Electrically reconfigurable polarization control with double tri-layer black phosphorus heterostructures Authors: Samuel K W Seah, Souvik Biswas, Claudio U Hail, George R Rossman, Harry A Atwater Published: 2026 ABSTRACT AbstractAccessing diverse polarization states across the Poincaré sphere via electrical control is highly desirable in optical communications, bio-imaging and quantum information processing, where fast continuous switching, compactness, and ease of integration into photonikic circuits are essential. Layered anisotropic 2D materials enable polarization modulation in a fast and compact manner, thereby overcoming the speed and size limitations of liquid-crystal based spatial light modulators and electro-optic Pockels cell modulators. Here, we report a Fabry-Perot cavity that integrates two cross-aligned black phosphorus layers, with each layer being independently gated to access the inherently two-dimensional range of polarization states across the Poincaré sphere surface. Our heterostructure design predicts electronic access to 86% of the Poincaré sphere at its S-band operating wavelength. We experimentally validate the design by fabricating and testing such a device in reflection, where we demonstrate independent, two-parameter electronic control of the output polarization.Subject terms: Nanophotonics and plasmonics, Metamaterials Introduction Polarization is an essential property of light that plays a vital role in the nature of light propagation in media. The ability to control polarization states on-demand has applications ranging from optical telecommunications, bio-imaging, and quantum information processing1–3. In these applications, multiplexing operations are used to efficiently encode information in the polarization basis, so there is a need for polarization modulators that can access the entire Poincaré sphere at high speeds, while maintaining a low form factor4–8. Conventional polarization modulators consist of either liquid-crystal based spatial light modulators that are limited to kHz speeds, or electro-optic modulators based on the Pockels effect that are generally bulky or lack access to the full Poincaré sphere9–12. Tunable anisotropic 2D materials, such as MoO3, ReS2, and black phosphorus (BP), offer an alternative and highly promising platform to control polarization13–16. Their atomically thin nature, along with a lack of lattice matching constraint due to its van der Waals nature, allows for easy integration into photonic circuits, while their carrier-mediated tunable optical properties allow for electronic control of polarization at GHz speeds17–19. BP in particular possesses a thickness dependent bandgap (750 nm to 4 μm) that covers the near-infrared telecommunications band, making it an attractive candidate for polarization tuning applications20. However, existing polarization modulators exploiting BP have so far been limited to traversing a 1D trajectory along the Poincaré sphere as a function of gating voltage21. This limitation results from gating a single BP sheet, which offers only a single degree of freedom, whereas polarization states on the surface of a normalized Poincaré sphere require minimally two degrees of freedom to access an arbitrary state of polarization. BP devices to date cover a limited range of polarization states on the sphere, as the operating wavelength is shifted to change the degree of birefringence imparted by the BP. Further comparisons to past works are included in Supplementary Notes 17 and 18. For most applications, however, monochromatic light at a fixed incident wavelength is preferred, where the output polarization state depends solely on actuations applied to the device itself. This requires a second voltage control parameter to access a broader range of states on the Poincaré sphere. While this two-degrees-of-freedom approach to controlling a two-dimensional parameter output has been explored in metasurfaces and liquid-crystal devices22,23, active photonics devices with 2D materials have been limited to single controls and single active layers, limiting functionality. Unlike thicker materials like indium tin oxide or liquid-crystals, atomically thin 2D materials cannot possess a large charge carrier density difference on its upper and lower interfaces without breaking down. This can be overcome by the use of two separate anisotropic 2D material layers separated by dielectrics, with two additional gating contacts sandwiching the 2D layers, to achieve a two-parameter control with wide dynamic tuning range. To this end, we have designed a dual-gated BP heterostructure composed of two orthogonally aligned BP layers, with each layer separately and independently gated. The heterostructure was incorporated into a Fabry-Perot cavity to maximize the light-matter interaction of the exciton. Furthermore, the cavity was designed to be marginally over-coupled at the charge neutral point, with the prospect of accessing the critically coupled and under-coupled regimes through gating. This allowed for access of up to 86% of all possible polarization states on the Poincaré sphere. We fabricated a proof-of-concept device and tested it to validate our design, displaying a two-parameter electronic control over polarization on the Poincaré sphere. This 2D heterostructure dual-gating approach can be generalized to allow for active simultaneous modulation of multiple constitutive properties of light, and for generating higher dimensional structured light. Results Simulations Figure 1a illustrates the double tri-layer BP polarization modulator design. It consists of a Fabry-Perot cavity, with reflective gold mirrors, incorporating a pair of three-atom thick BP (3L-BP) layers that have their active armchair (AC) axes orthogonally aligned (twist angle = 90°). Hexagonal boron nitride (hBN) acts as a dielectric medium between the gold and 3L-BP layers, allowing the top and bottom BP layers to be capacitively gated, independently of each other. hBN is also the primary optical medium of the cavity, which is set to a thickness of λ/4n, where λ is the operating resonance wavelength within the telecommunication S-band, and n the refractive index of hBN. The relative thicknesses of the three hBN layers are tuned such that there is significant spectral overlap between the electric-field resonances along both principal axes (x, y) of the device (which the top and bottom BP layers’ AC axes are aligned to, respectively). The central hBN layer is also thicker than the sum of the two outer hBN layers to prevent electrical breakdown at the maximal top and bottom operating voltages. Finally, a top layer of PMMA is added to give some adjustability to the cavity’s resonant wavelength, and to also further encapsulate the oxygen-sensitive BP during fabrication (see supplementary Fig. 1). Schematic and performance of an ideal dual-gated tri-layer black phosphorus (3L-BP) device (not to scale). a Schematic for the dual-gated BP device incorporating two orthogonally aligned BP layers (twist angle = 90°) into a critically coupled Fabry-Perot cavity. The crystallographic orientation of the active armchair axes of the BP layers is labelled by the white ‘AC’ arrows. Using hexagonal boron nitride (hBN) as spacer layers, each BP layer is capacitively gated against its closest gold contact (Au) to control its doping density (nBP-top & nBP-bot), allowing a two-parameter control over the output polarization state. A Poly(methyl methacrylate) (PMMA) layer is spun on top to give some adjustability to the cavity’s resonance wavelength during fabrication, and for additional encapsulation. The lab coordinate axes are in the bottom left corner, with the positive x-direction (aligned with the AC axis of the top 3L-BP layer) defining S1 = 1. b, c Calculated 1D trajectory across the normalized Poincaré sphere when the top (bottom) BP is gated while keeping the other BP layer grounded. d Combined normalized Poincaré sphere coverage when gating both BP layers in concert. The top (bottom) BP doping density (nBP-top(bot)) is represented in border (fill) color of the plot points (from black to peach). The optical response of the device is modelled using the transfer matrix method (details of which are in Methods). We assume an illumination with 45° linearly polarized light at normal incidence and probe the device in reflection. At the operating wavelength of 1479 nm, when only the top or bottom 3L-BP layer is gated one at a time, two independent 1D trajectories across the Poincaré sphere are formed (Fig. 1b, c). When gating both BP layers in concert, their respective 1D trajectories are multiplicatively combined, enabling a wide 2D range of output polarization states covering 86% of the Poincaré sphere (Fig. 1d). The broad Poincaré sphere coverage can be understood by looking at the amplitude and phase spectra of the output electric field along the principal x and y directions (Fig. 2a, b). The top gold mirror thickness was designed such that at low carrier densities (without any gating), the device is slightly overcoupled, as seen by the solid curves in Fig. 2a, b. As the gating voltage is increased, the carrier density increases, which screens the excitons present in the BP, thereby weakening its oscillator strength. This results in a reduction in the intra-cavity absorption, which causes the device to pass the critically coupled point and enter the undercoupled regime at higher carrier densities. By judiciously choosing the operating wavelength (1479 nm), we can find some intermediate combination of the top and bottom carrier densities (nBP-top and nBP-bot) where the device is critically coupled. Figure 2c shows the S0 parameter map as a function of nBP-bot and nBP-top. Since the S0 parameter maps the intensity of the reflected beam, the point where the BP layers are critically coupled to the cavity is determined from the minima of the plot. Around this point, both the phase and amplitude of the Ex and Ey fields become extremely sensitive to small changes in doping density. As a result, the azimuthal and ellipticity gating maps (Fig. 2d, e) show large variations about this critically coupled point, accounting for the broad polarization state coverage seen earlier in Fig. 1d. We also observe that the critically coupled point manifests as a saddle point in the azimuthal map and as a zero in the ellipticity map. This is a direct consequence of the optical phase singularity resulting from the critically coupled resonator24,25. Simulated performance of a critically coupled dual-gated BP device. a, b Reflected amplitude and phase response of the device. The solid blue (orange) lines represent the reflected amplitude/phase of the electric field along the armchair direction of the top (bottom) BP layer when no gating is applied. The dotted blue and orange lines are the same reflected amplitude/phase at maximum gating (limited by the dielectric breakdown of the hBN spacer layers). The operating wavelength of the device is indicated by the dotted black lines. Good spectral overlap between the Ex and Ey curves is obtained by carefully choosing the thicknesses of the top and bottom hBN layers. c S0 parameter map of the device at the operating wavelength, as a function of the top and bottom carrier densities. The S0 minima (marked with the white cross) correspond to the critical coupling point where the device flips between the under-coupled and over-coupled regimes. d, e Respective azimuthal and elliptical gating maps as a function of top and bottom carrier densities. The critical coupling point, marked with the black cross, is visible as a saddle point in the azimuthal plot and as a zero in the elliptical plot. Around this point, polarization tuning is maximized. Experiments The double 3L-BP polarization modulator was fabricated by performing the hot pick-up technique twice26, followed by electron beam lithography and evaporation for the top gold mirror (see “Methods”). An optical image of the fabricated device is shown in Fig. 3a. The bottom 3L-BP layer was connected to a thicker auxiliary BP flake that electrically connects it to a gold finger contact. Similarly, the top 3L-BP layer was attached to a thin graphene ribbon to connect it to its nearest gold contact (see supplementary Fig. 2). After each heterostructure transfer step, Raman microscopy was performed on the BP flake to determine its crystallographic orientation (Fig. 3b). By fitting a cos² θ function to the polarization sensitive Ag2 Raman peak, the relative armchair separation angle between the two 3L-BP layers was found to be 84° (see “Methods”). After AFM determined the top gold mirror thickness to be 15.5 nm, a thin 460 nm layer of PMMA was spun on top of the device. Fabrication and experimental results of the dual-gated BP device. a Optical image of the device, with the top gold mirror, top BP layer, and bottom BP layer outlined in purple, blue, and orange, respectively. Inset shows the cross-sectional schematic. The top BP layer makes electrical contact with the north-western gold finger via a graphene ribbon (not shown), while the bottom BP layer is contacted by the western gold finger by an attached 6L-BP (not shown). Coordinate system for polarization basis is shown in the bottom right, with S1 = 1 corresponding to the positive x-direction. White scale bar represents 20 μm. b Raman measurements to determine the AC axis of the two BP layers, aligned to fit the optical image in (a). Blue (orange) data points demarcate the Raman intensity of the Ag2 peak of the top (bottom) BP layer, with solid lines used to fit the data points. The AC axis separation of the two BP layers was determined to be 84°. The black double headed arrow marks the input polarization angle of 17° used in subsequent measurements. c Normalized reflectance of the device at different input polarization angles, showing a minimum at 1470 nm. The small difference between the curves verify that the orientation of the two BP flakes was near-orthogonal. d, e Spatial maps showing changes in azimuthal and elliptical coordinates when spectrally tuning the operating wavelength from 1463 nm to 1483 nm. The black box marks the pixel where the device was found to be active along both coordinates, and was further investigated via dual-gating. The white scale bar represents 5 μm. Passive optical characterization was performed on the device by first acquiring reflectance spectra at the center of the two overlapping 3L-BP flakes (Fig. 3c). Multiple input polarization angles were used to probe the optimal input angle where reflectivity was minimized, which corresponds to the angle bisector of the top and bottom BP AC axes, where the polarization modulation depths are maximized. It was observed that the input linear polarization angle had little effect on the reflectance minima (5.9% vs 6.6%). Since the BP exciton absorbs most strongly along the AC axis, this was further confirmation that the two 3L-BP flakes were indeed almost orthogonally aligned. Spectral reflectance data for a full polarization angle sweep is available in the supplementary note 3. We also note that the optical modulation efficiency reported here is fundamentally limited by the lossy gold mirrors of the Fabry-Perot cavity, but can be improved by replacing them with distributed Bragg reflector mirrors instead, as reported in ref. 27. After determining the optimal input linear polarization angle giving the lowest reflectance minimum, spectral polarization tuning maps were acquired by mapping the area of interest at wavelengths below and above the reflectance dip (1463 nm and 1483 nm, respectively), and taking the azimuthal angle differences (Fig. 3d) and elliptical angle differences (Fig. 3e) of these spatial maps at differing operating wavelengths. This allowed us to pinpoint areas of the device that were active along both the azimuthal and elliptical coordinates for gating (marked by the black square in Fig. 3d, e). The non-uniformity of the maps vis-à-vis the 3L-BP flake locations comes from blisters, bubbles, and/or impurities between the flakes, which occur during the stacking process. This was expected for the 5-stack heterostructure, as spatial polarization variations within a singular BP flake have been observed in less complex heterostructures21,28,29. Next, active gating was performed on the device while optically probing the pixel of interest at an operating wavelength of 1470 nm. Voltage maps of the output azimuthal and ellipticity angles are shown in Fig. 4a, b. The color gradient in both plots advances in nearly orthogonal directions, indicating the independent polarization control exerted by Vtop and Vbot; Vbot predominantly controls the azimuthal angle, while Vtop exerts more control over the ellipticity angle. Figure 4c represents the output polarization data on the ellipse parameter plane as a function of doping density. Here, the same pattern of nBP-bot controlling the azimuthal angle and nBP-top controlling the ellipticity angle can be viewed as gradients of the border and fill color, respectively. Further gating experiments were also performed to determine the presence of gating hysteresis, and to confirm the optimal operating wavelength for maximal polarization control (see Supplementary Note 4). An additional device was also fabricated to show reproducibility of the two-parameter tuning results (see Supplementary Note 9). Active polarization control across the Poincaré sphere. a, b Measured azimuthal and elliptical coordinates as a function of top and bottom gate voltage at an operating wavelength of 1470 nm. The color gradients advance in different directions, indicating a two-parameter degree of freedom in controlling output polarization. c Normalized Poincaré sphere plot showing 2D polarization coverage over Poincaré sphere. The top (bottom) BP doping density is represented in border (fill) color of the plot points (from black to peach). Discussion To evaluate the device performance against simulations, a black phosphorus model from ref. 27 was used to determine the expected reflectivity and Poincaré sphere coverage for the given device parameters. The exciton resonance for black phosphorus along the AC direction was modelled as a Lorentzian, parameterized by an exciton resonance frequency ω0, resonance linewidth Γ, and oscillator strength f0, with an optical conductivity given by: where σ0 is the universal conductivity. The BP carrier density was calculated using a simple capacitance model: where e is the elementary charge value, and dhBN is the thickness of the gate dielectric. Here, the maximum applicable potential difference V is limited by the dielectric breakdown of hBN30. Experimentally obtained values from ref. 21 were used to correlate calculated values of nBP to f0, ω0, and Γ. The complex permittivity, dominated by interband absorptions, can then be calculated using the sheet conductivity via: with the high-frequency limit permittivity ε∞ being 12.5 in the AC direction and 10.8 in the ZZ direction31, and dBP being the thickness of the BP used (0.53 nm per layer)32. xPerforming transfer matrix calculations with the modelled BP resulted in a reflectance minimum of 2.8%, compared to the experimentally measured 5.9% (Fig. 5a). In this model, the only source of optical loss within the cavity stemmed from absorption losses in the BP, with no loss between interfaces or through the hBN dielectric. However, multi-layered heterostructures often exhibit stacking imperfections and contamination, visible as blisters or bubbles, that introduce additional loss33,34. Small variations in hBN thickness across the finite illumination spot size are also expected to introduce a spectral broadening effect that increases both reflectance and linewidth. These structural imperfections were further incorporated into the model as an imaginary refractive index k of 0.006 in the hBN, simulating a distributed loss throughout the cavity. This resulted in a close reflectance spectrum fit between simulations and experiment. Consequently, at this degree of coupling, good agreement of Poincaré sphere coverage afforded by gating was also found (Fig. 5b). The discrepancy between the simulated coverage and experimental data is attributed to device non-idealities, and is further expounded upon in Supplementary Note 11. Comparison of experimental results to simulations, with projected potential Poincaré sphere coverage. a Spectral reflectance comparison of experimental results (black) to simulations (colored). Additional small intracavity losses (corresponding to a k-value of 0.006 in the hBN) were introduced into the simulation model to match reflectance values (orange solid). Due to these losses, the optimal top gold mirror thickness to achieve critical coupling is considerably lower than initially anticipated (blue dashed). b Poincaré sphere coverage comparison, with experimental results represented by the peach/black plot points, and simulation coverage shown by the black borders. Simulations here assume small cavity losses with a 15.5 nm top gold mirror thickness (corresponding to the orange solid curve in (a)). c Projected Poincaré sphere coverage, assuming small cavity losses, with a 11.0 nm top gold mirror thickness (corresponding to blue dashed curved in (a)). Operating under the critically coupled condition at 1506.8 nm allows for electronic access of up to 70% of the Poincaré sphere. A consequence of the higher-than-expected optical loss was the degree of undercoupling in the cavity. The deposited 15.5 nm of gold was aimed at achieving near critical coupling, assuming no additional cavity losses, but simulations indicate that with these losses, a significantly thinner 11.0 nm layer was sufficient. At this optimal thickness, up to 70% of the Poincaré sphere is accessible through gating alone, even with the added intracavity loss (Fig. 5c). Further analysis on the Poincaré sphere coverage as a function of cavity loss is included in Supplementary Note 10. We note that these cavity losses are difficult to estimate a priori, and that the deposited gold thickness cannot be adjusted without redoing the 2D heterostructure. Nonetheless, the fit between simulations and experiment confirms the suitability of the BP parameters used in the model and suggests that the quality of the carrier contacts and 2D materials used was sufficient in potentially achieving comprehensive two-parameter polarization control. Further improvements to the device’s circuit design to minimize resistance and parasitic capacitance for high-speed modulation are possible, but remain beyond the scope of this work (see refs. 17–19 for ns-scale rise and fall times in BP). To summarize, a proof-of-concept, electronically-modulated polarization control device capable of accessing a range of states across the Poincaré sphere was fabricated and tested. This device improved over previous BP polarization modulators by achieving a 2D coverage over the Poincaré sphere through electronic gating alone without the need to change the operating wavelength. Therefore, our work demonstrates independent, active two-parameter control via excitonic tuning in an optical cavity. While the small polarization tunability in the fabricated device is attributed to cavity undercoupling, simulations confirm that with an optimal top gold mirror thickness enabling critical coupling, up to 86% of the Poincaré sphere can be accessed with this design. More importantly, our results underscore the potential of 2D heterostructures possessing multiple degrees of freedom imparted by independently controlled gates. To illustrate this further, a device design enabling complex amplitude/360° phase modulation is presented in Supplementary Note 12. Spatial patterning of the gates can also confer an additional degree of control that enables beam steering and polarization beam splitting27, opening further prospects to structuring light with a fast and compact modulator. Methods Device fabrication SiO2 (285 nm)/Si chips were sonicated in acetone and isopropanol for 15 min each, then placed in an oxygen plasma asher (100 W, 300 mTorr, 10 min). hBN was exfoliated onto these chips at 100 °C. hBN flake thickness was measured using AFM, and its optical path thickness determined by FTIR using a transfer matrix model. The optical path thickness could then be used to estimate the refractive index of hBN. BP was exfoliated onto polydimethylsiloxane (PDMS) stamps in an inert N2 glovebox environment (with <0.1 ppm of water and oxygen inside). Their thickness was estimated using contrast microscopy, using a background compensated formula for measuring contrast (3.7% contrast per layer of BP, see Supplementary Note 5). The zigzag axis of BP was estimated by looking for microscopic tears along a preferential direction of the exfoliated BP flake, since BP preferentially fractures along the zigzag axis21. The zigzag orientation was confirmed using Raman microscopy once the BP was encapsulated in hBN and taken out of the glovebox. Gold contacts were printed using electron beam lithography (100 kV EBPG 5000+, 200 nA) followed by electron beam evaporation (5 nm Ti, 110 nm Au, 1 Å/s, <8 × 10−8 Torr). The heterostructure was transferred onto the contacts using the hot transfer technique26, where flakes were picked up using a polycarbonate film on PDMS between 80 and 110 °C, and dropped off onto the substrate at 180 °C. The film was then washed off by soaking in chloroform for 4 h, followed by soaking in IPA for 1 min. The final heterostructure was built in two transfers, with the first consisting of the bottom BP layer encapsulated between two hBN flakes, and the second consisting of the top BP layer with the top hBN flake. This allowed for the bottom BP layer to be analyzed using Raman microscopy first before it was covered by the top BP layer. The top gold mirror was printed using electron beam lithography (0.1 nA) and evaporation again (1 nm Ti, 15.5 nm Au). The final thickness of the top gold mirror was confirmed using AFM, before the device was spin-coated in PMMA. The PMMA thickness was estimated using a commercial film thickness measurement tool (Filmetrics paired with an optical microscope) on areas of the chip with only PMMA on SiO2/Si. For actively gated devices, a window was opened in the PMMA along the gold contact pads by swabbing carefully with acetone around the pads. This allowed wire bonding of the pads to the chip carrier. Raman spectroscopy A commercial polarization-resolved Raman microscope (Renishaw inVia Raman) was used to confirm the area and crystallographic orientation of the transferred BP flakes. The device was placed on a rotatable stage and analysed with a 514 nm laser at 5° intervals for the Ag2 peak intensity at approximately 467 cm−1. The data is then fit to a cos2 θ function to determine the armchair/zigzag axes, since the mode peak intensity occurs along the in-plane atomic motions of the armchair axis35. Measurement configuration The full schematic of the optical experimental setup is shown in Supplementary Note 6. A tunable near-infrared laser (Santec TSL-210) was used to cover the wavelength range (1410–1520 nm) used in optical measurements. A wire-grid polarizer and half-waveplate mounted on a rotatable stage controlled the input linear polarization angle. The laser was focused onto the sample using an infinity corrected objective (Mitutoyo, 20× magnification, NA = 0.40), then imaged onto a near-infrared camera (Xenics) and polarimeter (PAX1000IR2) using a tube lens. Spatial mapping was achieved by using a motorized translation stage for the sample (MT3-Z8). Gating was done by using two Keithley 2400 sourcemeters, linked to a common ground. Python scripts were used to automate data acquisition. All measurements were normalized against a gold reference area of the device composed of 460 nm PMMA on 110 nm gold. Transfer matrix model For a N-layer vertical stack, the transfer matrix for each individual ith layer is given by ref. 36: where q is the wavevector, and z is the layer thickness. The total transfer matrix relating the transmission and reflection coefficients is given by: where the complex reflection and transmission coefficients can be calculated via: For illumination at a polarization angle θ relative to the AC axis, the overall Jones matrix can be obtained by applying a rotational transform: where rAC/ZZ and φAC/ZZ are the amplitude and phase pickup along the armchair/zigzag direction, respectively, calculated via the transfer matrix method above. For Poincaré sphere plots, the Stokes parameters are calculated from the Jones matrices using where Ex=Axeiϕx, Ey=Ayeiϕy. The Stokes parameters are further normalized by dividing each parameter by S0. Ellipse parameters are evaluated from the Stokes parameters via the following relations: where atan2 is the four-quadrant inverse tangent. Where relevant in the manuscript, Poincare sphere representations were used to illustrate the extent of polarization coverage over the orthonormal basis, while ellipse parameters were used to more explicitly show the dependence of polarization on the tuning parameters (nBP-top/bot). Supplementary information Supplementary Information Transparent Peer Review file Acknowledgements We gratefully acknowledge the critical support and infrastructure provided for this work by the Kavli Nanoscience Institute at Caltech. Author contributions S.K.W.S., S.B. and H.A.A. conceived the project. S.K.W.S. performed the simulations, fabricated the devices with support from S.B., performed experiments, analyzed the data, and drafted the manuscript. S.B. designed the device and did preliminary testing on early test devices. C.U.H. advised on fabrication and prepared the optical setup. G.R.R. supervised the Raman measurements. H.A.A. supervised the project. All authors discussed the results and edited the manuscript. Peer review Peer review information Nature Communications thanks Thomas G. Folland, who co-reviewed with Nihar Sahoo; and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available. Funding The authors disclose support from the Department of Energy-Office of Science under grant DE-FG02-07ER46405. S.K.W.S. discloses support from the DSO National Laboratories Postgraduate Scholarship. S.B. discloses support from the Department of Energy Office of Science under grant DE-FG02-07ER46405. C.U.H. discloses support from the Swiss National Science Foundation through the Early Postdoc Mobility Fellowship grant #P2EZP2_191880 and the Postdoc Mobility grant #P500PT_214452. Data availability Relevant data supporting the key findings of this study are available within the article and the Supplementary Information file. All raw data generated during the current study are available from the corresponding authors upon request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-73458-6. References Associated Data Supplementary Materials Supplementary Information Transparent Peer Review file Data Availability Statement Relevant data supporting the key findings of this study are available within the article and the Supplementary Information file. All raw data generated during the current study are available from the corresponding authors upon request.