Silicon-based photonic integrated circuits (PICs) represent a noteworthy advancement in optoelectronics and offer advantages such as high speed, low power consumption, and enhanced integration density [1–8]. These innovations are transforming the landscape of optical communication, sensing, and interconnectivity. Among the various materials employed in PICs, silicon nitride (SiN) is used extensively owing to its exceptional thermal stability, negligible two-photon absorption effects, significantly lower thermo-optic coefficient compared with silicon, and low-loss characteristics across a broad wavelength range [9–16]. SiN can be deposited via plasma-enhanced chemical vapor deposition (PECVD) at low temperatures or by low-pressure chemical vapor deposition (LPCVD) at elevated temperatures. The optical properties can be customized by adjusting the silicon-to-nitrogen (Si/N) ratio to satisfy the requirements of different PICs applications [13,17]. However, several manufacturing challenges degrade the performance of PICs, as characterized by increased insertion loss and wavelength shifts, thereby impeding their large-scale applications.

Microring resonators (MRRs), which are vital components in PICs, are utilized in various sectors, including optical communication, sensing, switching, and filtering [18–22]. The fabrication of MRRs is sensitive to minute variations in the waveguide dimensions, with discrepancies as small as 1 nm resulting in frequency shifts of approximately 100 GHz. Moreover, MRRs primarily depend on traditional active tuning methods and various post-fabrication trimming techniques to fine-tune critical parameters including the quality factor (Q-factor), extinction ratio (ER), and resonance wavelength shifts, ensuring optimal performance for applications in PICs. Among the trimming techniques employed are thermal and electrical tuning [23,24], ultraviolet (UV)-laser modification [11,25], trimming of cladding layers such as oxides and chalcogenides [26–28], and Ge or Si particle injection into the core layer [29–32]. However, these methods are associated with unstable outcomes, high energy consumption, substantial operational footprints, and increased complexity. Therefore, developing more efficient and precise trimming methods is imperative.

Femtosecond lasers offer unique advantages, including non-thermal effects, high instantaneous intensity, ultrashort pulse durations, and contactless processing, thus rendering them highly suitable for precise post-fabrication trimming [33–36]. Studies showed that femtosecond lasers can effectively restore and enhance the performance of silicon-based MRRs, thus facilitating phase correction through either the amorphization of single-crystal silicon waveguides or surface milling [34,35]. This precise manipulation of material properties allows one to fine-tune the optical characteristics of MRRs. Additionally, femtosecond laser treatment can induce microstructural changes in PECVD SiN films, thus resulting in the formation of amorphous and nanocrystalline Si nanoclusters, which significantly affects the optical properties of the films [17,37,38]. Consequently, further investigation into the modification of SiN-based photonic devices using femtosecond lasers is warranted. However, a drawback of PECVD-SiN is the presence of nitrogen–hydrogen bonds, which may adversely affect the performance of photonic devices, particularly in the conventional band (C-band). Conversely, LPCVD-SiN films offer ultralow propagation losses, primarily due to the high temperatures employed during their deposition process, which effectively dissociate hydrogen bonds. Moreover, it is imperative to investigate the microstructural evolution in LPCVD-SiN subjected to femtosecond laser treatment based on the performance advantages and application demands of LPCVD-SiN in photonic devices. Such insights could provide a viable post-fabrication trimming solution for the mass production of PICs incorporating LPCVD-SiNs.

In this study, we investigate the modification mechanisms of femtosecond laser interactions with MRRs on a 400-nm LPCVD-SiN photonics platform, thus enabling post-processing trimming in second-order MRRs and MRR-based four-channel wavelength-division multiplexing (WDM). We analyze the mode variations of 400-nm SiN waveguides at a wavelength of 1550 nm, focusing on waveguide widths ranging from 0.5 to 1.1 μm and by selecting an 800-nm transverse electric (TE) single-mode waveguide. Subsequently, we examine 10 MRRs with a radius of 50 μm placed in different positions on an 8-inch (1 inch = 2.54 cm) wafer and observe resonance wavelength shifts induced by fabrication errors. The SiN-MRRs underwent a series of laser treatments utilizing different levels of laser energy. This results in the formation of silicon nanoclusters and a redshift in the MRR resonance wavelengths. Subsequently, we analyze the effect of multiple laser modifications on the resonance wavelength shifts, ER, and Q-factor of the devices. Additionally, we apply this technique to enhance the performance of second-order MRRs and the four-channel WDM configuration. This study provides essential experimental evidence for the utilization of femtosecond lasers to correct LPCVD-SiN photonic devices and facilitates the advancement of post-fabrication trimming techniques for large-scale PICs.

In this study, we utilized the 400-nm LPCVD-SiN photonics platform to investigate the propagation conditions of various waveguide widths ($w_g$). The finite-difference method was employed to compute the relationship between $w_g$ and the effective refractive index ($n_{\mathrm{eff}}$) to facilitate TE single-mode optical propagation in the waveguides. The $n_{\mathrm{eff}}$ values for the SiN waveguide layer and ${\mathrm{SiO}}_2$ oxide layer were set to 1.99 and 1.445, respectively, at a wavelength of 1550 nm. Figure 1(a) illustrates the TE and transverse magnetic (TM) modes (${\mathrm{TE}}_0$, ${\mathrm{TM}}_0$, ${\mathrm{TE}}_1$, and ${\mathrm{TM}}_1$) corresponding to different $w_g$ values (ranging from 0.5 to 2 μm) and their respective $n_{\mathrm{eff}}$ indices. The fiber-to-waveguide coupler exhibited polarization-selective characteristics, permitting only TE-polarized light to enter the waveguide. To ensure ${\mathrm{TE}}_0$ mode transmission within the waveguide, the waveguide width was constrained to 0.5–1.1 μm, thus resulting in an 800-nm SiN waveguide. Figure 1(c) shows a cross-sectional diagram of the SiN waveguide, which highlights the effective confinement of the optical mode within the waveguide core at a width of 800 nm. Figure 1(b) illustrates the bending losses for a 90° bend as a function of bend radius, varying from 10 to 60 μm at $w_g=800\mathrm{nm}$. The minimum bending loss remained below 0.04 dB per 90° bend for larger bend radii exceeding 25 μm. Moreover, the spacing between the straight and bent waveguides affected their coupling coefficients; the waveguide spacing for single-mode transmission is shown in the inset of the figure. The scanning electron microscopy (SEM) image of the as-fabricated SiN-MRR structure is displayed in Fig. 1(d).

The sample was positioned on a micromechanical coupling platform equipped with a temperature-controlled stage to mitigate measurement errors caused by temperature fluctuations. Spectral measurements were performed using a spectrometer (LUNA OVA5100) within both C-band and long-wavelength band, as shown in Fig. 2(a). An amplified Ti:sapphire laser system (Astrella, Coherent, USA) operating at a repetition rate of 1 kHz and a central wavelength of 800 nm was used as the femtosecond laser source. The laser beam was focused onto the samples with a $50\times$ objective lens ($\mathrm{NA}=0.55$), producing a focal spot of approximately 2 μm in diameter. The laser power was precisely regulated using a pilot motorized attenuator and neutral-density filter, yielding a maximum single-pulse energy of 190 nJ and a pulse width of 40 fs. During the laser machining, the samples were positioned on a six-axis motorized processing platform with a spatial resolution of 20 nm and a scanning speed of 0.1 mm/s. The samples moved perpendicular to the polarization of the femtosecond laser pulses, resulting in the formation of a modification line, as illustrated in Fig. 2(b). Focused ion-beam milling was performed to fabricate ultrathin nanosheets from the samples, including the laser-modified areas, as shown in Fig. 2(c). Subsequently, these nanosheets were transferred onto a copper grid for transmission electron microscopy (TEM; Talos F200X, Thermo Fisher) analysis.

Prior to the trimming process, we investigated the interaction between femtosecond lasers and the LPCVD-SiN films to elucidate the mechanisms that alter the optical performance of the laser-modified SiN-based MRRs. Stoichiometric ${\mathrm{Si}}_3{\mathrm{N}}_4$ was selected as the primary SiN waveguide, as shown in Fig. 3(f). The as-fabricated SiN waveguide without laser treatment exhibits an amorphous structure, indicating its amorphous nature. The internal structures of the laser-modified SiN nanosheets were investigated using TEM, and the results are presented in Fig. 3. Magnified analyses of the laser-modified areas [Fig. 3(a)] are presented in Fig. 3(b), with emphasis on microstructural changes in regions A and B. Fast Fourier transform (FFT) pattern analysis of the laser-modified region shows several continuous rings with distinct diffraction spots, which correspond to the (111), (220), and (331) planes of crystalline silicon, thus revealing the polycrystalline nature of the SiN layer subjected to laser treatment, as shown in the inset of Fig. 3(b). The analysis of localized region A shows continuous lattice fringes with a spacing of 0.31 nm, which is consistent with the (111) crystallographic plane of silicon [Fig. 3(d)]. The clear lattice stripes suggest a stable crystal structure and high-quality crystallization. The accompanying FFT patterns show distinct diffraction points aligned with the Si $⟨110⟩$ family zone axis, thereby indicating a well-defined crystalline structure within the laser-modified area, as shown in the high-resolution TEM (HRTEM) image in Fig. 3(d). Region B [Fig. 3(e)] exhibits single-crystal silicon embedded within a substantial amount of disordered amorphous material, suggesting that the modification induced by the femtosecond laser varies with laser energy, leading to differences in internal defects and lattice disorders.

Further analysis using TEM-energy-dispersive spectroscopy (EDS) revealed the presence of Si, N, and O distributed across the laser-modified areas. Figure 3(c) shows the result of EDS mapping, which indicates a significant increase by an order of magnitude in the Si/N atomic ratio. This trend indicates that the laser processing of the SiN layer resulted in the formation of new amorphous silicon nanoclusters and their nanocrystalline counterparts. The experimental results suggest that the optical properties are primarily affected by changes in the stoichiometry of the waveguide owing to laser manipulation. The laser treatment of SiN-MRRs exemplifies this, as the formation of silicon nanoclusters increased the effective refractive index, thus resulting in a significant redshift in the resonance wavelength, as illustrated in Fig. 4.

Figure 4 shows the variation in the resonance spectra for 10 MRRs with identical design parameters situated at different locations on an 8-inch wafer. Each MRR, with a radius of 50 μm, exhibited a free spectral range (FSR) of approximately 3.93 nm. Owing to fabrication-induced variations in the dimensions of the resonators, the 10 MRRs showed a wavelength deviation of $\pm 0.63\mathrm{nm}$ around the central wavelength (which measured approximately 1550.9 nm). This variation arises primarily from size inconsistencies and minor structural deviations during the fabrication process. The structural properties of the MRRs are associated with model confinement. Consequently, alterations in the effective index can induce shifts in their resonance wavelengths, which can be mathematically represented as follows [35,36]: $$\mathrm{\Delta }{n}_{\mathrm{eff}}=\frac{\mathrm{\Delta }\lambda }{\lambda }\frac{2\pi R}{L}{n}_g,$$(1)where $\lambda$ signifies the resonance wavelength, $\mathrm{\Delta }\lambda$ is the shift in the resonance wavelength, $R$ denotes the microring radius, $n_g$ is the group index, and $L$ stands for the trimming length of the waveguide via laser irradiation.

To address the abovementioned issues, we employed a post-fabrication trimming technique to fine-tune the resonance wavelength using a femtosecond laser manipulation method. The SiN-MRRs underwent a series of laser treatments from the first to the fourth, utilizing varying levels of laser energy. To ensure the reliability of the experimental results, each group was repeated three times for accurate error assessment. Figure 4(b) illustrates the variation in resonance wavelength peak shifts corresponding to the laser treatments. At a laser energy of 1.9 nJ, the minimum fine-tuning achieved per exposure was 0.08 nm, accompanied by a vertical error bar of approximately 0.035 nm after the fourth exposure. As laser energy increased, the tuning exhibited stable consistency, with larger shifts in the resonance wavelength. This phenomenon could be attributed to the uniform microstructural modifications and the formation of monocrystalline silicon clusters induced by higher laser energy. A significant resonance wavelength redshift ($\mathrm{\Delta }\lambda$) of approximately $1.8\pm 0.025\mathrm{nm}$ was observed following the fourth exposure using a laser energy of 2.1 nJ. In subsequent experiments, we intend to optimize our processing techniques by systematically adjusting parameters such as laser energy, processing speed, processing angle, and pulse width, while integrating a real-time monitoring feedback system to enhance consistency and reliability of the laser modifications. It is noteworthy that higher laser energy or multiple laser modifications resulted in increased insertion loss for the SiN-MRRs, thereby reducing the Q-factor, as depicted in Fig. 4. This phenomenon is attributed to the formation of crystalline and amorphous silicon clusters within the SiN waveguide during laser modification, which introduced additional losses, consistent with previous reports on laser-modified PECVD-SiN materials [17,37,38].

During the second trimming of the SiN-MRRs with a laser energy of 2 nJ, an enhancement in performance was observed. The ER improved by approximately 11 dB near a wavelength of 1550 nm, and the average ER increased by approximately 11.8 dB within the C-band, as illustrated in Figs. 4(c) and 4(d). The performance of the SiN-MRRs through femtosecond laser tuning involves a delicate balance between round-trip loss and phase variation, wherein the ratio of the self-coupling coefficient ($t$) to the loss coefficient ($a$) is finely adjusted to optimize resonator coupling. The transmission spectra of an all-pass MRR are governed by the following equation [22,39]: $$T=\frac{I_{\mathrm{pass}}}{I_{\text{input}}}=\frac{a^2-2at\cos \phi +t^2}{1-2at\cos \phi +{(at)}^2},$$(2)where $a$ represents the loss coefficient, characterizing the amplitude attenuation of the optical field per round trip in the ring waveguide, while $t$ denotes the self-coupling coefficient, and $\phi$ is the round-trip phase of the microring.

The coupling mechanism of the SiN-MRRs is primarily dictated by the interaction between $t$ and $a$, both having a significant impact on the transmission spectrum characteristics. The original devices were in an overcoupled state ($a>t$), and despite the introduced losses post laser modification, they transitioned to a critically coupled state ($a=t$), enhancing SiN-MRR performance. Similar phenomena have been observed in studies involving high-energy laser modifications and multiple low-energy laser scribing of SiN-MRRs. Furthermore, post-fabrication trimming could necessitate a compromise between the extent of resonance tuning and device performance such as Q-factor and ER, alongside considerations for potential applications (e.g., modulators, filters, and optical switches). Additionally, rapid thermal annealing or localized laser annealing can serve as compensation methods [27,29,32,33], to mitigate internal defects and lattice disorders arising from nanocrystalline silicon particles formed within the laser-modified SiN layer, potentially enhancing the quality of SiN-MRRs. Minimizing femtosecond laser-induced losses in photonic devices while employing performance compensation techniques, such as laser and thermal annealing, showcases considerable potential for advancing this technology in the post-fabrication refinement of SiN-MRRs.

After evaluating the performance characteristics of laser-manipulated SiN-MRRs, we analyzed the effects of laser modifications on the MRR-based photonic devices. The structure of the second-order MRRs promoted a box-like spectral response characterized by broad channel bandwidths and steep passband edges, which are essential for the advancement of optical communication technologies. To further demonstrate the efficacy of the femtosecond laser technique for tuning LPCVD-SiN photonics, we performed post-fabrication trimming on two nominally identical second-order MRRs ($R=50\mathrm{\mu m}$), denoted as upper MRR1 and lower MRR2. Figure 5 shows the transmission spectra prior to tuning, which highlight significant wavelength mismatches and splitting due to unavoidable fabrication imperfections in both microrings. Figure 5 illustrates the variations in the transmission spectra at the through (th) and drop (dr) ports of the second-order MRRs as a result of laser treatment. As shown in Fig. 5(a), both second-order MRRs underwent laser modifications, which enabled the two resonance peaks of the second-order MRRs to be aligned successfully. Notably, minimal fine-tuning with high stability can align the resonance peaks when laser modification is applied to either one or two MRRs, warranting further investigation. Additionally, we observed a total resonance wavelength redshift of approximately 0.4 nm at the th port, an ER exceeding 20 dB, and a 3-dB bandwidth of approximately 0.58 nm at the dr port. Moreover, the upper MRR1 underwent modifications twice under the same low threshold energy, which yielded redshifts of approximately 0.42 nm at the th port, as shown in Fig. 5(b). This approach is promising for enhancing the precision and efficiency of the trimming process, thereby improving the performance and reliability of photonic devices.

Figure 6 illustrates the practical application of post-fabrication trimming MRR-based photonics on an LPCVD-SiN platform using femtosecond laser technology. The MRR radius was set to 25 μm, with a gap (G1) of 0.3 μm between the straight waveguide and microring, and a distance (G2) between the two microrings, as shown in Fig. 6(c). The responses of both the first- and second-order MRRs were analyzed, which revealed that the second-order MRRs exhibited a more distinct box-like shape, as determined using the finite-difference time-domain method [Fig. 6(a)]. Although higher-order MRRs enhance the spectral response characteristics, they introduce greater insertion losses owing to the added complexity and interactions among the microrings. Therefore, second-order MRRs are crucial for low-cost, high-integration WDM systems, which are vital for the advancement of optical communication technologies. A four-channel WDM device based on second-order MRRs [Fig. 6(c)] was designed to achieve an FSR of approximately 7.5 nm, as shown in Fig. 6(b). In this design, the radius of the first MRR was 25 μm (considering process tolerances), and each subsequent ring was larger than the previous ring in radius. This configuration ensures effective channel separation while addressing the process constraints.

Figure 6(d) shows an SEM image of the as-fabricated WDM device, with the parameters matching those of the simulation results. Light from a broadband source in the C-band was coupled to the input port, resonated at the corresponding wavelengths within each channel, and exited through the designated output ports (Ch1–Ch4). Figure 6(e) shows the transmission spectra of the as-fabricated four-channel WDM in the C-band. Notably, the waveguide widths and gap spacings of a fabricated device with an FSR of approximately 7.7 nm may deviate from the designed values, thus resulting in channel-spacing mismatches. To address these deviations, femtosecond laser tuning was used to adjust the resonance wavelength. Figure 6(e) displays the measured drop-port transmission spectra prior to any laser tuning, revealing the spacing between adjacent channels ($\mathrm{\Delta }\lambda$) to be approximately 2.35, 1.6, 2.3, and 1.45 nm, respectively. This tuning resulted in a considerable redshift and a decrease in the Q-factor of the SiN-MRRs. Thus, fine-tuning using lower femtosecond laser energies was performed only on Ch1 and Ch3 ports, which shifted the resonance peaks by approximately 0.48 and 0.4 nm, respectively. The device performance after tuning, as depicted in Fig. 6(f), resulted in a tuned channel spacing $\mathrm{\Delta }\lambda$ of $1.9\pm 0.05\mathrm{nm}$. This study demonstrates the flexibility of adjusting the resonance wavelengths of MRRs via femtosecond laser tuning, which enhances the transmission capacity and stability of four-channel WDM devices that use MRRs as fundamental components, showcasing significant potential for application in optical communication systems.

In this study, we examined the modification mechanisms involved in femtosecond laser interactions with MRRs fabricated on a 400-nm LPCVD-SiN photonic platform to address fabrication-induced variations in second-order MRRs and MRR-based four-channel WDM. Initially, we analyzed the mode variations of 400-nm SiN waveguides at a wavelength of 1550 nm, with emphasis on waveguide widths ranging from 0.5 to 1.1 μm and by designing an 800-nm TE single-mode waveguide. Subsequently, we investigated 10 MRRs positioned at various locations on an 8-inch wafer and discovered resonance wavelength shifts of $\pm 0.63\mathrm{nm}$ attributed to fabrication-induced variations. The interaction between the femtosecond lasers and LPCVD-SiN films resulted in the formation of silicon nanoclusters, which redshifted the resonance wavelengths of the MRRs. By applying varying levels of laser energy to the SiN-MRRs, we improved performance by tuning the overcoupled devices toward near-critical coupling, achieving an overall redshift of $0.4\pm 0.015\mathrm{nm}$ and enhancing the average extinction ratio by 11.8 dB in the C-band. Additionally, we demonstrated the practical application of this technique in correcting resonance mismatches in second-order MRRs. Moreover, this technique was employed to optimize the performance of the four-channel WDM configuration, which resulted in a finely tuned channel spacing $\mathrm{\Delta }\lambda$ of $1.9\pm 0.05\mathrm{nm}$. Overall, this study provides essential experimental evidence that elucidates the interaction mechanisms between lasers and LPCVD-SiN films, thus facilitating the future development of post-fabrication trimming techniques for large-scale PICs.