Silicon nitride stoichiometry tuning for visible photonic integrated components Open Access

Silicon nitride (Si₃N₄) is one of the most utilized and important materials in integrated photonics, acting as the core material for guiding light in different ranges of the spectrum,¹ extending from visible to near-infrared (NIR) due to its broad optical transparency.² Si₃N₄ is one of the platforms used for the development of photonic integrated circuits (PICs) around the world, together with indium phosphide (InP),³ silicon-on-insulator (SOI),⁴ and silicon dioxide (SiO₂).⁵ Among these materials, Si₃N₄ based PICs are fabricated by a CMOS compatible technology, which exploits the already mature microelectronics processes. Refractive index tunability, multi-layer stacking, isotropic behavior, low temperature sensitivity, and high tolerance to fabrication errors are some of the remarkable properties of this material.⁶ 

One of the main challenges faced by the Si₃N₄ photonic integrated community is to try to tailor the refractive index of the Si₃N₄ layer in order to achieve more or less confinement in the waveguide mode, tune the propagation phase, or exploit nonlinear effects,^7,8 with repeatability and reliability. Different studies changing the deposition conditions using plasma-enhanced chemical vapor deposition (PECVD) have been demonstrated, achieving ranges from 3.1 down to 1.7 in the refractive index value. However, the repeatability and reliability of the PECVD process is not as good as with low-pressure chemical vapor deposition (LPCVD).¹ LPCVD silicon nitride refractive index tuning has been demonstrated, achieving a value of 2.2 by changing the ratio of the precursor gases during the deposition.⁹ A different approach to vary the optical index is the ion implantation, doping the Si₃N₄ layer after the deposition.^10–12 

All these material studies have been exploited in integrated photonics and have been driven by telecommunication applications operating at 1550 nm and/or 1310 nm.¹³ The transparency of the Si₃N₄ allows the already mature technology to extend the range of operation toward the visible range, where applications, such as optogenetics,^14,15 imaging and display technologies,^16–19 quantum computing,²⁰ or underwater communication,²¹ amongst other applications,^22–28 have been exploited. A few research groups and companies are starting to develop different building blocks in this range of the spectrum.^29–39 Therefore, to optimize the photonic integrated applications in the visible range, one key parameter to be studied, apart from the already mentioned refractive index, is the auto-fluorescence of the Si₃N₄ layers.⁴⁰ 

Auto-fluorescence, or native fluorescence, in Si₃N₄ occurs when thin films acting as waveguide cores are confining light at a certain wavelength able to excite the material, emitting lower radiation energy in the UV, visible, or NIR spectral range. The emitted background luminescence⁴¹ may interact with the used wavelength producing back-scattering or interfering with it along the propagation channel. This is an undesired phenomenon for quantum applications, such as single photon emitters,^42,43 and also for chip-based optical systems, such as biosensors,⁴⁴ Raman spectroscopy,^45,46 optical coherence tomography (OCT),⁴⁷ and nanoscopy,⁴⁸ where the presence of photoluminescence (PL) from the guiding material introduces noise to the measurements.

The PL mechanism of different silicon nitride stoichiometries is still an object of research. Some groups associate the PL to the quantum confinement effect of Si nanoclusters,^49,50 while others keep the band tail luminescence of silicon nitride matrix itself as the responsible for the PL.⁵¹ 

In this work, two sets of p-type (100) 4 in. silicon wafer samples were fabricated. 300 nm of two different silicon nitride films, stoichiometric silicon nitride (SSN) and silicon-rich nitride (SRN), were deposited on wafers with 2.5 μm oxidation and on wafers without the oxidation, directly deposited on the Si substrate. Once the standard stack layers were fabricated for the different silicon nitrides, the wafers were cut in quarters, named A, B, C, and D, each submitted to a specific thermal annealing process, which can be found in Table I. In addition, the annealing processes were studied in two different atmospheres, argon and nitrogen. Two different annealing methods, rapid thermal annealing (RTA) and furnace annealing, were selected in order to distinguish between effusion and diffusion effects in the films. More information about fabrication and characterization is provided in the supplementary material.

One of the main objectives of the presented work is to analyze the auto-fluorescence in silicon nitride films employed for building photonic integrated structures in the visible spectral range. In order to elucidate the mechanism and find ways to reduce it, layer structures similar to the ones used at the IMB-CNM PIC platform⁵² will be considered, where Si₃N₄ waveguide structures with different functionalities are fabricated over a silicon oxide buffer. To analyze the thermal annealing effects on the interfaces, single silicon nitride layers deposited directly on silicon substrates will be also under study. Furthermore, a refractive index tuning study is also given. By changing the stoichiometry of the Si₃N₄ films, a variation in the PL is produced, but also in the refractive index. Both parameters are key for designing and fabricating photonic integrated circuits in the visible range of the spectrum.

In the as-deposited samples, the PL response for the nitride deposited on top of the SiO₂ is higher than for the layer directly deposited on the Si substrate, noticing that the interface between the SiO₂ and the different nitrides stoichiometry (SiN_x) has a contribution to the PL response, as can be seen in Fig. 1. The figure shows how the interface contribution, due to the presence of SiO₂, affects both types of nitride, having an increment of 13% and 22% for SRN and SSN, respectively. Defects associated with both nitrogen and silicon dangling bonds, with maximum emission peaks at around 410 and 520 nm, respectively, play a crucial role in the SSN PL response.^53–57 These samples show a peak located at 462 nm when SiO₂ is present and at 477 nm when it is not. For the SRN, the peaks are located at 640 nm (defects coming directly from the silicon excess) for the samples with SiO₂ and 700 nm for the sample without SiO₂.

For both nitrides, a red shift is produced when there is no SiO₂ in the stack, remarking that the defects coming from the nitrides deposited directly on the Si substrate make the silicon dangling bonds have a stronger effect in the PL response. Increasing the silicon content in the deposited film gives rise to a shift in the PL response of around 180 nm with SiO₂ in the stack and 220 nm without SiO₂.

To compare the PL spectra obtained for silicon nitride films with the two compositions after being subject to thermal treatments, Fig. 2 shows the highest and the lowest PL responses for the annealing process under nitrogen and argon atmospheres. The complete set of measurements for the SSN with all the annealing processes for the two different atmospheres can be seen in Fig. S2 in the supplementary material. In Fig. 2(a), there is clear evidence that the PL intensity is lower when the annealing is done in an argon atmosphere than in a nitrogen atmosphere, indicating that the latter seems to react with the SSN network, increasing the number of “dangling bonds” that are responsible for the PL emission. The highest PL response for argon atmosphere has a similar value as the smallest presented by the nitrogen atmosphere, making argon a better candidate for mitigating the PL in the SSN films, and making nitrogen a better candidate if the interest lies in the emission. In Fig. 2(b), a comparison of the maximum and minimum PL values for the SRN for the two atmospheres is presented. The complete set of measurements can be found in Fig. S3 in the supplementary material. In this case, for the SRN, the argon atmosphere presents the highest and lowest PL intensity values. The responses are more similar between the atmospheres compared with Fig. 2(a). In Fig. 2(b), for the SRN films, the PL intensity response is affected in a lower way by the gas atmosphere than for the SSN layer [see Fig. 2(a)], and the atmosphere alone seems not to be only responsible for the PL behavior. For the SRN, the selected annealing process has a bigger effect for argon atmosphere than it does for nitrogen. While the annealing in the former exhibits an intensity difference of around 50%, the latter shows around 17%.

Figure 3 shows a comparison of the absolute number of counts for each of the nitride compositions deposited on SiO₂ or directly deposited on the Si substrate for all the annealing processes (A, B, C, and D) or without annealing (N.A.). The PL spectra for each individual case can be found in the supplementary material in Fig. S4 for nitrogen atmosphere and S5 for argon. In Fig. 3(a), for nitrogen atmosphere annealing, the SRN is most stable over the different annealing processes, only differing between having an interface of SiO₂ or not with no annealing. For the SSN, the PL caused by the interface effect is notable for the material as deposited. However, when the annealing treatment is realized, this interface effect disappears and points out that the temperature of 1100 °C (processes B and D) decreases the PL response compared with the annealing at 950 °C (A and C).

In Fig. 3(b), for the argon atmosphere, the SRN is more stable again than the SSN. The SRN on top of the stack with SiO₂ fluctuates more in an argon atmosphere than in nitrogen. In addition, using argon, the interface effect increases compared with nitrogen. Again, for the SSN, processes B and D (1100 °C) decrease the PL response compared with A and C (950 °C). Even more, for the SSN on top of the SiO₂, the PL intensity is reduced, noticing that temperatures of 1100 °C (B and D) decrease the PL considerably, almost reaching the limit of detection of our system, independently of the annealing method. For annealing D, the PL peak is the same for the SRN and for the SSN on the Si substrate, making it unable to distinguish between the nitrides by analyzing only the number of counts of the PL peak. However, it is easy to distinguish the compositions with the peak shift between nitrides with different silicon contents.

In Fig. 4, the study of the wavelength shift between the PL peaks, $Δ λ$⁠, of the nitride compositions (SSN and SRN), with and without the interface effects, is plotted. The SRN presents, in all the cases and independently of the atmosphere and thermal annealing, a red shift in the PL response compared with the SSN. From Figs. 4(a) and 4(b), in all cases, the shift in the nitride on top of the Si substrate is bigger than using SiO₂, making clear that the excess of silicon is producing a red shift. Regarding the SiN_x deposited on top of the SiO₂ in the nitrogen atmosphere, as can be seen in the black dots of Fig. 4(a), the annealing processes B and D (1100 °C) produce a smaller shift compared with annealings A and C (950 °C), with around 15 and 25 nm differences, respectively. For both atmospheres and in the presence of SiO₂, the smallest shift is produced for process D, noticing that, in this annealing, the interface is having the biggest impact, as the maximum difference between the two stack structures presented can be found. In the presence of SiO₂, the highest shift is produced, for both atmospheres, in process C or without annealing. The rest of the annealing processes maintain an almost constant difference between the substrates, with SiO₂ and without SiO₂. For the optical stack (SiN_x + SiO₂ + Si substrate), in all cases, a shift between 140 and 190 nm is achieved, being able to tune the PL peak to mitigate or reduce the PL at the selected working wavelength.

Not only has a complete PL study been carried out in order to show the effect on propagation characteristics on waveguide structures but also the refractive index is evaluated at the wavelength of 633 nm, using the ellipsometer and model described in supplementary material, for both nitride compositions, as deposited or without annealing (N.A.), and using different thermal treatments with two different atmospheres, see Table II. In order to tune the optical properties of the mode, the refractive index difference between the SRN and the SSN (Δn) for the different thermal treatments is plotted in Fig. 5, for both nitrogen and argon atmospheres. For the nitrogen atmosphere, the annealing in processes A and C (950 °C) produces a higher refractive index shift, compared with the annealing in B and D (1100 °C), which means that the temperature plays a stronger effect than the thermal process. In all cases, the nitrogen atmosphere demonstrates an increase in the refractive index higher than in argon. In nitrogen, a shift range between 5.8% and 7.2% in the refractive index is achieved. In the case of argon, a refractive index variation between 4.8% and 6.2% is demonstrated. It has to be noted that the refractive index difference between SRN and SSN films as deposited (N.A.) is around 8.4%, demonstrating a higher difference than the variation achieved with annealing treatments.

In order to show the real potential of this platform in visible integrated photonics and in multi-layer stacking circuits, a vertical directional coupler using both SSN and SRN waveguides has been designed and simulated. As a matter of showing an example, one of the annealing treatments has been selected (annealing C, RTA at 950 °C in the nitrogen atmosphere). First, the SRN waveguide has been designed, in the standard studied photonic platform stack (SRN + SiO₂ + Si substrate), in order to satisfy the single-mode condition, obtaining a width of 350 nm and a slab of 120 nm, see Fig. 6(a). Afterward, a parametric sweep of different SSN-based waveguide dimensions has been carried out. In this case, the waveguide should satisfy two conditions: the single-mode and the phase-matching conditions.^58,59 The single-mode condition is set with a thickness of 500 nm, a width of 720 nm, and a slab of 300 nm, as can be seen in Fig. 6(b). Also, the bending losses have been simulated, obtaining a radius of 40 μm for the SRN waveguide and 140 μm for the SSN, achieving a loss per bend smaller than 0.13 dB, see Fig. 6(c). The SSN propagation losses are set to 3.6 dB/cm, corresponding to our well-studied stoichiometric material, while the SRN propagation losses are set to 4 dB/cm, as the mode has more interaction with the sidewalls.

The resulting vertical directional coupler can be seen in Fig. 7(a). For this device to work, the phase-matching condition between two asymmetrical waveguides must be satisfied, so both waveguide modes must have a similar propagation constant, i.e., a similar effective refractive index. These dimensions have also been settled in order to match the effective refractive index of the fundamental TE mode of the previous designed single-mode SRN waveguide, see Fig. 7(b). The SRN waveguide is placed at the bottom of the stack, and the SSN waveguide on top of it, with a 50 nm inter layer of SiO₂. Performing 3D simulations, the optimum vertical directional coupler is obtained for a coupling length (L_c) of 9.8 μm and a coupling efficiency of 95.8%, see Fig. 7(c).

In conclusion, the PL responses for two silicon nitride compositions, SSN and SRN, were analyzed regarding their native luminescence emission. Interface effects were studied by depositing materials on standard photonic stack (SiN_x + SiO₂ + Si substrate) or directly on the Si substrate, with different annealing processes in nitrogen or argon atmospheres. SSN showed lower PL intensity in argon compared to nitrogen, especially negligible in standard photonic stack at 1100 °C annealing. Conversely, SRN exhibited a higher refractive index and red shift in PL measurements compared to SSN, with stable PL response across annealing treatments. SSN annealed at 1100 °C showed a lower PL intensity compared to 950 °C, regardless of atmosphere. A red shift of 140–190 nm and a refractive index increase of 4.8%–7.2% was observed between SSN and SRN. Additionally, single-mode waveguides made of SSN and SRN, along with a vertical directional coupler have been designed and simulated, demonstrating the 3D integration capability with a 9.8 μm coupling length and 95.8% power coupling between waveguides. This study opens up the possibility of trying to maximize the refractive index contrast avoiding the PL in the interest region of the visible spectrum for photonic integrated circuits. These results show significant implications for chip-based optical systems using SiN, potentially leading to a substantial increase in their signal-to-noise ratio. This is particularly valuable in fields, such as chip-based Raman spectroscopy and nanoscopy, biosensors, OCT, and single photon emission, where mitigating background fluorescence noise is crucial for accurate analysis.

See the supplementary material for additional data and characterization relevant to this article and referenced in the main text, including Figs. S1–S5.

This work was funded by MICIU/AEI/10.13039/501100011033 and by the European Union NextGenerationEU/PRTR. (No. FJC2020-042823-I) and supported by the PLEC2022-009381 project from the Spanish Ministry of Science and Innovation.

The authors have no conflicts to disclose.

M. Blasco: Data curation (equal); Formal analysis (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). S. Dacunha: Data curation (supporting); Formal analysis (supporting). C. Dominguez: Conceptualization (lead); Funding acquisition (lead); Investigation (lead); Resources (lead); Writing – review & editing (supporting). J. Faneca: Conceptualization (lead); Data curation (lead); Funding acquisition (lead); Investigation (lead); Resources (lead); Supervision (lead); Writing – original draft (lead); Writing – review & editing (lead).

The data that support the findings of this study are available within the article and its supplementary material and from the corresponding author upon reasonable request.