Second-level fast sintering enables high-brightness light source for visible light communication

Abstract

Laser-driven light sources are critically important for lighting, display, and communication systems. However, the traditional phosphor ceramics used in these light sources suffer from low light conversion efficiency, low color rendering index and narrowly tunable color temperature. Here we present a approach of rapid pulsed high-temperature sintering technology to fabricate phosphor ceramics with both dense and grain refined microstructure simultaneously. It results in a 31% improvement in light conversion efficiency, markedly better color mixing uniformity, and 1.5-fold increased fracture toughness for Lu₃Al₅O₁₂:Ce ceramics. Moreover, this method enables the Lu₃Al₅O₁₂:Ce–CaAlSiN₃:Eu composite ceramics with a 95-fold enhancement in quantum efficiency, showing white light under blue laser excitation with a color rendering index of 93 and good reliability. An accurate emission wavelength prediction model was also established with an interval below 5 nm from 495 to 605 nm, and achieved a broad color temperature range (1996-9803 K). This research enhances the development of high-performance light sources toward lighting and indoor/underwater visible light communication.

This study reports a rapid sintering technique with machine-learning-guided design to fabricate phosphor ceramics within only 10 s. Such ceramics show excellent performance in the laser-driven light sources for visible light communication.

Introduction

High-brightness light sources have garnered significant attention for their broad applications in cutting-edge technologies, including automotive lighting, projection displays, night vision systems, and visible light communication^1–10. Activating phosphor materials with blue laser diodes is one class of efficient ways for this aim, and phosphor ceramics are particularly notable for these light sources due to their superior reliability, thermal conductivity, and robustness under laser irradiation^6–13. The performance of these ceramics, in particular, both optically and thermally, is influenced by microstructural characteristics, such as relative density and grain size. High densification enhances the thermal conductivity of the ceramics, and small grain sizes can improve the light scattering, resulting in higher light conversion efficiency and superior light quality. It also benefits the mechanical properties of the ceramics^14–16. However, it is challenging to achieve simultaneous densification and grain refinement simultaneously, because densification is typically accompanied by grain coarsening during the sintering process^17,18. Introducing a secondary phase can inhibit boundary migration and refine the grain size^19,20. However, it may cause inhomogeneities and reduce the transparency of the ceramics^21,22. Alternative strategies were explored by using techniques like cold sintering, spark plasma sintering, and full crystallization from glass^16,23–25. Nevertheless, these methods are often limited to specific material systems and cannot be universally applicable. Moreover, for composite ceramics (such as oxide–nitride systems), these approaches still trigger severe interfacial reactions that markedly degrade luminescent performance^26,27. To overcome these limitations, it is urgent to develop innovative sintering strategies to realize both densification and grain refinement, and further improve the optical, thermal, and mechanical properties of phosphor ceramics.

Recently, the pulsed high-temperature technology has revolutionized the rapid fabrication of diverse materials, including high-entropy nanoparticles, glasses, ceramics, alloys, and cermets^28–34. It can rapidly achieve ultrahigh temperatures (such as 3000 °C) within milliseconds, together with a short duration from milliseconds to seconds³⁵. This method meets the thermodynamic prerequisites for ceramic consolidation and also provides kinetic advantages, which hold great promise to achieve both densification and grain refinement, a key point in this field. Moreover, the rapid sintering can facilitate compositional optimization and also accelerate the construction of a large sample library that encompasses a wide range of compositions. Recent advances in machine learning have opened new avenues for predicting the properties of luminescent materials^36–42. For example, the emission wavelength serves as a critical performance metric for luminescent materials, and an accurate prediction is essential for materials design. For the current machine learning studies, compiling databases often relies on wavelength data aggregated from published literature, which is subsequently utilized to train prediction models^38,39. However, the disparities between experimental protocols and testing conditions can introduce systematic biases and lead to substantial prediction inaccuracies^43,44. By harnessing the rapid sintering to construct a comprehensive database, it is highly feasible to mitigate these systematic biases. This would help build an accurate emission wavelength prediction model and thereby contribute to phosphor ceramics with tailored emission profiles and peak wavelengths.

In this study, we propose a rapid sintering strategy to realize simultaneous densification and grain refinement of garnet phosphor ceramics within just 10 s sintering. It effectively overcomes the long-standing trade-off in traditional sintering methods (Fig. 1a). We extensively engineered the nitride-oxide composite phosphor ceramics by using fast sintering to mitigate interfacial reactions, reaching a 95-fold enhancement in quantum efficiency (Fig. 1b). Furthermore, we constructed an extensive compositional library across the garnet family and established an accurate emission wavelength prediction model based on machine learning. The model further enables a series of samples with emission wavelengths ranging from 495 to 605 nm with a small interval below 5 nm (Fig. 1c). By integrating these phosphor ceramics with a blue laser, the high-brightness white light with a broad tunability in chromaticity was obtained. These advancements represent a substantial step forward in the development of phosphor ceramics for high-performance laser-driven light sources, and contribute to the future lighting technologies, indoor and underwater visible light communication and other interdisciplinary applications.

Fig. 1. Schematic of second-level fast sintering enables the fabrication of high-performance phosphor ceramics.

a Rapid sintering facilitates simultaneous densification and grain refinement in ceramics, with the green dots representing luminescence centers. b Rapid sintering controls interfaces in the Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics, resulting in a 95-fold increase in quantum efficiency. The red blocks denote the CaAlSiN₃:Eu particles. c The schematic illustrates how rapid sintering and machine learning help establish an accurate emission wavelength prediction model, producing phosphor ceramics with tunable emission wavelengths. The brown and blue circles represent the training and test sets in machine learning, respectively.

Results

The phosphor ceramics of Lu₃Al₅O₁₂:Ce were synthesized via an in-situ reaction sintering process by employing a pulsed high-temperature technique. The temperature was rapidly raised up to 1800 °C within milliseconds and then kept for several seconds to facilitate the sintering of Lu₃Al₅O₁₂:Ce. The evolution of phase and microstructure during the high-temperature holding period of 1–10 s was investigated (Fig. 2a, b, and Supplementary Figs. 1–3). During the initial time range of 1–4 s, scanning electron microscopy and transmission electron microscopy results indicate that Al₂O₃ and Lu₂O₃ remain confined within their respective precursor particles, with no observable overlap, suggesting minimal interaction at this stage. At the 5-s mark, solid-state interdiffusion is detected at the contact points between Al₂O₃ and Lu₂O₃, as evidenced by the formation of intermediate phases confirmed by X-ray diffraction in Supplementary Fig. 1. After 6 s, a distinct spatial overlap between Al and Lu elements is observed, along with the evident formation of the Lu₃Al₅O₁₂ phase. Beyond 10 s, energy-dispersive X-ray elemental mapping demonstrates complete elemental homogenization. The lattice fringes and diffraction patterns confirm the complete transformation of the precursors into a single-phase Lu₃Al₅O₁₂. At the same time, a dense microstructure with a relative density of 98.1% was achieved in just 10 s. As the holding time increased from 13 to 25 s, the grain size was adjusted in the sub-micrometer range, varying from 0.43 to 1.37 μm (Supplementary Fig. 4). In contrast, ceramics prepared through traditional sintering, with hour-scale holding times ranging from 0.5 to 6 h, exhibited much larger grain sizes in the range of 4.57–6.17 μm (Supplementary Fig. 5). Based on the grain sizes observed for different holding times, the grain growth exponents were determined to be 3 and 4 for rapid and traditional sintering methods, respectively (Supplementary Figs. 6, 7, and Supplementary Note 1). The results indicate that both processes involve pore drag, but differ in their diffusion mechanisms⁴⁵. Rapid sintering is governed by volume diffusion, while traditional sintering is dominated by surface diffusion. Both mechanisms contribute to densification and grain growth.

Fig. 2. Fabrication of Lu₃Al₅O₁₂:Ce phosphor ceramics using a fast pulsed high-temperature sintering technique at the second level.

a Phase content as derived from X-ray diffraction and corresponding scanning electron microscopy images in secondary electron mode during the high-temperature holding period of 1–10 s. Scale bars indicate 500 nm, and error bars represent standard deviations. b Energy-dispersive X-ray elemental mapping images and high-resolution transmission electron microscopy results obtained during in situ reaction sintering at holding times of 1, 6, and 10 s, with the corresponding phases marked. c In-line and total transmittance curves of Lu₃Al₅O₁₂:Ce phosphor ceramics prepared via rapid and traditional sintering methods. The inset shows a photograph of transparent Lu₃Al₅O₁₂:Ce prepared by rapid sintering, with a ruler placed below to indicate its size. d Spatial distribution of correlated color temperature (CCT) of the light mixture, consisting of residual blue light and phosphor-converted light. e Fracture toughness of Lu₃Al₅O₁₂:Ce phosphor ceramics prepared using rapid and traditional sintering methods. Error bars indicate standard deviations. Source data are provided as a Source data file.

Furthermore, we would like to clarify that the driving force remains consistent between rapid sintering and traditional sintering. In both methods, the raw materials undergo the formation of intermediate phases, which subsequently transform into garnet (Supplementary Figs. 1 and 8). In rapid sintering, garnet grains nucleate from these intermediate phases, with rapid high temperature facilitating numerous small nucleation events. The accelerated high temperature process results in faster grain bonding and reduced opportunities for grain growth (Supplementary Fig. 4). In contrast, traditional sintering, due to its extended duration, provides more opportunities for larger grains to engulf smaller ones, leading to grain growth and coarsening (Supplementary Fig. 5). It suggests that the rapid sintering technique offers a significant advantage in achieving microstructure densification together with grain refinement concurrently.

We subsequently evaluated the optical and mechanical properties of Lu₃Al₅O₁₂:Ce phosphor ceramics fabricated using rapid or traditional sintering techniques. The Ce³⁺ doping concentration in the Lu₃Al₅O₁₂ ceramics was set at 0.007%, as this concentration yielded the highest quantum efficiency and absorption efficiency under 450 nm excitation (Supplementary Fig. 9). The Lu₃Al₅O₁₂:Ce ceramics exhibit similar total transmittance values (89 and 92% at 800 nm), approaching the theoretical limit¹⁶. As illustrated in Fig. 2c, refining grain size can increase light scattering and lead to a lower in-line transmittance (13% at 800 nm) for the ceramics with sub-micrometer grain sizes, in contrast to the larger grain size counterparts (33% at 800 nm). The high total transmittance and low in-line transmittance synergistically enhance light conversion efficiency and uniform light-mixing for the Lu₃Al₅O₁₂:Ce ceramics when excited by blue laser light. Figure 2d shows the spatial distribution of the correlated color temperature (CCT) of the mixed light consisting of residual blue light and phosphor-converted light, upon irradiation with a 5 W blue laser, indicating that grain refinement significantly improves the Lambertian distribution (Supplementary Fig. 10). Additionally, refining grain size in the Lu₃Al₅O₁₂:Ce ceramics leads to a 31% increase in light conversion efficiency, because it promotes the absorption and conversion of blue laser light (Supplementary Figs. 11, 12, Supplementary Note 2 and Supplementary Table 1). Importantly, they show high quantum efficiency (92.0%) and absorption (90.0%) (Supplementary Figs. 13 and 14). Refining grain size also helps improve hardness and toughness. The ceramics with sub-micrometer grain sizes exhibit higher hardness (15.90 ± 1.88 GPa) compared to that with larger grain sizes (14.70 ± 1.30 GPa), as determined by Vickers hardness testing under loads ranging from 1.96 to 9.60 N (Supplementary Figs. 15 and 16). Importantly, the fracture toughness is over 1.5-fold increase from 1.08–0.67 to 1.35–0.88 MPa m^1/2 due to grain refinement (Fig. 2e and Supplementary Note 3). These results confirm the pivotal role of grain refinement and microstructure densification via rapid sintering in enhancing optical and mechanical properties.

The Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics were further prepared using the rapid sintering technique. The raw material powders, including Lu₂O₃, Al₂O₃, and CeO₂ for the in-situ reaction sintering of Lu₃Al₅O₁₂:Ce, were initially mixed with CaAlSiN₃:Eu powder. Green bodies were obtained following the shaping process, and then heated at 1650 °C for 10 s to form Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics with a relative density exceeding 98.0%, and no other phase was observed (Supplementary Fig. 17). Under blue light excitation, the composite ceramics show much broad emission spectrum, which along with the residual blue light result in white light output with a high color rendering index (>90) (Fig. 3a). Moreover, the emission spectra can be broadly tailored by simply adjusting the weight ratio of CaAlSiN₃:Eu to Lu₃Al₅O₁₂:Ce (Supplementary Fig. 18). In contrast, there exist impurities and significant interfacial reactions between the two phases for the composite ceramics prepared via traditional spark plasma sintering (Supplementary Figs. 17 and 19), resulting in a marked decrease in the red emission intensity for CaAlSiN₃:Eu. Meanwhile, the rapid sintering achieves a high quantum efficiency of 57.0%, much higher than that prepared by traditional methods (0.6%) (Supplementary Fig. 20). Microstructural analyses revealed that CaAlSiN₃:Eu grains, preserving the morphology of the original powder with a particle size of ~15 μm, were dispersed within the dense Lu₃Al₅O₁₂:Ce matrix, which had a particle size of about 300 nm (Supplementary Figs. 21 and 22). Clear interfaces between the two phases were observed (Fig. 3b). The thickness of the interface layer is several tens of nanometers according to the high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image (Fig. 3c). Elemental mapping images demonstrated that the interface layer is rich in aluminum, and CaAlSiN₃:Eu and Lu₃Al₅O₁₂:Ce are distributed on the two sides of the interface (Fig. 3d). High-resolution transmission electron microscopy (HRTEM) was employed to analyze the interface phase, revealing very sharp three-phase grain boundaries, with all three phases being well-crystallized (Fig. 3e and Supplementary Fig. 23). The interplanar spacings of 4.23 and 4.91 Å were attributed to the ($\overline{1}21$) and ($1\overline{2}1$) planes of Lu₃Al₅O₁₂, respectively. The interplanar spacings of 2.84 and 4.85 Å corresponded to the (002) and ($0\overline{1}\overline{1}$) planes of CaAlSiN₃, respectively. Additionally, the thin interface phase was identified as CaAl₄O₇ according to the interplanar spacings of 2.46 Å ($00\overline{2}$) and 2.71 Å (110).

Fig. 3. Characterization of Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite phosphor ceramics via a fast pulsed high-temperature sintering technique.

a Emission spectra of the phosphor ceramics prepared via rapid sintering and traditional spark plasma sintering (excited at 450 nm). The inset shows a photograph of the ceramic prepared by rapid sintering under blue light excitation. b Cross-sectional scanning electron microscopy image in secondary electron mode. c High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image, with the interface layer highlighted by yellow lines. d Energy-dispersive X-ray elemental mapping images, with the interface layer marked by yellow lines. The scale bar represents 100 nm. e High-resolution transmission electron microscopy (HRTEM) image of the three-phase grain boundaries, follow by HRTEM images of the three phases. Insets are the corresponding selected area electron diffraction patterns. f Schematic of the microstructure formation process of the composite ceramic, where Lu₂O₃ preferentially reacts with Al₂O₃ to form Lu₃Al₅O₁₂, and CaAlSiN₃:Eu also reacts with Al₂O₃ to form CaAl₄O₇ at the interface, inhibiting further reactions. g Light efficiency loss of the phosphor ceramic or phosphor-in-glass packaged with an 8 W blue-LED after aging for 6000 h. The inset presents a photograph of the packaged device. Source data are provided as a Source data file.

To clarify the formation process of the interface phase, the green bodies of Lu₂O₃-CaAlSiN₃:Eu and Al₂O₃-CaAlSiN₃:Eu composites were sintered at 1650 °C for 10 s. The results indicate that the interfacial reaction between Lu₂O₃ and CaAlSiN₃:Eu is notable, characterized by significant interdiffusion of Al and Lu elements, which completely quenches the red emission of CaAlSiN₃:Eu (Supplementary Fig. 24). In contrast, the interfacial reaction between Al₂O₃ and CaAlSiN₃:Eu phases is significantly diminished, allowing for intense red emission of CaAlSiN₃:Eu. The thickness of the interfacial layer is approximately several hundred nanometers and is identified as the CaAl₁₂O₁₉ phase (Supplementary Fig. 25). As shown in Fig. 3f, these findings indicate that in the Al₂O₃-Lu₂O₃-CaAlSiN₃:Eu three-phase reaction system, Lu₂O₃ preferentially reacts with Al₂O₃ to form Lu₃Al₅O₁₂, rather than undergoing an interfacial reaction with CaAlSiN₃:Eu. Concurrently, CaAlSiN₃:Eu also reacts with Al₂O₃ to form CaAl₄O₇ at the interface that inhibits further reactions. Moreover, the competition between CaAlSiN₃:Eu and Lu₂O₃ for Al₂O₃ leads to a thinner (tens of nanometers) interface layer of CaAl₄O₇ compared to the thicker Al-abundant CaAl₁₂O₁₉ (hundreds of nanometers) in the Al₂O₃-CaAlSiN₃:Eu binary system (Supplementary Note 4). Therefore, Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics were successfully prepared by effectively minimizing the interfacial reaction via the rapid sintering technique with a 95-fold enhancement in quantum efficiency. Furthermore, the optical properties of Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics were comparably studied with those of the traditional Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu phosphor-in-glass (PiG). The quantum efficiency and thermal conductivity of the ceramics are superior to those PiGs (Supplementary Figs. 26 and 27), suggesting that the ceramics are better suited for applications in high-power light sources. Upon excitation with a blue laser, the ceramics can withstand a high blue laser power density of 7.96 W mm⁻², much higher than that of PiGs (4.50 W mm⁻²) (Supplementary Fig. 28).

Next, the phosphor ceramics with diverse compositions were prepared using the pulsed high-temperature technique toward high-throughput experimentation. The second-scale heating characteristics of the rapid sintering process are demonstrated to effectively suppress the volatilization of elements, such as Ga, thereby broadening the compositional range (Supplementary Fig. 29)⁴⁶. This approach helps generate a substantial dataset that is crucial for machine learning algorithms aimed at developing prediction models for key luminescent properties (such as emission peak wavelengths). Taking the A₃B₅O₁₂ garnet structure as a representative case, where A occupies the dodecahedron site, and B locates at either the tetrahedral (B₁) or octahedral (B₂) site, we initially engineer 48 classes of Ce³⁺-doped garnet phosphors to establish a preliminary sample library to tailor the emission wavelengths through A/B site engineering. These samples were prepared using the rapid sintering method, and their luminescence spectra were recorded (Supplementary Fig. 30). Meanwhile, we performed first-principles calculations on the samples in our initial library to obtain their geometric and chemical features that influence the emission peak wavelengths. By employing machine learning, we established a correlation between the experimentally determined emission peak wavelengths and the computationally derived features, and constructed a prediction model for emission peak wavelengths (Fig. 4a). The model was then used to predict new compositions and thereby expand the sample library.

Fig. 4. Machine learning predictions for phosphor ceramic compositions and rapid experimental verification.

a Schematic of a closed-loop feedback system that integrates rapid experimental verification with machine learning predictions, utilizing geometric and chemical features obtained through first-principles calculations. b Pearson correlation coefficient matrix identifying ten uncorrelated features. c The four most important features: electron affinity (EA), Pauling electronegativity (χ), average coordination radius of the A site (avg_r_A), and first ionization potential (IP). d Training performance metrics of coefficient of determination (R²) and mean absolute error (MAE) for the four machine learning models. e The MAE decreases as the sample library expands from 48 to 64. f Prediction model of emission peak wavelength with a small error of 7.22 nm. g Broad tuning of emission peak wavelengths from 495 to 605 nm with a small interval of 5 nm derived from experiments. The solid lines indicate the initial sample library, while the dashed lines represent the predicted new samples. Source data are provided as a Source data file.

To highlight the key features of the model, we utilized the Pearson correlation coefficient matrix to identify 10 uncorrelated features (Fig. 4b, Supplementary Fig. 31, and Supplementary Table 2). Among them, electron affinity (EA), Pauling electronegativity (χ), average coordination radius of the A site (avg_r_A), and first ionization potential (IP) emerged as the four most important features (Fig. 4c). According to phenomenological luminescence theory, these features play a crucial role in determining the covalency of the host materials (EA and χ), the crystal field strength around Ce³⁺ (avg_r_A), and the host’s band structure (IP), which further affect the emission peak wavelengths⁴¹. Four machine learning models were selected to train our dataset, among which the random forest (RF) model outperforms the others in terms of training performance (Fig. 4d and Supplementary Table 3). This model achieved a coefficient of determination (R²) of 0.84, a mean absolute error (MAE) of 7.22 nm, and a root mean square error of 8.67 nm on the test set (Supplementary Fig. 32). These metrics indicate that the model has a high prediction accuracy (R² > 0.8) with a low error (MAE < 7.5 nm). By establishing a closed-loop feedback system that integrates machine learning predictions with rapid experimental verification, we expanded our sample library from 48 to 64 samples. The iterative process progressively reduced prediction error from 9.79 to 7.22 nm (Fig. 4e). The model successfully predicted and validated four new compositions, confirming its high prediction accuracy (Fig. 4f and Supplementary Table 4). These results enabled us to rapidly tune the emission peak wavelengths of Ce³⁺-doped A₃B₅O₁₂ garnet phosphor ceramics across a broad spectral range from 495 to 605 nm with a small interval of 5 nm (Fig. 4g and Supplementary Table 4). Importantly, the quantum efficiency is as high as at shorter wavelengths, and it is all above 70% for most of the wavelengths owing to the grain refinement by rapid sintering (Supplementary Table 5).

The application of as-prepared phosphor ceramics in visible light communication was demonstrated by integrating a 15 W blue laser to produce high-brightness and chromaticity-tunable light sources. Our initial step involves constructing an optical signal communication setup, which includes a signal sending PC end, an encoder, a light source device, a decoder, and a signal receiving PC end (Fig. 5a). The optical signal communication operates as follows: the textual information is initially input into the XCOM port of the signal sending PC, and then encoded into a light signal by modulating the driving current of the light source. This light signal is subsequently collected and decoded back into the original text information. Since the information was carried by visible light, its propagation can be blocked by setting obstacles (Supplementary Fig. 33 and Supplementary Movie 1), which facilitates visual protection of the signal information. To assess the stability of our optical signal communication setup, we ran it continuously for 108 h (Fig. 5b). The signal waveforms recorded at 20-h intervals demonstrated that the setup remains stable with no information distortion during transmission. The identical text at both sending and receiving ends confirms high reliability with zero error rate for the transmitted information (Supplementary Fig. 34).

Fig. 5. Applications for visible light communications.

a Schematic of the communication system. b Signal waveforms recorded at 20-h interval during continuous operation over 108 h. c Optical quality (inset) and communication distance of the white light source based on Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics. d Underwater optical communication distance of the green light source based on Lu₃Al₅O₁₂:Ce ceramics, with the inset showing a photograph of the underwater measurement. e Comparison of communication distances between the reported literature^47–56 and this study. Photographs (f) and chromaticity coordinates (g) of white-light sources with tunable correlated color temperatures (CCTs) utilizing the designed A₃B₅O₁₂:Ce ceramics. Source data are provided as a Source data file.

To investigate the feasibility of indoor lighting and visible light communication, we evaluated the optical quality and communication distance of the white light source based on Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics (Fig. 5c). The color coordinates are located at (0.39, 0.36) with a color temperature of 3758 K and a color rendering index of 93, meeting the chromatic requirements for indoor lighting. The white light source supports the light signal communication over a distance of 4.8 m (Supplementary Movie 2), indicating its potential for dual-purpose of lighting and communication. Furthermore, the underwater optical communication was evaluated by using the high-brightness green light source based on Lu₃Al₅O₁₂:Ce ceramics (Fig. 5d), achieving a significant communication distance exceeding 6.0 m, even after illuminance attenuation by water (Supplementary Movie 3). The laser-driven light sources, produced by combining blue lasers with phosphor ceramics, offer advantages of lower cost and longer communication distances compared to other light sources such as light-emitting diodes (Fig. 5e and Supplementary Table 6)^47–56. Additionally, we fabricated white light sources with tunable CCTs ranging from 9803 to 1996 K by employing the designed A₃B₅O₁₂:Ce ceramics to cater to different scenarios (such as aircraft, medical facilities interconnection and emergency scenes) (Fig. 5f, g, and Supplementary Table 7). Taking the white light source with CCT of 1996 K as an example, the indoor communication distances were measured to be 5.0 m, fulfilling the specifications for visible light communication (Supplementary Fig. 35). These findings highlight the significant potential of high-brightness white light sources for long-distance visible light communication.

Discussion

We demonstrated a rapid pulsed high-temperature technique to fabricate Lu₃Al₅O₁₂:Ce phosphor ceramics with both densification and grain refinement. The optimized microstructure led to a synergistic enhancement in both optical and mechanical properties for the phosphor ceramics, showing a 31% increase in light conversion efficiency and a nearly 1.5-fold increase in fracture toughness. The Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite phosphor ceramics were further engineered by effectively reducing the interfacial reaction layer with a 95-fold enhancement in quantum efficiency. Under an 8 W blue light excitation, these composite ceramics produced white light with a high color rendering index of 93 and good reliability with only a 14.4% luminance degradation after 6000 h irradiation. By integrating rapid sintering with machine learning, we further constructed an extensive compositional library across the garnet family and developed an accurate model for predicting emission wavelengths from 495 to 605 nm with a small interval of 5 nm. This further allows a wide tuning of color temperature from 1996 to 9803 K. These light sources demonstrated a zero bit-error-rate in both long-distance and prolonged communication tests, highlighting their potential for multi-scenario illumination-communication applications.

Methods

Chemical materials

Lu₂O₃, Y₂O₃, Tb₄O₇, and CeO₂, all with purities of 99.99%, were obtained from Baotou Research Institute of Rare Earths. MgO, SiO₂, HNO₃, C₆H₁₇N₃O₇, NH₄HCO₃, Si(OC₂H₅)₄, and (C₈H₁₄O₂)_n, also with purities of 99.99%, were sourced from Aladdin Reagent Co., Ltd. Gd₂O₃ (99.99%) was acquired from Alfa Aesar Chemical Co., Ltd., while Ga₂O₃ (99.99%) was purchased from Zhongnuo New Materials Co., Ltd. Al₂O₃ (99.99%) was obtained from Taimei Chemical Co., Ltd. Additionally, CaAlSiN₃:Eu and Lu₃Al₅O₁₂:Ce were purchased from Beijing Nakamura Yuji Co., Ltd. Anhydrous ethanol was sourced from Shantou Dahao Fine Chemicals Co., Ltd., and Al₂O₃ grinding balls (99.50%) were supplied by Qidong City Honghong Instruments and Equipment Factory. The glass frit is composed of SiO₂, Al₂O₃, P₂O₅, B₂O₅, K₂O, Na₂O, Li₂O, BaO, and TiO₂, with a glass transition temperature of 385°C and a softening temperature of 400 °C.

Preparation methods

Nano-sized Lu₂O₃ powder was synthesized using a co-precipitation method. A mixture of 20 mL of nitric acid and 30 mL of deionized water were prepared in a beaker to obtain a total volume of 50 mL. Subsequently, 6 g of micron-sized Lu₂O₃ powder was added and dissolved under stirring for 30 min to obtain a clear solution. The solution was then transferred into a 100 mL autoclave and subjected to a hydrothermal reaction at 120 °C for 24 h, resulting in a Lu(NO₃)₃ precursor solution with a specific molar concentration. Meanwhile, a 1.5 mol L⁻¹ NH₄HCO₃ solution was prepared as the precipitant. The precursor solution was added dropwise into the NH₄HCO₃ solution at a rate of 3 mL min⁻¹ to initiate the precipitation reaction, while maintaining the pH of the reaction solution at 7.5 ± 0.3. Upon completion of the reaction, the resulting complex was aged for 24 h. The precipitate was washed by centrifugation three times with deionized water and three times with ethanol, followed by drying at 100 °C for 12 h. The dried product was ground using a mortar, yielding crystalline nano-sized Lu₂O₃.

Other raw materials such as Y₂O₃, Gd₂O₃, Tb₄O₇, and Al₂O₃ were used as commercially available. The raw materials of 10 g were weighed according to stoichiometric ratios, incorporating 0.1 g of ammonium citrate (C₆H₁₇N₃O₇) as a dispersant and 50 μL of tetraethyl orthosilicate (Si(OC₂H₅)₄) as a sintering aid. Anhydrous ethanol was used as a solvent at a mass ratio of 1:1 relative to the powder. The mixture was combined with Al₂O₃ grinding balls at a ball-to-powder mass ratio of 12:1 in a 250 mL nylon jar, using grinding balls of 5 mm and 2 mm diameters in a 4:1 ratio. The milling was conducted at 321 rpm for 33 h. The resulting slurry was dried at 80 °C, sieved through a 200-mesh screen, and 0.1 g of (C₈H₁₄O₂)_n was added and mixed with ethanol for wet milling. This mixture was then divided into portions of 0.16 g each, pressed under 40 kN for 2 min to form green pellets with a diameter of 10 mm and a thickness of 1 mm. These pellets were pre-sintered at 800 °C for 8 h to remove organic materials, followed by cold isostatic pressing at 250 MPa for 3 min, achieving a green relative density of 55–60%. Rapid sintering was conducted using a pulsed high-temperature sintering apparatus (JSJ150-II, Anhui Jiashuo Vacuum Technology Co., Ltd.). The sample was placed between two layers of carbon cloth (2 cm × 5 cm) in direct contact with the heating element, with a contact area of about 1 cm². A DC power supply facilitated rapid heating under a vacuum pressure of 10⁻³ Pa, with temperatures ranging from 1750 to 1800 °C and a holding time of 1–30 s. The temperature was calibrated using an infrared pyrometer focused on the sample surface, set at an emissivity of 0.85. The holding time was measured from the moment power was applied, ranging from 1 to 30 s, followed by natural cooling. The resulting ceramics exhibited a diameter of about 7 mm and a thickness of around 0.7 mm. For comparative study, the green bodies were also sintered using traditional methods in a vacuum furnace (MCGS II, Shanghai Chenhua Technology Co., Ltd.), under a vacuum pressure of 10⁻³ Pa, at temperatures ranging from 1580 to 1780 °C for 1–3 h, with both heating and cooling rates maintained at 10 °C min⁻¹.

The ball milling mixture of Lu₂O₃, Al₂O₃ and CeO₂ was then mixed with commercially available CaAlSiN₃:Eu in a mortar at varying weight ratios ranging from 7:1 to 13:1 for the sintering of Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu composite ceramics. For rapid sintering, the powder mixture was initially pressed under a load of 40 kN for 2 min to form green bodies, followed by cold isostatic pressing at 250 MPa for 3 min. They were then subjected to rapid sintering in an environment of high-purity nitrogen at atmospheric pressure, at a temperature of 1650 °C for a duration of 1–30 s. For spark plasma sintering (SPS-211H, Fuji Electric), the mixture was placed into a graphite mold with an inner diameter of 10 mm, lined with high-purity graphite paper. To maintain the sintering temperature, the outer surface of the graphite mold was insulated with a layer of carbon felt. Prior to sintering, the chamber was maintained at atmospheric pressure using high-purity nitrogen. The heating rate was set at 250 °C min⁻¹ until reaching a sintering temperature of 1650 °C, which was held for 30 min under a uniaxial pressure of 60 MPa. After natural cooling, the resulting ceramic sample was mechanically thinned to a thickness of 0.7 mm.

For the sintering of Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu phosphor-in-glass, commercially available Lu₃Al₅O₁₂:Ce and CaAlSiN₃:Eu phosphor powders were mixed with glass frit in a weight ratio of 1:1. The resulting mixture was pressed under a load of 40 kN for 2 min to form green bodies. They were subsequently placed in a crucible and transferred to a muffle furnace (NBD-M1600, Zhengzhou Nobadi Material Technology Co., Ltd.). The heating profile was set to a rate of 5 °C min⁻¹, reaching a peak temperature of 490 °C, which was maintained for 30 min, followed by furnace cooling to obtain the Lu₃Al₅O₁₂:Ce-CaAlSiN₃:Eu phosphor-in-glass.

Characterization

Relative density was measured using the Archimedes method. The dry weight of the ceramic sample was recorded with a high-precision electronic balance (BS210S, Beijing Sartorius Balance Co., Ltd.). The sample was then boiled in distilled water for 30 min, after which the wet weight was measured. The suspended weight was obtained by placing the sample in a density determination kit. The microstructure of the phosphor ceramics was examined using a field-emission scanning electron microscope (SU8600, Hitachi) equipped with energy-dispersive X-ray spectroscopy, at an accelerating voltage of 10 kV and an emission current of 15 μA. To obtain optimized surface details, a combination of the upper detector and lower detector was employed to capture secondary electron signals, providing both high resolution and excellent topographic contrast. Scanning electron microscopy images were captured in secondary electron mode. The phase and crystal structure were identified with an X-ray diffractometer (D8 ADVANCE, Bruker), operated at a scanning rate of 2° min⁻¹ with a step size of 0.02°. High-temperature in-situ phase analysis was conducted using a high-temperature X-ray diffractometer (Panalytical, Empyrean), with measurements taken between 1000 and 1500 °C at 100 °C intervals. Transmission electron microscopy, high-resolution transmission electron microscopy, high-angle annular dark-field scanning transmission electron microscopy, selected area electron diffraction, and elemental mapping were performed using a transmission electron microscope (Talos F200s, FEI) at an accelerating voltage of 200 kV. Internal quantum efficiency and absorption were measured using a home-made integrating-sphere spectroradiometer setup, which comprised a single-mode laser diode (50 mW, 450 nm), a 15 cm integrating sphere, and a spectrometer (USB2000+, Ocean Optics). Photoluminescence spectra were acquired with a spectrofluorometer (FLS1000, Edinburgh Instruments) with a 450 W Xe lamp. Thermal quenching was assessed using a temperature controller (T96-S, Linkam Scientific Instruments Ltd.), a heating stage (TS1500V, Linkam Scientific Instruments Ltd.), a cooling water pump (WCP, Linkam Scientific Instruments Ltd.), a vacuum pump (DUO 3, Linkam Scientific Instruments Ltd.), and a spectrometer (QE65 Pro, Ocean Optics). Transmittance spectra were measured using a spectrophotometer (UV-3600Plus, Shimadzu). Spatial distribution of CCT and luminance of the light mixture was evaluated with a multi-angle illuminance detection device, which consists of an arc track with a radius of 30 cm, a photodetector, and a spectral irradiance colorimeter (SPIC-200, Everfine). Hardness and fracture toughness were assessed using a digital micro-Vickers hardness tester (1–1000 gf, Beijing Time High Technology Ltd.). Thermal conductivity (κ) was calculated using the formula:

$$k=\alpha \cdot C_p\cdot \rho$$ 1

where α represents thermal diffusivity, C_p denotes specific heat capacity, and ρ indicates bulk density. The thermal diffusivity was measured using a laser flash apparatus (LFA447, ETZSCH-Gerätebau GmbH). Samples were prepared as disks with a diameter of 12.7 ± 0.2 mm and a thickness of 0.5–0.6 mm, with both sides coated in a thin layer of graphite. Measurements were conducted from room temperature to 200 °C. The bulk density was determined using the Archimedes method, while specific heat capacity was obtained via a differential scanning calorimeter (DSC 204 F1, Netzsch-Gerätebau GmbH) over the same temperature range. Three identical blue laser diodes (NUBM0F, Nichia), each with a power output of 5.0 W at a driving current of 3.0 A, were integrated into a laser diode array, resulting in a total power output of 15.0 W with a laser spot size of 0.79 mm². These diodes served as the excitation source for the phosphor ceramics within the light source device of the visible light communication system. An oscilloscope (4 series, Tektronix) was employed to visualize waveform changes during optical signal transmission. Photographs were captured with a digital camera (A7S3, Sony). The long-term reliability was evaluated by mounting the phosphor ceramics or phosphor-in-glass on a 450 nm blue light-emitting diode chip, which was continuously driven at an input power of 8 W, and recording the luminescent efficacy over time. The visible light communication system utilized an STM32F103C8T6 microcontroller on a minimum system board. The serial data from the MCU was fed into an LED driving circuit, which modulated the light output of a blue LED accordingly to convert the electrical signal into an optical signal. The MCU was programmed via the SWD interface using a J-Link debugger. The modulation format was OOK, generated by toggling the LED based on the serial data stream. The development environment was STM32CubeIDE with the standard peripheral library. On the receiver side, the same STM32F103C8T6 platform was used for signal decoding. A photodetector captured the optical signal and converted it into an electrical signal. This signal was then fed into a comparator circuit, which compared it with a reference voltage to recover the original digital waveform. The resulting digital signal was directly transmitted to the UART receive pin of the MCU. The decoded data was sent to the PC via the USB virtual COM port.

Author contributions

S.L. conceived the idea. L.Q., C.H., and G.W. conducted the experiments. G.W. analyzed the data and drafted the initial manuscript. W.X. performed first-principles calculations and machine learning. Y.L. carried out long-term reliability tests. Y.H. contributed to the machine learning. R.-J.X. provided valuable discussions and guidance through the project. S.L. revised the manuscript.

Peer review

Peer review information

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

Funding

This work is supported by the National Key Research and Development Program of China (2023YFB3506600, 2023YFB3506601), the National Natural Science Foundation of China (52572182, 52272165), the Fujian Provincial Natural Science Foundation of China (2024J09009), the Fundamental Research Funds for the Central Universities (20720250120), and the Xiamen Major Science and Technology Project (3502Z20231050).

Data availability

All relevant data generated in this study can be found within the manuscript and/or Supplementary materials. Source data are provided with this paper.

Code availability

The code that supports the findings of this study is provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains Supplementary material available at 10.1038/s41467-026-74741-2.