Widely tunable semiconductor lasers have garnered significant attention due to their potential applications in optical communication, sensing, and light detection and ranging (LiDAR)^[1–3]. To achieve wide wavelength tunability, various approaches have been investigated. Compared to monolithically integrated devices, such as distributed feedback (DFB) arrays and distributed Bragg reflector (DBR) lasers^[4–8], the external cavity lasers (ECLs) based on hybrid integration^[9–15] offer the flexibility to individually optimize both active and passive components, along with having improved thermal management through chip separation. Furthermore, by incorporating low-loss passive external cavity waveguides, these lasers are capable of attaining high quality (Q-factors) wavelength filters, facilitating broadband tuning in a wide free spectral range (FSR) with superior side-mode suppression ratios (SMSRs) and narrow linewidths, even for isolator-free operating^[16].

The passive external cavity primarily comprises a mode-selective filter and a partial reflector. The mode selector, typically utilizing cascaded micro-ring resonators (MRRs) leveraging the Vernier effect, enables wide wavelength tuning ranges with high Q-factors. The partial reflector, on the other hand, provides the necessary feedback for laser oscillation, with its reflectivity significantly influencing key laser attributes like threshold current, output power, wall-plug efficiency (WPE), and linewidth^[17–19]. Conventional waveguide-based reflectors, including Bragg gratings^[8,10,12], ring reflectors^[12,14], and Sagnac-loop mirrors (SLMs)^[9,17], have fixed reflectivity determined by design parameters upon fabrication. They are also susceptible to manufacturing inconsistencies, leading to reflectivity and wavelength shifts, which can disrupt the laser’s optimal operating state. To address this issue, an effective approach is to integrate an SLM with a Mach–Zehnder interferometer (MZI) switch to realize an adjustable reflector^[15,18], allowing for flexible reflectivity adjustment by adjusting the phase difference between the MZI arms. Next, the bandwidth of the reflector emerges as another pivotal concern, especially during ultra-wide range wavelength tuning. If the reflector’s bandwidth falls short of the tuning range, frequent adjustments become necessary to ensure consistent laser performance across various wavelength channels, causing considerable inconvenience. Regrettably, adjustable reflectors incorporating an MZI switch and an SLM typically designed with traditional directional couplers (DCs) or multi-mode interferometers (MMIs)^[15,18], suffer from limited bandwidth and struggle to encompass a wide wavelength tuning range of nearly 100 nm. This significantly constrains their application in widely tunable lasers.

In this paper, we propose and demonstrate a hybrid-integrated wavelength and reflectivity tunable ECL using a cascade of double MRRs in the external cavity, coupled with a wide-bandwidth adjustable Sagnac-loop reflector (WASR) based on 800-nm-thick ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguides from a standard multi-project wafer (MPW) foundry process. The WASR employs bent DCs in the MZI-based adjustable reflectors, enabling precise control of feedback intensity by adjusting the phase difference between the MZI arms. Notably, the feedback intensity remains nearly constant across the ECL’s wavelength tuning range for a fixed electrical power applied to the MZI shifter. Experimental results showcase the ECL’s broadband wavelength tuning capability, accompanied by high SMSRs and narrow linewidths throughout the entire tuning range.

A schematic structure of the proposed hybrid integrated ECL is presented in Fig. 1(a). It is composed of a commercial reflective semiconductor optical amplifier (RSOA) chip butt-coupled with a low-loss ${\mathrm{Si}}_3{\mathrm{N}}_4$-based external cavity chip. The RSOA contains a 1-mm-long multi-quantum well (MQW) gain waveguide with an effective group index of 3.6 and a 3-dB gain spectrum bandwidth of about 80 nm. To reduce the undesired light back-reflection between the two-chip facets, an anti-reflection (AR) film is coated at the RSOA coupling facet and the waveguide is tilted at a 6° angle to the facet normal, and thus the angled facet reflectance is less than 0.01%. The other normal facet of the RSOA is coated with a 95% high reflectivity (HR) film to form one of the mirrors for the laser cavity. An SLM forms the other mirror of the laser cavity in the ${\mathrm{Si}}_3{\mathrm{N}}_4$ external cavity waveguides, which has a thickness of 800 nm and was fabricated in the standard MPW process. The standard single-mode waveguide is 1 µm in width with the propagation loss being less than 0.2 dB/cm. The ${\mathrm{Si}}_3{\mathrm{N}}_4$-based external cavity chip mainly includes a tilted spot size converter (SSC), a Vernier cascaded double micro-ring filter, and an integrated MZI-based WASR for reflective power tuning.

The waveguide width in the SSC is gradually widened to 2.5 µm in the lateral direction so that the efficiency of the mode overlap with the RSOA waveguide reaches over 95%. A phase shifter is positioned after the SSC to fine-tune the lasing longitudinal mode. A Vernier filter made of two cascaded add-drop MRRs with slightly different FSRs is inserted before the SLM to select the lasing wavelength. Based on the transfer-matrix method (TMM), the calculated transmission spectra of two individual MRRs and the round-trip reflection spectra of the cascaded double MRRs (assuming the reflectivity is 1 across the entire wavelength band) are shown in Figs. 1(b) and 1(c). The FSRs of 2.335 and 2.4 nm for ${\mathrm{MRR}}_1$ and ${\mathrm{MMR}}_2$ are designed so that the corresponding overall tuning wavelength range reaches above 85 nm due to the Vernier effect. It should be noted that the value of the power coupling coefficient (PCC) of the MRR would affect the loss and mode suppression ratio (MSR) of the Vernier filter. In this design, the PCC of the MRR is set as 0.075, configured with a symmetrical coupler^[20], and it can be seen from Fig. 2(b) that the simulated MSR is approximately 15 dB and the 3-dB bandwidth of the aligned peak is $\sim 17.6\mathrm{pm}$. Notably, the FP longitudinal mode spacing of the entire composite resonant cavity can be estimated^[12] to be about 80 pm, exceeding the 3-dB bandwidth of the aligned peak, enhancing the stability of the single-longitudinal-mode lasing in the ECL. Additionally, to decrease the effect of fabrication error, such as the roughness of the sidewalls and the fluctuation in waveguide width, the MRR is designed with a racetrack configuration^[21], and thus the straight waveguide parts, excluding the bent waveguide, are designed as a multi-mode waveguide of 1.8 µm width. The through port of the ${\mathrm{MMR}}_2$, which is connected to the phase shifter directly, is used as a monitor port (marked as AUX) to execute the coupling auxiliary between the RSOA and the ${\mathrm{Si}}_3{\mathrm{N}}_4$ chip because it is relatively less sensitive to the wavelength compared to the OUT port. The waveguides associated with the designated OUT, AUX, and UT ports (where the UT waveguides feature a U-shaped configuration specifically designed to facilitate the coupling of fiber arrays) have been tilted to prevent unwanted back reflections.

The WASR consists of an SLM and an MZI with two bent DCs, which are arranged in a point symmetrical manner to keep these two MZI arms balanced, as shown in Fig. 1(a). The reflectivity of the light can be modulated by controlling the phase difference ($\phi$) between the MZI arms. Based on the TMM, the reflectivity of the WASR is mathematically represented as $$R={|2ab[b^2(e^{i\phi }+e^{i2\phi })-a^2(1+e^{i\phi })]|}^2,$$(1)where $a$ and $b$ are the cross and bar coupling coefficients of the electric field magnitude of the bent DC (if the loss is ignored, $a^2+b^2=1$). Figure 1(d) illustrates how the reflectivity of the WASR changes with the phase shift in the MZI arm for different PCC values of the couplers and if the PPC of the coupler is 0.5, i.e., a 3-dB coupler is used, the minimal phase change of $\pi /2$ suffices to shift reflectivity from 0 to 1, indicating minimal power consumption. Figure 1(a) shows the configuration of the bent DC, where $R_1$ and $R_2$ are the radii of the inner and outer bent waveguides, respectively, and $\theta$ is the angle length of the coupling region. Both inner and outer bent waveguides have identical widths of 1 µm. According to the coupled mode theory^[22], the power splitting ratios of the bent DC are given by$$T_{\text{cross}}=\frac{k^2}{(k^2+{\delta }^2)}{\sin }^2\left(\sqrt{k^2+{\delta }^2}{L}_c\right),$$(2)$$T_{\mathrm{bar}}=1-T_{\text{cross}},$$(3)where $k$ is the coupling coefficient as determined by an overlap integral of the electric fields, $L_c=(R_1+R_2)*\theta /2$ is the length of the coupling region, and $\delta =({\beta }_1-{\beta }_2)/2$ represents the phase mismatch between the fundamental modes in the two waveguides. It is apparent that the maximum power splitting ratio is constrained by the phase mismatch $\delta$. To achieve wavelength-insensitive operation, the coefficient $k^2/(k^2+{\delta }^2)$ is selected to correspond to the desired cross-coupled ratio, and this is achieved at the peak of the sine term where the gradient is zero. In the bent DC, phase mismatch arises due to the differing radii of the two bent waveguides. By carefully selecting the bending radii $R_1$, $R_2$, and the angle length $\theta$, a splitting ratio close to 50% : 50% can be achieved. Following parameter optimization, we set $R_1=155\mathrm{\mu m}$, $R_2=156.35\mathrm{\mu m}$, and $\theta =12.6°$. The designed bent DC exhibits a power splitting ratio varying from 45% : 55% to 50% : 50% across the wavelength range of 1475–1631 nm via 3D-FDTD simulation, as depicted in Fig. 1(e). Consequently, when these bent DCs are introduced into the WASR, it can be seen from Fig. 1(f) that the back reflectivity demonstrates broadband operation exceeding 100 nm for three distinct MZI arm phase shifts as an example.

Here, we have utilized two commercial RSOA chips for butt-coupling with the ${\mathrm{Si}}_3{\mathrm{N}}_4$ chip to showcase the capabilities of the proposed ECL. Before assessing their performance, we initially measured the amplified spontaneous emission (ASE) spectrum of each RSOA chip using an optical spectrum analyzer (OSA, YOKOGAWA, AQ6370D-12) at varying injection currents, as shown in Figs. 2(a) and 2(b). These two RSOAs exhibit different central wavelengths in their ASE spectra, i.e., RSOA1 centers around 1540 nm, while RSOA2 is approximately 1620 nm. Additionally, as the injection current was incrementally increased, the 3-dB gain bandwidth expanded, accompanied by a blue shift in the central wavelength. The observed ripple patterns are attributed to the residual reflection occurring at the end of the receiving fiber. Notably, the RSOA chips were bonded on a carrier using a AuSn solder and then installed onto a high-precision six-dimensional adjustment frame. To maintain a stable operating temperature, a thermoelectric cooler (TEC) was positioned beneath the RSOA chip for efficient heat dissipation, while a nearby thermistor monitored the chip’s temperature in real time. However, given the constraints of the frame’s temperature control capabilities, the RSOA gain chip was cautiously operated with currents not exceeding 100 mA to prevent overheating and potential damage.

Given the complexities in directly assessing the reflection spectrum of the ${\mathrm{Si}}_3{\mathrm{N}}_4$ external cavity, we instead analyze the spectrum observed at the OUT port. Figure 2(c) presents the normalized transmission spectra of both the AUX and OUT ports of the fabricated ${\mathrm{Si}}_3{\mathrm{N}}_4$ external cavity chip, as measured using a commercial tunable laser source (Santec-570, with 0.5 pm of resolution) and a power meter (MPM-210 H). The AUX port reveals an ${\mathrm{MMR}}_2$ with a measured FSR of 2.407 nm. Meanwhile, the OUT-port spectra yield an MSR of approximately 13.5 dB and a total FSR of roughly 85 nm, attributable to the Vernier effect, both of which are consistent with the design. The measured 3-dB bandwidth of the aligned peak is about $\sim 18.5\mathrm{pm}$, as shown in the enlarged view of Fig. 2(d), which corresponds to a loaded Q-factor of $8.47\times {10}^4$.

As shown in Fig. 3, the proposed ECL integrates the RSOA and the ${\mathrm{Si}}_3{\mathrm{N}}_4$ PIC chip through butt-coupling. The RSOA rests on a six-dimensional adjustment stage (with 50 nm displacement precision), while the ${\mathrm{Si}}_3{\mathrm{N}}_4$ is mounted on a fixed test bench. Both components feature TECs for temperature regulation to minimize measurement errors caused by temperature variations. A lensed fiber aligned with the output port of the ${\mathrm{Si}}_3{\mathrm{N}}_4$ circuit captures the emitted light, which is then split by a 3-dB power splitter: one beam goes to an OSA, the other to an optical power meter. During the coupling, the lensed fiber is aligned with the AUX port to monitor the coupling efficiency between the RSOA and the ${\mathrm{Si}}_3{\mathrm{N}}_4$. Once the output power from the AUX port reaches its maximum, the fiber is then repositioned to the OUT port for lasing characteristic evaluation. This ensures optimal performance and accuracy in the evaluation of the hybrid ECL. The coupling efficiency between the two chips using a test waveguide reveals a loss of about 1.5 dB.

Figure 4(a) shows the experimental analysis of the ECL’s AUX port spectra under two distinct injection currents. At $I=20\mathrm{mA}$, the ASE spectrum of the RSOA is evident at the AUX port, with a dip caused by ${\mathrm{MMR}}_2$ resonance. Increasing the current to 30 mA reveals a distinct laser emission, indicating that the ECL’s laser oscillation threshold lies between 20 and 30 mA. Subsequently, the lensed fiber is repositioned to the OUT port and the injection current is then increased to 50 mA. The lasing wavelength of the laser is coarsely adjusted by the MRRs and fine-tuned by the phase shifter. Furthermore, the MZI phase shifter and the ECL’s phase shifter are tuned to maximize output power. Figure 4(b) displays the lasing spectrum at 1568.52 nm, achieved with a WASR heating power ($P_{\mathrm{WASR}}$) of 48.8 mW and a phase shifter power ($P_{\mathrm{PS}}$) of 136.5 mW, yielding a 1.5 dBm output power and an SMSR of 61.06 dB. Notably, if the subsequent butterfly packaging of the ECL is implemented, it will hold the potential to effectively address the temperature control issue of the RSOA. In that case, it becomes feasible to further boost the injection current of the SOA, which in turn leads to an increase in the output power of the ECL, thereby further enhancing its performance.

By adjusting the heater power of one MRR, the lasing wavelength changes in a step of one FSR of the other MRR. As depicted in Fig. 4(c), by tuning only ${\mathrm{MRR}}_2$ (the smaller one), we measured the superimposed lasing spectra. The thermal tuning efficiency was about 0.252 nm/mW, enabling wavelength tuning across a broad range of 81 nm, from 1518 to 1599 nm. A benefit from the broadband response of the WASR, the $P_{\mathrm{WASR}}$ was kept at 48.8 mW during measurements. Figure 4(d) shows the measured SMSR changing with the lasing wavelength, maintaining values above 55 dB across the entire tuning range.

Subsequently, we substituted RSOA1 with RSOA2 as the gain chip to evaluate its broadband wavelength and reflection tunability. Figure 5 presents the measured superimposed output lasing spectra and corresponding SMSRs of this ECL. With an injection current of 60 mA for RSOA2 and a fixed heater power of $\sim 87\mathrm{mW}$ in the WASR, a wavelength tuning range of 90 nm from 1576 to 1666 nm was achieved by tuning ${\mathrm{MRR}}_2$. As shown in Fig. 5(b), the SMSRs exceeding 50 dB were measured for all channels. The fluctuations observed in the SMSR and the power are mainly caused by the mechanical jitter of the coupling stage and parasitic reflections within the optical system. The former can be enhanced through further packaging optimization, while the latter can be mitigated by applying antireflection coating to the chip facet and incorporating isolators at the fiber receiving end.

At last, we have estimated the intrinsic linewidth of the ECL utilizing the delayed self-heterodyne method^[23,24].

A schematic of the experimental setup is shown in Fig. 6(a). The output of the ECL feeds into an unbalanced MZI constructed with two 50/50 fiber splitters, featuring a 20-m arm length difference. On the shorter arm, a fibered acousto-optic modulator (AOM) shifts the laser frequency by 40 MHz to prevent homodyne interference. A balanced photodetector (BPD) converts the interferometer output into an electrical signal which is then processed by an ALPHA250 (@Koheron) acquisition board. The unbalanced MZI exhibits a sub-coherence time delay, during which the fluctuations in laser frequency are translated into phase variations in the beat note. By analyzing the phase noise spectral density (PNSD) of the heterodyne beat note, we can discern both the frequency noise and the intrinsic laser linewidth. Figure 6(b) shows the typical frequency noise spectrum for RSOA1 and RSOA2 with a lasing wavelength of 1568.52 and 1615.8 nm, respectively. The laser frequency noise exhibits a 1/f-noise characteristic at the low frequency bands and becomes white noise at higher frequencies. The spikes observed in the frequency noise are likely to originate from either electronic sources or RF-pickup in the cables. At about 2 MHz, the white frequency noise levels off at $5807{\mathrm{Hz}}^2/\mathrm{Hz}$ for RSOA1 and $7585{\mathrm{Hz}}^2/\mathrm{Hz}$ for RSOA2. Using the relationship between the intrinsic linewidth and the white noise level $\mathrm{\Delta }v=\pi S_v^0$^[24], the calculated intrinsic linewidths are 17.5 and 23.8 kHz, respectively. Furthermore, we also measured the laser noise spectral density across most wavelength channels, yielding the corresponding intrinsic linewidths as illustrated in Fig. 6(c). Across the entire wavelength tuning range for both RSOA1 and RSOA2, the laser intrinsic linewidth ranges from 17.5 to 23.8 kHz, which is mainly due to the varied location of the laser longitudinal mode in the resonance curve of the dual MRRs and the gain spectrum of the RSOA. We believe that through advanced packaging technology to optimize heat dissipation and improved mechanical stability, the linewidth can be further compressed.

A performance comparison of our work with the previously reported ECLs is presented in Table 1. Our ECL exhibits a relatively low threshold current, while its intrinsic linewidth and thermal tuning efficiency are also quite competitive. Furthermore, our laser demonstrates a wider tuning range, and the design of its broadband adjustable reflector allows the laser to operate within such a broad tuning range without the need for frequent adjustments to the reflector to maintain a constant reflectivity across all wavelength channels.

In summary, we have proposed and experimentally demonstrated a widely tunable hybrid-integrated ECL utilizing 800-nm-thick ${\mathrm{Si}}_3{\mathrm{N}}_4$ waveguides from a standard MPW foundry process. To address challenges related to uncontrollable feedback intensity and limited bandwidth in on-chip reflectors, we incorporated a WASR with an SLM and a balanced MZI featuring bent DCs. Consequently, the feedback intensity can be precisely controlled without frequent adjustments of the reflector over a broadband tuning range, enabling a stable and efficient laser operation of the ECL. In the experiment, the fabricated ECL exhibited a wide wavelength tuning range of more than 81 and 90 nm when separately butt-coupled with two distinct RSOAs. An intrinsic linewidth superior to 23.8 kHz and the SMSRs greater than 50 dB across the entire tuning range were also measured. We believe further packaging will be highly beneficial in enhancing the output power and operational stability of the laser, as well as reducing phase noise and intensity noise. Furthermore, the demonstrated ECL can be potentially used in coherent detection systems such as optical communication networks and LiDAR.