## Introduction

Polaritons are hybrid electromagnetic waves originating from the coupling of light and dipole-like excitations in matter and include phonon polaritons (PhPs), plasmon polaritons, and exciton polaritons1,2,3,4. Among these polaritonic modes, PhPs with strong field confinement, low loss, and a long lifetime can enable a wide range of applications in light transmission, electro-optical modulation, sub-diffraction focusing, and imaging, and molecular detection5,6,7,8. However, to best utilize PhPs, external control of their properties is needed, which is a difficult task owing to the wide bandgap (~2–10 eV) in polar crystals, making the common electrical-gating method impractical.

Although a few attempts have been made9,10, effective tuning of PhPs remains challenging. Recently, PhP manipulation has been investigated by controlling the dielectric environment through the fabrication of vertical heterostructures of van der Waals (vdW) crystals (hexagonal boron nitride (h-BN)/black phosphorous11 and MoS2/SiC12), polaritonic metasurfaces in combination with phase change materials (h-BN/VO213,14,15, germanium antimony telluride (GeSbTe)/quartz16 and h-BN/GeSbTe17), or polar materials suspended in air18. The latter enables improvement of the PhP figures of merit but imposes limitations on the device architecture and functionality.

Very recently, in-plane anisotropic and ultra-low-loss PhPs in the natural vdW semiconductor crystal α-phase MoO3 (α-MoO3) have been reported19,20. As a typical transition metal oxide, α-MoO3 possesses variable chemical and physical properties and is widely studied for use in catalysis, energy storage devices, and optoelectronic devices21,22,23. In particular, the high chemical compatibility of α-MoO3 enables direct tuning of its optical properties by metal-atom intercalation24. Intercalation has been previously explored in nanometer-thick MoO3 slabs25, where metallic Sn atoms were introduced into the crystal-lattice interlayers, effectively modifying the surface propagation of PhPs. Nevertheless, realizing intercalation tuning PhPs in a controlled and reversible way is still challenging due to the large volumes of the intercalators and the solvent-based character of the process.

Here, we successfully demonstrate chemical switching of PhPs in nanometer-thick α-MoO3 by precisely modifying its crystal structure via hydrogen intercalation (so-called hydrogenation or protonation). Unlike changing external dielectric constants, our hydrogenation approach directly acts on the polaritonic material itself. The crystal lattice of MoO3 is rearranged owing to the formation of type I HxMoO3 (0.2 < x < 0.4), giving rise to perturbed phonon oscillations and the disappearance of PhP propagation. Importantly, following a deintercalation (also known as dehydrogenation or deprotonation) process, the crystal structure can fully recover to its original state without significant damage, enabling quantitatively reversible switching of low-loss PhPs. By precisely controlling the intercalation time, in-plane nanostructures that can reflect and launch PhPs are obtained in intermediate states. Furthermore, by combining our intercalation method with photolithography, we realize spatially controlled switching of PhPs. Our findings provide a chemical strategy for effective control of PhPs to develop reconfigurable polaritonic devices or topological photonic structures on a large scale.

## Results

### Structural analysis during reversible hydrogenation

As a typical vdW material, α-MoO3 is composed of double layers of linked and distorted MoO6 octahedra26, as shown in Fig. 1a. The nanometer-thick α-MoO3 (original α-MoO3, hereafter termed O-MoO3) flakes studied in this work were synthesized by a chemical vapor deposition (CVD) method, followed by a mechanical exfoliation process (see Methods and Supplementary Fig. 1, in which different characterizations of the as-prepared α-MoO3 flakes confirm their orthorhombic crystal structure). The hydrogen intercalation process was conducted by treating the flakes with a hydrogen plasma. To avoid irreversible damage to the crystal, the exposure time was controlled at 10 s (see Methods and Supplementary Fig. 2). Four typical hydrogen molybdenum bronzes (HxMoO3, 0 < x ≤ 2) with different compositions and structures can be obtained after α-MoO3 hydrogenation27. In our experiment, we obtained type I HxMoO3 (orthorhombic, 0.2 < x < 0.4) in hydrogenated α-MoO3 (hereafter, termed H-MoO3), as revealed by Raman and X-ray diffraction (XRD) analysis. In the MoO6 octahedron, the oxygen atoms can be divided into three types according to their site (Fig. 1a). As shown in Fig. 1b, c, the Raman shifts at 667, 818, and 994 cm−1 are assigned to the stretching vibration modes of the Mo–O3’, Mo–O2’, and Mo–O1 bonds, respectively. After hydrogenation, owing to a structural rearrangement, the peaks at 818 and 994 cm−1 shift to 760 (broadband) and 1008 cm−1 (black arrows in Fig. 1b), respectively, agreeing well with the Raman spectrum of type I HxMoO328. Another stretching vibration peak of the Mo–O3 bond at 441 cm−1, which is hardly observable in O-MoO3, emerges after hydrogenation. Given that the Raman peaks at 667, 818, and 994 cm−1 remain discernible in H-MoO3, even those with weaker intensities (Supplementary Fig. 3), the hydrogenated product is a mixture of type I HxMoO3 and α-MoO3. In addition, the XRD spectra of O-MoO3 (Fig. 1d,e) correspond to the Pbnm space group of orthorhombic α-MoO3 (ICDD PDF: 05–0508) with lattice constants a = 3.97 Å, b = 13.86 Å, and c = 3.71 Å. After hydrogenation, new peaks arise at 12.6°, 25.3°, and 38.3° (black arrows), which are attributed to the (020), (040), and (060) planes of orthorhombic H0.31MoO3 (ICDD PDF: 70–0615). This result unambiguously proves the formation of type I HxMoO3 with calculated lattice constants a = 3.93 Å, b = 14.02 Å, and c = 3.74 Å.

The as-prepared HxMoO3 is stable at room temperature and can last for several months in the ambient atmosphere (Supplementary Fig. 4). However, we can apply a simple thermal treatment (see Methods) to induce deintercalation, which proceeds as

$${\mathrm{H}}_{{x}}{\mathrm{MoO}}_3 + {\mathrm{O}}_2 \to {\mathrm{MoO}}_3 + {\mathrm{H}}_2{\mathrm{O}}$$
(1)

Both the Raman and XRD spectra of the recovered α-MoO3 (hereafter, termed R-MoO3) agree excellently with those of O-MoO3, demonstrating that the H-MoO3 crystal structure is reconfigured to α-MoO3 after dehydrogenation. Moreover, X-ray photoelectron spectroscopy (XPS) characterizations (Supplementary Fig. 5 and Supplementary Table 1) provide solid evidence for the reversibility of the chemical composition changes that occur during the intercalation/deintercalation process. Compared with the metal-atom intercalation, our approach thus affords an effective and relatively non-destructive strategy to reversibly modify the composition and structure of nanometer-thick transition metal oxide flakes.

### Reversible switching of PhPs

The structural reversibility during hydrogenation/dehydrogenation processes enables direct manipulation of the vibration-dependent PhPs in nanometer-thick α-MoO3 slabs. To study the polaritonic response of our samples, we used scattering-type scanning near-field optical microscopy (s-SNOM, Fig. 2a), which enables direct mapping of the near field in terms of the amplitude and phase and the sample topography by illuminating the sample with mid-infrared light at different frequencies. First, we measured the samples in the lower Reststrahlen band (L-RB, 818–974 cm−1) of α-MoO3, where PhPs are known to propagate with a hyperbolic wavefront along the [100] direction19,20,25. Figure 2b shows s-SNOM amplitude images (third-harmonic signals, s3) at frequency ω = 890 cm−1 together with optical micrographs of O-MoO3, H-MoO3, and R-MoO3 flakes with the thicknesses of 220 nm. As expected, for O-MoO3, we observe interference fringes in the s-SNOM image, revealing the excitation of PhPs propagating parallel to the [001] direction with a polariton wavelength λp = 1.61 ± 0.01 μm, which well matches previous work19. After intercalation with hydrogen (H-MoO3), we observe a weaker contrast in the optical image, suggesting a varied optical behavior. More importantly, no regular interference fringes can be observed in the s-SNOM image (Supplementary Fig. 6), implying that PhPs are not excited. After dehydrogenation, the initial contrast of the flake in the optical image is recovered, and parallel fringes in the s-SNOM image are again observed, revealing the excitation of PhPs.

For a quantitative analysis of the PhP behavior, we compared the group velocities (vg), propagation lengths (L), and lifetimes (τ) of PhPs in the original and recovered α-MoO3 flakes. First, we extracted the PhP dispersions (ω(Re(k)), where k is the complex-valued wavevector) along the [100] direction (dashed lines in Fig. 2b) at different frequencies by fitting the line scans obtained from s-SNOM amplitude and phase images using the function29:

$$\sigma \left( x \right) = A\frac{{{\mathrm{e}}^{i2{\mathbf{k}}x}}}{{\sqrt x }} + B\frac{{{\mathrm{e}}^{i{\mathbf{k}}x}}}{{x^d}}$$
(2)

where σ is the complex-valued optical signal, d (~1) is variable decay, A and B are the parameters for tip- and edge-launched PhPs, respectively. We used the function y = axb to fit the dispersion data (solid lines in Fig. 2c, see Supplementary Table 2) and then performed a numerical derivation process on the fitting curve, as the group velocity can be extracted by vg = ∂ω/∂Re(k)19. The resulting group velocities of PhPs in the original and recovered α-MoO3 flakes are shown in Fig. 2d. We observe similar values of 4.42 × 10−3 c and 4.45 × 10−3 c at ω = 903 cm−1 for the original and recovered flakes, respectively, where c is the speed of light in vacuum. Then we calculate the polariton lifetime by τ = L/vg (see Supplementary Note 1 for more details). Accordingly, the line profiles at ω = 903 cm−1 and the fitting results are plotted in Fig. 2e, where the propagation lengths of PhPs in the original and recovered α-MoO3 are 2.14 ± 0.61 μm and 2.09 ± 0.22 μm, respectively. The calculated lifetime of PhPs in O-MoO3 (1.61 ± 0.46 ps) is comparable to that in R-MoO3 (1.57 ± 0.16 ps), suggesting high-fidelity recovery of hyperbolic PhPs after dehydrogenation. The PhP behavior remains almost unchanged even after three intercalation/deintercalation cycles, as shown in Fig. 2f (Supplementary Fig. 7).

In addition to PhPs in the L-RB, PhPs with elliptical propagation in the upper Reststrahlen band (U-RB, 960–1010 cm−1) of α-MoO3 can also be reversibly switched through hydrogenation while maintaining their extraordinary properties (strong field confinement and ps lifetimes19). As shown in Fig. 3a, PhPs propagate along both the [100] and [001] directions in O-MoO3 and R-MoO3 owing to the elliptical PhP propagation. We compared the performances of PhPs in the original and recovered α-MoO3 flakes (140 nm) along the two directions (Fig. 3b–d). In the [100] direction (dashed line 1 in Fig. 3a), the PhP group velocities and lifetimes are 0.36 × 10−3 c and 10.93 ± 2.96 ps (original) and 0.35 × 10−3 c and 10.38 ± 1.71 ps (recovered) at 990 cm−1, respectively. Comparable characteristics are also found in the [001] direction (dashed line 2), where the group velocities (lifetimes) in the original and recovered slabs are 0.43 × 10−3 c (12.25 ± 1.94 ps) and 0.42 × 10−3 c (12.86 ± 1.67 ps), respectively. These results demonstrate that the hydrogen intercalation/deintercalation method enables robust, permanent (several months) and reversible switching of PhPs in α-MoO3 both in the L- and U-RB.

### Analysis of intermediate states during hydrogenation

To study the hydrogen intercalation process and the evolution of the PhPs excited in nanometer-thick α-MoO3, we prepared H-MoO3 flakes with various hydrogen proportions by precisely controlling the intercalation time (i.e., hydrogen plasma treatment time, Supplementary Fig. 8). The XPS results (Fig. 4a and Supplementary Table 3) show a slowly increasing proportion of Mo5+ (HxMoO3) with prolonged intercalation time. After intercalation, we used s-SNOM to investigate the effect on the PhP properties in these samples. Figure 4b shows normalized s-SNOM amplitude (top) and phase (bottom) images of a 120 nm-thick α-MoO3 flake before (O-MoO3 (0 s)) and after 2-, 5-, 7-, and 10-s hydrogen intercalations (H-MoO3). The parallel interference fringes in both the amplitude and phase images reveal that the excitation of PhPs gradually vanishes with increasing intercalation time, which can be further verified by the extracted amplitude line traces (Fig. 4c) and fast Fourier transform (FFT) results (Supplementary Fig. 9). The quantitative comparison indicates that upon gradual hydrogenation, the PhP dispersion has no obvious difference, while the propagation length and lifetime gradually decrease. By comparing the Raman spectra in hydrogenated intermediate states, we attribute this chemical switching of PhPs to the disappearance of optical phonons (L-RB and U-RB) of the polar material during hydrogenation (Supplementary Note 2 and Supplementary Fig. 10).

Interestingly, we found regular acicular nanostructures in intermediate H-MoO3 (2 and 5 s) compounds. For a better analysis, we prepared an intercalated flake (2+2 s) by a step-by-step method (Supplementary Fig. 11). We observe needle-like nanostructures (Fig. 4d) with a weaker optical contrast extending along ~57 ± 2 ° with respect to the [100] direction (on the (010) surface), which can be assigned to the <203> direction30. Similar nanostructures are also observed in nanometer-thick α-MoO3 discs (Supplementary Fig. 12), indicating that the hydrogenation direction determined by the intrinsic crystal structure is the main reason for their nucleation orientation instead of the specific geometry of the flake. Indeed, by performing Raman intensity mapping, we discover that these nanostructures consist of type I HxMoO3 (Supplementary Fig. 12).

To better understand the nucleation of these nanostructures, we studied the diffusion of hydrogen atoms in type I HxMoO3 by performing calculations using density functional theory (DFT) (see Methods and Supplementary Note 3). The O2 sites (with an adsorption energy of 3.58 eV) are the most energetically favorable for hydrogen adsorption compared with the O1 (3.06 eV) and O3 (1.13 eV) sites (Supplementary Table 4). In addition, hydrogens are alternately located along the zig-zag chains of intralayer oxygen atoms ([001] direction) within MoO6 octahedra31. By comparing the calculated thermodynamic stabilities for hydrogen diffusion along the [203] direction and along the [001] direction, we find the former to be slightly more energetically favorable (15 meV), possibly because of the cationic repulsion (Supplementary Fig. 13). We can thus conclude that the three-dimensional hydrogen diffusion in HxMoO3 at low hydrogen concentration occurs along the [203] direction, as schematically shown in Fig. 4e, giving rise to the observed needle-like nanostructures oriented along 57 ± 2 ° with respect to the [100] direction.

Further characterization by s-SNOM of these nanostructures shows that while in terms of the topography (Fig. 4d and Supplementary Fig. 14), they are relatively hard to identify (probably because of their atomic expansion of ~4 nm caused by the rearrangement of MoO6 octahedra, which can be recovered after deintercalation), in the amplitude image, they show a strong near-field contrast compatible with a high degree of metallicity because of the greatly enhanced conductivity of H-MoO3 (Supplementary Fig. 15). Clear parallel interference fringes revealing excitation and/or reflection of PhPs are observed on both sides of the nanostructure (Fig. 4f). In fact, by extracting amplitude line scans perpendicular to the [203] direction (i.e., [302]) between two parallel nanostructures (Fig. 4g) in the red rectangular region in Fig. 4d, we obtain resonator-like oscillations32 with frequency-dependent wavelengths similar to PhPs along that direction. We also compare λp at different frequencies in O-MoO3, R-MoO3, and four typical regions in H-MoO3 (rectangles in Fig. 4d). As displayed in Fig. 4h, the λp of PhPs along the [100] direction remains similar in O-MoO3, R-MoO3, and H-MoO3. However, PhPs along the <302> direction in H-MoO3 exhibit decreased polariton wavelengths, which are much shorter than those of PhPs along the same direction in O-MoO3. The FFT results along the [302] direction in Supplementary Fig. 16 show two prominent peaks, which is attributed to PhPs launched by both the tip and acicular nanostructures19; thus, we can conclude that these hydrogenated nanostructures act as efficient reflectors and launchers of PhPs (see Supplementary Fig. 17 for U-RB) while forming part of the crystal (as defined chemically and not physically), which can be highly beneficial for the development of planar devices in nanophotonic technologies (Supplementary Note 4).

### Spatially controllable switching of PhPs

Finally, we studied the scalability and feasibility of spatially selective hydrogenation within α-MoO3 flakes. As illustrated in Fig. 5a, we combined the hydrogenation method with a conventional photolithography technique. We used direct photolithography to cover a 225 nm-thick α-MoO3 flakes by a photoresist with designed patterns (see Methods and Supplementary Fig. 18 and 19 for details). In this way, upon hydrogenation, only open regions of α-MoO3 flakes are exposed, and thus, H-MoO3/α-MoO3 in-plane heterostructures with various geometries can be achieved. Figure 5b, c shows a flake where the bottom half region has been hydrogenated and thus PhPs are switched off in both the L-RB and U-RB. In contrast, in the hydrogen-free region (α-MoO3), parallel interference fringes indicating PhPs are observed.

Owing to its in-plane anisotropy, α-MoO3-based in-plane nanostructures with different orientations are attractive for nano-optics studies and can be easily prepared using our method by designing the masked region as shown in Fig. 5d, where the boundary of the heterostructure is perpendicular to the PhP propagation direction. Furthermore, the thickness of the hydrogenated regions shows no obvious difference from that of the unhydrogenated regions, whereas the optical signal is very different, as confirmed by the extracted line traces (Fig. 5d). Figure 5e shows the potential application of our chemical switching method. By using customized patterns in photolithography, discs, squares, and more complicated arrays with various PhP dispersions can be constructed. Notably, we observe a clear reflection of PhPs at the sharp boundaries along the <203> direction (Fig. 4d and square in Fig. 5e). However, no significant reflection occurs at the boundaries along other directions (Fig. 5d and disc in Fig. 5e) because these boundaries are composed of much smaller <203>-oriented HxMoO3 nanostructures, at which PhPs are irregularly scattered, leading to unobservable reflection fringes.

## Discussion

We have demonstrated an effective chemical approach to manipulate PhPs in α-MoO3 by the hydrogen intercalation-induced perturbation of lattice vibrations. Reversible switching of PhPs within the α-MoO3 flake is achieved. The figures of merits for recovered PhPs after multiple intercalation/deintercalation cycles remain quantitatively comparable to those for the original PhPs. By precisely controlling the intercalation time, we reveal the relationship between PhPs and the crystal structures of the polar material, and in-plane nano-antennas that can reflect and launch PhPs are obtained. We also achieve spatially controllable intercalation and selective switching of PhPs by combining our intercalation process with photolithography (or possibly a shadow mask), through which in-plane heterostructures are constructed, which is difficult to realize in other PhP tuning methods (Supplementary Note 5 and Supplementary Table 5).

Our methodology establishes a proof of concept for chemically manipulating polaritons, offering opportunities for the growing nanophotonics community. We envision that in operando switching of PhPs could be realized during s-SNOM nanoimaging via tip-induced hydrogen intercalation (the so-called spillover technique33). The plasmonic resonance34, newly observed ferromagnetic properties35, and high conductivity of hydrogen-doped MoO3 enable analysis of the interaction of PhPs with plasmon and magnetic fields. Furthermore, using other intercalators, such as ions (Li+, Na+)36,37, metal atoms (Co, Fe)38,39, and polymers40, programmable in-plane heterostructures and flat metasurfaces could be constructed, and applications combining scalable planar optics with photomagnetics, photoelectronics and topological photonics could be realized.

## Methods

### Preparation of α-MoO3 flakes

α-MoO3 bulk crystals were grown by a CVD method. In brief, 0.1 g of MoO3 particle (Alfa Aesar) was placed in the middle of a quartz tube (2 inches in diameter), where the temperature was 780 °C. Ar (200 standard cubic centimetres per minute (sccm)) and O2 (50 sccm) were used as carrier gases to reduce the oxygen vacancies. α-MoO3 bulk crystals were collected at the downstream position where the temperature was ~550 °C. Nanometer-thick α-MoO3 flakes were mechanically exfoliated and then transferred to a SiO2 (300 nm)/Si substrate.

### Reversible hydrogen intercalation/deintercalation processes

The hydrogen intercalation process was conducted by a plasma-enhanced CVD method. Typically, α-MoO3 flakes on a SiO2/Si substrate (1 × 1 cm2) were placed in the middle of a quartz tube (four inches in diameter), where the temperature was 150 °C. A radio-frequency plasma generator (17 mW, 13.56 MHz, VERG-500) was placed upstream of the sample at a distance of ~ 50 cm. H2 (5 sccm) was used as the gas source under a pressure of 15 mTorr. By controlling the reaction time, hydrogen-intercalated flakes with various hydrogen levels were obtained. The deintercalation experiment was carried out by a thermal treatment process at 350 °C for 120 min. Ar (200 sccm) and O2 (50 sccm) were used as carrier gases at atmospheric pressure to reconfigure the crystal structure and reduce oxygen defects. Notably, to obtain HxMoO3 flakes with different hydrogen distributions, two intercalation strategies were carried out: continuous intercalation and step-by-step intercalation.

### Spatially controllable hydrogen intercalation process

Spatially controlled hydrogenation in nanometer-thick α-MoO3 flakes was realized by combining intercalation with photolithography. First, a layer of photoresist (~1 μm, AZ 1512, AZ Electronic Materials) was coated on top of the surfaces of α-MoO3 flakes and then patterned by direct-write lithography (SF100 XPRESS). After the development process, the unexposed regions of α-MoO3 flakes were covered and protected by the photoresist. Then, a similar hydrogenation process was conducted, during which hydrogen was inserted into the exposed regions of flakes. Following a lift-off process, selectively intercalated flakes (i.e., in-plane H-MoO3/α-MoO3 heterostructures) with designed geometries were obtained.

### Theoretical calculations

We investigated the adsorption of hydrogen on α-MoO3 using DFT calculations as implemented in the Vienna ab initio simulation package. The Perdew-Burke-Ernzerhof form of the generalized gradient approximation was used to describe electron exchange and correlation. The plane-wave kinetic energy cutoff was set to 600 eV. A semiempirical function developed by Grimme (DFT-D3) was employed to describe dispersion forces41. All structures were relaxed until the ionic forces were < 0.01 eVÅ−1. For accurate calculations of the electronic structures, we used 9×3×9 Γ-centered grids for sampling the Brillouin zone to calculate the preferred adsorption site of one hydrogen on α-MoO3 and 5 × 5 × 3 Γ-centered grids to simulate the hydrogen diffusion process in a 6 × 1 × 6 α-MoO3 supercell.

### Characterizations

Optical observations and Raman tests were conducted using a confocal microscope system (WITec, alpha 300 R). TEM and SAED images were obtained from a FEI Tecnai T20. HAADF-STEM images were acquired by a JEOL JEM-ARM300F TEM. XRD and XPS spectra were collected by a Bruker D8 Advance and a Nexsa surface analysis system (Thermo Scientific), respectively. Surface topography and near-field images were captured using a commercial s-SNOM (NeaSNOM, NeaSpec GmbH) setup with an atomic force microscope tip (NanoWorld, Arrow-NCPt) in tapping mode (~270 kHz frequency and ~70 nm amplitude). To eliminate edge-excited polaritons, flakes were rotated till the incident polarization was parallel to the edges.