresden, published in Communications Materials 2 and used focused-ion-beam milling to carve down mesoscopic NbP crystals along different crystallographic orientations, and measure the resistivity in each direction. They found a clean, predictable anisotropy: samples scaled along the favourable axis (the c-axis) kept their low resistivity far better than wires aligned along the b-axis, and the measured difference matched the predicted shape of NbP’s electronic structure.
In other words, NbP is a directional conductor: electrons move easily along some crystal axes and poorly along others, so aligning the wire with the favourable direction minimises the surface scattering that typically affects narrow-gauge copper cables.
NbP became a benchmark, not the ultimate solution
In January 2026, a consortium spanning NORDITA (Stockholm), Northwestern University, IBM Watson, Academia Sinica, and the National Yang Ming Chiao Tung University, used machine learning combined with density-functional theory to screen roughly 3,000 candidate topological conductors for nanoscale interconnects.3
The result: titanium sulphide (TiS), zirconium diboride (ZrB₂), and the mononitrides of molybdenum, tantalum, and tungsten (MoN, TaN, WN) all matched or exceeded copper and the benchmark topological semimetals NbP and NbAs in computed surface transmission. In doing so, NbP is effectively emerging as the new reference material against which alternatives are measured, much like the role copper plays in the broader interconnect literature.
This is good news for the field. Several of the new candidates fit existing semiconductor manufacturing far better than NbP does. Tantalum nitride and tungsten nitride are already routine in fabs as diffusion barriers; using their conductive properties as primary wiring would extend an existing process rather than introduce a new one. The community is no longer betting on a single material, but mapping a class.
An industrial signal already appeared
IBM has been the most visible large company to deepen its focus on topological semimetals, and its two key patents trace the maturation of the field. The first, granted in 2023, covers back-end-of-line interconnects built from Weyl semimetals (NbAs, NbP, TaAs, TaP, and others) with conductor thicknesses from 1 to 100 nanometres, projecting more than 50% reductions in resistance-capacitance product at the 5-nanometre node compared with copper 4. The second, granted in October 2025, tackles the principal weakness of the first: topological semimetals are excellent at avoiding the surface-scattering losses that cripple copper at small dimensions, but they have a far lower density of free electrons per cubic centimetre. The new patent proposes wrapping the topological-semimetal wire in thin layers of high-carrier-density materials (Ta, TaN, Ru, Pt and similar metals already routine in fabs) that inject extra charge carriers into the wire’s surface states, with claimed conductivity gains of orders of magnitude over an undoped topological wire 5. Both patents were led by Ching-Tzu Chen, the same IBM researcher who co-authored Khan et al.’s Science paper on NbP. The patent line and the academic line are connected through one person.
A patent-landscape analysis published in April 2026 by PatSnap surveyed sub-10 nm back-end-of-line interconnect patents from TSMC, Samsung, IBM, and Applied Materials. Ruthenium is the material doing the heavy lifting in current filings across all four companies. IBM’s topological-semimetal patents sit one step beyond, framed in the analysis as a longer-horizon alternative for the post-ruthenium generation 6.
Two “topological” stories that are used to get mixed up
When you read in the general press that “tech companies are racing to build topological chips,” the article is almost always conflating two different things. The confusion is easy to make and matters for understanding the field, so I make a quick explanation here.
There are two distinct research tracks, both using ideas from the mathematics of band-structure topology, but applied to entirely different materials, conditions, and applications.
Track A: topological semimetals for classical interconnects. This is the NbP and related materials that have been the focus of this article. The materials are Weyl and Dirac semimetals: NbP, NbAs, TaP, CoSi, MoP, and the newly screened candidates above. What is exploited are the protected surface states (Fermi arcs) on these materials, which become the dominant conduction channel when films are thinner than the bulk electron mean free path, meaning the electrons flow along the surface rather than scattering off it. These materials work at room temperature. The application is classical on-chip wiring at very small geometries, replacing the copper that connects billions of transistors. The industrial protagonist is IBM.
Track B: topological superconductors and exotic quantum states for quantum computing. This is the Microsoft and Nokia Bell Labs lead, and the materials used are not topological semimetals. Microsoft’s Majorana 1 chip,7 announced in February 2025, uses an indium-arsenide / aluminium hybrid called a “topoconductor”, where the active material is a topological superconductor. Nokia Bell Labs8 is pursuing a different topological qubit based on a gallium-arsenide two-dimensional electron gas at temperatures below 100 millikelvin and magnetic fields above one tesla, where the fractional quantum Hall state hosts non-Abelian anyons. What is exploited is the protection of quantum information against local errors, and the application is to qubits.
In other words, Track A makes electrons flow better; Track B keeps quantum states from being scrambled. Both use the mathematics of topology, but they exploit different physical properties at different temperatures for different purposes: a Majorana 1 chip is not designed to wire up a GPU, and a niobium phosphide film is not designed to store a qubit.
Why topological semimetals matter for energy systems
According to S&P Global Market Intelligence,9 the combined 2026 capex guidance from Alphabet, Amazon, and Microsoft, disclosed during their fourth-quarter 2025 earnings calls and directed largely at AI-data-centre infrastructure, reached $495 billion. That is 61% above 2025 levels and roughly six times what these companies were spending in 2020. S&P Global Ratings estimates the capital cost of building this capacity at $25 to $30 billion per gigawatt of data-centre power. Looking across the four largest hyperscalers, the International Data Corporation (IDC) projects a combined capex of approximately $600 billion in 2026, a 70% year-over-year jump.10 And the same forecast puts data-centre semiconductor revenues at $477.1 billion in 2026, climbing to $843.2 billion by 2030, for context this is nearly half of the entire semiconductor market.
What that money is buying, in physical terms, is silicon… and the wires connecting that silicon. As transistors shrink past 5 nanometres, the wires connecting them shrink too. And when copper wires get thin enough, their resistance starts to climb sharply because electrons spend more time bouncing off the surface than flowing along it. Every watt lost in those wires has to be supplied by the grid and removed as heat from the data centre. It is one of several reasons hyperscalers are now siting new facilities next to dedicated power generation.
NbP and its emerging peers address exactly this constraint. A topological semimetal whose surface states become more conductive as the wire gets thinner is, in some sense, the opposite of copper at the nanoscale.
The realistic timeline is the one suggested by previous interconnect transitions. Cobalt entered production at Intel’s 10-nanometre node around 2019-2020, almost a decade after sustained research began, though its role has since been narrowed in favour of enhanced copper. Ruthenium is now being integrated for sub-10-nanometre back-end-of-line wiring at the 2-nanometre node and beyond, where TSMC began mass production in 2025–2026. Topological semimetals are at the early-to-middle stage of a similar arc, with production integration plausible in the early 2030s. For the AI buildout already in motion, that is just over the horizon… but it is on the horizon.
A note on supply chains. Niobium production is heavily concentrated in Brazil, where CBMM is estimated to hold around three‑quarters of the global market,11 and the European Union sources roughly 90% of its niobium from Brazil. Phosphate, on the other hand, was added to the United States Critical Minerals List already in 2025. These are real strategic dependencies in other applications, like steel microalloying and battery anodes. But for interconnect-grade NbP, the mass required per chip is essentially negligible: an atomic-thin film is a tiny amount. The binding constraint, if NbP were to scale, would be precursor chemistry and high-purity sputter targets, not directly mining. That is a different (and more tractable) bottleneck than the headlines about critical minerals would suggest.
Editorial Note
This article is a state-of-the-art synthesis as of May 2026, based on peer-reviewed papers, preprints, patents, and conference programmes published within the previous 16 months, with two foundational pre-window references retained for context. I did not include any unpublished proprietary work by foundries (TSMC, Samsung, Intel), which may well be ahead of the public literature. In the supply-chain discussion, I treat interconnect-grade NbP as a hypothetical future demand category, but no production demand exists yet, so volume reasoning is structural rather than empirical.
For a deep dive into possible pathways for copper substitution, readers can explore this article.
How to cite
Caniglia, G. (2026). “Niobium phosphide, one year on: what happened to the material that beat copper.” Raw Science. Available at raw-science.org.
References
- Khan, A. I., Ramdas, et al. Surface conduction and reduced electrical resistivity in ultrathin noncrystalline NbP semimetal. Science 387, 62–67. doi: 10.1126/science.adq7096.
- Mariani, G., et al. Orientation-dependent resistivity scaling in mesoscopic NbP crystals. Communications Materials 6, 106. doi: 10.1038/s43246-025-00828-w.
- Tyner, A. C., at al. Accelerated Discovery of Topological Conductors for Nanoscale Interconnects. Advanced Science. doi: 10.1002/advs.202520535.
- US Patent 11,749,602, IBM, “Topological semi-metal interconnects” (2023).
- US Patent 12,451,432, IBM, “Multi-layer topological interconnect with proximal doping layer” (2025).
- PatSnap, “Ruthenium interconnects at sub-10nm BEOL nodes” 16 April 2026.
- Microsoft Azure Quantum, Aghaee, M., et al. (2025) Interferometric single-shot parity measurement in InAs–Al hybrid devices. Nature 638, 651–655. doi: 10.1038/s41586-024-08445-2.
- Fitchard, K. (2025). Topological quantum computing: The quest for a quality qubit. Nokia Bell Labs
- S&P Global Market Intelligence (March 2026), “Hyperscaler earnings quarterly: What price inference?”
- IDC, (April 2026) “Semiconductor Market to Surge Past the Trillion-Dollar Threshold; AI Infrastructure Drives Market Growth,”
- IntelMarketResearch, (2025) Niobium Market Growth Analysis, Dynamics, Key Players and Innovations, Outlook and Forecast 2025-2032