Thin-Film Dark Matter Targets Whose Response Is Measured, Not Assumed

Some dark matter searches look for particles far lighter than a proton. A collision with such a particle deposits only a fraction of an electronvolt, so the detector must be extremely sensitive. One of the most sensitive detectors in use is a superconducting nanowire: a strip of superconductor a few nanometers thick, folded into a meander, that switches to a resistive state when even a tiny amount of energy lands in it. The wire is both the target the dark matter hits and the sensor that records the hit. To turn a count of events into a limit on dark matter, an experiment must know how readily the wire's electrons absorb small amounts of energy and momentum. Today that response is taken from bulk-crystal models. Yet a film a few nanometers thick is mostly surface, its electrons are squeezed in the thin direction, and its band structure need not resemble the bulk.

We propose to measure the low-energy electronic structure of the actual detector films with angle-resolved photoemission spectroscopy (ARPES). ARPES shines ultraviolet light on a clean surface and records the energy and exit angle of the electrons it knocks out, which maps electron energy against crystal momentum. We would build the dark matter scattering response from those maps and test it against two independent electrical and optical measurements on the same films. The key point is that ARPES's best-known weakness becomes an advantage for this target. It probes only the top 5–10 Å, and it cannot pin down momentum perpendicular to the surface. Both are defects when studying bulk crystals, but a nanometer film is a quasi-two-dimensional object, so ARPES measures nearly the whole region that does the detecting.

Core Claim

At sub-eV thresholds in a thin metallic or superconducting target, the dark matter–electron scattering rate and its directional dependence are controlled by the film's Fermi surface. That is the set of electron states at the boundary between filled and empty levels, together with the speeds of electrons on it. Thinning the film reshapes both, by splitting bands into confined "quantum-well" sub-bands and by adding surface states. These features are directly visible in ARPES and absent from the bulk density functional theory (DFT, the standard first-principles band-structure method) used in current rate calculations.

We claim that a response function built from ARPES on the real films, and tied to the films' measured kinetic inductance, changes predicted rates and directional modulation enough to matter for limits. The result could go either way. We want to know whether the change is large, and we can show it is small if it is.

The rate per unit deposited energy $\omega$ in the dielectric formalism is

$$\frac{dR}{d\omega} \propto \int d^3q \; q^2\, |F_{\rm DM}(q)|^2\, \mathrm{Im}\!\left[\frac{-1}{\epsilon(\mathbf{q},\omega)}\right] \eta\big(v_{\min}(q,\omega)\big),$$

where $\mathbf{q}$ is the momentum transfer, $F_{\rm DM}$ is the dark matter form factor (set by the mass of the particle that carries the force), $\epsilon$ is the material's dielectric function, and $\eta$ is the dark matter velocity integral above the minimum speed $v_{\min}$ needed to deposit $(\mathbf{q},\omega)$. The factor $\mathrm{Im}[-1/\epsilon]$ is the energy loss function (ELF), the material's appetite for absorbing energy $\omega$ at momentum $\mathbf{q}$. For a metal at $\omega$ of 0.1 to a few eV, the ELF is dominated by transitions across the Fermi surface, screened by the electron gas. The Fermi-surface shape and velocities are exactly what ARPES measures, and they determine both the overall rate and how it changes as Earth's rotation sweeps the dark matter wind across the crystal axes.

Relation to Prior Work

The detector side is motivated by QROCODILE. That experiment runs a superconducting microwire single-photon detector as both target and sensor at a 0.11 eV threshold and constrains dark matter–electron scattering down to 30 keV masses. It derives directional sensitivity from the anisotropic dielectric response of the thin layer. QROCODILE takes that response from a model rather than measuring it on its film, and that is the gap we fill.

The method for turning band structure into rates comes from Migdal Effect in Semiconductors. That paper computes the ELF from bulk DFT for Si and Ge, keeps only diagonal dielectric components, and treats the omitted pieces as an order-one uncertainty. We reuse its ELF machinery but replace bulk DFT inputs with measured, thickness-dependent bands. Phonon-mediated Migdal effect in semiconductor detectors extends such calculations down to 10 MeV under an isotropic, monatomic, zero-temperature lattice. It also points to narrow-gap and Dirac materials for lower masses. We share its interest in going beyond silicon, but we target metallic thin films, where anisotropy is the point rather than a nuisance.

The warning that bulk theory can miss surface physics comes from A metallic room-temperature d-wave altermagnet. There, ARPES on the $(10\bar{1}0)$ cleave of CrSb shows ribbon-like surface Fermi pockets absent from bulk calculations. That observation was made on a bulk crystal. In a nanometer film the surface is a large fraction of the target.

Geometric miniaturization enters in two ways.

  1. Shrinking the target. The RF antenna literature shrinks devices by changing their effective electrical structure rather than their outline. Examples are meander lines that add inductance (meander-loaded metasurface antenna) and shorting walls combined with complementary split-ring resonators (Huiming Yao et al.). Thinning a superconducting film does the electronic analogue: the target gets smaller and its effective electronic structure changes.
  2. Calibration. Superconducting films carry kinetic inductance, the extra inductance from the inertia of the paired electrons. Kinetic inductance resonators (Low Tc Hafnium Kinetic Inductance Device, spinodal-decomposition high-kinetic-inductance materials) use this inductance in exactly the circuit relation $f_0 = 1/(2\pi\sqrt{LC})$ that RF miniaturization exploits. The resonance frequency of a meander made from the film therefore measures its superfluid weight, which in the clean limit equals a Fermi-surface average of squared electron velocity. That gives an independent electrical check on the ARPES model.

We deliberately do not adopt the fractal-wire idea suggested by the RF fractal work (Minkowski fractal antenna). The reasons are given in the "Why not just…?" section.

Research Program

Films. We grow epitaxial superconducting films at three or four thicknesses spanning roughly 3 nm to a bulk-like 30 nm. Candidates are crystalline materials already used in kinetic-inductance or nanowire devices, such as hafnium. Amorphous films have no crystal momentum to resolve and are out of scope; that is a stated limitation. Each film is transferred under vacuum to ARPES without breaking ultra-high vacuum, so the surface measured is the surface used.

ARPES maps. At each thickness we measure:

  • the Fermi surface;
  • band dispersions within about 1 eV below the Fermi level;
  • quantum-well sub-band spacings;
  • line widths, which give quasiparticle lifetimes.

Response model. We fit a tight-binding model to each thickness, constrained by the ARPES dispersions and widths. A tight-binding model is a compact description of electrons hopping between atomic orbitals. The empty states that dark matter excites into lie within a few tenths of an eV above the Fermi level. ARPES cannot see them directly, so they come from the fitted model's continuation across the Fermi level. Where feasible we check this continuation by alkali surface dosing, which shifts empty bands below the Fermi level. From the model we compute the ELF in the random-phase approximation, the standard screened-electron-gas treatment, following the Knapen et al. formalism. We then insert the ELF into the rate expression above for heavy and light mediators.

Detectors. The same films are patterned into two device types:

  • simple meander kinetic-inductance resonators, whose resonance frequencies give the superfluid weight;
  • nanowire photon detectors, whose detection efficiency between 0.1 and 1 eV probes the low-momentum limit of the response.

Baselines

Naive reference. A free-electron (Lindhard) dielectric function with the film's measured carrier density, applied unchanged at every thickness. This is the "do nothing" model a quick thin-film analysis would use. It sets the floor: if our ARPES-derived model does not predict resonator frequencies and photon response better than this, the project has failed.

Strongest existing method. The bulk-DFT ELF pipeline of Migdal Effect in Semiconductors, run for the bulk film material with the same exchange-correlation choice, k-point density, and diagonal-only approximation as in that paper. This is the strongest competitor because it is the established route from band structure to rate. It is also the kind of input that dielectric-response analyses like QROCODILE's rely on.

Strongest simple alternative. Slab DFT: first-principles calculations of a finite-thickness film with real surfaces, which captures confinement without any measurement.

Fairness. All three models are scored against the same held-out measurements:

  • resonator frequencies at each thickness;
  • spectral photon-detection efficiency;
  • where available, optical reflectivity.

None of these measurements is used in fitting. ARPES enters only the tight-binding fit, and the DFT models receive the same compute budget for convergence tests.

What counts as beating them. The ARPES model must predict superfluid weight and the low-momentum photon response with smaller error than both DFT variants at the thinnest thickness, while agreeing with them at 30 nm, where all should converge.

Ablations. Two ablations isolate the key ingredient:

  • the ARPES model with quantum-well sub-bands and surface states removed, keeping only the bulk-like bands seen in the thick film;
  • the ARPES model with lifetimes set to zero.

Any improvement should vanish in the first ablation if confinement and surfaces are what matter.

Reported reference numbers. QROCODILE reports a 0.11 eV threshold, a 30 keV mass reach, 415.15 hours of exposure, and 15 unclassified pulses (QROCODILE). We found no reported result on Lacuna for a cross-section limit value, or for any measured ELF or kinetic inductance of a nanowire detector film. Our rate comparison will therefore be expressed as the ratio of predicted rates, ARPES model to bulk DFT, applied to a fixed exposure. Absolute limits are not claimed.

Why not just…?

Why not just use bulk DFT, as existing rate calculations do? It is cheap and well understood, and for thick films it is probably adequate. The evidence against it in our regime is that bulk calculations miss surface Fermi pockets that ARPES sees (CrSb altermagnet), and that in a few-nanometer film such states are a large share of the electrons. Our thickness series settles this directly. If bulk DFT predicts the resonator frequencies and photon response as well as the ARPES model at 3 nm, then bulk DFT is good enough, and we will say so.

Why not just run slab DFT and skip the beamtime? This is the real threat. Slab DFT captures confinement and surfaces in principle. It can fail for oxidized or reconstructed real surfaces, interface strain from the substrate, and lifetime broadening. We treat it as a full baseline rather than a straw man. If slab DFT matches ARPES band by band, the outcome is a validated slab-DFT recipe for detector modeling, which is still useful. ARPES then becomes a one-time validation step rather than a routine input.

Why not just measure the loss function directly with electron energy loss spectroscopy, or with optics? Optical data and photon-detection efficiency constrain only the low-momentum limit of the ELF. Dark matter–electron scattering samples momentum transfers of a few keV, comparable to the size of the Brillouin zone (the range of distinct crystal momenta). Momentum-resolved electron energy loss spectroscopy on films this thin at sub-eV resolution would be a strong complement. Where it is available we would add it as an external check rather than argue it away. ARPES's distinct contribution is that it resolves which states and directions generate the response, which is what the daily directional modulation depends on.

Why not just increase exposure, or adopt fractal wire geometries? Exposure does not help a mis-modeled response: a biased ELF biases every limit derived from it, and larger exposure only shrinks the statistical error around the wrong answer. Fractal meanders would raise wire length per area. The fill factor of simple meanders is already high, though, and sharp fractal corners crowd current and lower the switching current. The RF fractal work itself attributes simulation–measurement mismatch to etching tolerance (Minkowski fractal antenna). The geometric lever that changes the target response is thickness, not in-plane fractal shape, so that is the one we vary.

Proposed Experiments and Decisive Checks

Electrical check. We predict the superfluid weight from the ARPES Fermi surface and lifetimes, then compare with resonator-derived kinetic inductance at each thickness. Disorder corrections use the ARPES line widths. Success means the ARPES model's error falls well below both DFT baselines at 3–5 nm, with all models converging at 30 nm.

Optical check. We compare the predicted near-threshold photon absorption with the measured spectral detection efficiency of nanowire detectors made from the same films.

Physics deliverables. Once the model passes both checks, we report:

  • the ratio of predicted dark matter–electron rates to the bulk-DFT prediction, as a function of mass and thickness, for heavy and light mediators;
  • the amplitude of the daily directional modulation, for comparison with the anisotropy argument of QROCODILE.

Interpreting the outcome. A rate ratio within the order-one uncertainty that Migdal Effect in Semiconductors already assigns to its approximations would mean bulk modeling is safe for thin-film targets, which is a useful negative result. A ratio well outside that band, together with passing electrical and optical checks, would mean current thin-film limits need measured response functions.