Digitally encoded dual-narrowband photodetectors for secure optical wireless communication Authors: Zilong Ye, Haoyu Huang, Wei He, Kai-Ning Tong, Chengcheng Wu, Tengfei Li, Dan Zhao, Yuan Li, Kanghui Ke, Kefei Shi, Huan Li, Zhuhua Xu, Xin Wang, Liping Song, Chang Shu, Hao Yan, Yuze Lin, Zonglong Zhu, Feiyu Kang, H Y Fu, Guodan Wei Published: 2026 ABSTRACT AbstractNext-generation optical wireless communication requires photodetectors that offer both high spectral selectivity and strong security against interception. However, conventional broadband devices remain vulnerable to spectral crosstalk and eavesdropping. Here we show a digitally encoded dual-narrowband organic photodetector that intrinsically integrates optical filtering with algorithm-assisted encryption to enable secure, high-fidelity optical wireless communication. Operating without an external power supply, the self-powered device employs a Fabry-Pérot cavity with a carefully designed organic spacer Liq to achieve selective detection at wavelengths of 485 nm and 910 nm, along with an ultrafast response time of 440 ns. By combining chaotic encryption with hardware-level wavelength selectivity, our hardware–software co-design system achieves an ultra-low bit-error rate of 9.17 × 10−5 at 1.25 Mbps while demonstrating strong resilience to eavesdropping and external interference. Furthermore, precise cavity engineering allows the dual-narrowband response to be extended into the short-wave infrared region (>1230 nm), offering a scalable route toward multi-wavelength secure transmission, high-resolution spectroscopy, and intelligent photonic networks.Subject terms: Electronic devices, Nanocavities Introduction The rapid advancement of data-driven artificial intelligence and next-generation wireless technologies, including 5 G and emerging 6 G, has significantly increased the demand for high-speed, electromagnetic-interference-free wireless communication systems which are inherently secure and resilient1,2. In autonomous and intelligent systems, where practical users continuously interact with dynamic environments, the vulnerability in communication links can directly translate to operational failure or severe privacy breaches3. Concurrently, the miniaturization and lightweight integration of photonic systems have become essential for emerging applications across communication, sensing and wearable electronics2,4–6. Optical wireless communication (OWC) has emerged as a compelling alternative to radio-frequency technologies by leveraging visible and near-infrared (NIR) light to enable high-bandwidth, directionally confined, anti-interference data transmission7. Benefiting from an enormous available optical bandwidth of approximately 400 to 800 THz, OWC can achieve data rates exceeding 400 Gbps, making it highly attractive for next-generation visible light communication, infrared free-space optical communication, and quantum-secure optical channels8–11. Despite these advantages, the security of most conventional OWC systems remains largely an afterthought, typically addressed through upper-layer cryptographic protocols rather than at the physical layer itself12–14. As a result, an eavesdropper equipped with a simple broadband detector can readily intercept the transmitted optical signal, while a jammer can disrupt communication by injecting high-power broadband noise15. These intrinsic vulnerabilities stem from the lack of spectral discrimination at the receiver, creating a critical bottleneck for deploying OWC in security-sensitive applications within next-generation intelligent networks7. Addressing these physical-layer security challenges requires a paradigm shift towards spectrally selective photodetection. As the core receiving components in OWC systems, photodetectors (PDs) play a decisive role in determining system security and signal-to-noise ratios (SNRs), as well as in rejecting undesired ambient interference11,16,17. Conventional inorganic PDs based on Si, Ge, InGaAs, and HgCdTe typically exhibit broad spectral absorption, which are not only rigid, expensive, and incompatible with flexible or wearable platforms, but also possess the risk of high background noise and security vulnerabilities18–21. Here, organic photodetectors (OPDs) present a compelling advantage over their conventional inorganic counterparts, providing high responsivity, mechanical flexibility, biocompatibility, and low fabrication cost22–24. OPDs can be engineered for intrinsic spectral selectivity without the need for external optical filters, thereby simplifying system design and improving optical isolation14,25. Especially, narrowband OPDs, which are a specialized class of PDs designed to detect only a confined spectral region while rejecting light outside that range, offering an intrinsically selective approach, suppressing out-of-band signals and improving the SNR22–24,26,27. Strategies for achieving narrowband detection in OPDs include the use of narrowband organic semiconductors28–30, charge-transfer absorption at specific wavelengths16,31, thick-film absorption filtering32,33, and optical resonant structures such as Fabry-Pérot (FP) cavities34–36. Among these strategies, the FP cavity-based approach, which consists of two semi-reflective mirrors that generate constructive interference at specific wavelengths, stands out for its intrinsic spectral selectivity, ultra-narrow full width at half maximum (FWHM), high-speed response, and spectral tunability, making it particularly suitable for secure OWC, high-resolution imaging, and encrypted photonic systems37,38. By tuning the cavity thickness, mirror reflectivity and the refractive index of spacer materials, FP-based OPDs can achieve precise spectral control across the visible-to-NIR range, enabling ultra-narrowband detection with FWHM values below 30 nm34–36,39,40. Recent researches have demonstrated substantial progress in FP cavity-based narrowband OPDs for multi-channel detection applications, enabling applications such as secure OWC, biometric authentication, and encrypted imaging13,17,41,42. In conventional OWC systems, information is commonly encoded through intensity modulation or wavelength division multiplexing, where multiple wavelengths function as independent data channels. However, the integration of existing PDs into such architectures remains intrinsically vulnerable to signal interception. Most conventional PDs exhibit broadband spectral responses, allowing undesired optical signals to be detected. To date, the majority of studies have emphasized standalone optoelectronic performance, treating security as a secondary consequence of spectral selectivity rather than a fundamental design target. There has been a lack of a comprehensive approach to improve both communication security and spectral efficiency by incorporating digital encoding into spectrally selective OPDs43. In particular, the absence of a synergistic system that unifies physical-layer barrier security, enabled by device-level spectral filtering, with algorithmic encryption and active countermeasures represents a critical gap1,44,45. Addressing this challenge by establishing an integrated hardware–software security system is therefore a key step toward fully unlocking the potential of OPDs for next-generation secure OWC systems. In this study, we present a digitally encoded dual-narrowband OPD platform that integrates physical-layer spectral selectivity with system-level digital encryption for secure OWC. Using an out-coupling FP cavity with 8-quinolinolato lithium (Liq) as the spacer, the device achieves two ultra-narrow detection channels at 485 nm (visible) and 910 nm (NIR) with an FWHM of 23 nm. This hardware-level wavelength selectivity provides a physical barrier against signal interception. Chaotic encryption is further embedded at the system level, enabling secure transmission and decoding exclusively via the matched OPD. The hardware–software co-design yields a low bit-error rate (BER) of 9.17 × 10−5 at 1.25 Mbps, a linear dynamic range (LDR) > 140 dB, and a response time of 440 ns. Importantly, this integrated security platform is inherently scalable and tunable. By tuning the cavity thickness, operation can be extended to the short-wave infrared (SWIR) region over 1230 nm, supporting multi-band secure communication. Thus, narrowband OPDs evolve from passive PDs into active security systems for multi-channel OWC. Results Design and fabrication of dual-narrowband OPDs Unlike conventional broadband PDs, which are susceptible to interception, herein we propose a unique dual-narrowband OPD that integrates spectrally selective photodetection with digital encoding, significantly enhancing the physical-layer security of OWC. The dual-narrowband OPD employs an out-coupling FP resonant cavity, enabling a dual-wavelength response at two wavelengths (λ₁ and λ₂) under zero-bias operation. The secure digital encoding mechanism operates through three key steps as Fig. S1 shows: (1) Dual-channel encoding, where λ₁ and λ₂ represent distinct wavelengths for spectrally encoded transmission; (2) Secure optical transmission, where two different light source selectively emit λ₁ and λ₂, respectively, ensuring that only an authorized receiver with precise spectral sensitivity can decode the signal; (3) Decoding and data stream reconstruction, where the dual-narrowband OPD selectively detects only the encoded spectral components, filtering out undesired ambient radiation. This design introduces a mutual dependency between the two channels: information carried by either channel alone is incomplete and cannot be reconstructed without the corresponding signal from the other channel. Therefore, any interceptors would have to know the exact two detection peaks transmitted data to decode the information. This system enhances security through an ingenious combination of hardware-level intrinsic spectral filtering and algorithm-level encryption and active defense, which will be elaborated in the following contents (Fig. S2). The architecture of the dual-narrowband OPD for secure OWC is schematically illustrated in Fig. 1a. This device incorporates an out-coupling FP resonant cavity into a conventional photodiode structure, enabling an intrinsic dual-narrowband response at two specific wavelengths (λ₁ and λ₂) under zero-bias operation, which is different from electrically switchable dual-band PDs in terms of operating mechanisms12,46. The fabricated dual-narrowband OPD consists of the multilayer structure Ag/Liq/Ag/ITO/PEDOT:PSS/PTB7-Th:COTIC-4F/Phen-NADPO/Ag, with the detailed fabrication process shown in Fig. S3. The cross-sectional SEM image confirms the integrity of the layered architecture, measuring an optimized Liq thickness of ~235 nm inside the resonant cavity. The active layer comprises a bulk heterojunction (BHJ) formed by blending the electron donor polymer PTB7-Th and the small-molecule electron acceptor COTIC-4F. Their molecular structures are depicted in Fig. S4. The narrow bandgap nature of COTIC-4F (HOMO = −5.26 eV, LUMO = −4.17 eV) results in a broad spectral absorption spanning from 400 to 1200 nm, extending beyond the silicon absorption range into the NIR region (Fig. 1b)47,48. Design of self-powered dual-narrowband OPD. a Schematic illustration of dual-narrowband organic photodetector (OPD) for digitally encoded optical wireless communication (OWC) system. b Normalized absorption spectra of PTB7-Th, COTIC-4F, and Liq in film state. The inset was chemical structure of Liq. c Simulated absorption spectrum of PTB7-Th:COTIC-4F BHJ device without and with cavity at a length of 235 nm. d Simulated transmission spectrum of dependency of resonance wavelength on the angle of light incidence at 200 nm cavity thickness. The FP resonant cavity was constructed using two semi-transparent Ag films as reflective mirrors and a widely used organic material, Liq, as the cavity spacer49. The adoption of Liq as the cavity layer is a deliberate engineering optimization based on surface smoothness, optical constant, narrowband feature, and precise control of wavelengths. The high optical transparency of Liq beyond 450 nm enables strong optical transmittance and sharp wavelength selectivity, thereby facilitating efficient dual-wavelength optical filtering. The performance of the FP resonant cavity is governed by both the optical properties of the spacer material and the precise control of the cavity geometry. Liq was selected as the cavity spacer due to its broad transparency over the visible-to-NIR range, excellent compatibility with thermal vapor deposition, and robust environmental stability, which collectively enable accurate thickness control and reliable cavity formation (Fig. S5). Furthermore, the resonant enhancement effect of the cavity relies on maintaining a highly smooth cavity interface, as confirmed by atomic force microscopy (AFM) measurements (Fig. S6). The ultra-low surface roughness (Ra) values of 0.815 nm for the Ag mirror layer and 0.453 nm for the Liq layer indicate a highly uniform surface, enabling optimal light modulation within the cavity for efficient resonance. By leveraging this engineered FP cavity, the dual-narrowband spectral response was achieved, with performance specifications detailed in the literature and Table S140. The adoption of Liq as the cavity layer enables low-crosstalk, highly selective and interference-resistant photodetection, making it particularly suitable for secure OWC applications. When light vertically enters through the resonant cavity, the resonance peak is determined solely by the cavity length, assuming the spacer material remains fixed. The relation is described by where Lm and λm represent respectively the cavity length and the corresponding resonance peak wavelength, modulated by the resonant mode number m. As illustrated in Fig. S7, for any resonance order, broadband incident light passes through the cavity; however, only light at the resonance peak exhibits high transmittance, resulting in a narrowband response. The spectral selectivity of the resonant cavity is characterized by the FWHM of the resonance peak, which directly reflects the optical quality of the cavity. To further investigate light modulation within the cavity, we employed finite- difference time-domain (FDTD) simulations, utilizing optical constants measured in Fig. S8. The squared electric field intensity, which corresponds to the optical field distribution50, was simulated to optimize the thickness of the mirror layer (Fig. S9). While increasing L enables wavelength selectivity across the visible-NIR-SWIR range, transmittance could be reduced due to higher-order mode losses and mirror imperfections, as longer paths increase photon scattering in the spacer, and thicker cavities amplify interfacial roughness losses. Additionally, to evaluate how the resonant cavity modulates light absorption in the active layer, we compared the absorption intensities of PTB7-Th:COTIC-4F BHJs with and without the resonant cavity, as depicted in Fig. 1c. The pristine 110-nm-thick active layer, without cavity modulation, exhibited a broadband absorption profile spanning 400–1200 nm, peaking around 700 nm. However, with the integration of a 235-nm-thick resonant cavity, the absorption spectrum was effectively modulated, generating a first-order resonance peak at 910 nm and a second-order peak at 485 nm. These results closely align with the simulated electric field intensity distribution within the resonant cavity (Fig. S10), confirming the cavity’s role in achieving dual-narrowband photodetection. Within the device, these two resonance wavelengths successfully penetrate the cavity and reach the organic BHJ layer, where incident photons generate excitons. These excitons diffuse to the donor-acceptor interfaces, undergo charge separation into electrons and holes, and subsequently transport toward the terminal electrodes. This process ensures efficient dual-wavelength photodetection, further validating the effectiveness of our FP resonant cavity design in achieving high-performance dual-narrowband OPDs. A minor peak near 350 nm stems from the third-order FP mode of the same cavity, which is negligible in this work due to its relatively low intensity. The angular dependence of the cavity is shown in Fig. 1d, showing that the first-order transmission peak shifts slightly toward shorter wavelength as the light incidence angle increases from 0° (perpendicular to the cavity) to 60°. This indicates that the cavity operates optimally when light is incident perpendicularly. Practical countermeasures against angular effects include using collimating optics to maintain θ < 10° and implementing algorithm-based correction. These measures ensure the FP cavity’s dual-band selectivity is preserved, making it promising for directed OWC implementations, such as indoor networks. Optoelectronic performances of dual-narrowband OPD Fundamental optoelectronic parameters and performances of the self-powered dual-narrowband OPD are presented to highlight its high SNR at specific wavelengths, which is a comprehensive metric for evaluating a PD’s ability to effectively extract signals from noise. As shown in Fig. 2a, the OPD utilizes PEDOT:PSS as the hole transport layer (HTL) and Phen-NADPO as the electron transport layer (ETL), ensuring favorable energy level alignment for efficient carrier transport. The noise spectral density51 was mainly determined by shot noise, thermal noise, and 1/f noise, of which the former two influencing factors were maintained at a stable level under ambient-temperature condition, while the latter 1/f noise generally decreases with higher frequencies27. The synergistic effect of these noise components resulted in a low noise spectral density IN of 1.9 × 10−14 A/Hz1/2 at a 2 kHz modulation frequency (Fig. 2b), which effectively enhances the SNR of the device. Based on the spectral responsivity R and noise spectral density, the specific detectivity (D*) of OPD, which represents its sensitivity to weak photons, was calculated as where A and Δf refer respectively to the effective device area and the electrical bandwidth of the noise measurement (Δf is normalized to 1 Hz in the calculation of D*). As shown in Fig. 2c, at zero bias, the D* of the dual-narrowband OPD reaches a peak value of 8.6 × 1011 Jones at λ1= 910 nm for the first-order resonance peak and 6.5 × 1011 Jones at λ2= 485 nm for the second-order resonance peak, which exhibits profound capability in detecting the desired spectral response. Notably, the two very narrow resonant peaks at λ1= 910 nm with an FWHM as narrow as 25 nm, and atλ2= 485 nm with an FWHM as narrow as 23 nm have been successfully obtained, respectively. The rejection ratio (RR) of a narrowband OPD quantifies its ability to suppress photoresponse to out-of-band illumination, which is critical for applications like secure OWC and spectral imaging. According to the calculation, in the visible channel (485 nm), RR ≈ 115; in the NIR channel (910 nm), RR ≈ 155, which is relatively high compared with values reported in the literature on narrowband OPD52. This result indicates that the response of the dual-narrowband peaks are more than one hundred times larger than those of the suppressed regions, which shows high rejection of ambient background noise. Optoelectronic performances of dual-narrowband OPD. a Energy level diagram of the corresponding material structures: ITO/PEDOT:PSS/PTB7-Th:COTIC-4F/Phen-NADPO/Ag. b Noise spectral density of the device. c Specific detectivity of the dual-narrowband OPD. d I–V characteristics of the device under dark and light at 485 nm (visible) and 910 nm (NIR) at 1.3 mW. e Photo-response for varying light intensity measured at 0 bias at 485 nm and 910 nm. f Photocurrent measured over days under air environment without encapsulation. g, h I–t characteristics of the device at g 485 nm and h 910 nm. i Comparison FWHM and rise time of reported narrowband OPDs57–62. The current-voltage (I–V) curves for the photocurrent and dark current at these two incident peaks under an optical power of 1.3 mW are shown in Fig. 2d, indicating typical diode characteristics. Through space-charge limited current (SCLC), Mott-Schottky (M-S), electrochemical impedance spectroscopy (EIS) measurements and dark current fitting, the results indicates the high quality of the as-fabricated OPD with an ideal diode structure, which significantly suppresses the potential leakage current across the constructed heterojunction53–55 (The detailed calculation process is given in Fig. S13 and Table S2). The dependence of the photocurrent and the incident light power intensity can be given by LDR, which are calculated by where Pmax and Pmin are respectively the maximum and minimum power densities of a linear range between photocurrent and incident light power. As shown in Fig. 2e, the as-obtained device shows a high LDR with 110 dB at λ=485nm and 140 dB at λ=910nm, respectively, indicating that both the visible and NIR channels have wide linearity regions for incident light power intensities. Time-dependent current stability was also measured to show the long-term performance of the device (Fig. 2f). In an ambient-air environment without any encapsulation, the device functions well, with photocurrent barely changing for more than one month, indicating very promising potential for highly stable visible-NIR dual-band applications. To further analyze the frequency response of the device as a crucial metric for data acquisition and OWC, the response times of the dual-narrowband OPD at λ=485nm and 910 nm peaks are measured in Fig. 2g, h. Ultrafast rise/decay times are measured towards nanoseconds at 580/585 ns at 485 nm and 440/530 ns at 910 nm, respectively, which are among the top level values for reported OPDs56. The 3-dB bandwidth of both wavelengths reaches high standards of 300 kHz at 485 nm and 550 kHz at 910 nm (Fig. S14), referring to the modulation frequency where the responsivity of the device is half of that at steady-state conditions, which is among the highest recorded values for reported OPDs. Compared with reported narrowband OPDs in Fig. 2i, not only did we obtain the narrowest peak with strong selectivity, but also demonstrated ultrafast response speeds, indicating the high performance of the self-powered dual-narrowband OPD (Table 1)57–62. All crucial parameters of the dual-narrowband OPD are listed in Table S3. Although the absolute external quantum efficiency (EQE) is moderate due to the introduction of the FP cavity, our device achieves an excellent SNR through the synergistic combination of an on-off ratio reaching 104, which enables clear signal-noise differentiation, and a low FWHM that effectively rejects broadband background noise. Future strategies to mitigate energy loss include replacing metallic mirrors with dielectric distributed Bragg reflectors (DBRs) to reduce absorption losses while maintaining high reflectivity, as well as integrating avalanche gain layers to amplify the photocurrent without sacrificing spectral purity38,63. Table 1Comparison of optoelectronic performances for narrowband OPDsStructureλ (nm)R (mA/W)FWHM (nm)Rise time (ns)RefAg/Liq/Ag/ITO/PEDOT:PSS/PTB7-Th:COTIC-4F/Phen-NADPO/Ag4855023580This work9106725440Ag/Bphen:Cs/DCV5T-Me:C60/BPAPF:NDP9/Ag53015072125057Ag/ZnS/Ag/ITO/CuSCN/PM6:BTP-eC9/C60/BCP/Ag82027980215058ITO/PEDOT:PSS/PTB7-Th:6TBA/LiF/Ag7407948700059ITO/PEDOT:PSS/TP-Si/LiF/Al380182601.52 × 10460ITO/P3HT:PC71BM/PH1000/Ag33030527 × 105616004880ITO/ZnO/PM6/L8-BO/MoOx/Ag53021701.2 × 1066264025580 Comparison of optoelectronic performances for narrowband OPDs Design of the digitally encoded dual-narrowband OWC system We present a high-security visible-NIR dual-narrowband encrypted OWC system enabled by this resonant dual-narrowband OPD, which operates without an external bias. In conventional OWC systems employing intensity modulation with direct detection, the original data stream is encoded into binary sequences that directly modulate the instantaneous optical power of the transmitter. The transmitted signal is subsequently detected by a broadband PD. As depicted in Fig. 3a, this broadband OWC system poses risks to information security and signal fidelity, because it lacks system integration for enhancing anti-eavesdropping and anti-interference abilities and operates without encryption algorithm methods. Following the illustration of the dual-narrowband OWC system in this work, as depicted in Fig. 3b, the binary sequence was encoded into a four-level pulse amplitude modulation (PAM4) symbol sequence. Each PAM4 symbol was then further decomposed into two PAM4 symbols based on an encryption scheme utilizing the second decimal place of the corresponding element in a Lorenz sequence, which was generated using a built-in codebook. The detailed working principle is given in Fig. S15 and the physical setup is pictured in Fig. S16. Two laser diodes (LDs), emitting at 485 nm (visible) and 910 nm (NIR), were employed to transmit their respective PAM4 sequences through an arbitrary waveform generator (AWG). The optical signals were received by the dual-narrowband OPD and converted into discrete electrical signals via trans-impedance amplifier (TIA) and an oscilloscope (OSC). As illustrated in Fig. S17, a 600 × 600-pixel colored photograph of a campus auditorium was used as the original binary sequence source. This image was first decomposed into three monochromatic images corresponding to the red, green, and blue (RGB) channels. Each pixel’s intensity in these monochromatic images was quantized into 256 discrete levels, equivalent to 8-bit encoding, resulting in a total of 8,640,000 bits in the original binary sequence. Illustration of dual-narrowband encrypted OWC system. a Illustration of traditional broadband OWC system. b illustration of dual-narrowband OWC system. c Bit-error-rate (BER) curve at different data rates of 910-nm and 485-nm channel. The dashed line corresponds to the hard-decision forward-error-correction (HD-FEC) value. d, e Communication eye diagrams of 0.5 Mbps and 1 Mbps at d 485 nm and e 910 nm. The structural similarity index measure (SSIM) is a perceptual metric that quantifies image quality degradation by comparing luminance, contrast, and structural information between two images, with a value of 1 indicating perfect similarity. The bit-error rate (BER) is the ratio of incorrectly received bits to the total number of transmitted bits, serving as a key measure of communication link reliability. At a data transmission rate of 20 kbps, the reconstructed image exhibited exceptional quality, with an ultra-low BER of 2.06 × 10−5 and SSIM of 0.9998, without the need for additional emphasis or equalization64. This result demonstrates the strong capability of our device to transmit a high-fidelity color image. To evaluate the upper limit of the data transmission rate, we measured the BER of the dual-narrowband OPD at different data rates at 485 nm and 910 nm, as recorded in Fig. 3c. The eye diagrams of the two wavelengths under 0.5 and 1 Mbps are shown in Fig. 3d, e. For the NIR channel, the BER at 1.25 Mbps reached an exceptionally low value of 9.17 × 10−5, well below the 7% hard-decision forward error correction (HD-FEC) limit of 3.8 × 10−3. At 2 Mbps, the BER remained relatively low at 5.76 × 10−3, below the 20% soft-decision forward error correction (SD-FEC) limit of 2 × 10−2, verifying the feasibility of high-speed communication. Similarly, the visible channel demonstrated a low BER of 1.78 × 10−3 at 1.25 Mbps, as shown in the eye diagrams depicted in Figs. S18 and S19. It is important to distinguish our approach from traditional multicolor OWC systems, in which spectral overlap between channels is a major source of noise and signal interference65. In this work, optical crosstalk is physically mitigated at the device-design level through the dual-narrowband OPD architecture. The hardware itself does not physically separate the 485 nm and 910 nm signals from each other, which is the essential principle of dual-narrowband OWC. The two detection channels are centered at 485 nm and 910 nm, and separated by an exceptionally large spectral gap exceeding 400 nm, rendering optical crosstalk effectively negligible (Fig. S20). In contrast, conventional light-emitting diode (LED)-based OWC systems often suffer from severe inter-channel interference, particularly between blue and green channels, due to their broad emission spectra and long spectral tails. Our device instead exhibits an ultra-narrow FWHM, which allows for dense wavelength assignment with a channel spacing below 50 nm while maintaining minimal spectral overlap. Furthermore, according to the LDR curve, the device remains below saturation even under simultaneous illumination at both channels when the light sources in the OWC system operate under normal working conditions and possess output powers within a pre-defined range. As a result, the digitally encoded dual-narrowband OPD supports high-fidelity and high-speed OWC with intrinsically enhanced resistance to eavesdropping and signal interception, thereby establishing a robust platform for secure multi-channel OWC. Demonstration of anti-eavesdropping OWC system In this encrypted OWC system, robust resistance to eavesdropping and signal interception is achieved without relying on computationally intensive algorithms or complex signal transformations. The complete workflow for eavesdropping detection and signal compensation is illustrated in Fig. S21. This anti-eavesdropping capability originates from a hardware-algorithm co-designed physical-layer security architecture, in which the dual-narrowband OPD is an indispensable component. Specifically, the as-obtained OPD is physically required for the encryption scheme to operate in practice and serves a unique role as a physical “spectral key” within the system. To implement encryption, the original data stream is decomposed using a Lorenz chaotic system, a deterministic nonlinear dynamical model governed by these coupled differential equations: Here, x, y, and z represent the state variables of the chaotic system, while σ, ρ, and β are system parameters. Due to the sensitivity of chaotic dynamics to initial conditions, it is mathematically infeasible for an eavesdropper to reconstruct the chaotic sequence or recover the original signal without precise knowledge of the initial parameters used by the transmitter, even if the intercepted waveform is recorded with high fidelity. Furthermore, when an optical signal at a single wavelength is intercepted, secure communication is maintained without interruption. The intercepted signal exhibits an elevated BER and a reduced fidelity relative to the original data stream, rendering it ineffective for information extraction. This intrinsic resilience against interception highlights the effectiveness of combining spectrally selective photodetection with chaotic encryption, thereby enabling a physically secured OWC system with strong anti-eavesdropping capability. Moreover, we implement transmitter-side signal intensity compensation to counteract a fixed percentage of beam-splitting eavesdropping or interception, thereby maintaining the fidelity of the received signal and ensuring uninterrupted OWC. At the transmitter side, we periodically insert and send detection sequences, i.e., pilot sequences, of a pre-defined length, regardless of potential eavesdropping or interception. These sequences are based on a codebook of Lorenz seeds shared by both the transmitter and receiver, allowing the receiver to accurately distinguish between authentic signals and intercepted data. This real-time assessment enables continuous monitoring of signal fidelity. At the receiver, pre-received detection sequences are used to compute and track signal intensity attenuation. If a fixed proportion of the optical signal is intercepted, the transmitter compensates for the power loss by adjusting the signal intensity, ensuring that the received data maintains high fidelity. When there is no eavesdropping, the signals obtained by the receiver are complete (Fig. 4a), thus the RGB components of the transmitted image (Fig. 4b) and the reconstructed photograph are of high fidelity (Fig. 4c). When an eavesdropper intercepts the 485-nm optical channel, the RGB components of the transmitted image are incomplete, leading to a distorted colored image (Fig. 4d–f). Meanwhile, the receiver, though experiencing partial information loss (Fig. 4g–i), still benefits from the encryption scheme. The SSIM of the reconstructed image for the eavesdropper is 0.11, whereas that for the receiver achieves 0.53. However, after transmitter-side compensation, the SSIM improves to 0.99, restoring the image quality to its original state. Anti-eavesdropping demonstration of OWC system. Signals obtained by eavesdropper/receiver was transferred into red (R), green (G), and blue (B) components and reconstructed into a colored photograph. Without eavesdropping, the reconstructed photograph after transmission is original photograph. When facing eavesdropping, the reconstructed images are distorted with incomplete information, which is evident by Structural similarity index measure (SSIM). When compensated through algorithms or conditions without eavesdropping, the reconstructed image was complete with high fidelity and SSIM of 0.99. a, d, g Obtained signals by eavesdropper or receiver; b, e, h RGB components; c, f, i Reconstructed colored photograph. The unique mapping between binary digits and Lorenz elements ensures that each binary sequence corresponds to both 485-nm and 910-nm PAM4 signals. Consequently, even if an optical signal at one band is eavesdropped on at a fixed ratio, it remains impossible to reconstruct the original binary sequence, significantly reducing the fidelity of the intercepted signal. A crucial security feature of our system is that the Lorenz sequence used for signal decomposition is exclusively generated at the transmitter-side and is periodically refreshed within the memory modules. This prevents unauthorized interceptors from accessing the essential decomposition parameters, making it infeasible to reverse-engineer the original PAM sequence through de-mapping. As a result, even if a single-wavelength signal is intercepted, the communication channel remains secure, ensuring high confidentiality and integrity. This anti-eavesdropping OWC strategy demonstrates robust resilience against interception and deliberate signal interference, effectively safeguarding high-security optical communications. Demonstration of anti-interference OWC and imaging A key advantage of the dual-narrowband OPD over conventional broadband OPDs is its high spectral selectivity and strong anti-interference capability, enabling error-resistant optical transmission even under complex ambient light conditions. Leveraging this property, we developed an anti-interference OWC system and a single-pixel imaging system. As illustrated in Fig. 5a, we use a 910-nm NIR light to transmit a monochromatic picture of a campus gate, while additive white Gaussian noise (1 kHz) from an off-the-shelf white LED was introduced to simulate the environmental radiation noise. For comparison, we use both a dual-narrowband OPD and a cavity-free broadband OPD to receive the signals, with their response spectra shown in Fig. S22. The system setup and its photograph obtained using the two different OPDs are presented in Figs. S23 and S24. Under noisy conditions, the dual-narrowband OPD demonstrated superior resilience, achieving 42% lower pixel intensity distortion and 54% lower mean-squared error (MSE) compared to the broadband OPD. These results highlight its significant potential for high-fidelity OWC in complex environments (Fig. 5b). A comparative analysis of the broadband and dual-narrowband OPDs is summarized in Table S4. Furthermore, using the transmission wavelength of the second-order peak at 485 nm, we conducted a video demonstration (see Supplementary video). When a broadband OPD was used as the receiver, signals became significantly distorted under torchlight illumination. In contrast, the signals obtained through the cavity-tuned dual-narrowband OPD remained remarkably stable, even under intense white LED illumination. Thus, while the absolute signal current is modest, the noise current from out-of-band sources is dramatically reduced, resulting in an improvement in SNR compared to broadband PDs. Anti-interference OWC and imaging based on dual-narrowband OPD. a Schematic illustration of the anti-interference OWC system. b Comparison of Intensity distortion per pixel and mean-squared error (MSE) of dual-narrowband OPD and broadband OPD. c Schematic illustration of the anti-interference single-pixel imaging system. d, e Reconstructed image of “TSINGHUA” under illumination of 910 nm and 685 nm. Beyond secure communication, the dual-narrowband OPD also demonstrates strong potential for anti-interference image sensing via NIR light66. To achieve a high-resolution NIR image with cost-effective method, we developed a single-pixel scanning imaging system based on the first-order peak at λ=910 nm in Fig. 5c. The as-obtained OPD was mounted on a mobile X-Y stage and photocurrent signals were recorded as a fixed light source illuminated a shadow mask featuring the word “TSINGHUA” (the abbreviation for Tsinghua University). To assess imaging contrast, we compared monochromatic LED illumination at λ= 910 nm and λ=685 nm. As the dual-narrowband OPD has a first-order peak spectral response at λ= 910 nm, it generated a significantly higher photocurrent than that at λ= 685 nm (Fig. S25). Through this method, a 20 × 300-pixel image of “TSINGHUA” is demonstrated in Fig. 5d, e with a pixel size at 2 × 1 mm. The photocurrent of the illuminated regions under λ= 910 nm illumination was 1000 times higher than that of the blocked areas, resulting in a highly distinguishable image. In contrast, under 685-nm illumination, the reconstructed image was barely visible, demonstrating that the effective SNR is substantially higher in our device due to its spectral selectivity. These results confirm the exceptional anti-interference capability of the dual-narrowband OPD, demonstrating its effectiveness in high-fidelity OWC and imaging applications. This approach paves the way for the development of secure and interference-resistant optical systems. Spectral tunability of dual-narrowband OPDs The tunability of the resonant peak in dual-narrowband OPDs can be achieved by varying the thickness of the organic spacer, as described by Eq. (1). Lumerical FDTD simulations, as shown in Fig. S26 demonstrate that increasing the Liq thickness from 200 nm to 250 nm shifts the peak optical field intensity from 820 nm to 980 nm, while maintaining a fixed Ag thickness of 30 nm. This confirms that adjusting the cavity thickness effectively tunes the resonant transmission peak. A more detailed transmission spectrum simulation (Fig. S27) further supports this concept, showing that when the cavity thickness increases from 180 nm to 280 nm, both the first-order and second-order resonant peaks shift to longer wavelengths, with a decrease in transmittance at the first-order peak as it moves toward longer wavelengths. To experimentally validate this tunability, we fabricated and measured a series of resonant cavities with varying thicknesses. As shown in Fig. 6a, by adjusting the cavity thickness from 80 to 300 nm, the resonant cavities exhibit narrowband transmittance with excellent tunability and high selectivity, covering the visible-to-NIR region. This demonstrates great potential for the development of arbitrarily tuned narrowband OPDs. Corresponding simulations in Fig. 6b also confirm that increasing the cavity thickness from 160 nm to 300 nm shifts both the first-order and second-order peaks to longer wavelengths. These results provide valuable insights into the fundamentals of resonant cavities, aiding in the design and fabrication of tunable dual-narrowband OPDs, particularly for NIR applications. Spectral tunability of dual-narrowband OPDs. a Normalized measured transmittance of first-order resonant peaks while changing cavity thickness from 80 to 300 nm. b Simulated transmittance of resonant peaks while changing cavity thickness from 160 to 300 nm. c–e D* of narrowband OPDs using different materials as photoactive materials and varied cavity thickness. c PM6:BTP-4F. d PTB7-Th:COTIC-4F. e PTB7-Th:L2. f Material structures of used organic semiconductors. g Comparison of external quantum efficiency (EQE) and full-width at half maximum (FWHM) as a function of wavelength from previously reported cavity narrowband OPDs37,38,41,58,69,70. To explore the versatility of this approach, we fabricated OPDs using different donor-acceptor materials with maximum absorption in different spectral regions. To achieve strong NIR absorption, we carefully selected three BHJ combinations: PM6:BTP-4F, PTB7-Th:COTIC-4F and PTB7-Th:L267,68. The detailed device structure, based on energy level matching and absorption spectrum shown in Fig. S28, indicates that PM6:BTP-4F possesses a higher absorption at the spectral region of 300–900 nm while PTB7-Th:COTIC-4F and PTB7-Th:L2, with their ultra-low bandgap, enhance absorption beyond 1100 nm, making them ideal for NIR detection (Fig. S29). By integrating resonant cavity tuning, we achieved high-selectivity narrowband OPDs. As shown in Fig. 6c–e, PM6:BTP-4F-based OPDs have 735–860 nm tuning with FWHMs of 23–26 nm and a detectivity up to 1.2 × 1012 Jones; PTB7-Th:COTIC-4F-based OPDs have 820–1040 nm tuning with a peak detectivity of 1 × 1012 Jones; PTB7-Th:L2-based OPDs have 1025–1215 nm SWIR tuning with a 40-nm FWHM at 1215 nm. Furthermore, the I-T curve of Fig. S30 demonstrates that the on-off ratio exceeds 103 at 1215 nm, with the FWHM of narrowband OPDs in Fig. S31, showing that despite the moderate EQE, the device maintains clean signal discrimination capability essential for OWC applications. Other optoelectronic performance metrics are given in Figs. S32–S35. To provide direct evidence of communication capability in the SWIR-adjacent region, we utilized the device with a peak response at 1080 nm (using PTB7-Th:L2 with a cavity thickness at 265 nm) and performed OWC experiments using a commercial 1064-nm laser diode. As shown in Fig. S36a, under 100-kHz square wave signal modulation, the current remained apparent with a high SNR. The frequency response measurement in Fig. S36b reveals a 3-dB bandwidth of approximately 120 kHz at 1064 nm, sufficient for moderate-speed OWC links. The eye diagrams at transmission data rate from 10 to 500 kbps (Fig. S36c–e) all exhibit clearly open eyes, confirming reliable data transmission with low inter-symbol interference. Only when the data rate exceeds 2.5 Mbps, the eye become distorted. These results demonstrate that our approach is not limited to the visible/NIR region but can be extended toward SWIR wavelengths while maintaining communication functionality. The success of this approach across various OPD material systems can largely be attributed to its versatility, primarily through the tuning of cavity dimensions and the targeted selection of wavelength-optimized active materials. All the chemical structures of the active materials used in this work are given in Fig. 6f. As shown in Fig. S37, the comparison of optoelectronic performance across different materials reveals that higher detection peaks in the NIR region can be achieved by altering the active materials. For instance, PM6:BTP-4F focuses on wavelengths below 830 nm, PTB7-Th:COTIC-4F targets wavelengths up to 1040 nm, and PTB7-Th:L2 is optimized for the range between 1040 nm and 1215 nm. These results demonstrate the potential for peak-tunable, high-speed narrowband detection. When compared with other resonant cavity-based tunable narrowband OPDs reported in the literature, our approach not only achieves comparable EQE values in the NIR region, but also offers the narrowest FWHM with superior selectivity (Fig. 6g)37,38,41,58,69,70. This highlights the exceptional capability of our approach in developing tunable narrowband OPDs. Hence, the performance of dual-narrowband OPDs is intrinsically linked to their digitally decoded applications, as their spectral selectivity and high detectivity enable secure, high-fidelity OWC system. Moreover, the digital encoding capability of this system relies on the dual-narrowband OPD’s ability to preserve signal integrity under interference and eavesdropping conditions. This capability is critically important for next-generation applications such as embodied intelligence and the Internet of Things, because these systems involve dense networks of autonomous agents and devices operating in dynamic, noisy real-world environments. The proposed digitally encoding scheme with hardware–software co-design ensures reliable and secure communication by intrinsically rejecting ambient optical interference and significantly raising the physical-layer barrier against eavesdropping, which is fundamental for trusted and resilient machine-to-machine and human-to-machine interactions. Additionally, the demonstrated application of single-pixel infrared imaging showcases the potential of this technology for high-contrast, digitally encoded imaging systems, where selective wavelength detection enhances object recognition and encrypted imaging. The proposed dual-narrowband OPD architecture not only offers unique functionality for secure OWC but also does so using a manufacturing blueprint that is inherently scalable and cost-competitive, thereby enhancing its potential for real-world adoption. Discussion In this study, we have developed a high-performance dual-narrowband OPD with high spectral selectivity, that is capable of simultaneous visible and NIR detection without the need for external optical filters. By integrating a strategically engineered FP resonant cavity with tailored organic active materials, we achieved intrinsic dual-narrowband photodetection at 485 nm (visible) and 910 nm (NIR), exhibiting a fast response time of 440 ns and an LDR exceeding 140 dB. The digitally encoded OWC system enabled by this dual-narrowband OPD demonstrated robust anti-eavesdropping capabilities, achieving a low BER of 9.17 × 10−5 at 1.25 Mbps. Unlike conventional broadband PDs, which suffer from undesired ambient radiation and require external optical filtering, our approach provides intrinsic spectral selectivity, minimizing spectral crosstalk and interference to significantly enhance the SNR and detection accuracy. Additionally, the device exhibited strong resilience against ambient noise and signal interception, ensuring high-fidelity OWC even in complex communication scenarios. Beyond optical communication, the tunable architecture of the dual-narrowband OPD offers broad spectral adaptability, extending into the SWIR region through precise cavity thickness modulation and molecular engineering. By demonstrating an unparalleled combination of spectral tunability, anti-interference capability, and digital encoding, this work paves the way for intelligent optoelectronic systems, setting a benchmark for the future of secure, high-performance photodetection and optical signal processing. Future directions lie in enhancing the response of OPDs in the SWIR region, including non-fullerene acceptor molecular engineering and the optimization of active-layer thickness and cavity-mirror reflectivity using dielectric DBRs, which can be potentially validated in multi-wavelength encrypted OWC systems71–73. Methods Materials The materials used in the experiments included chlorobenzene (99.9%, Sigma-Aldrich), chloroform (99.8%, Energy Chemical), isopropanol (99.8%, Macklin), acetone (99.8%, Aladdin), ethanol (99.7%, Aladdin), Poly[4,8-bis(5-(2-ethylhexyl)thiophen-2-yl)benzo[1,2-b;4,5-b’]dithiophene-2,6-diyl-alt-(4-(2-ethylhexyl) −3-fluorothieno[3,4-b]thiophene-)−2-carboxylate-2-6-diyl] (PTB7-Th) (99%, 1-Material), 2,2’-((2Z,2’Z)-(((4,4-bis(2-ethylhexyl)−4H-cyclopenta [2,1-b:3,4-b’]dithiophene-2,6-diyl)bis(4-(2-ethylhexyloxy)thiophene-5,2-diyl)) bis (methanylyli-dene))bis(5,6-difluoro-3-oxo-2,3-dihydro-1H-indene-2,1-diylidene))dimalononitrile (COTIC-4F) (99%, 1-Material). PM6, BTP-4F, PDINN, Phen-NADPO, 1-CN were purchased in 1-Materials. L2 was synthesized by Lin et al. and reported in previous publications68. Liq was purchased in Xi’an Yuri Solar Cor., Ltd. Other commercialized materials were purchased through official avenues. Device fabrication The schematic fabrication process of dual-narrowband OPD based on resonant cavity was shown in Fig. S3. The substrate is ITO/Glass substrate which are cleaned by ultrasonic cleaner in the acetone, ethanol, distilled water with one drop of cleanser essence (3 times) and ethanol for 20 min, respectively. ITO substrate was then treated with oxygen plasma for 5 min before use. PEDOT:PSS was spin coated on the surface of ITO/Glass for 30 s with speed at 3000 rpm, and annealed at 150 °C for 15 min. The PTB7-Th:COTIC-4F solution (dissolved in chlorobenzene of 20 mg/ml with the proportion of 1:1.5 and 2% vol% 1-Chloronaphthalene with overnight stirring) was then spin-coated for 30 s at 2000 rpm speed in glove box, following the annealing at 100 °C for 5 min. Phen-NADPO (dissolved in isopropanol with 0.5 mg/mL) was then spin-coated on the surface of active layer at the same parameters. Finally, the silver electrode with the thickness of 100 nm was deposited by thermal evaporation deposition. The resonant cavity was sequentially deposited through thermal evaporation deposition on the back of ITO/Glass substrate. The deposition speed of thermal evaporation deposition is lower than 1 Å/s. To produce smooth Ag mirrors, 3 nm MoO3 and 1 nm Au were used as seed layers. As for the device structure of ITO/PEDOT:PSS/PM6:BTP-4F/PDINN/Ag and ITO/PEDOT:PSS/PTB7-Th:L2 /Phen-NADPO/Ag, the fabrication process was similar to those mentioned above. PM6:BTP-4F and PTB7-Th:L2 were dissolved in chloroform with concentration of 16 mg/mL and 15 mg/mL at the ratio of 1:1.2 followed with stirring for 3 h. Characterization and measurements The cross-sectional SEM image was obtained using scanning electron microscopy (JEOL-7401). AFM surface morphologies were imaged using atomic force microscopy (Keysight 5500). UV-visible-NIR transmittance spectra were obtained using the HP 8453 spectrophotometer. All I–V and I–t curve measurements were carried out with a self-built system at room temperature. The optical part includes SC-pro and AOTF-PRO produced by OYSL and can produce 430–1450 nm light which provide the illumination needed by OPDs. The electrical part consists of a probe and a Keithley 2600B and can detect the current of the OPDs. The response speed was obtained using a frequency response system, which uses an arbitrary waveform generator (AWG, RIGOL DG812) to produce a square wave, which was received by the OPDs and amplified through a trans-impedance amplifier (TIA, ELECDEMO, IV-AMP 1814L). The signals were exhibited and stored through an oscilloscope (Tektronix MDO3054). The noise current was tested using an SR830 lock-in amplifier (Stanford Research Systems) and an SR570 low-noise current preamplifier (Stanford Research Systems). Optical simulation The optical simulation was performed by using Lumerical FDTD. Optical constants of refractive index (n) and extinction coefficient (k) of all layers were obtained from ellipsometry measurement (J. A. Woollam M-2000 spectroscopic ellipsometer platform). In the software, first import the optical parameters of the material’s refractive index and extinction coefficient and perform fitting. Then, the device model to be calculated is constructed by setting the geometric and optical parameters for each layer. The FDTD simulation region and light source are then configured, with boundary conditions and the wavelength and type of incident light specified. Monitors are added to record changes in parameters such as electric and magnetic fields. Additionally, a finer mesh can be incorporated to enhance the accuracy of the computation. Demonstration of encrypted OWC system 485-nm visible and 910-nm NIR lasers are utilized simultaneously to transmit their PAM4 signals in the linear addition mode. Under such arrangement of contribution, the payload sequences for the two lasers are both noticeably different from the original PAM4 sequence because they carry the components but not the complete original PAM4 signals. Thus, the information obtained by eavesdropping on the signals emitted by one of the lasers is meaningless. In other words, it is unnecessary for the secure OWC system to suspend communication when it is detected that the light signals from one of the lasers are eavesdropped on. In terms of eavesdropping detection and compensation, it is assumed that, in this work, eavesdropping on light signals results in the loss of the corresponding response voltage contribution in a fixed proportion. The pre-designed detection sequences are inserted between the payload sequences for the two lasers so the eavesdropping can be detected by monitoring the power loss proportion of the corresponding response voltage contribution of each light source. It also helps calculate the compensation of the corresponding response voltage when one light source is eavesdropped on or intercepted. In addition, the Lorenz sequence used for generating the contribution coefficients and the pilot sequences is based on the Lorenz seeds in the codebook shared by the transmitter and receiver sides. Following this scheme, we can realize secure OWC based on this dual-narrowband OPD. Demonstration of anti-interference OWC The intensity of each pixel of a 600 × 600-pixel monochrome image is quantized into 256 discrete levels, i.e., 8 bits. Thus, the square colored photograph with a side length of 600 pixels can be represented by 864,000 binary numbers. These binary numbers were arranged as a sequence rather than a matrix for easy processing, which is called the “plaintext sequence”. The transmitter and receiver contain two identical codebooks, containing several Lorenz chaos system seeds. The transmitter selects one of the seeds to generate the corresponding Lorenz sequence, which is equal to the length of the plaintext sequence. After the above two sequences are aligned, the bitwise exclusive OR (XOR) is performed to obtain a ciphertext sequence with the same length of the plaintext sequence. A training sequence of a certain length is added in front of the ciphertext sequence, which can be used to train the equalizer at the receiver side. The binary sequence is coded into an unequally spaced PAM4 symbol sequence, where the voltage levels corresponding to the symbols 0, 1, 2, 3 are 0, 0.25, 0.75, 1, respectively. After receiving the signals produced by the OPDs and extracting the ciphertext sequence and the training sequence, the least mean square equalizer was applied to the training sequence. In the ciphertext sequence, the symbols with amplitudes greater than or equal to 0.875 are decided to be 1, those between 0.5 and 0.875 are decided to be 0.75, those between 0.125 and 0.5 are decided to be 0.25, and those less than or equal to 0.125 are decided to be 0, which correspond to 10, 11, 01, 00, respectively. Therefore, the demodulation of unequally spaced PAM4 was completed and a binary sequence is obtained, which is a simple ciphertext sequence. According to the arithmetic characteristics of bitwise XOR, after aligning this ciphertext sequence with the same Lorenz sequence as that of the transmitter, the bitwise XOR is carried out, and the new sequence generated is the plaintext sequence, which is converted from binary to decimal and then arranged into a matrix. Finally, the intensity information of a square monochrome image or the intensity information of a certain channel of a colored image can be obtained. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary video Transparent Peer Review file Source data Source Data Acknowledgements The authors thank the support of Dr. Yueyue Wang and Prof. Hong Meng (Peking University-Shenzhen) for EQE measurements, Zehao Wang, Prof. Xiaomin Xu (Tsinghua-Shenzhen) for M-S and EIS measurements. The authors discloses support for the research of this work from the National Natural Science Foundation of China (grant numbers 52027817 and 52572175); Shenzhen Science and Technology Innovation Committee (grant number GJHZ20210705143204013); Guangdong Basic and Applied Basic Research Foundation (grant number 2025A1515011923). Author contributions Z.Y. and H.H. contributed equally to this work. G.W., H.Y.F. and F.K. conceived the ideas and supervised the project. Z.Y. carried out device fabrication. H.H. set up the system of OWC and performed data collection. Z.Y., W.H. and K.S. performed optical simulations. Z.Y., K.N.T., C.W. and X.W. designed and fabricated resonant cavities. T.L. and Yuan L. provided materials for universalization. D.Z. and Z.Y. carried out anti-interference imaging. K.K., H.L., Z.X., L.S. and C.S. performed device testing and characterization. Thanks for the help of H.Y., Yuze L., Z.Z. and F.K. for discussing and revising this manuscript. G.W. and H.Y.F. guided the experiments. All authors commented on the manuscript. Peer review Peer review information Nature Communications thanks Zuxing Zhang 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 that support the findings of this study are provided in the main text and the Supporting Information. More data are available from the corresponding author upon reasonable request. Source data are provided with this paper. Code availability The custom code used to collect and analyze the data has been deposited in a community repository “Code Ocean” (URL: https://codeocean.com/capsule/8958608, Capsule ID: 8958608). The codes used for the optical simulation and OWC are available from the corresponding author on reasonable 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. These authors contributed equally: Zilong Ye, Haoyu Huang. Contributor Information Zonglong Zhu, Email: zonglzhu@cityu.edu.hk. Feiyu Kang, Email: fykang@sz.tsinghua.edu.cn. H. Y. Fu, Email: hyfu@sz.tsinghua.edu.cn Guodan Wei, Email: weiguodan@sz.tsinghua.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-74011-1. References Associated Data Supplementary Materials Supplementary Information Description of Additional Supplementary Files Supplementary video Transparent Peer Review file Source Data Data Availability Statement The data that support the findings of this study are provided in the main text and the Supporting Information. More data are available from the corresponding author upon reasonable request. Source data are provided with this paper. The custom code used to collect and analyze the data has been deposited in a community repository “Code Ocean” (URL: https://codeocean.com/capsule/8958608, Capsule ID: 8958608). The codes used for the optical simulation and OWC are available from the corresponding author on reasonable request.