High-power and narrow-linewidth laser on thin-film lithium niobate enabled by photonic wire bonding Open Access

Integrated photonics shows promise for the realization of low-cost, energy-efficient, and scalable solutions for numerous applications across communications, computation, and sensing. In the past decade, the thin-film (TF) lithium niobate (LN) platform has positioned itself as a promising alternative to silicon on insulator, boasting many material properties that other platforms lack: its large electro-optic coefficient has made it an excellent host for efficient high-speed, wide-bandwidth modulators; its large second-order nonlinearity and ability to be periodically poled allow for efficient nonlinear three-wave mixing processes; and its wide transparency window allows for low-loss propagation over a wide range of wavelengths.¹ Recently, a slew of TFLN components have been demonstrated, including high-performance electro-optic modulators,^2,3 broadband supercontinuum^4,5 and Kerr comb^6,7 generation, and nonlinear frequency conversion,^8,9 confirming it as an excellent platform for large-scale and diverse photonic systems. Before fully integrated systems on TFLN can be realized, a key element to drive on-chip systems remains an outstanding challenge: a tunable, narrow-linewidth, and high-power laser that removes the need for an off-chip, bulk-optic source.

As integrated photonic systems continue to scale up, the co-integration of lasers and various TFLN devices would eliminate the large cumulative insertion losses from using bulk components, increasing the signal-to-noise ratio and resulting in more compact, energy-efficient, and higher-performance on-chip systems.¹⁰ Due to the lack of a direct bandgap in lithium niobate, different laser integration approaches have been pursued. Aside from rare-earth doped lithium niobate lasers,^11,12 the majority of TFLN laser development has been focused on integration methods with semiconductor gain materials, including heterogeneous^13,14 and hybrid integration.^15–18 However, a significant improvement is still needed to achieve the combination of high performance with scalability. Devices using heterogeneous integration approaches require complex, wafer-scale fabrication strategies and have yet to show high-power operation.¹⁹ Meanwhile, traditional hybrid strategies relying on precise, sub-wavelength alignment accuracy of components have shown promising performance but have yet to be effectively adopted at the wafer scale due to these stringent alignment requirements.²⁰ While the required performance metrics are application-specific, as a benchmark for laser performance, we use current commercial bulk-optic systems, which offer in the order of 100 mW fiber-coupled output power, with gain-wide tunability and linewidths in the sub-100-kHz range. Besides the performance metric, here “scalability” refers not only to a fabrication and integration process that allows for high-volume production, but also to the capability of straightforward integration of many separate components, from optical amplifiers and modulators to fiber arrays.

An emerging integration technology that could address scaling issues without compromising performance is photonic wire bonding (PWB), which enables three-dimensional and low-loss optical connections between different photonic chips via polymer waveguides.²¹ Importantly, the PWB formation process not only can take into account the relative position of different facets but can also be adapted to match different optical mode profiles. Therefore, it can overcome large misalignment and mode mismatch between a whole range of components within a system, including but not limited to amplifiers, waveguides, and fibers. Previous work has already heralded the strength of PWB combined with silicon photonics, showing its applicability for high confinement platforms.^22–24 However, the TFLN platform, with its many functionalities, enables PWB to tailor to a much wider range of applications.

Here, we present a photonic wire bonded, extended cavity laser that integrates indium phosphide (InP) amplifiers with a lithium niobate feedback circuit [Fig. 1(a)]. As an initial demonstration of the flexibility of the photonic wire bonding integration process, we realize upscaling in performance and integration through the PWB of a second amplifier waveguide to one cavity [Fig. 1(b)], allowing for power scaling in such lasers with a possibility of adding additional amplifiers as needed.

A major advantage of hybrid integration is the separate fabrication of optical amplifiers and photonic feedback circuits. This benefit remains for our laser, allowing the amplifiers to be fabricated in a dedicated semiconductor foundry for uncompromised performance. Similarly, our lithium niobate fabrication process needs no adaptation for the laser integration process. After the fabrication of the individual components, the integration process starts by coarsely placing and aligning the individual amplifier, TFLN, and fiber array on a common submount. The tool applying the photonic wire bonds (Vanguard Sonata 1000) offers a large write field, allowing for a wide tolerance of the optical interface alignment of ∼320 × 325 × 250 μm³ for the x-, y-, and z-directions, respectively (PWB is written along the x-direction, and z is the height). The position and angle of the optical components are detected using a confocal microscope and a machine vision algorithm. The topside of the InP is partially obscured due to a metal layer used for applying current to the amplifier, reducing the detection accuracy when compared to the TFLN waveguide and fiber core. Therefore, photonic wire bond alignment on the amplifier side typically requires a refined calibration of the photonic wire bond position using experimentally obtained transmission data. Such a calibration is required only once for a set amplifier material stack and amplifier waveguide geometry. Using the location of the facets and user input on the mode profiles at each interface, the PWB structure is consequently written in a negative photoresist using two-photon polymerization with a pulsed near-infrared (IR) laser. The PWB structure is routed to maximize the transmission, by ensuring a large bend radius (>60 μm) and using sufficiently long tapers (>100 μm per taper). After the development of the resist, for endured stability and increased transmission, all PWB interfaces are cladded with a UV-cured epoxy. A scanning electron microscope (SEM) image of a photonic wire bond to a TFLN spot size converter²⁵ before cladding is shown in Fig. 1(c). The single-pass PWB loss between the amplifier and the LN is measured at 4.1 dB, while the LN to fiber PWB loss is estimated at 2.9 dB (see the supplementary material). To our knowledge, these are the lowest photonic wire bonding losses reported so far for the thin-film lithium niobate platform interfacing with a fiber or a semiconductor amplifier.

The TFLN waveguide circuit provides the cavity extension and frequency selection in the laser using a Vernier filter consisting of two tunable, micro-ring resonators.²⁶ A full schematic of the dual-gain laser design is shown in Fig. 2(a), which depicts the Vernier filter with two sequentially coupled micro-ring resonators with a difference in the free spectral range (FSR) of ∼2 GHz. The Vernier filter allows for coarse tuning of the wavelength, while the effective length of the laser cavity, hence cavity resonance, can be fine-tuned using the thermo-optic phase shifter. Optimizing the output coupling of the laser allows for maximum extraction of optical power at a set current and Vernier filter setting; this is achieved using a tunable outcoupler. In our design, this outcoupler is realized through two cascaded tunable Mach–Zehnder interferometers (MZIs), giving full control to the splitting of feedback to the amplifier and power to the output waveguide [Fig. 2(b)]. The rings, couplers, and cavity length are actively tuned using Ti–Pt thermo-optic heaters [Figs. 2(c)–2(e)]. Separate characterization on a Mach–Zehnder test structure shows a phase shifting efficiency of about 125 mW/π and full control over the power splitting between 0% and 100%. Heaters at the ring resonators allow for shifting the resonance frequency over the full FSR (measured at around 113 GHz), with an efficiency of about 150 mW/π. Using the measured transmission spectra of ring resonators fabricated on the same chip, we infer a low propagation loss in the TFLN circuit of 5.8 dB/m. The low loss is mainly attributed to the high confinement of the optical mode and low sidewall roughness. After photonic wire bonding all optical components, the packaging is completed using gold wire bonds for the electrical connections and a Peltier element with an on-chip thermistor for temperature control (see the supplementary material).

With the laser in operation, we find a laser threshold at 100 mA and an approximately linear increase in output power with current [Fig. 3(a)]. At each current, the Vernier filter central wavelength is optimized together with the cavity phase section to obtain a maximum output power with high side mode suppression (SMSR) as measured on an optical spectrum analyzer (Yokogawa AQ6370) (see the supplementary material). To achieve the highest power measurement, different currents were applied to the amplifiers, 200 and 800 mA, at which a high, on-chip output power of 76.2 mW was achieved at single longitudinal mode operation with 51 dB side mode suppression (SMSR). At 375 mA current, the laser oscillates with the highest recorded SMSR of 61 dB with the spectrum shown in Fig. 3(b). The laser wavelength can be tuned coarsely by varying the effective length of one of the ring resonators in the Vernier filter using a thermo-optic heater. Here, the laser is widely tuned in wavelength around 1530 nm over 43.7 nm, which is beyond the 3-dB amplified spontaneous emission (ASE) bandwidth of both amplifiers (see the supplementary material). The superimposed spectra collected at different settings for the Vernier filter central wavelength are shown in Fig. 3. A high optical power, 15 mW on average, and a single mode operation with >52.7 dB side mode suppression across the entire tuning range are measured. We note that while this measurement focuses on the coarse wavelength tuning, earlier work on similar Vernier-based lasers has shown that finer steps in heater currents set the laser at any intermediate wavelength, and these lasers provide reliable mode-hop-free tuning up to about 100 GHz.^27,28

Equally important to output power and wavelength tunability is the passive long and short-term frequency stability of the laser to enable low-noise devices, necessary, for example, in microwave photonics applications.¹⁰ Here, the short term frequency stability can be characterized by measuring the intrinsic linewidth of the laser.^26,29 Due to the long, extended cavity of our laser design, we expect a low intrinsic linewidth, calculated to be in the order of 100 Hz. However, the resolution of the optical spectrum analyzer (20 pm or 2.5 GHz) used for the measurements shown in Fig. 3 is not sufficient to resolve the linewidth of the laser. To resolve such an ultra-low intrinsic linewidth, a delayed self-heterodyne setup is employed³⁰ with a 19.6 m length delay. The measured frequency noise power spectral density, computed from five 100-ms long oscilloscope (Keysight UXR1102B) traces of the heterodyne beat note,³¹ is shown in Fig. 4(a). For this measurement, the amplifiers of the laser are set to, respectively, 225 and 150 mA and the laser produces an on-chip power of 21.3 mW at a wavelength of 1527.8 nm. The frequency noise flattens at the quantum-limited noise around 2 MHz offset frequency, at a level that corresponds to an ultra-narrow intrinsic linewidth of 550 Hz. A similar linewidth, within the experimental error, is measured by recording the beat note in the frequency domain using a long delay of 20 km and performing a Voigt fit to the line shape. We note that for both cases, we are not limited by the noise floor of our measurement.

In contrast to mechanical, acoustic, and thermal alignment drifts of bulk-optic extended cavity lasers, integrated photonic circuits typically show excellent passive and long-term stability, an essential benefit for the scaling to large, fully integrated systems. Previous demonstrations of chip-based external cavity diode lasers on silicon nitride have realized a stable mode-hop-free operation of 5.7 h for stage-coupled lasers³² and up to 120 h for actively stabilized, integrated lasers.³⁰ To investigate the passive, long-term frequency stability of the laser without any active stabilization, we operate and monitor its performance continuously over multiple days. The current on both amplifiers is set just above threshold at 125 mA to minimize thermal fluctuations, corresponding to an initial on-chip output power of 2.6 mW. After the laser is turned on and has thermally stabilized over a period of about an hour, the heater conditions are finalized. During the measurement, the laser output is monitored continuously using an optical spectrum analyzer (Yokogawa AQ6370), a powermeter (Thorlabs S155C), and a high resolution wavelength meter (High Finesse WS6-200 IR, resolution of 4 MHz) to assess the spectral characteristics, optical power, and frequency drift, respectively. Following initial start-up and when thermal equilibrium is reached, the laser maintains stable and mode-hop-free over a 58-hour period [Figs. 4(b)–4(d)]. The laser emission spectrum over time is shown in a cumulative spectral plot in Fig. 4(b), with a sustained, high side mode suppression ratio of 55.5 ± 0.4 dB during the mode-hop-free operation. The on-chip optical power over time is shown in Fig. 4(c), with less than ±0.2 mW variation. Finally, Fig. 4(d) shows the frequency drift of the laser extracted from the wavemeter measurement. We note excellent passive frequency stability, with a trend in the frequency drift of 4.4 MHz/h. We hypothesize that the mode hop after 58 hours was caused by a change in ambient temperature, as mechanical perturbations and aging of the laser do not seem to affect the laser stability.

In conclusion, we demonstrate a high-power, passively stable, and ultra-narrow linewidth extended cavity laser integrated with photonic wire bonding. The maximum on-chip output power exceeds the existing TFLN extended cavity diode lasers by 1–2 orders of magnitude,^13,15 due to the power scaling of the dual amplifier design enabled by photonic wire bonding. To our knowledge, our on-chip power exceeds all single-gain ECLs across all photonic platforms, with performance comparable to or approaching that of other dual-gain ECLs demonstrated on the more mature silicon and lower-loss silicon nitride platforms.^26,33,34 Currently, our laser shows the lowest intrinsic linewidth of any on-chip laser source demonstrated using TFLN. We attribute the narrow linewidth mainly to the low-loss and long extended cavity, with an effective length of about 24 cm. The highly frequency selective Vernier design allows the laser wavelength to be tuned over twice the range of what has been reported before.¹⁵ As far as we know, this work is a first demonstration of long-term frequency stability for TFLN-based extended cavity diode lasers. Our lithium niobate laser also shows comparable or improved passive stability to similar Vernier-based lasers demonstrated using silicon nitride.^30,35

Although we find a high side mode suppression ratio at almost every current (54.3 dB on average, see the supplementary material), we found that at higher currents (>525 mA), the laser is more likely to oscillate in multi-mode, similar to what has been observed in other recent studies.³⁶ The reason for the decreased stability regime at higher powers is likely due to the coupling loss between the InP and TFLN waveguides, which could be improved upon in future iterations. Furthermore, it became apparent that the laser performance had degraded with time, seen as a lower maximum power for measurements taken after the initial 76.2 mW measurement. We attribute this to gradual moisture diffusion into the top oxide cladding, leading to increased absorption and optical loss. Currently, high-quality low pressure chemical vapor deposition (LPCVD) oxides are difficult to integrate with the lithium niobate fabrication process due to the high temperatures required for deposition. As a consequence, we resort to a lower temperature plasma enhanced chemical vapor deposition (PECVD) SiO₂. To suppress the moisture diffusion into the top oxide, an aluminum oxide sealing layer, deposited by atomic layer deposition (ALD), was used to slow the degradation process down, although it does not eliminate it fully. To assess the quality of the oxide cladding and the efficacy of the sealing layer, optical Q measurements on the TFLN chip are used to evaluate loss. Immediately after processing, the inclusion of the PECVD oxide cladding leads to a 20% reduction in the optical Q compared to the uncladded case, pointing to reduced optical quality of our current oxide. Moreover, with the ALD layer, the optical Q degrades by 15% over two weeks of aging in environment compared to >50% with no sealing layer. A long hot plate bake (>85 °C) before integration can recover any degradation of the Q but is currently unfeasible for the fully integrated device. We believe that higher optical powers could be achieved reproducibly with the implementation of higher density oxides that are compatible with the lithium niobate fabrication process.

Looking forward to higher power applications, the slope efficiency and maximum laser power can be improved significantly by increasing the PWB transmission and reducing the LN propagation loss to <3 dB/m,^37–39 which would lower the intra-cavity loss. We expect that in future demonstrations, PWB loss can be improved through further optimization of the bond alignment and improvement in the facet quality. Moreover, losses could be reduced further with the improvement to the TFLN coupler design. In further experiments, the PWB showed no degradation with up to 1 W of optical power in the bond, verifying that high intra-cavity powers could be supported. Furthermore, photonic wire bonding to tapered amplifiers could open the path toward optical powers in the Watt-level regime.

In addition to completing the TFLN platform with a tunable, narrow-linewidth, and high-power laser, our work verifies photonic wire bonding as a scalable solution for performance and integration. The power scaling through photonic wire bonding can be used to enable different nonlinear processes, while a combination with fast modulation can be useful for microwave photonic applications.¹⁰ Future work involves integration with optical isolators⁴⁰ for a fully protected laser cavity, as well as other functionalities of the TFLN platform, such as high performance electro-optic modulators,^2,41 electro-optic frequency comb sources,^42,43 and high-speed on-chip detectors.⁴⁴ 

Experimental data on the photonic wire bond transmission can be found in the supplementary material, Fig. SM-1. The amplified spontaneous emission power and spectra for the individual optical amplifiers are shown in Fig. SM-2. For all currents reported here, the spectra and side mode suppression ratio as a function of current are shown in Figs. SM-3 and SM-4, respectively. An additional picture of the packaged and photonic wire bonded laser is shown in Fig. SM-5.

The authors thank Jeremiah Jacobson for the photograph in Fig. 1(a); Donald Witt for the image in Fig. 1(c); Xinrui Zhu and Yaowen Hu for providing a characterization chip for the LN-fiber PWB loss; Allen Chu and Henry Garrett for initial PWB characterization; and Yunxiang Song, Lisa V. Winkler, Letícia Magalhães, and Amirhassan Shams-Ansari for the support during the project. Lithium niobate fabrication was performed in the Center for Nanoscale Systems at Harvard University.

The authors acknowledge the support from the Defense Advanced Research Projects Agency (DARPA) under Grant No. HR0011-20-C-0137, Office of Naval Research (ONR) under Grant No. N00014-22-C-1041, and National Science Foundation (NSF) under Grant Nos. EEC-1941583 and OMA-2137723.

The views, opinions and/or findings expressed are those of the authors and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.

V.R., J.M., and L.J. are involved in developing III–V technologies at Freedom Photonics. R.C. and M.L. are involved in developing lithium niobate technologies at HyperLight Corporation.

C.A.A.F. and R.C. contributed equally to this work.

Cornelis A. A. Franken: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Visualization (equal); Writing – original draft (equal). Rebecca Cheng: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Visualization (equal); Writing – original draft (equal). Keith Powell: Investigation (equal); Writing – review & editing (equal). Georgios Kyriazidis: Investigation (equal); Writing – review & editing (equal). Victoria Rosborough: Investigation (equal); Writing – review & editing (equal). Juergen Musolf: Investigation (equal); Writing – review & editing (equal). Maximilian Shah: Investigation (equal); Software (equal). David R. Barton III: Investigation (supporting); Writing – review & editing (equal). Gage Hills: Supervision (supporting); Writing – review & editing (equal). Leif Johansson: Funding acquisition (equal); Supervision (equal); Writing – review & editing (equal). Klaus-J. Boller: Funding acquisition (lead); Project administration (lead); Supervision (lead); Writing – original draft (supporting). Marko Lončar: Funding acquisition (lead); Project administration (lead); Supervision (lead); Writing – original draft (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.