The a-SiN_xO_y films (∼500 nm) were deposited on polished Si substrates in a plasma-enhanced chemical vapor deposition (PECVD, OXFORD Plasmalab 80PLus, Oxford, UK) system with a silane, ammonia, and nitrogen gas mixture The gas flow ratios R (R = NH₃/SiH₄) and the related fabrication parameters were well controlled through the whole fabrication processes, both of which were described in our previous work last year in detail [23]. After fabrication, the samples were subsequently oxidized in situ, and then post-treated by combining thermal annealing with pulsed laser annealing.

The structure of the a-SiN_xO_y films was investigated through XPS and FTIR measurements. To intensively investigate the atom scale structure defects of a-SiN_xO_y thin films, we measured the EPR spectra under different R conditions at room temperature with a Bruker EMXplus in the X-band (microwave frequency f∼9.85 GHz, microwave power 20 mW). The temperature dependent time resolved PL (TD-TRPL) properties of a-SiN_xO_y films with different R were measured with a Fluorolo-3 system (Jobin Yvon) and a HP4284 LCR meter in a computer-controlled Delta 9,023 oven, using a FLS980 (Edinburgh Instrument) equipped with an EPL375 pulse diode laser (pulse width ∼53 ps, repetition rate = 20 MHz, λ _exc = 375 nm, pumping fluence W _PF = 5 mJ/cm²) as light sources. A time-correlated single photo counting (TCSPC) system (time resolution ∼100 ps) were used to record the characteristics of TD-TRPL decay properties. Here we consider that the initial decay of the PL intensity coincides with the excitation signal which is origin from the instrumental response (the instrumental response is about 100 ps). Thus, to deduct the impact of the initial excitation signal, we carefully deconvolved the measured ns-PL decay curves and modified the τ average .

The principle of TCSPC is shown in Figure 1, where a sample is first excited with a narrow laser pulse, and the sample emits fluorescent photons after stimulus. It is assumed that the excitation laser pulse is weak enough so that the sample produces only a single fluorescent photon after each pulse. The time it takes for the first fluorescent photon emitted by the sample to reach the optical receiver (which can also be seen as the time t when a single photon appears) is then measured. This time is proportionally converted into a corresponding voltage pulse by TAC, which is then fed into a multichannel analyzer via A/D conversion. Multiple counts are then performed, and in the multi-channel analyzer, these output pulses are sequentially fed into each channel for cumulative storage. Since the probability of a photon being detected in a certain time interval is proportional to the intensity of the fluorescence emission, repeated measurements can obtain a P(t) histogram of the probability distribution of fluorescent photons that is essentially the same as the original waveform. The histogram measured in this case is equivalent to the I(t) curve of the decay of fluorescence intensity over time after excitation has stopped. This is as if a beam of light (many photons) passing through a small hole creates the same diffraction pattern as a single photon passing through a small hole over a long period of time.

In general, the PL of solid films is a non-equilibrium radiation process, and the luminescence can continue for a period (>10⁻¹¹ s) after the excitation stops. When the excitation stops, the PL intensity I(t) decays exponentially over time. The time required when I(t) drops to 1/e of the maximum intensity at excitation is called PL lifetimes (τ) of the excited state, which indicates the average time that a particle exists in the excited state. τ average at a specific temperature can be generally obtained according to the formula I t = I 0 ⁡ exp − t / τ T [1]. The PL lifetimes of luminescent materials are related to their own structure, and this phenomenon mostly occurs from the nanosecond to the microsecond time range, which is in the time scale of molecular movement. Therefore, the property changes of the systems and the intermolecular interaction processes can be directly understood by the PL lifetimes determination. In particular, when fitting fast fluorescence lifetimes, two major types of fitting may be used: tail fitting and reconvolution fitting.

The stretched exponentials are used widely and accepted as good markers of light recombination in disordered systems [16, 18, 20]. For comparison, we firstly checked the recombination processes in the controlled a-SiN_x films by using stretched exponential model and obtained nice fitting results. However, in the fitting processes of a-SiN_xO_y systems, we found that if we use stretched exponential model, the error is large and the fitting results are unstable. For band tail related a-SiN_x films, the excited carriers relax to the deeper tail states then through the thermalization and radiative recombination to give luminescence. With the increasing excitation photon energy (E_exc), the excited carriers occupied the higher states in the band tail [18]. In our a-SiN_xO_y systems, we have confirmed that the light emission is mainly originated from the N-Si-O-related defect centers in our previous work [21, 23]. The defect luminescence model has been confirmed by the Stokes Shift between Urbach edge (E_{U Edge}) and PL energy (E_PL) ∆ E stokes = E U   Edge – E PL from PLE measurements. We found that the E_PL Stokes Shift showed a near constant value of about 0.75 eV, which were independent on the optical band gap. There are two steps in the PL recombination process of a-SiN_xO_y films. Firstly, the excited electronics are relaxed down to the band tail states meanwhile thermally ionized to the defect states in the band gap through non-radiative processes, then recombine via transition between the defect states and valence band tail states to give luminescence [21]. The recombination processes of a-SiN_xO_y films are different from band tail related a-SiN_x films. Thus, here we choose the double exponential decays [11] instead of stretched exponentials to fit the PL lifetimes of a-SiN_xO_y films.

The presence of Si, N, and O is measured from the binding energies of the Si 2p, N 1s, and O 1 s peaks in the XPS spectrum [23]. After the top layer (∼60 nm) was removed by Ar ion beam, the O concentration changed slightly, and the average oxygen content in a-SiN_xO_y films for different R was about 3.5%. As the sputtering time increases, the O concentration tends to stabilize, which is direct evidence of the incorporation of O into a-SiN_x. In order to further verify the N-Si-O bonding configuration, we measured the Fourier infrared absorption spectra (FTIR) of a-SiN_xO_y films and the corresponding a-SiN_x films.

As shown in Figure 2, the vibration peaks of the Si-Si disordered structure (460 cm⁻¹), the Si-N stretching mode (840 cm⁻¹), the N-H rocking mode (1,170 cm⁻¹), the Si–H stretching mode (2,150 cm⁻¹), and the N–H stretching mode (3,350 cm⁻¹) are clearly visible in both of a-SiN_xO_y films and the controlled a-SiN_x films. However, the Si-O stretch mode (1,070 cm⁻¹) was not seen in all FTIR spectra, suggesting that the oxygen atoms were incorporated only as trace impurities. It is worth noting that, from the vibration peaks of the Si-N stretching mode, we found that the Si-N stretching peaks of a-SiN_xO_y films are slightly shifted to the direction of higher wavenumbers than those of a-SiN_x, and there is already an obvious shoulder peak (such as R = 1). This shoulder is clearly derived from the N-Si-O bonding configuration formed in a-SiN_x by the incorporation of O.

Figure 3 shows the measurements of the EPR curves of a-SiN_xO_y films with different R. As can be seen from Figure 3, there is a strong resonance absorption peak near the x-axis with magnetic induction intensity of 3,515 G, indicating the presence of unpaired dangling bonds in a-SiN_xO_y films. Since the structure of a-SiN_xO_y films is regarded as an intermediate state with both a-SiN_x and a-SiO_x, it has been found that the suspensory bond defect states in the bandgap of a-SiN_x and a-SiO_x films may coexist in a-SiN_xO_y films. We calculated the value range of the zero-crossing g-factor under different R of a-SiN_xO_y films is 2.0025–2.0040, and the peak line width is 8.5–12.0 Gauss.

According to previous analysis [21], we selected the trivalent Si DBs and the N-Si-O configurations related N_x defect states, and performed Lorentz function fitting on the experimental EPR curves. The fitting results show that the trivalent Si DBs and N_x defect states coexist in the band gap of the a-SiN_xO_y film, and the relative dangling bond concentrations of the two defect centers are also obtained by fitting. Combined with the PL spectra of a-SiN_xO_y films of different R [23], we found that the PL integrated intensities (I_PL) of a-SiN_xO_y films were proportional to the concentrations of N-Si-O related N_x dangling bond defect states. At the same time, we also found that the relative concentration of Si DBs in a-SiN_xO_y films was much higher than that in the N_x defect states. However, due to the low PL efficiency of Si DBs defects in a-SiN_xO_y thin films [20], they can even be regarded as non-radiative recombination centers.

We measured the PL spectra of a-SiN_xO_y films with different R ratios at 8 K under He-Cd continuous laser excitation at a wavelength of 325 nm. As shown in Figure 4, by adjusting the flow ratio R, we achieved a fluorescence emission of a-SiN_xO_y films with modulated luminescence wavelengths (λ_PL) in the visible range. It is worth noting that as R increases from 0.3 to 5, the PL peak blue shifts from 590 nm to 425 nm as R gradually increases; the corresponding PL integrated intensity (I_PL) first increases and reaches saturation (R = 1.5), and then gradually decreases.

To further investigate the relationship between the PL properties of a-SiN_xO_y films and the defect states of N-Si-O-related N_x dangling bonds, we systematically measured the ns-transient fluorescence PL characteristics of a-SiN_xO_y films (R = 1) at different detection wavelengths. As a comparison reference, we also measured the related properties of a-SiN_x films with the same flow ratio.

Figures 5A, B show the ns-TRPL decay spectra and the relative fitting decay curves for a-SiN_x and a-SiN_xO_y thin films at R = 1, respectively. According to the formula I t = A 1 ⁡ exp − t / τ 1 + A 2 ⁡ exp − t / τ 2  with  τ T = ∑ i = 1 n A i τ i 2 A i τ i [11], we obtain the corresponding ns-PL lifetimes ( τ average ) at different detection wavelengths, as shown in Figure 5C.

From Figure 5C, we can see that for a-SiN_x films, the PL lifetimes of the samples continue to increase as the detection wavelength increases. This is a typical radiative recombination dynamics process of the band tail carrier transition, which is consistent with the previously reported results [18]. When the excitation photon energy ( E exc ) is less than the optical band gap ( E opt ) ( E exc   <   E opt ), the absorption only occurs at the band tail, and with the decrease of E exc , the luminescence peak E_PL gradually shifts to the deeper energy level of the band tail, that is, the E_PL red shift. At this time, the excited state carriers relax to the deeper energy level of the band tail, and emit light through thermal ionization inter-band recombination, resulting in a longer corresponding PL lifetime.

However, for a-SiN_xO_y films, with the increasing detection wavelength, the fluorescence lifetime of the sample first increases to saturation, and then gradually decreases, and its change trend is obviously different from that of a-SiN_x films, but consistent with the change trend of defect density distribution, and the corresponding fluorescence lifetime value is proportional to the concentration of N_x defect states. Therefore, we believe that the N_x luminescence defect states related to the N-Si-O configuration dominate the radiation recombination process of a-SiN_xO_y films. The ns-TRPL properties of the a-SiN_xO_y films mentioned above verify that our high-efficiency light emission originates from the N-Si-O-related defect luminescence centers of the ns-order radiative recombination processes.

Next, to gain a deeper understanding of the radiative and non-radiative recombination mechanisms of high-efficient light emission from N-Si-O-related defect luminescence centers, we studied the PL kinetics of a-SiN_xO_y films at different test temperatures. Figure 6 shows the PL decay curve for the ns time range of a-SiN_xO_y thin films (R = 1) at a measured temperature range of 8 K–300 K. As shown in Figure 6A, the PL decay curve profiles remain unchanged at the test temperature range below 100 K, indicating that radiative recombination dominates the entire recombination process. When the test temperature is increased from 120 K to 300 K, it can be seen from the PL decay curve that the PL lifetimes of a-SiN_xO_y films become shorter, and the non-radiative recombination effects become more and more obvious, as shown in Figure 6B.

At the same time, from the PL decay curve of a-SiN_xO_y films, we can further see two different ns fluorescence radiation recombination processes, which we call the fast ns-PL lifetimes and the slow ns-PL lifetimes, respectively. The PL decay curves of a-SiN_xO_y films were well fitted with a double exponential function, and the fitting results for all PL lifetimes are shown in Figure 7. The obtained ns-PL lifetimes τ_average of the samples were reduced from 11.9 ns (less than 100 K) to 7.8 ns As shown in Figure 7, τ_average tends to be stable (about 11.9 ns) and does not change with temperature in the low temperature range (less than 100 K), indicating that the PL lifetimes are mainly due to the radiation recombination processes of the carriers. However, the non-radiative recombination increases with increasing temperature, resulting in a continuous decrease in τ average , which drops to 7.8 ns at 300 K. It is worth mentioning that, the variation trend of the PL lifetimes at a certain temperature τ meas (T) is consistent with the PL integration intensity of the steady-state temperature dependent PL spectra [23]. This phenomenon once again verifies the typical PL properties of luminescent N-Si-O-related defect states.

As Section 3.5 mentioned above, we use the ns-PL lifetimes τ average to determined the radiative and the non-radiative recombination lifetimes. Thus, the PL internal quantum efficiency at different temperatures (η(T)) should be determined as [5]: η T = k r T k r T + k n r T = 1 τ r T 1 τ r T + 1 τ n r T = τ a v e r a g e T τ r T , with  1 τ a v e r a g e T = 1 τ r T + 1 τ n r T . (1)

Here k r T and k nr T denote the radiative and the non-radiative recombination rates, and τ r T and τ nr T denote the radiative recombination lifetimes and the non-radiative recombination lifetimes. τ r T and τ nr T are expressed as [5]: τ r T = τ a v e r a g e T η T , and  τ n r T = τ a v e r a g e T 1 − η T (2)

Based on our previous work [23], we obtained the PL IQE values from TD PL spectra and the directly measured PL QYs. Thus, by combining the resulting PL IQE values with the ns-PL lifetimes, we can further determine the radiative recombination lifetimes and non-radiative recombination lifetimes of the sample according to Equation 2. Figure 8 shows the radiative recombination lifetimes and non-radiative recombination lifetimes of a-SiN_xO_y films with different R at different detection wavelengths. The detailed of the calculated τ r , τ nr , k r and k nr by using the PL IQE (η) and the obtained lifetimes τ average under room temperature for a-SiN_xO_y films with various R are listed in Table 1. We note that the trends of radiative recombination lifetimes are also consistent with the trend of the densities of the N_x defect states. When τ nr T is much greater than τ r T (R = 1, 1.5, 2.5), the radiative recombination dominates the whole recombination process, and the PL IQE is higher (∼70%). When τ nr T is less than τ r T (R = 0.3, 0.5, 5), the effect of non-radiative recombination is relatively obvious, resulting in a decrease in PL IQE (∼40%).

Since the fluorescence lifetimes at low temperature are hardly affected by non-radiative recombination, we also can use that as the radiative recombination lifetimes ( η ∼ 1 , and τ average   ∼ τ r in the low temperature range) to calculate the non-radiative recombination lifetimes at room temperature. We compared the radiative and non-radiative recombination lifetimes obtained by these two different calculation methods, and obtained the similar results at last. The average τ_r is basically unchanged and tends to be stable at ∼10 ns under different R conditions. These results show that the radiative recombinations of the luminescent N-Si-O defects remain in the order of 10⁸ s⁻¹ and keep unchanged throughout the whole recombination processes, which can be compared with those direct bandgap materials (e.g., typical cadmium selenide NCs, kr∼10⁸ s⁻¹).