Low-temperature molten-salt enabled synthesis of highly-efficient solid-state emitting carbon dots optimized using machine learning

Synthesis and optical properties of the full-color SSF CDs

As illustrated in Fig. 1a, the SSF CDs were synthesized with different precursors in a molten salts system which is composed of NaCl, KCl, and ZnCl₂ (Table S1). By regulating the precursors and reaction factors, the well-dispersed solid-state powder of B-, G-, Y-, and NIR-CDs can be synthesized (Fig. S1). As illustrated in Fig. 1b–e, these B-, G-, Y-, and NIR-CDs present blue, green, yellow, and near-infrared SSF under 365 nm UV lamp and exhibit a SSF peak around 432 nm, 517 nm, 594 nm, and 660 nm with distinct UV-vis absorption. Herein, the B-, G-, Y-, and NIR-CDs present different absorption bands, which are corresponding to the π-π* transition of sp² carbon, and n-π* transition of surface heteroatom functional groups. The corresponding SSF spectra reveal a significant color transition from blue to red region in a Commission Internationale de l’Eclairage (CIE) coordinate (Fig. S2), which is consistent with the observation in naked-eyes. As shown in Fig. 1f–i, the excitation-emission mapping of the B-, G-, Y- and NIR-CDs present a maximal SSF center at (335 nm, 430 nm), (430 nm, 517 nm), (560 nm, 594 nm) and (640 nm, 660 nm). Herein, the SSF emission of the B-, G-, Y-, and NIR-CDs illustrate a decreased excitation-dependence. The PL QYs of the B-, G-, Y-, and NIR-CDs are measured with integrating sphere under their optimal excitation wavelength (Fig. S3) and the results are calculated as 62.77%, 82.51%, 63.46%, and 8.62%, which are ever relative high values among the reported SSF CDs (Table S2). The time-resolved PL decay of these SSF CDs can be well fitted with a biexponential curves (Fig. S4) and the results reveal an average lifetime of 6.47 ns, 16.16 ns, 10.18 ns, and 1.07 ns for the B-CDs, G-CDs, Y-CDs, and NIR-CDs, respectively. As illustrated Fig. S5a, the optical properties of the products located at different reaction area were investigated to evaluate the homogeneity of the CDs in different reaction locations. As shown in Fig. S5b–e, the as-prepared G-CDs samples located at Position 1, Position 2, Position 3, and Position 4 exhibited similar excitation-emission matrix with a maximal center at (418 nm, 501 nm), (420 nm, 499 nm), (418 nm, 501 nm), and (414 nm, 503 nm), respectively. Meanwhile, these products present a similar PL QY of 89.7%, 89.13%, 90.22% and 91.96% (Fig. S5f–i). These surveys clearly indicate the excellent homogeneity of the CDs in different reaction location and demonstrate the outstanding features of molten salts strategy for the synthesis of SSF CDs.

On the condition, the influence of molten salts on the synthesis of SSF CDs was investigated. Firstly, we have prepared CDs without molten salts by dry-burning the same organic precursors of B-, G-, Y- and NIR-CDs, which were named B-ws-, G-ws-, Y-ws- and NIR-ws-CDs, respectively. As illustrated in Fig. S6a–d, the obtained well-dispersed solid powder of B-ws-, G-ws-, Y-ws-CDs exhibited different fluorescence under 365 nm UV lamp. And the excitation-emission matrix of the B-ws-, G-ws-, and Y-ws-CDs exhibit an excitation-independent PL emission with a maximal PL emission center at (350 nm, 432 nm), (440 nm, 525 nm), and (480 nm, 536 nm). Notably, there is nearly no PL emission from the NIR-ws-CDs, which is obviously different from the initial B-, G-, Y-, and NIR-CDs (Fig. S6e–h). The as-obtained B-ws-, G-ws-, Y-ws- and NIR-ws-CDs present PL QYs of ~46.08%, ~35.91%, ~2.24% and ~0%, which are far lower than the B-, G-, Y- and NIR-CDs (Fig. S6i–k). The comparison between the preparation of CDs with and without molten salts clearly demonstrates that the space confinement effect can improve the SSF emission of the CDs. Additionally, the different organic precursors are also employed to prepare possible SSF CDs. As shown in Fig. S7a, the organic precursors of biuret, Rhodamine 6 G, M-phenylenediamine, and phloroglucinol are employed to be dry-burned in the molten salts, as similar as the B-, G-, Y-, and NIR-CDs. As a result, the obtained products from these precursors exhibit different fluorescence under 365 nm UV lamp (Fig. S7b–e). Herein, the products from the biuret and Rhodamine 6 G in NaCl-KCl-ZnCl₂ molten salt exhibited pink SSF with a maximal excitation-emission center at (572 nm, 650 nm) (Fig. S7f). The products prepared from phloroglucinol and m-phenylenediamine in NaCl-KCl-ZnCl₂ molten salts exhibited yellow SSF with a maximal excitation-emission center at (585 nm, 610 nm) (Fig. S7g). After further heating treatment at 200 °C for 20 min, the obtained yellow SSF CDs in NaCl-KCl-ZnCl2 molten salts present an SSF spectral redshift from 610 to 705 nm with a maximal excitation-emission center at (415 nm, 705 nm) (Fig. S7h). In addition, we also investigated whether changing the molten salt matrix could also obtain SSF CDs. With the novel NaOH-KOH molten salts, the CDs derived from citric acid and urea can still demonstrate bright blue SSF with a maximal excitation-emission center at (337 nm, 430 nm) (Fig. S7i). With these surveys, it can be deduced that the molten salt method can be employed to prepare solid-state CDs with full-color emission and high PLQY from various precursors or molten salt systems, and thus the molten salts strategy is universal to enable the synthesis of SSF CDs.

Structure and components of the full-color SSF CDs

To investigate the structure and components of the full-color SSF CDs, these CDs powder were treated with dialysis to remove the molten salts (Fig. S8). As a result, the obtained purified G-CDs present quite low Zn ionic precipitation (<2.75 mg L^–1) in the aqueous solution (Fig. S9), demonstrating the complete removing of molten salt from the SSF CDs. As illustrated in Fig. 2a–d, the transmission electron microscopy (TEM) images reveal that the purified SSF B-, G-, Y-, and NIR-CDs contain well-dispersed spherical particles, and the high-resolution TEM image indicate an interplanar lattice spacing of 0.21 nm for all the CDs, which is corresponding to the (100) plane of graphitic carbon^38,39,40,41. The TEM images of the B-, G-, Y-, and NIR-CDs reveal the approximate size distribution around 3.41, 3.18, 2.46, and 2.63 nm, and therefore there is no obvious size-dependent PL emission (Fig. S10). For these initial SSF CDs powder, the X-ray diffraction patterns present an obvious a characteristic amorphous carbon peak around 25.6° and the characteristic peaks of Zn-related w-ZnO (wurtzite ZnO) and o-Zn(OH)₂ (orthorhombic Zn(OH)₂) (Fig. 2e), indicating the carbonization and polymerization of the precursors and the in situ confinement effect of the molten salts^19,38. Thus, the surface functional groups and elemental compositions of the SSF B-, G-, Y-, and NIR-CDs were analyzed with Fourier-transform infrared spectroscopy, Carbon-13 nuclear magnetic resonance (¹³C NMR), mass spectrum, and X-ray photoelectron spectroscopy (XPS). Herein, all these purified CDs samples present the C = C vibration peak at 1572 cm^–1, the C = O vibration peak at 1630 cm^–1, and –OH vibration peak at 1406 cm^–1, and the NIR-CDs present obvious C–N vibration peak (1113 cm^–1) and N–H vibration peak (1494 cm^–1 and 754 cm^–1), implying the N-related long-wavelength emission (Fig. 2f)^42,43,44. The ¹³C NMR reveals different carbon atom bonding types on the surface of CDs, including the –COOH (155 ~ 163 ppm), O–C–O (90 ~ 102 ppm) and RN = CN (115 ~ 136 ppm) (Fig. 2g). As shown in mass spectrum, the purified B-, G-, Y-, and NIR-CDs reveal the same component of CH₂CH₂[(COO)₂Zn]OH⁺ (m/z = 198) and [(COO)₂Zn]OH⁺ (m/z = 170). These results can demonstrate that zinc ions are coordinated through the carboxyl groups on the surface of CDs (Fig. S11). The full XPS survey of these SSF CDs with dialysis reveal the elements of C (~284 eV), N (~400 eV), O (~531 eV), and Zn (~1022 eV) (Fig. 2h and Table S3), implying the existence of surface Zn ions coordination^45,46. Meanwhile, the high-resolution C1s spectrum exhibits three components at 284.8 eV (C–C/C = C, sp² hybridized carbon), 286.2 eV (C–O/C–N), and 288.7 eV (C = O) (Fig. 2i–l). Correspondingly, the O1s spectrum reveals oxygen species at 531.8 eV (C = O) and 533.2 eV (C–O) (Fig. 2m–p), further implying the surface ion exchange between the CDs and molten salts. The N1s spectrum further confirms the presence of three nitrogen configurations: pyridinic N (399.5 eV), pyrrolic N (400.5 eV), and graphitic N (402.1 eV) (Fig. S12), clearly demonstrating the mixed sp²/sp³ hybridized nature of these CDs with alterable N edge states.

The formation and light emission mechanism of the SSF CDs

The structure and effect of the zinc ions in CDs are further investigated. As illustrated in Fig. 3a–d, the high-angle annular dark-field (HAADF) TEM images and corresponding element mapping of the purified CDs are conducted. As a result, it can be clearly observed that the zinc atoms are uniformly distributed on all the surface of CDs. As shown in Figs. S13,  S14, when the G-CDs are prepared in molten salts without ZnCl₂ or with other zinc salts, the products present different SSF performance. Herein, the sample prepared without ZnCl₂ exhibits significantly reduced FL intensity and the samples prepared with zinc sulfate, zinc nitrate, or zinc acetate exhibit approximate SSF emission as the SSF CDs, suggesting that zinc ions are the key factor in enhancing the SSF of CDs. On the other hand, after removing the molten salts, the obtained B-, G-, Y-, and NIR-CDs solutions exhibit maximal excitation-emission centers at (365 nm, 450 nm), (445 nm, 495 nm), (475 nm, 510 nm), and (450 nm, 520 nm), and the corresponding PL QY is calculated as 14.93%, 57.72%, 1.28% and 4.11% (Fig. S15), demonstrate the in situ space confinement effect from molten salt. The different mass of B-, G-, Y- and NIR-CDs powder were dispersed into the DMF with different volume fraction of deionized water to investigate the possible solvent effects or dynamic processes. Consequently, the B-, G-, Y-, and NIR-CDs exhibit concentration-dependent and solvent-dependent luminescence behaviors, highlighting their unique solvent effects and intraparticle interactions (Fig. S16). Summarizing the structure and component analysis, it can be deduced formation of zinc ions coordination in the SSF CDs.

The density functional theory (DFT) calculation was employed to investigate the influence of zinc ions coordination on the reaction progress and optical properties of the CDs. For the precursors of citric acid molecules, there is a strong p-π conjugation interaction between the hydroxyl oxygen and carbonyl oxygen on the carboxyl group, and this interaction can delocate the π electrons and thereby increase the polarity of the O-H bond, endowing them prone to breakage (Fig. 3e). When the reaction system is transformed into a molten state after heating, the molten salts can provide a low-viscosity and highly diffusible ionic liquid environment for the cleavage of H⁺. On the condition, H⁺ is more inclined to combine with the amino group on the urea molecule to form ammonia gas and rapidly volatilize in a high-temperature environment, enabling the disappearance of N element in the as-prepared SSF CDs. On the other side, the original position of the N atom on the urea molecules prefers to be oxidized and replaced by an oxygen atom to form a –COOH group when the reaction occur in an atmospheric environment. Since the oxygen atoms have extra lone pairs of electrons and the spatial orientation of –COOH matches the orbital spatial configuration of zinc ions, the precursors of citric acid and urea can undergo the molecular polymerization through the coordination between the –COOH and zinc ions. In the absence of molten salts, the citric acid and urea precursors will complete the polymerization process to form amide bonds from the dehydration condensation between carboxyl and amino groups (Fig. 3f). We have employed DFT to compare the difference of Gibbs free energy (ΔG) in the processes of the molecule polymerization for the reaction with and without the molten salts. As a result, the ΔG for the reaction with the molten salts is −27.64 eV, while the ΔG of the reaction without the molten salts is 0.52 eV. Therefore, when the molten salt is involved in the reaction system, a lower ΔG can enable the reaction to be more easily proceed. Additionally, the negative ΔG of the reaction with molten salts implies that the molten salt can promote reduce the reaction barrier, endowing a lower temperature threshold for the formation of CDs. As illustrated in Fig. S17, the reaction temperatures of B-, G-, Y- and NIR-CDs was monitored by a thermometer, and the highest recorded temperature of the B-, G-, Y-, and NIR-CDs are as low as 100 °C, 110 °C, 120 °C, and 142 °C, respectively. The extremely low reaction temperatures agree with the theoretical calculation results, approving that the molten salts can reduce the potential barrier of reaction and promote the formation of CD nanoparticles.

To further understand the specific effects of zinc ion coordination, the graphene-like CDs with two carboxyl groups (Model 1), four carboxyl groups (Model 2), six carboxyl groups (Model 3), two zinc ion coordination (Model 4), four zinc ion coordination (Model 5), or six zinc ion coordination (Model 6) were established (Fig. 3g). The time-dependent DFT (TD-DFT) calculations were employed to further investigate their band structures. Firstly, the vibrations were analyzed on these models and the respective molecular vibration frequencies were calculated. As shown in Fig. 3h, the vibration frequency of the Model 1, Model 2 and Model 3 is higher than that of Model 4, Model 5 and Model 6, while the Model 4, Model 5 and Model 6 present more vibration modes than the Model 1, 2 and 3. Meanwhile, the average vibration frequency (VF_avg) of the Model 4, 5 and 6 are lower than that of the Model 1, 2 and 3 (Fig. 3i). Furthermore, a more pronounced reduction in VF_avg was observed with increasing zinc ion coordination in these models, implying that the zinc ion coordination can suppress the vibrational deactivation pathway of the S1 state and reduce the non-irradiation to improve the SSF of CDs. Besides, the highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energy level distributions of these models were calculated. As depicted in Fig. 3j and Fig. S18, the Model 1, Model 2 and Model 3 present HOMO energy level of −5.73, −6.03 and −6.27 eV, and LUMO energy level of −2.24, −2.61, and −2.83 eV, and the Model 4, Model 5 and Model 6 present HOMO energy level of −5.84, −6.18 and −6.49 eV, and LUMO energy level of −2.42, −2.87 and −3.07 eV. Thus, the Model 1, Model 2 and Model 3 present bandgap of 3.49, 3.42, 3.44 eV, and the Model 4, Model 5 and Model 6 present bandgap of 3.42, 3.31, 3.42 eV. The decreased bandgap of graphene-like CDs with zinc ion coordination imply that the zinc ions can also promote the redshifted emission of CDs, which are corresponding with the experimental survey. Hence, we can observe a red shift in the spectrum and an increase in luminescence efficiency as CDs transition from a solution state to a solid state.

On the other side, we have carefully investigated the interactions between molten salts and CD particles in these SSF CDs. Firstly, the molten salts were prepared by directly heating the mixed salt system and the products exhibit ultraweak fluorescence under 365 nm UV lamp. Their optical properties clearly indicates that the molten salts can’t contribute to such novel SSF of these CDs (Fig. S19a–S19c). Similarly, the possible molten salts and impurities (Ms/impurities) were separated from the G-CDs, and these Ms/impurities exhibit ultraweak fluorescence under 365 nm UV lamp. The low PL QY indicates the little contributions of the Ms/impurities to the PL emission of these SSF CDs (Fig. S19d–f). To further investigate the contributions of molten salts, the pure CDs (pCDs) were separated from the SSF G-CDs and the as-obtained starch-like pCDs present slight fluorescence under 365 nm UV lamp. The different optical properties of the pCDs and SSF G-CDs imply that the in situ confinement of molten salts can reduce the nonradiative recombination and improve the PL emission of CDs (Fig. S20). On the condition, the pCDs were assembled with the molten alts via different strategies to investigate the interactions between the CDs and molten salts. Herein, the pCDs/molten salts (pCDs/Ms) were prepared by direct mixed and grinded, the self-assembled pCDs and molten salts (pCDs-Ms) were prepared by dissolving and drying the pCDs and molten salts in deionized water, and the heated pCDs and molten salts (pCDs-Ms-H) were prepared with the pCDs and molten salts as the approach to the synthesis of SSF G-CDs (Fig. S21). The assembled products can increase the PL QY of the pCDs but can’t contribute to the high PL QY of the SSF G-CDs, implying that the orderly assembly of molten salts and Zn-CD particles occur in the reaction process and contribute to the intraparticle interaction to effectively inhibit the non-radiative recombination. We have also prepared the CDs-ws by only dry-burning the organic precursors of the SSF G-CDs and further assemble the CDs-ws and the molten salts to prepare the CDs-Ms in aqueous solution (Fig. S22). In comparison with the CDs-ws, the decreased PL QY of the salt-like CDs-Ms demonstrate that the important effect of zinc ion surface coordination on the PL emission of SSF CDs.

As a result, the B-CDs, G-CDs, Y-CDs to NIR-CDs exhibit different relative amount of sp² C (C–C/C = C) and different relative amount of pyridinic/pyrrolic N and graphitic N, which means the introduction of N edge state in these CDs (Figs. 2 and S12). As previous reports^47,48,49,50, controlled graphitization and surface functionalization can tune the HOMO − LUMO gap and midgap states of the CDs, resulting in the tunable PL emission. Considering the different precursors and reaction factors, the different graphitization and surface functionalization can lead to the part optical properties of these SSF CDs. Meanwhile, the in situ confinement of molten salts and zinc ion coordination further regulate the intraparticle interaction and surface functionalization (Fig. S23), resulting in the significant differences in the luminous efficiency among the samples. For the NIR-CDs, the precursor of o-phenylenediamine is different from the precursors of B-, G- and Y-CDs, which may be difficult to be carbonized to form sp²-hybrid carbon core and there are little surface groups for the surface zinc ion sites, jointly resulting in the lowest luminescence efficiency. Furthermore, according to the energy gap law, the reduction of excited state energy levels significantly enhances non-radiative decay and quenching rates, consequently diminishing the quantum efficiency of long-wavelength emission^{51,52,53,54,55}.

Machine learning-assisted improvements of the SSF CDs

As the previous survey, the in situ confinement of molten salts and surface Zn coordination jointly endow the extraordinary PL QY and therefore the possible residual organic precursors indeed may affect the luminescent properties. As illustrated in Fig. S24 and Table S4, the G-CDs were prepared with different mass of precursor in the molten salts. As a result, these G-CDs present different PL QY, demonstrating the influence of residual organic precursors on the SSF CDs. On account of the SSF enhancement from molten salts and reduced influence of residual organic precursors, we believe that the reaction factors can influence the SSF performances of these CDs and thereby the SS QYs can be further improved. To achieve this hypothesis, the machine learning techniques are employed to optimize the reaction parameters. With the mass of citric acid (CA), the mass of urea (UA), the mass of NaCl (NaCl), the mass of KCl (KCl), the mass of ZnCl₂ (ZnCl₂), the reaction Time (Time), and the temperature of the heating equipment (Temp) as features, and the measured PLQYs as a label, the dataset consisting of 120 samples was constructed. After constructing the dataset in the laboratory, the dataset was preprocessed before training to ensure the quality of the dataset and the training effect. Firstly, the mutual information between each feature and the label was evaluated. As illustrated in Fig. 4a, the facture of ZnCl₂ presents no significant information which contributes to the label prediction. Therefore, the facture of ZnCl₂ was removed from the feature set. On the condition, the Pearson correlation coefficient matrix was employed to evaluate the correlation between these features. As shown in Fig. 4b, there is a moderate negative correlation between the feature of time and temperature, implying that the shorter reaction time was required under a higher temperature for the G-CDs with the same emission wavelength. This correlation coefficient is consistent with the currently recognized mainstream view that the temperature and time jointly determine the carbonization and polymerization degree of CDs. On the other hand, there is a strong positive correlation between the feature of CA and UA, implying that the optical properties of the CDs were intensely affected by the proportion of CA and UA. Since a strong correlation is prone to cause the problem of multicollinearity, the feature transformation was performed to reduce the feature correlation. Considering the requirements of prediction results in actual synthesis in laboratory, the transformed features need to be able to infer the specific dosage of the precursor. Therefore, the mass sum of citric acid and urea (Sum) and the mass Ratio of citric acid to urea (Ratio) were taken as the new features. Herein, the Pearson correlation coefficient matrix was calculated and the results revealed that all the feature pairs exhibited low linear correlation after feature transformation, approving the effectiveness of feature selection (Fig. 4c). On the condition, the dataset was divided into the training set and test set (5:1). After the model learned the potential features of the data in the training set, the test set was used to evaluate the differences between the prediction results and real labels. Four models, namely RF, XGB, SVM and ELN, were selected for training on the training set, respectively and the grid search was adopted to optimize the model parameters and improve the model performance. The R² and RMSE of the model on the test set were adopted as the specific indicators to evaluate the model performance. Among these models, the R² is a statistical indicator to measure the goodness of fit for the regression model, and provides information about the interpretation degree from the independent variability to the dependent variability. The closer the value is to 1, the better the interpretability of the model for the data. The RMSE is a commonly used indicator to measure the difference between the predicted value and actual value, and its core idea is to measure the error size of the prediction result. A lower value corresponds to higher prediction accuracy in the model. In this work, from the specific performances of RF, XGB, SVM and ELN, the XGB presented the highest R² score and the smallest RMSE, which was superior to other models (Fig. 4d). To prevent overfitting of XGB, the differences between the true values and the predicted values was examined on the test set and the training set, respectively. As depicted in Fig. 4e, the data in the training set were basically coincided with the X = Y line, indicating the excellent learning effect of XGB on the training set. Meanwhile, the data in the test set also largely maintain the X = Y trend, approving the excellent performance of XGB on unknown datasets. As a result, the XGB has the highest prediction accuracy and generalization ability. Therefore, the XGB model was further employed to perform the PLQY prediction task. To explore the maximum label value and its corresponding features in the unknown data space, we constructed a feature grid containing 134,198,064 groups and input it into the XGB for PL QY prediction (Table S5). Since the feature of ZnCl₂ in the data preprocessing step was removed, the specific value of ZnCl₂ could not be input into the XGB prediction. When the XGB output was used for the laboratory preparation, the mean value of ZnCl₂ in the dataset was taken and the XGB presented a predicted maximum PL QY for the G-CDs when the reaction parameters were Ratio = 1.6, Sum =  2.1, NaCl = 1.5, KCl = 2.5, and ZnCl₂ = 4.1 (Fig. S25). As shown in Fig. 4f, Fig. S26, and Table S6, the as-prepared CDs exhibit a near-unity solid-state PL QY of 99.86%, as independently verified by multiple laboratories, representing the highest value ever reported for solid-state CDs (Table S7). The laboratory synthesis conditions can be readily scaled up for mass production of SSF-CDs, yielding ~1338.5 g per batch with an efficiency of ~92.66% (Fig. 4g and Fig. S27), indicating the promising potential of molten salt-assisted SSF CDs in industrial production. Meanwhile, different batches of the G-CDs prepared with the same reaction factors present similar excitation-emission matrix and similar PL QYs (Fig. S28), demonstrating the minor variations of the properties of the SSF CDs from batch to batch. In addition, after continuously illumination under 365 nm lamp, the as-prepared G-CDs powder present almost the same SSF emission (Fig. 4h, i), demonstrating their excellent photostability.

Multi-color LED and backlit displays application of the SSF CDs

The full-color light emission wavelength and high PL QY inspire us to develop the CDs as phosphors in light-emitting diodes (LEDs). Herein, the SSF CDs powder was first evenly mixed with polydimethylsiloxane and then coated onto a 365 nm UV or 400 nm blue LED chips (Fig. S29). The as-prepared LEDs from B-, G-, Y-, and NIR-CDs are named as B-, G-, Y-, and NIR-LEDs, respectively. Figure S30 shows the actual photographs of these LEDs, and Fig. S31 presents the emission spectra of these LEDs, indicating the light emission conversion from the SSF CDs. The corresponding CIE coordinates also demonstrate the success light conversion (Fig. S32). As shown in Fig. 5a–d, the B-, G-, Y-, and NIR-LEDs present a maximum luminous efficiency of 23.46, 156.29 and 15.26 and 0.03 lm W^–1, approving the potential applications of these SSF CDs in light conversion. On the condition, the luminous efficiency of these LEDs devices was measured to evaluate the luminescent performance. To obtain LED devices with the optimal performance, the detailed information about the relationship between the packing ratio and device performances were investigated. Herein, Seven G-LEDs are prepared by different concentrations of G-CDs and 420 nm chip. As depicted in Fig. 5e and Fig. S33, these devices present different light emission color and EL spectra. When the G-CDs concentration is lower than 0.84 g mL^–1, the LEDs present the higher light emission intensity of G-CDs and lower EL intensity of 420 nm chip with the increased G-CDs concentration. When the G-CDs concentration reached a threshold of ~0.84 g mL^–1, the increased loading concentration of G-CDs will not cause changes of EL emission. Meanwhile, the G-LEDs present an increased luminescence with the increased G-CDs concentration when the G-CDs concentration is lower than 0.84 g mL^–1, and a decreased luminescence intensity when further increasing the loaded G-CDs concentration (>0.84 g mL^–1) (Fig. 5f). The current and luminescence efficiency of the LEDs exhibits the same tendency. As shown in Fig. 5g, an excessive loaded G-CDs concentration can cause a decreased luminous efficiency. Since the luminescence of LEDs is dependent on the factors of the phosphors, the excessive loaded G-CDs can lead to numerous self-absorption and thus result in energy loss due to the overlap between the absorption and PL emission of G-CDs (Fig. S34). Meanwhile, the luminous efficiency of the initial LED chip gradually decreased as the current was further increased when the LEDs reached maximum luminous efficiency and therefore the EL emission of G-LEDs also be influenced by the saturated injected electrons in LED chip and exhibit a decrease luminous efficiency when the current are too high (Fig. S35). As a result, the G-LEDs are prepared with a threshold load concentration (0.84 g mL^–1) and thus the LED device present a maximum luminance of 455,694 cd cm^–2 at a driving current of 52.7 mA, and a maximum luminous efficiency of 272.65 lm W^–1 at a driving current of 15.2 mA. This is the highest performance value of CDs-based LEDs as far as we know, and it is even comparable to some rare earth-based LEDs devices (Table S8). In addition, the new G-LEDs were used for aging test and continuously lit them at an initial luminance of 10748 cd m^–2. Consequently, the T₉₅ of the device was measured to be 15.58 h (Fig. 5h). According to the formula L₀ⁿT₉₅ = constant (n is a laboratory factor with a value of 1.79), the T₉₅ for an initial luminance of 1000 cd m^–2 and 100 cd m^–2 is 1093 and 67397 h, demonstrating the excellent stability of the devices.

In recent years, high-brightness mini-LED backlight displays have become increasingly popular for high-end applications and thus the optimized G-CDs were further used as the phosphors in mini-LED. Subsequently, a 450 nm blue mini-LED chip-on-board (COB) were prepared with a green-emitting and red-emitting conversion film of G-CDs and (Sr,Ca)AlSiN₃ (Figs. S36,  S37). As a result, two color conversion films were combined and successful to create a mini-LED white backlight (Fig. S38). As shown in Fig. 5i, the mini-LED backlight was integrated with a commercial TFT-LCD panel to design a complete display device (Figs. S39,  S40). The EL emission spectrum of the mini-LED white backlight can cover the entire visible light region (Fig. 5j) and present obvious white color and the CIE coordinate is calculated as (0.31, 0.29) (Fig. S41). With an initial luminance of 10599.8 cd cm^–2, the T₉₅ (the time that the luminance intensity decreases to 95% initial value) was 10.69 h (Fig. 5k). With the formula L₀ⁿT₉₅ = constant (n is the laboratory factor with a value of 1.79), the T₉₅ values were estimated as 731.6 h at initial 1000 cd cm^–2 and 45108.7 h at initial 100 cd cm^–2, respectively. The long lifespan of the mini-LED white backlights demonstrates their durability for long-term display applications. In addition, as a simple demonstration, the display exhibits a clear blue sky, green plants, and red flowers, showcasing the vibrant color saturation and high luminance of the CDs-based device (Fig. 5l).