Hydrogen atom collisions with a semiconductor efficiently promote electrons to the conduction band

The study addresses when and how the Born–Oppenheimer approximation (BOA) breaks down during atom–surface collisions, focusing on semiconductors, a class distinct from insulators and metals. Prior work showed BOA holds for insulators like Xe and largely holds (with perturbative corrections via electronic friction) for metals and semi-metals such as graphene. The question here is whether collisions of fast hydrogen atoms with a semiconductor surface can generate electronic excitations across the bandgap (valence band to conduction band), akin to optical excitation, and under what conditions this non-adiabatic process dominates energy transfer. The purpose is to directly probe sub-picosecond collision dynamics and test BOA validity using H atom scattering from Ge(111)c(2×8), combining experiments with first-principles molecular dynamics.

- Insulators (e.g., Xe): Prior high-resolution H scattering experiments with full-dimensional PES MD reproduced energy loss, supporting BOA validity when low-lying electronic excitations are energetically inaccessible. - Semi-metals (graphene): Similar measurements for H and D showed no BOA breakdown signatures despite possible low-energy electron–hole pair (EHP) excitations. - Metals: Energetic H atoms invariably excite EHPs; weak-coupling electronic-friction approaches (e.g., LDFA) successfully account for modest BOA failure perturbatively. - Semiconductors: Earlier Xe-on-semiconductor studies detected transient currents at 3–10 eV due to phonon heating followed by EHP generation (a delayed mechanism), with energy loss describable by adiabatic models—providing limited insight into immediate collision dynamics. This work fills the gap by probing direct non-adiabatic VB→CB excitation during H–Ge collisions and assessing the limits of BOA and electronic-friction descriptions.

- System: Reconstructed Ge(111)c(2×8) semiconductor surface; surface bandgap ~0.49 eV (determined at 30 K; reconstruction and similar bandgap expected at 300 K). - Experimental design: Scattering of energetic atomic hydrogen beams from Ge(111)c(2×8); measurement of translational energy-loss distributions and energy-resolved angular distributions (polar plots of final energy Ef versus scattering angle θf). Surface temperature Ts = 300 K. - Incidence conditions: Translational incidence energies Ei spanning below and above the surface bandgap (e.g., 0.37, 0.99, 1.92, 6.17 eV). Incidence angles θi = 30°, 45°, 60° (with additional scans in 5° increments; e.g., 0° to 75° for some datasets; separate energy scans at θi = 45°). - Data acquisition: In-plane scattering flux recorded; distributions normalized to incident H flux. Bimodal features identified based on energy-loss relative to the bandgap threshold; branching ratios between channels estimated by integrating in-plane flux over Ef and θ. - Theoretical simulations: First-principles electronically adiabatic molecular dynamics (MD) trajectories on a newly developed high-dimensional neural-network potential energy surface (NN-PES) for H/Ge(111)c(2×8). Also tested MD including electronic friction via local density friction approximation (LDFA). Representative simulation geometry for comparison used θi = θf = 45° along [110] surface direction (for Extended Data), Ts = 300 K. - Comparative models: Line-of-centres binary collision formula used to indicate kinematic energy limits in polar plots. Bandgap marked to demarcate adiabatic vs non-adiabatic (VB–CB) channels. - Analysis: Identification of adiabatic channel (low energy loss, reproduced by MD) and non-adiabatic VB–CB channel (onset at bandgap, absent in adiabatic/EF MD). Angular full width at half maximum (FWHM) extracted for both channels (Table 1). Average energy losses and their dependence on Ei and θi summarized (Table 2, described in text).

- Below the Ge(111)c(2×8) surface bandgap (0.49 eV): Only a single scattering channel is observed; adiabatic MD reproduces energy-loss distributions quantitatively. MD with electronic friction (LDFA) does not improve agreement and can fail to capture distributions. - Above the bandgap: Energy-loss distributions are bimodal. The high-loss channel has an onset at ~0.49 eV (within uncertainty), consistent with valence-to-conduction band (VB–CB) excitation; this feature is absent in adiabatic and EF MD simulations, indicating BOA failure. - Incidence-energy dependence: The VB–CB channel fraction increases strongly with Ei, accounting for ~90% of scattering at Ei = 6.17 eV; the adiabatic channel remains with small energy losses. - Angular distributions: Both channels peak near the specular angle, indicating non-thermal, sub-picosecond scattering. The VB–CB channel exhibits notably narrower angular FWHM than the adiabatic channel. Example FWHM (experimental, Table 1): at Ei = 0.99 eV, θi = 30°/45°/60°, VB–CB = 24°/31°/24° vs adiabatic >56°/>70°/34°; at Ei = 1.92 eV, VB–CB = 24°/>70°/24° vs adiabatic >56°/>73°/34°. - Average energy loss: For the adiabatic channel, ⟨Ei − Ef⟩ is small and nearly constant at 10 ± 5% of Ei across conditions. For the VB–CB channel, the fractional energy loss increases sharply as Ei decreases due to the fixed bandgap threshold; average energy loss depends weakly on θi for both channels. - Interaction timescale: Angle-dependent most-probable scattering demonstrates sub-picosecond interaction for both channels (no thermalization). Evidence of some sticking is present; overall signal suggests sticking probability decreases with increasing Ei. - Theoretical comparison: Adiabatic MD reproduces only the low-loss adiabatic channel (energy loss magnitude, angular dependence on θi). The VB–CB channel is not captured by adiabatic MD or by MD with LDFA electronic friction, implying a non-perturbative non-adiabatic mechanism beyond current models.

The experiments reveal two distinct scattering pathways for H atoms on a semiconductor surface: (1) an adiabatic, mechanically dominated channel consistent with the BOA and well reproduced by MD on an NN-PES, and (2) a non-adiabatic channel with an energy-loss onset at the surface bandgap, indicative of direct promotion of electrons from the valence band to the conduction band (VB–CB). The latter increases rapidly with incidence energy and has a narrower angular distribution, demonstrating a distinctly different interaction. The dependence of the VB–CB onset on the 0.49 eV bandgap and its absence in adiabatic/EF MD show that electronic-friction approaches, which successfully describe metals via low-energy EHP excitations near the Fermi level, are inadequate here. The results indicate that fast atom collisions can generate time-varying fields capable of driving higher-energy electronic transitions across the semiconductor bandgap on sub-picosecond timescales. These observations differ qualitatively from insulators (no BOA failure) and metals (simultaneous phonon and low-energy EHP excitation treated perturbatively), thus exposing a new non-adiabatic mechanism in semiconductors. The consistency of branching ratios across angles and energies and the specularly peaked angular distributions corroborate that neither channel is due to thermalized scattering. The findings call for new theoretical frameworks that explicitly include strong, transient non-adiabatic couplings capable of VB–CB excitation during atom–surface collisions.

This work demonstrates that collisions of energetic hydrogen atoms with a semiconductor surface, Ge(111)c(2×8), efficiently promote electrons across the surface bandgap, revealing a strongly non-adiabatic VB–CB excitation channel that dominates at higher incidence energies. Below the bandgap, scattering is adiabatic and quantitatively reproduced by BOA-based MD. Above the bandgap, bimodal energy-loss distributions appear with a high-loss channel whose onset equals the bandgap and reaches ~90% probability at Ei = 6.17 eV. The narrower angular distribution and absence of this channel in adiabatic or electronic-friction MD underscore the failure of current perturbative treatments and the need for new non-adiabatic theories for semiconductor surfaces. Future research should develop first-principles non-adiabatic dynamics methods for atom–semiconductor interactions, identify the specific electronic states and coupling mechanisms involved, assess material and surface-structure dependence, and explore implications for surface chemistry, photocatalysis, and energy conversion technologies leveraging collision-induced electronic excitation.

- The in-plane detection geometry precludes direct determination of absolute integrated sticking probabilities; out-of-plane scattering fractions may vary with incidence conditions and channel. - The surface bandgap value (0.49 eV) is taken from low-temperature (30 K) measurements; while the reconstruction persists at room temperature, small temperature-dependent shifts cannot be ruled out. - MD simulations, both adiabatic and with LDFA electronic friction, do not capture the VB–CB channel, indicating missing physics (explicit electron dynamics or strong non-adiabatic couplings) in the models. - Experimental branching ratios are scaled relative to a reference condition, and some datasets rely on angle-resolved integrations with finite angular step sizes. - Measurements focus on a single surface (Ge(111)c(2×8)) and a finite set of energies and angles; generality across other semiconductors and reconstructions remains to be established.