# Fileset

[3-12-Revised Manuscript without Any Highlighting.docx](https://mdr.nims.go.jp/filesets/566276f5-877f-4254-afa8-a6994e5eebfc/download)

## Creator

[Hong Ming](https://orcid.org/0000-0003-0118-1897), Yayun Zhou, [Maxim S. Molokeev](https://orcid.org/0000-0002-8297-0945), Chuang Zhang, Lin Huang, Yuanjing Wang, [Hong-Tao Sun](https://orcid.org/0000-0002-0003-7941), [Enhai Song](https://orcid.org/0000-0003-1666-0532), [Qinyuan Zhang](https://orcid.org/0000-0001-6544-4735)

## Rights

This document is the Accepted Manuscript version of a Published Work that appeared in final form in ACS Materials Letters, copyright ©  2024 American Chemical Society after peer review and technical editing by the publisher. To access the final edited and published work see　https://doi.org/10.1021/acsmaterialslett.4c00263.[In Copyright](http://rightsstatements.org/vocab/InC/1.0/)

## Other metadata

[Machine-Learning-Driven Discovery of Mn<sup>4+</sup>-Doped Red-Emitting Fluorides with Short Excited-State Lifetime and High Efficiency for Mini Light-Emitting Diode Displays](https://mdr.nims.go.jp/datasets/2e11728e-fc84-46cb-9963-35512cf610e4)

## Fulltext

Machine Learning Driven Discovery of Mn4+-Doped Red-Emitting Fluorides with Short Excited-State Lifetime and High Efficiency for Mini Light-Emitting Diode DisplaysHong Ming,1,2,3 Yayun Zhou,1,2,4,* Maxim S. Molokeev,5,6,7 Chuang Zhang,1,8 Lin Huang,9 Yuanjing Wang,1,8 Hong-Tao Sun,3 Enhai Song,1,2,* Qinyuan Zhang1,2,8,*1State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, South China University of Technology, Guangzhou 510641, China2School of Materials Science and Engineering, South China University of Technology, Guangzhou 510641, China3International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science (NIMS), 1-2-1 Sengen, Tsukuba 305-0047, Japan4Guangdong-Hong Kong-Macao Joint Laboratory for Intelligent Micro-Nano Optoelectronic Technology，School of Physics and Optoelectronic Engineering, Foshan University, Foshan 528225, China.5Laboratory of Crystal Physics, Kirensky Institute of Physics, Federal Research Center KSC SB RAS, Krasnoyarsk 660036, Russia6Laboratory of Theory and Optimization of Chemical and Technological Processes, University of Tyumen, Tyumen 625003, Russia7Institute of Engineering Physics and Radioelectronics, Siberian Federal University, Krasnoyarsk 660041, Russia8School of Physics and Optoelectronics, South China University of Technology, Guangzhou 510641, China9Fujian Key Laboratory of Surface and Interface Engineering for High Performance Materials and College of Materials, Xiamen University, Xiamen 361005, China*Corresponding author E-mail: msehsong@scut.edu.cn (Enhai Song), zhouyayun@fosu.edu.cn (Yayun Zhou), qyzhang@scut.edu.cn (Qinyuan Zhang)AbstractThe discovery of high-efficiency Mn4+-activated fluoride red phosphors with short excited-state lifetimes (ESLs) is urgent and crucial for high-quality wide color gamut display applications. However, it is still a great challenge to design target phosphors with both short ESL and high luminescence efficiency. Herein, we propose an efficient machine learning approach based on a small dataset to establish the ESL prediction model, thereby facilitating the discovery of new Mn4+-activated fluorides with short ESLs. Such a model can not only accurately predict the ESLs of Mn4+ in fluorides, but also quantify the impact of structure features on ESLs, therefore elucidating the “structure-lifetime” correlations. Guided by the correlations, two new Mn4+-doped tetramethylammonium (TMA)-based hybrid fluorides (TMA)2BF6:Mn4+ (B = Sn or Hf) with both short ESLs (τ ≤ 3.7 ms) and high quantum efficiencies (internal QEs > 92%, external QEs > 55%) have been discovered successfully. A prototype displayer with excellent performance (~124% National Television Standards Committee (NTSC) color gamut) is assembled by employing a (TMA)2SnF6:Mn4+-based white Mini-LED backlight module, demonstrating their practical prospects in high-quality displays. This work not only brings promising candidates for Mn4+-doped fluoride phosphors, but also provides a valuable reference for accelerating the discovery of new promising phosphors.Mn4+-activated fluorides are potential red converter candidates for phosphor-converted white light-emitting diodes (pc-WLED) due to their fascinating luminescent properties (e.g., narrow-band emission) and low-cost advantage of being rare-earth-free, which endow them with great competitiveness in high-quality wide color gamut display.1, 2 However, they generally suffer from long excited-state lifetimes (ESLs ˃ 5 ms, e.g., ~8 ms for commercial K2SiF6:Mn4+ (KSF)) due to the spin- and parity-forbidden characteristics of Mn4+: 2E→4A2 transition.3 Such long ESLs will not only induce the non-thermal optical saturation under high-power excitation, but also lead to the red trailing smear or motion blur problems, showing adverse effects on the image quality of backlight display (particularly in high-definition displays based on Mini-LEDs).4, 5 The long ESL has become the Achilles’ heel of Mn4+-doped fluoride red phosphors. Therefore, there is an urgent need to discover Mn4+-activated fluorides with short ESLs.Currently, there have been some reports on Mn4+-activated fluorides with short ESLs, which are mainly designed by heterovalent doping of Mn4+ into the fluoride matrixes represented by K2NbF7:Mn4+.6 However, their internal and external quantum efficiencies (IQEs and EQEs) are inferior due to the luminescence quenching caused by charge compensation defects.7 In other words, this short ESL is achieved at the expense of luminescence efficiency. Compared to heterovalent doping systems, the equivalent counterparts represented by KSF do not suffer from charge compensation defects, so they generally possess higher IQEs and EQEs but longer ESLs.8 Therefore, it is still a huge challenge to design Mn4+-activated fluorides with both high efficiency and short ESL.Considering the principle that structure determines properties, the prerequisite for achieving a short ESL is to reveal the “structure-lifetime” correlations between fluoride matrixes and Mn4+. In recent years, data-driven machine learning (ML) has been proven to be an effective tool for accelerating the discovery of new materials.9, 10 The state of the art in phosphor research has also begun to focus on ML for the prediction of properties and the optimization of synthesis parameters.11 This has led to the accelerated discovery of new phosphors with desirable luminescent properties. Nevertheless, utilizing ML to unveil the “structure-property” relationship between host matrixes and activators has not received deserved attention. Given that the electronic transitions of Mn4+ rely on the crystal field environment provided by the host structure,12 we therefore took the view that such “structure-lifetime” relationship could greatly facilitate the discovery and screening of new promising Mn4+-doped fluorides with short ESLs. In addition, our previous research on Mn4+-doped organic-inorganic hybrid (oxy)fluorides has demonstrated that the key to achieving high luminescence efficiency is the highly isolated octahedral lattice framework in the matrix, which can effectively avoid the agglomeration of Mn4+ and suppress the energy transfer of Mn4+-Mn4+ and Mn4+-to-defects.3, 13 Combining this structural feature with the “structure-lifetime” correlations to be unveiled, we can rationally design Mn4+-activated fluorides that possess both short ESL and high QEs.Nowadays, the community of Mn4+-activated fluoride phosphors,14 including all-inorganic and organic-inorganic hybrid systems, has accumulated enough cases for ML.15 Accordingly, in this work, we employ the ML strategy onto Mn4+-doped fluoride phosphors to unveil the “structure-lifetime” correlations between fluoride matrixes and Mn4+, thereby guiding the discovery of new Mn4+-activated fluorides with both short ESL and high QEs. By training on a small dataset of 40 cases, we build a reliable ML model that is not only able to accurately predict the ESLs of Mn4+ in fluorides, but also reveals the “structure-lifetime” correlations between the fluoride matrix and Mn4+. Moreover, the relationship between energy levels and ESLs of Mn4+ was studied in detail, where the luminescence decay mechanism for the long/short ESL controlled by the structural parameters was unraveled. Guided by these correlations, two new Mn4+-activated organic-inorganic hybrid fluoride (TMA)2BF6:Mn4+ (where TMA stands for tetramethylammonium, and B = Sn or Hf) phosphors with short ESLs (τ ≤ 3.7 ms) have been discovered and synthesized successfully, confirming the reliability of the proposed ML strategy. Benefiting from both the structure characteristic of highly isolated [BF6]2- octahedrons and the heavy Mn4+ doping amount, their quantum efficiencies (IQEs > 92%, EQEs > 55%; specifically, EQE = 58.1% for (TMA)2SnF6:Mn4+) are almost comparable to that of the commercial KSF. Finally, based on the (TMA)2SnF6:Mn4+, a high-efficiency homemade white Mini-LED backlight module with wide color gamut was fabricated as the flat-panel light source to demonstrate their potential in high-quality Mini-LED displays. This work not only brings promising Mn4+-equivalent-doped fluorides with short ESLs to the sight of the phosphor community, but also can be a useful reference for the future application of ML to the development of luminescent materials.The overall process of discovering new Mn4+-doped fluoride red phosphors with short ESLs using the ML strategy was organized into the following six steps (Figure 1): (I) Collecting data of structural features and the corresponding ESLs from literature (mainly) or own experiments (supplementary) to build the ML dataset; (II) Selecting the Random forest (RF), an ensemble ML method (Supporting Information), to train the dataset and construct the RF model that can predict the ESL of Mn4+ in fluorides; (III) Clarifying the “structure-property” relationship, i.e., the correlations between the structure and ESL, through the RF model; (IV) Searching for unexplored host compounds according to the revealed “structure-lifetime” correlations; (V) Using the RF model to predict the ESL of Mn4+ in the screened potential hosts; (VI) Synthesizing the Mn4+-doped target phosphors experimentally and characterizing their ESLs, and comparing it with the predicted value from RF model. If the measured ESL value significantly differs from that predicted by the RF model, then the initial dataset needs to be enhanced to further improve the RF model, until the target phosphor material is successfully obtained.Figure 1. Schematic machine learning-driven strategy for discovering Mn4+-doped fluoride phosphors with desired ESLs. An iteration includes six steps, i.e., I) Data collection, II) Machine learning, III) Getting feature importance, IV) Screening target hosts from structure databases, V) Model prediction, and VI) Experimental verification.In order to obtain a reliable RF model that can predict the ESLs of Mn4+-doped fluoride AxBFy (A = alkali metal, alkaline earth metal, or organic ammonium cations; B = bivalent, trivalent, tetravalent, pentavalent or hexavalent cations such as Zn2+, Al3+, Si4+, Ge4+, Nb5+, and Mo6+; x = 1, 2, or 3; y = 6 or 7) phosphors, we collected 65 cases of Mn4+-doped fluoride red phosphors from published articles as the original ML dataset, encompassing their structural information and ESLs. More cases were added into the dataset by experiments, including hexagonal-phase KSF (Figure S1) and organic-inorganic hybrid fluoride (TMA)2SiF6:Mn4+ (Figure S2 and Table S1) phosphors. Additionally, the crystal structures of two all-inorganic fluorides, Rb2TiF6 and Cs2TiF6, were resolved through single-crystal X-ray diffraction (SCXRD) (Table S1), aiming to complement the missing structure features for Rb2TiF6:Mn4+ and Cs2TiF6:Mn4+ cases in the original dataset. Their crystallographic data, as well as that of (TMA)2SiF6, is deposited in the Cambridge Crystallographic Data Centre (CSD 2315758 for Rb2TiF6, CSD 2315760 for Cs2TiF6, and CCDC 2315757 for (TMA)2SiF6). Due to differences in synthesis methods and measurement conditions of Mn4+-doped fluoride phosphors, the original dataset was rigorously screened into 40 cases, comprising 36 all-inorganic compounds and 4 organic-inorganic hybrid compounds (Table S2), adhering to the following principles: (1) To eliminate the influence of temperature-induced electron-phonon coupling on the ESL, only the cases whose ESL measured at room temperature were retained; (2) To exclude poor experimental synthesis and ensure the efficient photoluminescence of phosphors as much as possible, only cases with luminescence decay curves strictly conforming to single-exponential decay characteristics were retained. In addition, some cases (Table S2) may have several ESL values from different sources, and their average value (Table S3) was used for ML process.The electronic transitions of Mn4+ are heavily affected by the crystal field environment provided by the host matrix.5 Consequently, to better predict the target property, i.e., the ESL, we selected 7 structural parameters, “IR(A)”, “IR(B)”, “Symmetry”, “d(B-F)”, “d(B-B)”, “Distortion”, and “Vasym”, that are highly correlated to the Mn4+ local environment for feature construction of the dataset. Detailed definitions of these parameters are listed as follow:“IR(A)”: the ion radii of A cations in general formula AxBFy;“IR(B)”: the ion radii of B cations in general formula AxBFy;“Symmetry”: the symmetry order of the local symmetry of B cation in the lattice;“d(B-F)”: the average values of B-F bond lengths from the first coordination sphere of B ion;“d(B-B)”: the average values of B-B bond lengths from the first coordination sphere of B ion;“Distortion”: the distortions of isolated [BFy] polyhedrons as calculated by the formula (Supporting Information);“Vasym”: the cell volume of asymmetric part of the unit cell.In addition, a feature classification parameter “O/I” was added to exclude the influence of compound type, where: 0 represents inorganic compound, 1 represents organic-inorganic hybrid compound. Finally, the constructed dataset of Mn4+-doped fluoride phosphors for ML is listed in Table S3. The statistical plots of the dataset are summarized in Figure S3. Preliminary data analysis shows that the ESL values of phosphors (i.e., cases) are almost uniformly distributed (Figure S3), which means that all representative compounds were selected.RF is an ensemble learning method based on decision trees for classification, regression, and other tasks, which is known as one of the most accurate algorithms since a large number of trees give a more robust model.9 Importantly, RF can be used to rank the importance of features (Figure S4), which is a very important tool for analyzing and revealing the “structure-property” relationship.16 Therefore, the RF was used to build the ML model capable of accurately predicting the ESLs of Mn4+ in fluorides, namely RF model. The RF model was constructed by a simple self-written python script named RandomForest.py (see Supporting Information) using the python 3.6 programming language.17 The standard libraries, including numpy, pandas, sklearn, matplotlib, and mpl_toolkits, were used in the program. Due to the stochastic nature of RF algorithm, we used its average performance across ten repetitions of cross-validation during modeling. Each time, the dataset was divided into two random sub-datasets: one for the training process (70% of total data), and another for testing (30% of total data). The obtained mean absolute errors (MAEs) of ESL for training set and test set were 0.182 ms and 0.432 ms, respectively (Figure 2a). In addition, we have performed a 5-fold cross-validation test on the entire dataset, showing a result of MAE = 0.509 ± 0.185 ms, which is less than ~10% of the average value (5.45 ms) of all case ESLs. The precision of prediction can be further improved by increasing the quantity of observation data, that is, the number of phosphor cases. The predicted ESL values of all cases are listed in Table S4, and their differences with the observed ones are less than 0.4 ms for more than 75% of cases (Table S4 and Figure S5), and their overall good fit can be checked in Figure 2a. These results demonstrate the success in obtaining a reliable RF model that can predict the ESL of Mn4+ in fluorides.Figure 2. (a) Comparative plot of observed “Lifetime” values per predicted “Lifetime” obtained from the RF model, red circles are the training dataset (70% of whole dataset) and blue circles are the test dataset (30% of whole dataset). Linear fit proves the correctness of model. Green line depicts the ideal place for circles from both datasets. Navy blue balls marked the new compounds under experimental investigation. (b) The case samples with high “Lifetime” values (red circles in highlighted area) are segregated from others in the 3D space spanned on three most important feature parameters. (c) Importance of all feature parameters on “Lifetime” values in the RF model. The “Symmetry” has the major influence compared to all others. (d) The ESL of Mn4+-doped fluoride phosphors as a function of the 4A2→4T2 transition energy. The red line is a linear fit of the data points. (e) Schematic energy level diagrams of Mn4+-activated systems with long and short ESLs, respectively.To analyze and reveal the “structure-lifetime” correlations between the fluoride matrix and Mn4+, the importance of each feature was extracted from the RF model. Feature importance refers to how much a feature contributes to prediction, and is obtained based on an algorithm called “permutation importance” (Figure S4). Its fundamental idea is to shuffle a column of data while keeping the data in other columns unchanged, then observe how much the predicted metric or loss changes, and determine the feature importance on the basis of the amount of variation.18 As shown in Figure 2c, “Symmetry” has the major influence (contribution up to 89.1%) on lifetime among all features, followed by “IR(A)”, “IR(B)”, and so on. The “O/I” feature contributes only 0.2% to the prediction of this RF model, indicating that the compound type has no impact on the prediction of the RF model. All cases were plotted in a three-dimensional (3D) space of these three most important features (Figure 2b), and it can be seen that cases with long ESL and those with short ESL are clustered into a small area respectively, showing a good separation between them. The main structure features of Mn4+-doped fluoride phosphors with long (or short) ESLs can be clearly understood through the 2D plot mapped by the 3D space (Figure S6). The main rules to obtain Mn4+-doped fluoride phosphors with long ESLs are: 1) Symmetry order of point group of B-site local symmetry should be greater than 45; 2) Size of A-site cations should be small and less than 2.0 Å; 3) Ion radii of B-site cations can be in the range of 0.4~0.6 Å. The major rules for Mn4+-doped fluoride phosphors with short ESLs are: 1) Symmetry order of point group of B-site local symmetry should be less than 15; 2) Ion radii of A-site cations should preferably be greater than 3.0 Å; 3) There is no obvious rule for the size of B-site cations. The resulting rules should be applicable to the compounds with the chemical formula AxBFy, which means that it covers a relatively broad range of fluorides. Besides, this simple ML approach can help to reveal “structure-property” relationship for other functional materials.Figure 3. Crystal structures of Rb2MnF6 (a), (TMA)2SnF6 (b), and (TMA)2HfF6 (c) obtained from SCXRD and the structural parameters extracted from them respectively.Despite the main rules (i.e., the “structure-lifetime” correlations) of long (or short) ESL obtained from the ML, it is still necessary to understand the hidden physical mechanism by which structural parameters affect the electronic transition of Mn4+, which can help to reasonably design new Mn4+-activated phosphors with desirable luminescent properties. Based on the above ML results (Figure 2c), the “Symmetry” feature has the highest prediction contribution (up to 89.1%) to this RF model, indicating that B-site symmetry is a key structural parameter that affects the ESL. In other words, the ESL can be regulated to the greatest extent by adjusting the B-site (occupation site of Mn4+) symmetry in crystal lattice. The other six structural features also have an impact on the Mn4+ ESL (10.7% of total contribution). Due to the complex 3D structure of the host matrix, there may even be undiscovered structural parameters that also influence the Mn4+ ESL, because this RF model temporarily cannot achieve perfect prediction for the ESL. However, it should be emphasized that we have obtained an RF model (Figure 2a) that can well predict the Mn4+ ESL using only 7 structural parameters, that is to say, they are responsible for the “structure-lifetime” correlations. They dominate the local crystal field environment of Mn4+, thereby affecting the electronic transition of Mn4+. To clarify this point, we plotted the ESL versus the 4A2→4T2 energy for many reported Mn4+-doped fluoride phosphors (Figure 2d, data are also listed in Table S5). These data show an interesting trend where the ESL becomes shorter as the 4T2 energy decrease. This obvious trend indicates that these structural parameters, especially the B-site symmetry, actually affect the 4T2 energy of the incorporated Mn4+, thus causing the shortening or lengthening of ESL. A proposed energy diagram for elucidating the mechanism of ESL has been drafted in Figure 2e featuring the difference of 4T2 energy in the long and short ESL systems, respectively. It can be seen that the overlapping of the electron clouds of 4T2 and 2E becomes larger as the energy gap decreases, allowing more mixing in the wave functions (electron clouds) of the two energy levels, which further relaxes the spin-forbidden selection rule, resulting in shortened ESL; conversely, the ESL will be longer. In general, these structural parameters greatly affect the relaxation of the spin-forbidden selection rule, thereby resulting in changes in the Mn4+ ESL.Figure 4. PLE and PL spectra, and PL decay curves of Rb2MnF6 (a, b), (TMA)2SnF6:Mn4+ (c, d), and (TMA)2HfF6:Mn4+ (e, f) measured at room temperature.Table 1. The IQE, AE, EQE, and corresponding excited-state lifetime (ESL) of some typical Mn4+-doped fluoride phosphors. Phosphor IQE (%) AE (%) EQE (%) ESL (ms) Reference K2SiF6:Mn4+ 90.4 86.5 78.2 7.6 8 K2SiF6:Mn4+ 93.3 72.3 67.5 8.1 8 K2SiF6:Mn4+ 92 47.8 44 8.3 19 KNaSiF6:Mn4+ 90 45.5 41 6.1 20 Rb2SiF6:Mn4+ 91.9 60.7 55.8 8.01 21 Cs2SiF6:Mn4+ 89.22 79.96 71.34 7.81 22 Na2GeF6:Mn4+ 60 / / 6.58 23 K2GeF6:Mn4+ 93 78 73 6.65 24 K2GeF6:Mn4+ 97.9 70.3 68.8 6.6 25 K2GeF6:Mn4+ 90 66.6 60 5.85 26 Cs2GeF6:Mn4+ 80 83.7 66.9 7.52 23, 27 K2TiF6:Mn4+ 93 54 50.22 5.7 1 K2TiF6:Mn4+ 94 69 64.86 6.35 28 K2TiF6:Mn4+@K2TiF6 96 69 66.24 6.25 28 Rb2TiF6:Mn4+ 76 / / 5.2 29 Rb2ZrF6:Mn4+ 75 / / 5.0 30 Rb2SnF6:Mn4+ 70.3 / / 5.22 31 K3AlF6:Mn4+ 88 57.5 50.6 5.2 32 K2NaAlF6:Mn4+ 58.4 / / 7.537 33 K2NaAlF6:Mn4+ 88 17 14.96 7.81 34 Na3GaF6:Mn4+ 69 / / 4.97 35 K2NbF7:Mn4+ 93.5 28.0 26.2 3.7 6 K2NaScF6:Mn4+ 70.3 18.2 12.8 3.5 36 K3GaF6:Mn4+ 46 / / 3.69 37 K2LiAlF6:Mn4+ 87.5 17.9 15.7 9.3 38 K2NaAlF6:Mn4+ 85 26 22.1 7.58 34 K2NaGaF6:Mn4+ 61 / / 5.6 39 K2NaScF6:Mn4+ 57.13 15.91 9.09 1.78 5 Rb2NaScF6:Mn4+ 54.98 18.88 10.38 3.69 5 Cs2NaScF6:Mn4+ 51.99 20.81 10.82 3.31 5 Rb2MnF6 81.4 78.0 63.5 8.23 This work (TMA)2SnF6:Mn4+ 92.0 63.1 58.1 3.733 This work (TMA)2HfF6:Mn4+ 94.4 58.7 55.4 3.602 This workNote: / denotes no reference data.To demonstrate the reliability of the constructed RF model and the correctness of the revealed “structure-lifetime” correlations, phosphor matrix candidates were searched from the Inorganic Crystal Structure Database (ICSD) or the CCDC. Based on the revealed structure rules of long (or short) ESL, an inorganic compound Rb2MnF6 was selected as the candidate for long ESL, and the other two novel tetramethylammonium (TMA)-based organic-inorganic hybrid compounds (TMA)2SnF6 and (TMA)2HfF6 as the candidates for short ESL. Their crystal structures were obtained by SCXRD analysis, as shown in Figure 3. Their main crystal data are shown in Table S6, and their crystallographic data have been deposited in the CCDC (CSD 2315756 for Rb2MnF6, CCDC 2315759 for (TMA)2SnF6, and CCDC 2315761 for (TMA)2HfF6). In addition, it can be seen that the inorganic [BF6] groups in the two hybrid structures are highly separated (d(B-B) ˃ 8.1 Å) thanks to the steric hindrance effect of large-sized organic cations, which is beneficial to the heavy doping of Mn4+ while avoiding the agglomeration of Mn4+ and suppressing the energy transfer of Mn4+-Mn4+ and Mn4+ to defects. Therefore, it can be expected that these two Mn4+-doped hybrid fluorides will both have high luminescence efficiency. In order to confirm whether these candidates possess the expected ESLs, we extracted the following structural parameters from their Crystallographic Information Files (CIFs): “IR(A)”, “IR(B)”, “Symmetry”, “d(B-F)”, “d(B-B)”, “Distortion”, and “Vasym” (Figure 3 and Table S3). These structural parameters and the classification feature “O/I” were uploaded to the constructed RF model, and the predicted ESL values of Mn4+ in these compounds were obtained using the conventional procedure of averaging the Forest voting: 7.958 ms, 3.877 ms, and 3.908 ms for Rb2MnF6, (TMA)2SnF6:Mn4+, and (TMA)2HfF6:Mn4+, respectively (Table S4 and Figure 2a). To verify the accuracy of this RF model prediction, we carried out experimental confirmation (Supporting Information), where Rb2MnF6 was directly used for characterization as it is a Mn4+-base compound crystals (Figure S7). XRD analysis (Figure S8) showed that a series of pure (TMA)2BF6:Mn4+ (B = Sn or Hf) phosphor samples doped with various Mn4+ concentrations (Table S7) were synthesized successfully. Rb2MnF6 and (TMA)2BF6:Mn4+ (B = Sn or Hf) both exhibit typical luminescence characteristics of Mn4+ in fluorides (Figure 4), where the concentration-dependent PL spectra of (TMA)2BF6:Mn4+ indicate that their optimal doping amounts of Mn4+ are about 20% (Figure S9). The IQEs (92.0% and 94.4%) and EQEs (58.1% and 55.4%) of (TMA)2BF6:20mol%Mn4+ have exceeded those of most reported Mn4+-activated fluoride red phosphors, and are almost comparable to that of commercial KSF with long ESL (Table 1). Their high QEs can be ascribed to the structure characteristic of highly isolated [BF6]2- octahedrons and the heavy Mn4+ doping amount. The Rb2MnF6 crystal phosphor also shows high IQE and EQE of 81.4% and 63.5%, respectively. Its high QEs results from the 100% Mn4+ content and the large single crystal particles. In the meantime, their luminescence decay curves all showed single-exponential decay behavior (Figure 4), which means that the PL decay starting from 2E excited state is mainly radiative, confirming they have excellent photoluminescence performances. Their actual measured ESLs were 8.23 ms (Rb2MnF6), 3.733 ms ((TMA)2SnF6:Mn4+), and 3.602 ms ((TMA)2HfF6:Mn4+), respectively; and their differences from the RF model predicted values were 0.272 ms (Rb2MnF6), -0.144 ms ((TMA)2SnF6:Mn4+), and -0.306 ms ((TMA)2HfF6:Mn4+), respectively (Table S4). All the differences were below the obtained MAE value (0.509±0.185 ms), demonstrating the good prediction credibility of RF model and the correctness of the revealed “structure-lifetime” correlations. Therefore, the constructed RF model can be used to screen unexplored compounds to find new Mn4+-doped phosphor candidates with long/short ESLs and allow researchers to select the desired materials before proceeding with conventional synthesis procedure. And more notably, unlike those previously reported Mn4+-doped fluoride phosphors, the discovered (TMA)2BF6:Mn4+ phosphors simultaneously achieve high QEs and short ESLs (Table 1). In addition, the short ESLs of (TMA)2BF6:Mn4+ can greatly suppress the red luminescence tailing phenomenon as illustrated in Figure S10, which is beneficial for fast-response backlight displays.5 Evidently, temperature dependence experiments (Figure S11) indicate that the (TMA)2BF6:Mn4+ phosphors are not suitable for high-power pc-WLED applications because their red emission is severely quenched at 150 °C (423 K).4 This quenching phenomenon is due to their relatively "soft" 0D crystal structure, which lacks structural rigidity compared with the KSF. Fortunately, the vast majority of their red emission can be preserved at 75 °C (e.g., 84.2% for (TMA)2SnF6:Mn4+), which is higher than the LED junction temperature (merely 70 °C) in backlight displays.6Figure 5. (a) Schematic diagram of the homemade Mini-LED display prototype in this work. (b) Photo of the internal blue Mini-LED backlight for the display prototype when powered on. (c) Photo of the PL material film coated by a mixture of (TMA)2SnF6:Mn4+ and green QD (~530 nm) taken under natural light. (d) Photo of the white backlight module assembled by directly placing the PL material film on the surface of the blue Mini-LED backlight when powered on. Output spectra collected from the screens of the homemade Mini-LED display prototype (e), a YAG phosphor-based commercial LCD display (f), and a QD-based commercial Mini-LED display (g) using a fiber optic spectrometer. Comparison of the images displayed by the homemade Mini-LED display prototype (h), the YAG phosphor-based commercial LCD display (i), and the QD-based commercial Mini-LED display (j).The high QEs, short ESLs, bearable thermal quenching, wide blue absorption, and narrow red emission make (TMA)2BF6:Mn4+ a promising emitter for high-quality wide color gamut Mini-LED display backlight displays. Based on the internal layout of the Mini-LED display (Figure 5a), we first need to prepare a PL material film with appropriate light transmittance that can be completely covered on the blue Mini-LED backlight (Figure 5b) to assemble a white backlight module. The small size of (TMA)2BF6:Mn4+ helps them to be coated into a uniform film (Figure S12). The PL material film was coated by a homogeneous mixture of the (TMA)2SnF6:Mn4+, commercial quantum dot (QD530) and isobornyl acrylate through a wire rod coating machine. As the photos shown in Figure 5c, the as-prepared yellow PL material film is uniform and translucent, and was cut to similar dimensions to a 27-inch monitor screen. Then, this film was placed on the blue Mini-LED backlight to assemble the white backlight module (Figure S13). A strong white light emission can be observed from it when powered on (Figure 5d). Moreover, due to the small heat generation of the mini blue chips and the non-contact design between the film and the chips, the maximum working temperature of the PL material film on the white backlight module is 31.4 °C (Figure S13), which is far lower than the thermal quenching temperature of this (TMA)2BF6:Mn4+ phosphors (T1/2 = 116 or 109 °C, Figure S11). In addition, due to the protection of the film by the polyethylene terephthalate (PET) substrates, no significant decrease in luminescent intensity was observed when the white backlight module was continuously lit in a 65% high-humidity environment for 48 hours (Figure S14), indicating good stability of the module. According to the configuration of the Mini-LED display mentioned above, this white backlight module was used with a commercial liquid crystal module, RGB filter film and display screen to assemble a homemade Mini-LED display prototype to demonstrate the display performance. The homemade prototype display screen emits standard cool white light with a correlated color temperature (CCT) of 7489 K and Commission Internationale de L’Eclairage (CIE) chromaticity coordinates of (0.3015, 0.3063), which is similar to the white light emitted by the commercial Y3Al5O12:Ce3+ (YAG) phosphor-based liquid crystal display (LCD) (CCT = 7418 K, CIE = (0.3039, 0.3007)) and QD-based Mini-LED display (CCT = 7552 K, CIE = (0.2974, 0.3185)) (Figure 5e-5g). Based on the filtered pure red, green and blue spectra of this homemade prototype display (Figure S15), its color gamut was determined to be ∼124% of National Television Standards Committee (NTSC) standard in the CIE 1931 color space, which is much higher than the commercial LCD display with broad emission spectral characteristics (64% NTSC, Figure S16). Furthermore, due to the narrower red emission of (TMA)2SnF6:Mn4+ compared to commercial QD630 (Figure S17), its color gamut is also larger than that of the QD-based commercial Mini-LED display (116% NTSC, Figure S18). Thanks to the wider color gamut brought by the homemade white backlight module, the (TMA)2SnF6:Mn4+-based homemade Mini-LED display prototype exhibits a more vivid picture of the red rose compared to the other two commercial displays (Figure 5h-5j). In addition, since it is impossible to measure the luminous efficiency (LE) of the large-size display devices, we also fabricated two conventional pc-WLEDs with standard white light emission based on the (TMA)2SnF6:Mn4+ phosphors (Figure S19). They all exhibit high LEs (122.42 lm/W or 135.99 lm/W), showing great practical prospects of the (TMA)2SnF6:Mn4+ phosphors in backlight displays. These results show that we have taken a substantial advance towards the commercial application of high-quality Mini-LED displays with the help of (TMA)2BF6:Mn4+ phosphors.In conclusion, we have demonstrated an efficient ML approach based on a relatively small dataset to discover new Mn4+-doped fluoride red phosphors with desired ESLs. Such a ML approach not only builds an RF model that can accurately predict the ESLs of Mn4+ in fluorides, but also elucidates the “structure-lifetime” correlations between the fluoride matrix and Mn4+. Feature importance results indicate that the B-site symmetry is a pivotal structural feature that determines the ESL of Mn4+. The underlying mechanism lies in the interesting intertwinements between these structural parameters (especially the B-site symmetry) and the 4T2 energy of Mn4+ that affects the ESLs by controlling the mixing of 4T2 and 2E wavefunctions. Under the guidance of “structure-lifetime” correlations and the RF model prediction, short ESLs (τ ≤ 3.7 ms) were achieved in two new Mn4+-doped fluoride (TMA)2BF6:Mn4+ (B = Sn or Hf) phosphors, respectively. Their EQEs are comparable to that of the commercial KSF (e.g., EQE = 58.1% for (TMA)2SnF6:Mn4+), but their ESLs are less than half that of KSF. Employing the (TMA)2SnF6:Mn4+, the prototype display with excellent performance (~124% NTSC) was assembled based on the homemade white Mini-LED backlight module, demonstrating the practical prospects of (TMA)2BF6:Mn4+ phosphors in high-quality displays. This work provides a deeper understanding on the correlations between structure and luminescence, as well as lays the groundwork for future advancement in employing ML to dramatically accelerate the discovery of new promising phosphors.ASSOCIATED CONTENTSupporting InformationThe Supporting Information is available on the ACS Publications website.Experimental details, ML method, supplementary text, and supporting figures and tables, including XRD, photoluminescence excitation (PLE), PL, ESL, scanning electron microscope (SEM), device performance and Photo, main crystal structural parameters for selected compounds, collected cases for the ML, and inductively coupled plasma optical emission spectrometer (ICP-OES) results.AUTHOR INFORMATIONCorresponding AuthorsEnhai Song - State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, South China University of Technology, Guangzhou 510641, China; orcid.org/0000-0003-1666-0532; E-mail: msehsong@scut.edu.cnYayun Zhou - State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, South China University of Technology, Guangzhou 510641, China; Guangdong-Hong Kong-Macao Joint Laboratory for Intelligent Micro-Nano Optoelectronic Technology, School of Physics and Optoelectronic Engineering, Foshan University, Foshan 528225, China; orcid.org/0000-0002-0952-1481; E-mail: zhouyayun@fosu.edu.cnQinyuan Zhang - State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, School of Materials Science and Engineering and School of Physics and Optoelectronics, South China University of Technology, Guangzhou 510641, China; orcid.org/0000-0001-6544-4735; E-mail: qyzhang@scut.edu.cnAuthorsHong Ming - State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, School of Materials Science and Engineering, South China University of Technology, Guangzhou 510641, China; International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science (NIMS), 1-2-1 Sengen, Tsukuba 305-0047, Japan; orcid.org/0000-0003-0118-1897Maxim S. Molokeev - Laboratory of Crystal Physics, Kirensky Institute of Physics, Federal Research Center KSC SB RAS, Krasnoyarsk 660036, Russia; Laboratory of Theory and Optimization of Chemical and Technological Processes, University of Tyumen, Tyumen 625003, Russia; Institute of Engineering Physics and Radioelectronics, Siberian Federal University, Krasnoyarsk 660041, Russia; orcid.org/0000-0002-8297-0945Chuang Zhang - State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, School of Physics and Optoelectronics, South China University of Technology, Guangzhou 510641, China;Lin Huang - Fujian Key Laboratory of Surface and Interface Engineering for High Performance Materials and College of Materials, Xiamen University, Xiamen 361005, China;Yuanjing Wang - State Key Laboratory of Luminescent Materials and Devices, Guangdong Provincial Key Laboratory of Fiber Laser Materials and Applied Techniques, and Guangdong Engineering Technology Research and Development Center of Special Optical Fiber Materials and Devices, School of Physics and Optoelectronics, South China University of Technology, Guangzhou 510641, China;Hong-Tao Sun - International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science (NIMS), 1-2-1 Sengen, Tsukuba 305-0047, Japan; orcid.org/0000-0002-0003-7941Author contributionsH.M., Y.Z. and Q.Z. conceived the idea and initiated the research. Q.Z. supervised the project. H.M. and M.S.M. performed the machine learning and analyzed the results. H.M., and C.Z. designed the experiments. H.M., and C.Z. performed the experiments and collected the data. H.M., Y.Z., C.Z., L.H., Y.W., E.S., H.S. and Q.Z. analyzed the data and discussed the results. Y.Z. conducted the construction of Mini-LED display devices and analyzed them. H.M. wrote the manuscript, and E.S., H.S. and Q.Z. revised and commented on it.NotesThe authors declare no conflict of interest. CCDC or CSD 2315756-2315761 contains the supplementary crystallographic data for this paper. These data can be obtained free of charge from the Cambridge Crystallographic Data Centre via www.ccdc.cam.ac.uk/data_request/cif.ACKNOWLEDGMENTSThis work was financially supported by the National Key Research and Development Program of China (2022YFB3503800), National Natural Science Foundation of China (Grants Nos. 52202170 and 52322208), Natural Science Foundation of Guangdong Province (No. 2022A1515140032) and Distinguished Youth Foundation of Guangdong Scientific Committee (No. 2023B1515020059). This work was also supported by the Tyumen Oblast Government, as part of the West-Siberian Interregional Science and Education Center's project No. 89-DON (3). H. Ming acknowledges the fellowship support from the China Scholarship Council (CSC No. 202206150038).REFERENCES(1) Zhu, H.; Lin, C. C.; Luo, W.; Shu, S.; Liu, Z.; Liu, Y.; Kong, J.; Ma, E.; Cao, Y.; Liu, R. S.; Chen, X. Highly efficient non-rare-earth red emitting phosphor for warm white light-emitting diodes. Nat. Commun. 2014, 5, 4312.(2) Senden, T.; van Dijk-Moes, R. J. A.; Meijerink, A. Quenching of the red Mn4+ luminescence in Mn4+-doped fluoride LED phosphors. Light Sci. Appl. 2018, 7, 8.(3) Ming, H.; Zhao, Y. F.; Zhou, Y. Y.; Molokeev, M. S.; Wang, Y. J.; Zhang, S.; Song, E. H.; Ye, S.; Xia, Z. G.; Zhang, Q. Y. Shining Mn4+ in 0D Organometallic Fluoride Hosts towards Highly Efficient Photoluminescence. Adv. Opt. Mater. 2022, 10, 2102141.(4) van de Haar, M. A.; Tachikirt, M.; Berends, A. C.; Krames, M. R.; Meijerink, A.; Rabouw, F. T. Saturation Mechanisms in Common LED Phosphors. ACS Photonics 2021, 8, 1784-1793.(5) Zhou, Y. Y.; Song, E. H.; Brik, M. G.; Wang, Y. J.; Hu, T.; Xia, Z. G.; Zhang, Q. Y. Non-equivalent Mn4+ doping into A2NaScF6 (A = K, Rb, Cs) hosts toward short fluorescence lifetime for backlight display application. J. Mater. Chem. C 2019, 7, 9203-9210.(6) Lin, H.; Hu, T.; Huang, Q.; Cheng, Y.; Wang, B.; Xu, J.; Wang, J.; Wang, Y. Non-Rare-Earth K2XF7:Mn4+ (X = Ta, Nb): A Highly-Efficient Narrow-Band Red Phosphor Enabling the Application in Wide-Color-Gamut LCD. Laser Photonics Rev. 2017, 11, 1700148.(7) Wang, Y.; Zhou, Y.; Ming, H.; Zhao, Y.; Song, E.; Zhang, Q. Luminescence Enhancement of Mn4+-Activated Fluorides via a Heterovalent Co-Doping Strategy for Monochromatic Multiplexing. ACS Appl. Mater. Interfaces 2021, 13, 51255-51265.(8) Zhou, Y.; Yu, C.; Song, E.; Wang, Y.; Ming, H.; Xia, Z.; Zhang, Q. Three Birds with One Stone: K2SiF6:Mn4+ Single Crystal Phosphors for High‐Power and Laser‐Driven Lighting. Adv. Opt. Mater. 2020, 8, 2000976.(9) Molokeev, M. S.; Su, B.; Aleksandrovsky, A. S.; Golovnev, N. N.; Plyaskin, M. E.; Xia, Z. Machine Learning Analysis and Discovery of Zero-Dimensional ns2 Metal Halides toward Enhanced Photoluminescence Quantum Yield. Chem. Mater. 2022, 34, 537-546.(10) Zhuo, Y.; Mansouri Tehrani, A.; Oliynyk, A. O.; Duke, A. C.; Brgoch, J. Identifying an efficient, thermally robust inorganic phosphor host via machine learning. Nat. Commun. 2018, 9, 4377.(11) Park, C.; Lee, J. W.; Kim, M.; Lee, B. D.; Singh, S. P.; Park, W. B.; Sohn, K. S. A data-driven approach to predicting band gap, excitation, and emission energies for Eu2+-activated phosphors. Inorg. Chem. Front. 2021, 8, 4610-4624.(12) Fang, M. H.; Wu, W. L.; Jin, Y.; Lesniewski, T.; Mahlik, S.; Grinberg, M.; Brik, M. G.; Srivastava, A. M.; Chiang, C. Y.; Zhou, W.; et al. Control of Luminescence by Tuning of Crystal Symmetry and Local Structure in Mn4+-Activated Narrow Band Fluoride Phosphors. Angew Chem. Int. Ed. Engl. 2018, 57, 1797-1801.(13) Ming, H.; Zhao, Y.; Zhou, Y.; Molokeev, M. S.; Wang, Y.; Song, E.; Zhang, Q. Chemical Group Substitution Enables Highly Efficient Mn4+ Luminescence in Heterovalent Systems. Adv. Opt. Mater. 2023, 11, 2300076.(14) Adachi, S. Photoluminescence spectra and modeling analyses of Mn4+-activated fluoride phosphors: A review. J. Lumin. 2018, 197, 119-130.(15) Jiang, L.; Jiang, X.; Wang, C.; Liu, P.; Zhang, Y.; Lv, G.; Lookman, T.; Su, Y. Rapid Discovery of Efficient Long-Wavelength Emission Garnet:Cr NIR Phosphors via Multi-Objective Optimization. ACS Appl. Mater. Interfaces 2022, 14, 52124-52133.(16) Choe, H.; Jin, H.; Lee, S. J.; Cho, J. Machine Learning-Directed Predictive Models: Deciphering Complex Energy Transfer in Mn-Doped CsPb(Cl1–yBry)3 Perovskite Nanocrystals. Chem. Mater. 2023, 35, 5401-5411.(17) Coelho, L. P.; Richert, W.; Brucher, M. Building Machine Learning Systems with Python: Explore machine learning and deep learning techniques for building intelligent systems using scikit-learn and TensorFlow. Packt Publishing Ltd., 2018.(18) Zhu, R.; Zeng, D.; Kosorok, M. R. Reinforcement learning trees. J. Am. Stat. Assoc. 2015, 110, 1770-1784.(19) Hou, Z. Y.; Tang, X. Y.; Luo, X. F.; Zhou, T. L.; Zhang, L.; Xie, R. J. A green synthetic route to the highly efficient K2SiF6:Mn4+ narrow-band red phosphor for warm white light-emitting diodes. J. Mater. Chem. C 2018, 6, 2741-2746.(20) Jin, Y.; Fang, M. H.; Grinberg, M.; Mahlik, S.; Lesniewski, T.; Brik, M. G.; Luo, G. Y.; Lin, J. G.; Liu, R. S. Narrow Red Emission Band Fluoride Phosphor KNaSiF6:Mn4+ for Warm White Light-Emitting Diodes. ACS Appl. Mater. Interfaces 2016, 8, 11194-11203.(21) Zhou, Y.; Ming, H.; Zhao, Y.; Wang, Y.; Song, E.; Xia, Z.; Zhang, Q. Preparation and Luminescent Properties of Rb2SiF6:Mn4+ Single Crystal for Laser Lighting. Chin. J. Lumin. 2021, 42, 1559-1568.(22) Song, E. H.; Zhou, Y. Y.; Yang, X. B.; Liao, Z. F.; Zhao, W. R.; Deng, T. T.; Wang, L. Y.; Ma, Y. Y.; Ye, S.; Zhang, Q. Y. Highly Efficient and Stable Narrow-Band Red Phosphor Cs2SiF6:Mn4+ for High-Power Warm White LED Applications. ACS Photonics 2017, 4, 2556-2565.(23) Lian, H.; Huang, Q.; Chen, Y.; Li, K.; Liang, S.; Shang, M.; Liu, M.; Lin, J. Resonance Emission Enhancement (REE) for Narrow Band Red-Emitting A2GeF6:Mn4+ (A=Na, K, Rb, Cs) Phosphors Synthesized via a Precipitation-Cation Exchange Route. Inorg. Chem. 2017, 56, 11900-11910.(24) Zhou, W.; Fang, M. H.; Lian, S.; Liu, R. S. Ultrafast Self-Crystallization of High-External-Quantum-Efficient Fluoride Phosphors for Warm White Light-Emitting Diodes. ACS Appl. Mater. Interfaces 2018, 10, 17508-17511.(25) Huang, L.; Liu, Y.; Si, S.; Brik, M. G.; Wang, C.; Wang, J. A new reductive dl-mandelic acid loading approach for moisture-stable Mn4+ doped fluorides. Chem. Commun. 2018, 54, 11857-11860.(26) Lang, T.; Fang, S.; Han, T.; Wang, M.; Yang, D.; Wang, J.; Cao, S.; Peng, L.; Liu, B.; Cai, M.; et al. Phase Transformation of a K2GeF6 Polymorph for Phosphors Driven by a Simple Precipitation-Dissolution Equilibrium and Ion Exchange. Inorg. Chem. 2020, 59, 8298-8307.(27) Wang, Z.; Yang, Z.; Wang, N.; Zhou, Q.; Zhou, J.; Ma, L.; Wang, X.; Xu, Y.; Brik, M. G.; Dramićanin, M. D.; Wu, M. Single‐Crystal Red Phosphors: Enhanced Optical Efficiency and Improved Chemical Stability for wLEDs. Adv. Opt. Mater. 2020, 8, 1901512.(28) Huang, D.; Zhu, H.; Deng, Z.; Zou, Q.; Lu, H.; Yi, X.; Guo, W.; Lu, C.; Chen, X. Moisture-Resistant Mn4+-Doped Core-Shell-Structured Fluoride Red Phosphor Exhibiting High Luminous Efficacy for Warm White Light-Emitting Diodes. Angew Chem. Int. Ed. Engl. 2019, 58, 3843-3847.(29) Sakurai, S.; Nakamura, T.; Adachi, S. Editors' Choice—Rb2SiF6:Mn4+ and Rb2TiF6:Mn4+ Red-Emitting Phosphors. ECS J. Solid State Sci. Technol. 2016, 5, R206-R210.(30) Wang, L. Y.; Song, E. H.; Zhou, Y. Y.; Deng, T. T.; Ye, S.; Zhang, Q. Y. An efficient and stable narrow band Mn4+-activated fluorotitanate red phosphor Rb2TiF6:Mn4+ for warm white LED applications. J. Mater. Chem. C 2018, 6, 8670-8678.(31) Jiang, C. Y.; Brik, M. G.; Srivastava, A. M.; Li, L. H.; Peng, M. Y. Significantly conquering moisture-induced luminescence quenching of red line-emitting phosphor Rb2SnF6:Mn4+ through H2C2O4 triggered particle surface reduction for blue converted warm white light-emitting diodes. J. Mater. Chem. C 2019, 7, 247-255.(32) Song, E.; Wang, J.; Shi, J.; Deng, T.; Ye, S.; Peng, M.; Wang, J.; Wondraczek, L.; Zhang, Q. Highly Efficient and Thermally Stable K3AlF6:Mn4+ as a Red Phosphor for Ultra-High-Performance Warm White Light-Emitting Diodes. ACS Appl. Mater. Interfaces 2017, 9, 8805-8812.(33) Wang, L. Y.; Song, E. H.; Deng, T. T.; Zhou, Y. Y.; Liao, Z. F.; Zhao, W. R.; Zhou, B.; Zhang, Q. Y. Luminescence properties and warm white LED application of a ternary-alkaline fluoride red phosphor K2NaAlF6:Mn4+. Dalton Trans. 2017, 46, 9925-9933.(34) Yi, X. D.; Li, R. F.; Zhu, H. M.; Gao, J.; You, W. W.; Gong, Z. L.; Guo, W.; Chen, X. Y. K2NaAlF6:Mn4+ red phosphor: room-temperature synthesis and electronic/vibronic structures. J. Mater. Chem. C 2018, 6, 2069-2076.(35) Deng, T. T.; Song, E. H.; Sun, J.; Wang, L. Y.; Deng, Y.; Ye, S.; Wang, J.; Zhang, Q. Y. The design and preparation of the thermally stable, Mn4+ ion activated, narrow band, red emitting fluoride Na3GaF6:Mn4+ for warm WLED applications. J. Mater. Chem. C 2017, 5, 2910-2918.(36) Ming, H.; Liu, L. L.; He, S. A.; Peng, J. Q.; Du, F.; Fu, J. X.; Yang, F. L.; Ye, X. Y. An ultra-high yield of spherical K2NaScF6:Mn4+ red phosphor and its application in ultra-wide color gamut liquid crystal displays. J. Mater. Chem. C 2019, 7, 7237-7248.(37) Deng, T. T.; Song, E. H.; Zhou, Y. Y.; Wang, L. Y.; Ye, S.; Zhang, Q. Y. Stable narrowband red phosphor K3GaF6:Mn4+ derived from hydrous K2GaF5(H2O) and K2MnF6. J. Mater. Chem. C 2017, 5, 9588-9596.(38) Zhu, Y. W.; Chen, D. Q.; Huang, L.; Liu, Y.; Brik, M. G.; Zhong, J. S.; Wang, J. Phase-transition-induced giant enhancement of red emission in Mn4+-doped fluoride elpasolite phosphors. J. Mater. Chem. C 2018, 6, 3951-3960.(39) Jiang, C. Y.; Brik, M. G.; Li, L. H.; Li, L. Y.; Peng, J.; Wu, J. N.; Molokeev, M. S.; Wong, K. L.; Peng, M. Y. The electronic and optical properties of a narrow-band red-emitting nanophosphor K2NaGaF6:Mn4+ for warm white light-emitting diodes. J. Mater. Chem. C 2018, 6, 3016-3025.2image4.tiffimage5.tiffimage6.tiffimage1.tiffimage2.tiffimage3.tiff