Electrically modulated plasmonic metasurfaces for light communication

Abstract

Electrically modulated metasurfaces manipulate light fields but suffer from high operating voltages, low tuning sensitivity, and a reliance on telecommunication bands, limiting their applications in light communication (LC). Here, we demonstrate electrically modulated plasmonic metasurfaces that enable continuous and reversible wavelength modulation with a tuning sensitivity up to ~ 1 nm/V at a CMOS-compatible voltage below 5 V. These metasurfaces consist of dimethyl sulfoxide (DMSO) immersed metal nanoparticle lattices and Au electrodes on transparent conductive oxide (TCO)/quartz substrate. Through simulations and experiments, we reveal that the wavelength shift is synergistically governed by the refractive index variation of the DMSO superstrate and the Seebeck effect of the TCO layer, which is further amplified by the lattice mode. Further, we propose two LC applications of these tunable metasurfaces: single-mode spectral shifting for image information transmission and multimode spectral shifts for a 1×3 encoder. The device paves the way for applying metasurfaces in optical communication and optoelectronic circuits.

Electrically modulated metasurfaces manipulate light fields but suffer from high operating voltages, low tuning sensitivity, and a reliance on telecommunication bands. This work shows designs of electrically modulated plasmonic metasurfaces that enable continuous and reversible wavelength modulation with a tuning sensitivity up to ~ 1 nm/V at a CMOS-compatible voltage below 5 V in the visible and near-infrared light regime.

Introduction

Light communication (LC, visible to near infrared wavelength) offers significant advantages over traditional communication bands, including an expansive bandwidth for high-speed data transmission, enhanced security from wall-contained signals, minimized electromagnetic interference, and seamless integration with existing LED lighting infrastructure^1–5. Metasurfaces composed of nanostructure arrays can serve as efficient, compact, and adaptable light manipulators, possessing significant potential for LC systems^3,4,6–8. Surface lattice resonance (SLR) metasurfaces, consisting of arrays of metal nanoparticles (NPs), combine characteristics of both localized surface plasmons and diffractive modes to support strong near fields and ultranarrow resonances^9–15. SLRs can realize high quality factor (Q-factor, the ratio of the resonance wavelength to its full width at half maximum (FWHM) in the transmission or reflection spectrum) resonance cavities for nanolasers, exhibiting great potential in LC applications^16–19. The resonance wavelength of SLRs can be designed across ultraviolet to near infrared ranges depending on structural parameters, including types of metal, NP geometries, lattice structures, and the refractive index (RI) environment²⁰. However, structural parameters of fabricated devices are difficult to change, which limits their adaptability for applications in optical communication, such as electro-optic modulators and tunable nanolasers. Microfluidic channels and stretchable substrates can be employed to create tunable SLRs, but the involved moving components will constrain their applications to optoelectronic integration^21,22.

Tunable metasurfaces without moving components have been achieved through active materials and applying external stimuli, such as electrical modulation, temperature changes, humidity variations, etc.,^23–26. In particular, electrically modulated metasurfaces can achieve the wavelength shift with precisely controlled voltages and integrated with electronics to realize optoelectronic integrated circuits. Higher environmental robustness and faster modulation speeds than humidity and temperature modulation methods make electrical modulation promising for the next generation of light communication applications²⁷. Phase-change materials exhibiting non-volatile refractive index changes enable the reconfigurable metasurfaces but suffer from an absence of continuous refractive index tuning and a limited number of operation cycles^28–31. Lithium niobate materials benefiting from the Pockels effect have been employed to develop electro-optic modulators characterized by high modulation bandwidths, rapid modulation rates (up to 100 GHz), complementary metal oxide semiconductor (CMOS) voltage compatibility, and low optical losses in the telecom regime, where wavelength modulation in the light band remains unachieved^32,33. Organic electro-optic (OEO) materials exhibit an electro-optic coefficient nearly three times greater than that of lithium niobate, but still suffer from the relatively high operation voltage ($\pm$17 V) and low tuning sensitivity (up to 0.16 nm/V)^34–37. Electrically modulated silicon metasurfaces with resonant wavelengths in the visible and near-infrared regimes have been realized using the photothermal effect of silicon, with modulation voltages below 5 V. However, its operating temperature is around 200 °C, which may affect the performance of other materials and limit the integration with CMOS devices³⁸.

In this work, we demonstrate electrically modulated plasmonic metasurfaces to achieve wavelength tuning sensitivity (defined as the wavelength shift divided by the operation voltage) up to 1 nm/V and reduce the operation voltage to a CMOS-compatible level below 5 V. We designed Ag and Al NPs on a TCO layer with high-quality SLRs in the visible and near-infrared light regimes (380–1000 nm). These metasurfaces show continuous and reversible electrical modulation of wavelength due to the change in RI of DMSO solution and the Seebeck effect of the TCO layer, as well as the amplification effect of SLR modes. All resonant modes are blue-shifted upon voltage applying with different amounts of shifts. To elucidate the electrical modulation mechanism, we simulated voltage-dependent temperature distributions and transmission spectra under varying RIs of DMSO and indium tin oxide (ITO). Moreover, metasurfaces incorporating diverse metal NP lattices on distinct TCO/quartz substrates confirm the universality of this strategy. Further, we demonstrate two proof-of-concept LC applications of these tunable metasurfaces: wavelength-selective image information transmission via single-mode shifts and optical encoding using multimode spectral modulation. Table S1 lists the device performance of electrically modulated metasurfaces using different active materials, key parameters including the wavelength, resonance linewidth, operation voltage, tuning sensitivity, modulation depth, and modulation speed. The proposed electrically modulated plasmonic metasurface shows a high tuning sensitivity of 1 nm/V at a CMOS-compatible voltage of below 5 V in visible and near-infrared regimes. Furthermore, the device operates at a relatively low temperature (below 60 °C), minimizing the risk of damage to integrated components.

Results

Device and application design of the electrically modulated metasurface

Figure 1 depicts the principal configuration of the tunable plasmonic metasurface and applications of image transmission and encoding. Plasmonic metasurfaces consist of Ag or Al NP lattices on the TCO/quartz substrate, two Au electrodes, and a drop of DMSO solution on the lattices (Fig. 1a). The TCO layer is introduced to generate waveguide-SLR hybrid modes while functioning as a conductive layer (please see Fig. S1–S5). The two Au electrodes serve as the source and drain for the bias input. The resonance peak position can be modulated by the external electric signals in real time. Figure 1b schematically shows changes in the device with the voltage applied. Upon the application of voltage, the heat generated by TCO leads to a temperature increase in DMSO, resulting in RI decrease of DMSO. Simultaneously, some electrons from the bottom of the TCO layer moved towards the interface with NPs as a result of the Seebeck effect, leading to a decrease in the RI of TCO. These electrically induced effects synergistically lead to a blue shift of all resonance wavelengths. The device exhibits the “0” state in the absence of voltage and transitions to the “1” state with wavelength shifts upon applying voltage, and removal of the voltage reverts it back to the “0” state. Further, two LC applications are demonstrated, where a single resonant mode can be used for image transmission, and multimode resonance with different shifts can serve as an encoder, respectively (Fig. 1c, d). By decoding the transmission spectra shifts of any of the modes in the metasurface into binary code, image information is transmitted (Fig. 1c). Metasurface featuring multimode resonances can exhibit different shifts at identical voltage, a 1 × 3 encoder with one input port of bias and three output ports of spectral shifts of three modes is proposed (Fig. 1d). Figure 1e summarizes the state-of-the-art electrically modulated metasurfaces utilizing active materials characterized by distinct operation voltages and tuning sensitivities, where the red star denotes our study with high tuning sensitivities (up to 1 nm/V) at the CMOS-compatible voltages (below 5 V)^{30,34–37,39–47}.

Fig. 1. Schematic illustration of the device structure, operation principle, application demos, and a summary of electrically modulated metasurfaces.

a The schematic structure of the electrically tunable plasmonic metasurface device. The device consists of metal NPs lattices, two Au electrodes on the TCO/quartz substrate, and a DMSO film on the lattices. White light incident is used to excite optical modes, and the red and blue light indicate the shift of transmission spectra. b Schematic diagram of the device without and with the voltage applied, movement of electrons, and corresponding spectrum shift. c Tuning a single resonant mode for the image information transmission. d Multimode manipulation for information encoding. e Operation voltages and tuning sensitivities of reported electrically modulated wavelength based on metasurfaces.

Preparation and characterization of the device

We first illustrate the electrical modulation using Ag NP lattices on an ITO/quartz substrate (Fig. 2). The metasurface consists of Ag NP lattices with 500 nm of array periodicity and 130 nm of diameter, and 80 nm of height on a 150 nm thickness of ITO (Fig. 2a). Ag NP lattices are prepared through modified solvent-assisted nanoscale embossing processes (please see Fig. S6). Figure 2b, c show scanning electron microscope (SEM) images of the top view and cross-sectional morphology of the metasurface. The inset shows the photo of the device with the size of the Ag NP lattices of 4 × 4 mm and the size of Au electrodes of 2 × 8 mm. The separation between the Ag NP lattices and Au electrodes is approximately 5 mm. The transmission spectrum of the metasurface shows multiple waveguide-SLR modes due to the addition of the ITO layer (Fig. 2d). Two modes with wavelengths at 750 nm and 790 nm (gray dashed lines) are defined as mode 1 and mode 2, respectively. Mode 1 and mode 2 exhibit blue shifts of 0.92 nm and 4.58 nm (Fig. 2e, f) at a bias voltage of 5 V, respectively. Temperatures of the device at various voltages are measured using an infrared thermal imaging camera (please see Fig. S7), which shows variations of 21.3 °C and 30.5 °C at the operation voltages of 3 V and 4 V. To further understand the temperature variation, we simulated the temperature distribution of the device at the bias voltages of 0 V, 3 V, and 5 V (Fig. 2g, h and please see Figs. S8–9), which are higher than the measured ones. This discrepancy may arise from the inability to accurately determine setup parameters, such as the thermal conductivity of the sample stage. Further, we simulated the temperature distribution of the sample in air and on the foam substrate to further confirm results on the stage (please see Figs. S10–11). We believe that the increase in temperature due to the applied voltage leads to a decrease in the refractive index of DMSO, as well as the gradient distribution of the temperature generates a thermoelectric field due to the Seebeck effect.

Fig. 2. Device preparation and temperature analysis of Ag nanoparticle lattices on indium tin oxide/quartz substrate.

a Schematic illustration of device structure and Ag NP lattices. b SEM image of the top view of Ag NP lattices. The inset shows a photo of the device. c Cross-section SEM image of Ag NP lattices. d Measured transmission spectra of Ag NP lattices on ITO/quartz substrate. The gray dashed lines label resonances as mode 1 and mode 2. The variation of shifts with temperature from 11 °C to 45 °C of e mode 1 and f mode 2. g The simulation of the temperature distribution of the device with a bias voltage of 0 V, 3 V, and 5 V. h Simulated local temperatures of the device at different voltages.

Resonance modulation principle analysis

To further verify the rationality of our hypothesis, we simulate transmission spectra of Ag NP lattices at various RI of DMSO at corresponding temperatures (Fig. 3). Compared to the Ag NP lattices without the ITO layer, which generates a single SLR resonance mode, the introduction of the ITO layer produces multiple hybrid waveguide-SLR modes (Fig. 3a). We focus on mode 1 and mode 2 as they exhibit higher Q-factors and resonance strength than other modes. Mode 1 and mode 2 show different near-field enhancements distribution charge distributions (please see Fig. S5). Figure 3b shows the transmission spectra of the Ag NP lattices at varying temperatures (temperature-dependent RIs of the DMSO measured by the Abbe refractometer (please see supplementary information Table S2). Mode 1 and mode 2 (gray dashed lines) are blue-shifted when the temperature rises from 10 °C to 50 °C, with mode 1 shifted by 6.75 nm and mode 2 shifted by 0.9 nm (Fig. 3c, d). It is worth noting that the experimentally measured shift of mode 2 is 2 nm larger than that of the simulation, which may be due to the variation of the electron concentration at the interface of Ag NPs and the ITO layer resulting from the Seebeck effect. Further, mode 2 exhibits larger shifts in the experiment and smaller shifts in the simulation than mode 1 at different RI of DMSO corresponding to varying temperatures. Consequently, we propose that in addition to the temperature-induced changes in the RI of DMSO, there is an additional thermoelectric Seebeck effect that leads to the wavelength shift.

Fig. 3. Modulation principle analysis of tunable metasurface.

a The simulated transmission spectrum of Ag NPs lattices. The insets show the near-field distribution of mode 1 and mode 2. b Simulated transmission spectra of Ag NP lattices on ITO/quartz substrate at varying temperatures. Wavelength shifts of the transmission spectra of c mode 1 with temperature from 10 °C to 60 °C. d Mode 2 with temperature from 10 °C to 60 °C. The inset shows the enlarged transmission spectra in (c, d). e Schematic diagram of the distribution of electrons at the interface without bias and with bias. f The simulated transmission spectra with ITO RI variation from 2 to 1.975. Wavelength shifts of the transmission spectra of (g) mode 1 with ∆n of ITO from 0 to 0.025. h Mode 2 with ∆n of ITO from 0 to 0.025. The insets illustrate the magnified view of transmission spectra for mode 1 and mode 2 at the ITO RI of 2, 1.99, and 1.98, respectively.

In the absence of bias voltage, electrons in the bottom ITO layer are uniformly distributed, while charge redistribution accrues at the metal-semiconductor heterojunction with a depletion-like layer between Ag NPs and the ITO layer at the interface (Fig. 3e and please see Fig. S12 for the energy band). A temperature gradient distribution from the ITO/Ag NPs interface to the bottom of the ITO is established with applied voltage, resulting in a vertical top-bottom electromotive force (the Seebeck effect). The movement of electrons in the ITO layer towards the interface is driven by the electric field generated by the temperature gradient, leading to an increase in the localized electron concentration of the interface, thus a decrease in the RI of ITO (please see Supplementary information). Additionally, the increased concentration of the resonant electrons at Ag NPs surface can elevate the plasma oscillation frequency of Ag NPs and lead to blue shifts of transmission spectra. However, the change in the frequency of metal plasma oscillations may be negligible due to the degeneracy of electrons in metals (please see Figs. S13–14). Simulated transmission spectra with different RIs of ITO are shown in Fig. 3f. Both mode 1 and mode 2 are blue-shifted with the RI reduction of ITO ranging from 0 to 0.02, with mode 1 exhibiting a shift of ~ 0.01 nm and mode 2 a shift of ~ 4 nm (Fig. 3g, h). Comparative analysis of simulations and experiments reveals that neither the temperature changes of DMSO nor the Seebeck-induced variation in ITO refractive index alone can be consistent with the experimental results. Thus, resonance shifts of metasurfaces are synergistically modulated by the RI variation of DMSO caused by the electrically induced temperature change of DMSO and the variation of carrier concentration of the TCO layer from the Seebeck effect. Notably, these resonance wavelength shifts arise from the electrical modulation of RIs and are further enhanced by lattice modes inherent to periodic NP arrays. The operation wavelength range of the proposed metasurface can be extended from visible light to the infrared and telecom regime through structural design. Based on this principle, we also design tunable plasmonic metasurfaces for different wavelength ranges, such as the telecommunication band (please see Fig. S15). The modulation range at a specific wavelength is constrained by DMSO, which may volatilize under prolonged energized heating, leading to modulation saturation. Thus, the liquid material can be well encapsulated to avoid evaporation or can be replaced by the solid materials to improve the structural stability.

Electrical modulation of different tunable metasurfaces

To demonstrate the applicability of the modulation principle to other metal materials on different TCO layers, we fabricate metasurfaces of Ag NP lattices on Zr, Ti, and Ga-doped indium oxide (ITGZO) layers and Al NP lattices on ITGZO layer. Figure 4a shows the electrical modulation performance of the Ag NP lattices with a period of 500 nm on the ITO layer under a bias voltage varying from 0 V to 5 V. Ag NP lattices with a period of 500 nm, and Al NP lattices with a period of 370 nm on ITGZO substrate exhibit similar modulation trends but distinct shift values (Fig. 4b, c). All three devices demonstrate reversible modulation performance, with shifts amplifying with the increase of the bias voltage and the spectra redshift back during the voltage reduction and revert to their original position at zero voltage (Fig. 4d–f). The continuous and reversible modulation of Ag NP lattices on ITO metasurface with a period of 500 nm is shown in the Supplementary Video 1. The maximum wavelength shifts observed for three devices are ~ 4.74 nm, ~ 4.58 nm, and ~ 2.55 nm with the voltage ranging from 0 V to 5 V. The Ag NP lattices on ITO and ITGZO layers exhibit a tuning sensitivity of ~ 1 nm/V, which is defined as the wavelength shift divided by the operation voltage. For Ag NP lattices on ITGZO, modulation depth is ~43% at wavelengths of ~800 nm calculated by $\frac{\mathrm{\Delta }T}{T}$, where ∆T is the change in transmission and T is the transmission at 0 V (Fig. 4e). The modulation depth of Ag NP lattices on ITO and Al NP lattices on ITGZO are 20% and 28%, respectively. Simulated and experimental results of the Al metasurface with a period of 370 nm on ITGZO layer also further confirm our conclusion that the wavelength shift is governed by the refractive index variation of the DMSO superstrate and the Seebeck effect of the TCO layer. (please see Figs. S16–20). Additionally, we characterized the electrical modulation performance of metasurfaces consisting of Ag and Al NP lattices with different diameters and thickness of TCO layer (please see Figs. S21–23). It is worth noting that since small changes in RI are amplified by the lattice mode, the shift of an array with a period of 370 nm is smaller than that of an array with a period of 500 nm at the same voltage and temperature. Consequently, the tuning sensitivity and operation voltages can be further optimized by designing plasmonic metasurfaces with larger periods. To test the reliability of the device under different voltages, we characterize the transmission spectra and wavelength shifts of Ag NP lattices on the ITGZO layer at voltages ranging from 0 V to 12 V (please see Fig. S24). The resonance at 722 nm shift exhibits an approximately linear relationship with voltage magnitude when the bias is higher than 5 V. The resonant wavelength shifts by approximately 7 nm, 8 nm, and 10 nm at voltages of 8 V, 10 V, and 12 V, respectively. We characterize the alternating current (AC) modulation performance of the Ag metasurface on ITGZO/quartz substrate using an AC voltage with a peak-to-peak value of 4 V produced by a functional signal generator (please see Fig. S25). The modulation speed of the device is at the Hz level, which results from the heat buildup and dissipation of the electrothermal effect. Further, this modulation approach can be partially applied to dielectric metasurfaces. We fabricated TiO₂ NP lattices with a period of 410 nm on ITGZO/quartz substrate with the same structures as the metallic metasurface. At 12 V, the transmission spectrum of the TiO₂ metasurface exhibited a ~0.75 nm blue shift at 720 nm, consistent with the simulated results with temperature variation of 30–40 °C (please see Figs. S26–27). Less resonance shifts in TiO₂ NP lattices are attributed to the absence of the Seebeck effect in the TiO₂ and TCO interface, which further confirms our modulation mechanism in metal NP lattices.

Fig. 4. Electrical modulation of tunable metasurfaces composed of different metal nanoparticle lattices on various transparent conductive oxide layers.

Electrically modulated transmission spectra of a Ag NP lattices on ITO layer with a period of 500 nm, b Ag NP lattices on ITGZO layer with a period of 500 nm, and c Al NP lattices on ITGZO layer with a period of 370 nm. The grey dashed line marks the resonance position at V = 0 V in (a–c). (d) Resonance shifts of Ag NP lattices on ITO with the voltage increases (dark blue line) and the voltage decreases (light blue line). e Resonance shifts of Ag NP lattices on ITGZO with the voltage increases (dark red line) and the voltage decreases (light red line). f Resonance shifts of Al NP lattices on ITGZO with the voltage increases (dark green line) and the voltage decreases (light green line). The insets show the resonance peaks with bias voltages of 0 V, 3 V, and 5 V in (d–f). The error bars in (d–f) represent the standard deviation of the five sets of test data.

Application demonstrations of the metasurface device for light communications

Further, we demonstrate two applications for LC using the electrically tunable metasurface, with a single mode for image information transmission and multimode resonances for the encoder. Figure 5a depicts the experimental setup for utilizing the tunable metasurface to achieve these two applications (please see Figs. S28–29). To demo the transmission of image, the letter “S” is encoded to the 9 × 12 ASCII code matrix, which is subsequently transformed into the digital signal with a voltage peak-to-peak (VPP) of 4 V and applied to the electrodes of the tunable metasurface (Fig. 5b). The alternation of “0” and “1” in the digital signals implies the absence and presence of the bias voltage, leading to a real-time blue shift and recovery of the transmission spectra relevant to the original position. The corresponding transmission spectra are collected and decoded according to the rule that the spectral blue shift to the maximum (4.5 nm) is defined as “1” state, while the spectral return to the initial position is defined as “0” state, with intermediate states existing between these two. The black blocks, white blocks, and the gray blocks in the output image denote the “0” state, “1” state, and intermediate state, respectively (Fig. 5c). Thus, the image of the letter “S” can be transmitted and reconstructed. The precision of image transmission can be improved by dividing the image into finer grid matrices or employing higher resolution spectrometers.

Fig. 5. Schematic of experiment setup and application demonstration of the metasurface device for light communications.

a Experimental setup for electrical modulation applications of the tunable metasurface device. b The process flow of the image information transmission via a single resonance mode of the tunable metasurface. The image of the letter “S” on the left is input, and the image on the right is output. c The magnitude view of the transmission spectra with shifts corresponding to the different color blocks in the output image, circled by the corresponding colors of transmission spectra. d The transmission spectra of the metasurface at bias voltages of 0 V, 2 V, 4 V, and 5 V. The grey dashed lines label the resonance wavelength of three modes. e Wavelength shifts and logic values of three modes with voltage variation from 0 V to 5 V. The red dashed line divides the figure into two parts, the pink part illustrates the logic “1” while the blue part illustrates the logic “0”. f The truth table of the 1 × 3 encoder with 1 input and 3 outputs, showing four logic code outputs at different input voltages.

The three modes of the Al NP lattices on ITGZO metasurface require distinct voltage thresholds for a specific shift, as each mode demonstrates unique shifts at the same bias voltage. We propose a logic encoder with one input and three outputs utilizing this Al metasurface; the bias voltage functions as the input, while the logic values of the three modes are outputs. All three resonant modes of the metasurface are shifted with bias voltages, and can be assigned to three logic states, M0, M1, and M2, according to the magnitude of the shifts (Fig. 5d). The mode with the largest shift is defined as M0, and the mode with the least shift is defined as M2. The transmission spectra of the Al metasurface at the bias voltages of 0 V, 2 V, 4 V, and 5 V exhibit three modes with distinct shifts, leading to different logic values (Fig. 5d). Figure 5e shows shifts and logic values of the three modes at different input voltages. The region delineated by the 0.8 nm shift, indicated by a red dashed line, designates the upper pink section of the figure as logic “1” and the lower blue section as logic “0”. The gray dashed line indicates the critical voltage values that result in the change of the logic states. Then we list the truth table of the encoder, indicating that the four input voltages govern the output logic values varying from “000” to “111” (Fig. 5f). To enhance the information transmission capacity, the states of input and output can be increased by incorporating additional resonance modes through the metasurface design.

In conclusion, we demonstrate an electrically tunable plasmonic metasurface with a high modulation sensitivity of 1 nm/V in the visible and near-infrared regime. These metasurfaces can operate at the CMOS-compatible voltage of sub-5 V for light communication. We clarify that resonances are synergistically modulated by the temperature of DMSO and the thermoelectric field generated by the Seebeck effect, and further enhanced by the lattice modes. The devices enable precise and continuous electrical control of wavelength shift with a high modulation depth, which can generate apparent responses owing to the high-quality surface lattice resonance cavities. We confirm the universality of this modulation method by preparing the plasmonic metasurfaces of various metal NP lattices on different TCO/quartz substrates, and the modulation wavelength can be flexibly designed by the lattice structure. Further, two applications for light communication via the electrically modulated metasurface with multimode resonance are demonstrated. The easily-accessible materials, low-cost and scalable fabrication, low operation temperature, and sub-5 V operation voltages of the electrically modulated metasurfaces are well compatible with the CMOS integrated circuits. We believe that the modulation speed needs to be further improved by optimized materials with higher temperature sensitivity (such as graphene, polymers, etc.) and coupling with other optical resonance modes (e.g., gap plasmon cavities). And it can be extended to a broader wavelength range by structure design, which will have a wide range of applications in electro-optic modulators, artificial optical synapses, electrically tuned nanolasers, and photodetectors. For instance, the integration of our device with nanolasers is expected to facilitate low-power and ultracompact light sources in optoelectronic integrated chips. Further, our strategy can also introduce dynamic features in wavefront manipulation and structural lights of metasurface to achieve electrical modification of amplitude, polarization, and phase, thereby facilitating high-capacity information transmission^48,49.

Methods

Device fabrication

Ag NP lattices and Al NP lattices on TCO-quartz substrates are fabricated using a nanofabrication process that combines sputter deposition, electron-beam deposition, modified solvent-assisted nanoimprint embossing (SANE), and etching⁵⁰. First, a 150 nm TCO layer is deposited by sputtering, and then the Al film or Ag film is deposited by electron beam evaporation, followed by photoresist (PR) S1805 spin-coated on the metal surface at 4000 rpm with an acceleration of 500 rpm/s for 30 s. Nanoimprinting is performed on the photoresist to generate periodic PR arrays with controlled size and period. The polydimethylsiloxane (PDMS) mask was rinsed with alcohol and then immersed in dimethylformamide (DMF) solution for 90 s. After drying with a nitrogen gun, it was pressed onto the sample coated with S1805 and left to stand for 1.5 h before being removed. Metal NP lattices are formed by ion beam etching (IBE) and removing the PR with acetone. We performed IBE etching using the HAASRODE-I200 etcher from Leuven Instruments. The etching gas Ar flow rate is 24 sccm, with an etching angle of 0 degrees, a temperature of 7 °C, and a voltage of 400 V. The etching times for Al and Ag are 300 s and 96 s, respectively. Finally, the Au electrodes are deposited via electron beam evaporation with a hard mask to protect the metal NP array (please see supplementary information Fig. S6).

Transmission simulation

Finite-Difference-Time-Domain (FDTD) calculations are performed to simulate the optical responses of Ag and Al NPs lattices on the TCO layer. Optical constants for Ag and Al are taken from Palik measurements (400–1000 nm). These calculations are performed with a uniform mesh size of 4 nm (x, y, and z). The refractive index of TCO is measured by ellipsometry (please see supplementary information Fig. S1). We set the refractive index of ITO to be a constant of 2 when calculating the transmission spectrum changes induced by DMSO at different temperatures. Further, the temperature is set to be 20 °C, and the refractive index of DMSO is a constant of 1.4755 for the calculation of the transmission spectrum variations caused by ITO with different refractive indices. The optical parameters of other materials, such as quartz, Ag, and Al, are derived from material databases and may differ from those of materials employed in experiments due to their preparation conditions.

Electric field simulation

The electrodes and ITO are defined as Schottky contacts, and the Gaussian density of states is employed to approximate the effective density of states of carriers in ITO. Additionally, carrier mobility is affected by high-field saturation. Here, we simplify the periodic metal array to four metal pillars, and we take the Ag pillar array as an example. The bandgap and electron affinity are approximately 3.2 eV and 4.16 eV for ITO⁵¹.

Temperature simulation

We simulate the temperature distribution of devices at different voltages by using the current module with the solid-state heat transfer module. Since the size of Al NPs and ITO relative to the overall size is very small, precise gridding is established. Consequently, these layers are simplified to thermal thin layers for the heat transfer analysis. The conductivity of Al is set as the contact impedance, while the thickness and the conductivity of the ITO are determined by the electrical shielding. We simulated the temperature distribution of samples on the plastic platform, on foam, and in air. For the sample on the plastic plate, the material within the plastic holes of the test platform is air, establishing heat conduction in the plastic and the holes. The boundary conditions at the bottom of the air column and the bottom of the plastic are set as natural convection boundary conditions. For the sample on the foam, periodic boundary conditions are used around the foam, and natural convective boundary conditions are applied to the bottom of the foam substrate. For the sample in air, the natural convective boundary condition is used. Natural convective boundary conditions of $T_0=273+15\mathrm{K}$ are used around the quartz. The electrical boundary conditions for all three models are electrically insulating due to the substrate approximation of the insulator. Gold electrodes with negligible thickness were set as the input of the current. Temperature distributions of samples at voltages from 0 V to 5 V are simulated (please see supplementary information Figs. S9–11).

Characterization of the transmission spectra

For transmission spectral measurements, the spectral sampling time we have set is 100 ms, the data update rate is 5 ms, and the number of scans averaged to 1. This integration time is set by comprehensively considering the modulation frequency distortion and ensuring the spectral transmittance does not exceed 100%. To ensure the precise capture of dynamic spectral data variations, we do not employ multiple scans followed by averaging, which leads to a bit of noise in the results.

Calculation of the temperature gradient in transparent conducting oxide

Since the TCO layer is only 150 nm thick, which is comparable to the phonon and electron mean free path at room temperature (~100 nm), the heat transfer in the TCO layer is ballistic, which does not follow Fourier’s law of heat diffusion. In this circumstance, the temperature distribution is linear in the thickness direction with abrupt temperature jumps at the boundaries⁵². However, it is quite difficult to quantitatively determine the exact temperature distribution in the TCO layer because it heavily depends on the TCO/quartz interface conditions, doping concentration of the TCO film, quality of the TCO and quartz films, surface roughness of the films, etc. Thus, we can only give a qualitative estimation of the temperature gradient in the TCO layer based on the detailed energy balance, which is

$$q_{in}^{″}+q_{Joule}^{″}=q_{out}^{″},$$ 1

where q^”_in is the input heat flux from the bottom of the TCO layer (TCO/quartz interface), q^”_Joule indicates Joule heating and q^”_out is the output heat flux from the top of the TCO layer (TCO/DMSO interface). As calculated by the multiphysical simulation, we can get $q_{in}^{″}=-485.61W/c{m}^2,$ $q_{out}^{″}=626.98W/c{m}^2.$

By invoking the 1D heat transfer equation, we have

$$q_{in}^{″}=\frac{T_{e0}-T_{e2}}{R_{SiO2\_\_bottom}+R_{holder}+R_{conv\_\_bottom}},$$ 2

$$q_{out}^{″}=\frac{T_{e1}-T_{e0}}{R_{DMSO}+R_{SiO2\_\_top}+R_{conv\_\_top}},$$ 3

where T_e0, T_e1, and T_e2 are the room temperature, top surface, and bottom surface temperatures of the ITO layer, respectively. The R_{SiO2_bottom}, R_holder, R_DMSO, R_{SiO2_top}, and R_{conv_top} are the conduction thermal resistances of the bottom quartz layer, plastic sample holder, DMSO layer, top quartz layer, while R_{conv_bottom} and R_{conv_top} are natural convection thermal resistances of the bottom and top surfaces contacting the air, respectively. Conduction thermal resistances can be directly calculated by d/k_cond, with d being the thickness of the layer and k_cond the corresponding thermal conductivity, while convection thermal resistances can be determined by empirical heat transfer correlations for horizontal plates⁵³. Roughly, R_{conv_bottom} is in the range of 0.2–0.5 m²K/W, and R_{conv_top} is in the range of 0.1–0.15 m²K/W, in which the latter is smaller due to higher surface temperature. With these considerations, we can estimate the temperature difference of the ITO layer, T_e1-T_e2, to be in the range of 2–14 K.

Drude model of transparent conducting oxide

The optical response of ITO can be modeled by a simple Drude model description:

$${\epsilon }_{ITO}={\epsilon }_{exp}\left(\omega \right)-{\left(\frac{{\omega }_p}{\omega }\right)}^2\frac{1}{1+i{\left(\omega {\tau }_D\right)}^{-1}}$$ 4

Where ω_p denotes the plasma frequency and can be expressed as:

$${\omega }_p=\sqrt{\frac{N{e}^2}{{\epsilon }_0{m}^{*}}}$$ 5

N is the free electron density, m^* = 0.4 m_e is the effective electron mass, and τ_D = 5×10⁻¹⁵ s is the Drude relaxation time of ITO^54,55. The shift of the plasmon resonance frequency is expressed as ${\mathrm{\Delta }\omega }_p=\left({\omega }_p/2N\right)\mathrm{\Delta }N$ for small changes in the carrier density ΔN⁵⁶. And the refractive index of the ITO is given by $n=\sqrt{{\epsilon }_{ITO}}$. According to the above equation, higher electron concentration in the ITO leads to higher plasma frequency and a decrease in the refractive index, which in turn leads to a blue shift in transmission spectra.

To explore the influence of the Seebeck effect on the dielectric function of ITO/ITGZO, we include the temperature-induced change in carrier concentration and build a simple temperature-dependent Drude model.

The electric field generated by the Seebeck effect is

$$E=S\nabla {\mathrm{T}}_e$$ 6

The Seebeck coefficient S can be written as

$$S\approx -\frac{k_B}{e}\left[\ln \left(\frac{N_C}{N}\right)+A\right]$$ 7

Where k_B, N_C, N, e denote the Boltzmann constant, effective state density of the conduction band, carrier concentration, and elementary charge. A is the transport coefficient, usually 1–2.

At steady state, the diffusion current is balanced by the drift current,

$$e{D}_n\frac{dN}{dx}=e{\mu }_{n}NE$$ 8

Where D_n and μ_n denote the diffusion coefficient and electron mobility.

According to the Einstein relationship,

$$D_n=\frac{k_B{\mathrm{T}}_e}{e}{\mu }_n$$ 9

Combined with the Seebeck coefficient, this gives

$$\mathrm{\Delta }N\approx \frac{eN∣S∣L\nabla {\mathrm{T}}_e}{k_B{\mathrm{T}}_e}$$ 10

L denotes the diffusion length.

When the temperature is considered, the carrier concentration of ITO/ITGZO in the Drude model needs to be modified for the N + ∆N.

The relationship between τ_D and T can be approximated as:

$${\tau }_D\left({\mathrm{T}}_e\right)\propto {{\mathrm{T}}_e}^{-\frac{3}{2}}$$ 11

The fitted Drude model can be written as:

$${\epsilon }_{ITO}={\epsilon }_{exp}\left(\omega \right)-{\left(\frac{{\omega }_p}{\omega }\right)}^2\frac{1}{1+i{\left(\omega {\tau }_D\left({\mathrm{T}}_e\right)\right)}^{-1}}$$ 12

The ω_p can be modified as:

$${\omega }_p=\sqrt{\frac{{\left(N+\mathrm{\Delta }Ne\right)}^2}{{\epsilon }_0{m}^{*}}}$$ 13

According to the Drude model, the refractive index n is

$$n\approx \sqrt{{ϵ}_{\infty }-\frac{N{e}^2}{{ϵ}_0{m}^{*}{\omega }^2}}$$ 14

Derivation for n gives

$$\frac{dn}{dN}=-\frac{e^2}{2{ϵ}_0{m}^{*}{\omega }^{2}n}$$ 15

For small changes, there are

$$\mathrm{\Delta }n\approx \frac{dn}{dN}\mathrm{\Delta }N$$ 16

Therefore, we approximate the relationship between the change in refractive index and the change in carrier concentration as

$$\mathrm{\Delta }N\approx -\frac{e^2}{2{ϵ}_0{m}^{*}{\omega }^{2}n}\mathrm{\Delta }n$$ 17

At the resonance wavelength of 790 nm, we take the value of N to be 6 × 10²⁰ cm⁻³, the ∆n is 0.005 to 0.01, and S is 1 × 10⁻⁴ V/K. We can calculate that ∆N is 1.35 × 10¹⁹ cm⁻³ to 2.4 × 10¹⁹ cm⁻³, the corresponding temperature gradient in 150 nm ITO is 5.85–10.75 K. The temperature gradient is consistent with the gradient of 2–14 K obtained through our thermal field calculation.

Author contributions

X.W. and S.D. conceived the ideas and designed the investigation. X.W. designed and prepared samples, carried out characterization, performed simulations, and analyzed data. H.Y. participated in the electrical measurements. H.Y., Y.W., K.Z., and X.W. participated in the theoretical analysis. E.G., P.Z., and B.W. contributed to the optical and temperature simulations. S.C. and F.M. contributed to the preparation of ITGZO. J.T. participated in the sample fabrication. Q.J. contributed to the setup of the optical measurement. X.W. and S.D. wrote the paper. All authors commented on and revised the manuscript.

Peer review

Peer review information

Nature Communications thanks Jongwon Lee and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data supporting the findings of this study are included in the main text, the Supplementary Information, and the Source Data files provided with this paper. Additional data are available from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71092-w.