The field of integrated nonlinear optics has grown dramatically during the past decade due to the development and commercial availability of thin-film lithium niobate [1,2]. In passive nonlinear devices, thin-film lithium niobate (TFLN) excels in efficient frequency conversion and quantum state generation from the visible to the infrared (IR) [3,4]. The strong mode confinement of single-pass, low-loss [5] nanophotonic waveguides and quasi-phase matched interactions utilizing lithium niobate’s largest second-order nonlinear optical tensor element have resulted in record-breaking efficiencies in applications such as second harmonic generation (SHG) [6,7], supercontinuum generation [8], difference frequency generation [9], parametric amplification [10], and parametric downconversion [11]. Similarly, TFLN active devices such as modulators [12], electro-optic frequency combs [13], and femtosecond pulse generators [14] show impressive performance in compact form-factors due to lithium niobate’s large electro-optic tensor elements. However, there is still significant room for lithium niobate’s use in ultraviolet (UV) photonics [15], with applications such as UV-visible spectroscopy, optogenetics, high-resolution microscopy, security banknote features, laser cooling [16], atomic clocks [17], and quantum computing [18].

Although lithium niobate has been extensively studied in the IR, and comparatively less so in the visible, UV devices have remained rare to date. The few reported lithium niobate devices for UV generation have been limited to metasurfaces [19], nanoparticles [20], and large micromachined or channel waveguides [21,22], and therefore do not take advantage of the sub-micron mode confinement and efficiency of lithium niobate in a nanophotonic platform. TFLN has yet to be well studied in the UV due to significant quasi-phase matching sensitivity to fabrication errors, the ultra-short poling periods required to overcome the high dispersion in waveguides at short wavelengths, and material and scattering loss. To date, variations in the thin film thickness of even 1 Å are sufficient to disrupt phase matching in visible SHG, limiting the effective interaction length and chip-to-chip repeatability [1,23]. UV generation should be possible up to lithium niobate’s bandgap at $\sim$330 nm (3.8 eV) [24], but an exponentially decaying Urbach absorption tail persists toward the visible due to defects in the crystal structure [25], and impurity ion (Cu$^{+}$, Fe$^{2+}$) resonances can cause additional loss [26,27]. Furthermore, losses at the waveguide sidewalls also increase at shorter wavelengths due to surface imperfection Rayleigh scattering, which scales as $\lambda ^{-4}$ [28]. In spite of these difficulties, there is much to gain by extending the spectral coverage of TFLN frequency conversion to the UV. Notably, near-IR laser diodes, which can be frequency doubled, are considerably more accessible than UV laser diodes and gas lasers [29]. Among other nanophotonic material platforms, only aluminum nitride (AlN) has been significantly investigated for waveguided second harmonic UV generation, despite AlN lacking ferroelectricity and therefore being incapable of periodic poling. The lateral polar structures used to achieve quasi-phase matching in AlN are highly scattering, resulting in much lower conversion efficiencies ($<1{\% }$) [30] compared with lithium niobate devices. Other potential UV platforms (lithium tantalate [31], BBO [32], LBGO [33]) have yet to be thoroughly explored in a thin film nanophotonic platform. Lithium niobate remains superior to these materials with its combination of low-loss waveguides, high second-order nonlinear response, ferroelectric poling for quasi-phase matching, and, for the case of thin-film lithium tantalate, commercial accessibility [34].

Here, we produce 30 $\mathrm {\mu }$W of efficient (197 $\pm$ 5%/W/cm$^{2}$) second harmonic generation of UV light (386.5 nm) with periodically poled lithium niobate (PPLN) rib waveguides [Fig. 1(a)]. The devices exhibit wavelength tunability through temperature and poling period, and are capable of UV SHG at the lowest wavelengths tested (710 nm frequency-doubled to 355 nm).

 

Fig. 1. (a) Schematic of the PPLN waveguide. (b) Mode profiles of the fundamental TE mode at the first and second harmonics. (c) Sensitivity of the SHG center wavelength to the thin film thickness as a function of the waveguide geometry for 600 nm (top) and 200 nm (bottom) thin film thicknesses; local areas that minimize the sensitivity also decrease the phase matching bandwidth. Two-dimensional sweeps of the (d) film thickness and etch depth and (e) top width and etch depth to vary the SHG center wavelength (contour lines). Note that perturbations in the contour lines are caused by mode crossings. (f) Poling period and effective refractive indices of the first and second harmonic fundamental TE modes as a function of wavelength.

Download Full Size | PDF

The waveguide geometry was designed to reduce sensitivity to the TFLN thickness without compromising the SHG bandwidth at the center wavelength of the laser used in this paper (773 nm). To quantify the phase matching sensitivity, the guided modes were simulated (Lumerical MODE) using the bulk Sellmeier coefficients of lithium niobate [35] and silicon dioxide [36] with the geometric parameters shown in Figs. 1(b) and 1(a) $60^\circ$ sidewall, which is consistent with the fabrication process. Only the fundamental quasi-transverse electric (TE) modes of the first harmonic (FH) and second harmonic (SH) [Fig. 1(b)] were considered since they access lithium niobate’s largest nonlinear tensor element (d$_{33}=25$ pm/V) [37]. Thicker waveguides (600 nm film thickness) were analyzed because they allow for reduced sensitivity to thickness errors and longer poling periods with a trade-off of reduced efficiency and increased susceptibility to slab leakage. The first-order wavelength sensitivity and SHG bandwidth were calculated using a numerical first derivative with respect to film thickness. A 600 nm film demonstrated a significant reduction in thickness sensitivity over a 200 nm film for nearly all waveguide geometry variations [Fig. 1(c)]. An etch depth of 375 nm and top width of 1.5 $\mathrm {\mu }$m [Figs. 1(d)–1(e)] were targeted to ensure operation within the flat region of the sensitivity plot. For these parameters, the wavelength sensitivity to film thickness is $\frac {{\rm d} \lambda }{{\rm d}T} =$ 0.08 nm/nm with a 3.4 pm SHG bandwidth, and the effective refractive indices ($n_{\text {eff,SH}}=2.32, n_{\text {eff,FH}}=2.10$) result in a quasi-phase matching poling period of $\Lambda =\lambda _{\text {SH}}/\Delta n_{\text {eff}}=1.8\,\mathrm {\mu } \text {m}$ [Fig. 1(f)]. The larger cross section of the 600 nm film thickness allows for this relatively long poling period, in comparison with a $1.1\,\mathrm {\mu }$m period for a 200 nm film with the same aspect ratio and wavelength.

The devices were fabricated from a 5% MgO-doped X-cut thin-film lithium niobate on insulator wafer (NANOLN), which consists of 600 nm of lithium niobate bonded to 2 $\mathrm {\mu }$m of silicon dioxide on a 0.4 mm silicon substrate. Periodically poled waveguides were fabricated following Ref. [10]. Each waveguide had a 7 mm poled length, with poling periods ranging from 1.35 to 2 $\mathrm {\mu }$m. The waveguide etch depth and sidewall angle were verified through atomic force microscopy, and the poled domain formation was measured with second harmonic microscopy [38].

The SHG from the PPLN devices was characterized using an optical setup depicted in Fig. 2(a). The output from a tunable continuous-wave (CW) single-frequency laser (Velocity TLD-6712, 765-781 nm) passed through an optical isolator and a variable neutral density filter to adjust the input power. An achromatic half-wave plate (Thorlabs AHWP10M-980) aligned the input polarization to the optical axis of the chip to maximize SHG power. The first harmonic was coupled to the waveguide using an aspheric lens (Thorlabs C140TMD-B). The waveguide output was collimated by another aspheric lens (Thorlabs C140TMD-A) and collected by a high-OH multimode fiber (Thorlabs M122L01). The device output was monitored using an optical spectrum analyzer (Yokogawa AQ6374) with a passband bandwidth of 5 nm around the second harmonic center wavelength to remove the residual first harmonic.

 

Fig. 2. (a) Optical setup for on-chip SHG characterization: ND, variable neutral density filter; HWP, half-wave plate; L, aspheric lens; MMF, multi-mode fiber; OSA, optical spectrum analyzer. (b) On-chip SHG output power measured as a function of the on-chip input power with experimental fit to calculate the efficiency. (c) Temperature tuning of the SHG center wavelength, measured as the weighted average of the spectrum with standard deviations and experimental fit. (d) Normalized spectra of the poling period swept from 1.35 to 1.75 $\mathrm {\mu }$m in 50 nm increments. (e) Second harmonic microscopy image of the periodically poled region.

Download Full Size | PDF

The off-chip power of the first harmonic was varied from 6 to 30 mW at 773.1 nm, and the resulting on-chip SHG power is plotted in Fig. 2(b). We achieved a maximum on-chip UV power of 30 $\mathrm {\mu }$W, and the slope of the least squares fit of the SHG output gives a power- and geometry-normalized efficiency of 197 $\pm$ 5%/W/cm$^{2}$ and power-normalized efficiency of 400 $\pm$ 10%/W. The power law scaling suggests that photorefraction and pump depletion are not significant at the maximum power tested. The calculated efficiency and on-chip powers take the coupling losses into account, which were 22% transmission per facet for the first harmonic and 10% transmission per facet for the second harmonic. These values were determined by measuring the overall transmission at 780 nm and 405 nm. Assuming that the input and output coupling losses are equal, this would result in an approximate propagation loss of 7.6 dB/cm at 405 nm.

The temperature and poling period dependence of the phase matched wavelength are plotted in Figs. 2(c) and 2(d), respectively. To measure the temperature dependence, a nonlinear crystal oven (HC Photonics TC038-PC) heated the PPLN waveguides from 21 to $100^\circ$C. The least squares slope (34 $\pm$ 2 pm/$^\circ$C) matches the theoretically calculated value of 32 pm$/^\circ$C. By varying the poling period, we also obtained SHG spectra extending down to 355 nm. A tunable CW Ti:Sapphire oscillator (Spectra-Physics Tsunami, 700-1100 nm) was used in a similar scheme to Fig. 2(a) to measure the SHG from several PPLN waveguides with poling periods as short as 1.35 $\mathrm {\mu }$m [Fig. 2(d)]. The fitted slope of the SHG center wavelength with the poling period (79.6 $\pm$ 0.2 nm/$\mathrm {\mu }$m) is in close agreement with the theoretical slope of 81.1 nm/$\mathrm {\mu }$m. This agreement in both the temperature and poling period demonstrate that the given Sellmeier coefficients [35] predict the temperature dependence and group velocity mismatch of lithium niobate in the UV relatively well, despite the limited refractive index data at shorter wavelengths. Unlike the initial measurements using the single-frequency laser, accurate efficiency data could not be extracted because the Ti:Sapphire oscillator linewidth (0.3 nm) is orders of magnitude larger than the phase matching bandwidth (3.4 pm). However, these spectra demonstrate that SHG is possible even closer to the 315 nm bandgap of lithium niobate than what has previously been demonstrated, and that the thin film lithium niobate platform is able to phase match the full range of a Ti:Sapphire laser.

Although the SHG wavelength tunability agrees well with theory, the experimental efficiency of 197 $\pm$ 5%/W/cm$^{2}$ measured from the single-frequency laser at 773.1 nm is lower than the calculated ideal theoretical efficiency [7] of 18 100%/W/cm$^{2}$. An immediate explanation is the poling quality of this device, which exhibited significant domain widening due to the short poling period. The duty cycle is estimated to be 90% from second harmonic microscopy images [Fig. 2(e)], which lowers the theoretical efficiency to 1810%/W/cm$^{2}$. Additional discrepancies can be explained by lateral leakage from the waveguide mode to the slab mode, second harmonic propagation loss, and asymmetry in the input and output coupling efficiency.

The experimental SHG spectrum [Fig. 3(a)] was measured by sweeping the first harmonic wavelength from 771 to 775 nm at a constant 30 mW input power. The spectrum exhibits multiple peaks over a 0.5 nm bandwidth, deviating from the theoretical sinc$^{2}$ line shape and 3.4 pm full-width at half maximum. The discrepancies in the efficiency, bandwidth, and spectral shape are likely caused by multimode effects, index variations, or poling variations. Although the waveguides in this work support multiple modes due to the large cross sections, the fundamental modes of the first and second harmonic are solely responsible for the phase matching of the UV generation due to large momentum mismatches or poor modal overlap of the higher-order modes. Furthermore, collecting the second harmonic with a single-mode fiber (Thorlabs P1-305A-FC-1) does not change the shape of the transfer function, which is expected to occur if higher-order modes were present. Potential sources of index changes are thickness variations, thermal gradients, induced absorption, or photorefraction, while poling defects could be caused by stitching errors in the lithography process. Without a straightforward method of distinguishing these possibilities, we chose to focus on index variations and assume that the poling is uniform. Index variations preserve the integral of the SHG transfer function, which allows us to determine the contribution of the index variations to the experimentally lower efficiency. An integral of the spectral efficiency yields an area of 31.8%/W/cm$^{2}\cdot$nm, and the transfer function with the same area and no index variations has a peak efficiency of 861%/W/cm$^{2}$.

 

Fig. 3. (a) Experimental and index variation-fitted [Eq. (1)] SHG spectra. (b) Phase mismatch error represented by Eq. (2) and corresponding thickness error against propagation distance.

Download Full Size | PDF

The magnitude and position of the index variations cannot be directly calculated without knowledge of the SHG phase relative to the first harmonic. However, the magnitude of the index variations can be estimated by approximating the phase mismatch error as a cubic polynomial to fit the SHG spectrum, given by the following expression for $\eta (\omega )$ [39], which assumes a 50% duty cycle and constant poling period:

In conclusion, we have demonstrated temperature-tunable UV light generation in an integrated thin-film lithium niobate waveguide. We have measured an SHG efficiency of 197 $\pm$ 5%/W/cm$^{2}$, with discrepancies from theory explained by the poling duty cycle and index variations. As of this publication, this is the shortest wavelength (355–386 nm) produced through second harmonic generation with CW-pumped periodically poled thin-film lithium niobate waveguides. Our work opens up opportunities to realize efficient frequency-doubled chip-scale ultraviolet laser diodes for UV integrated photonics, with applications spanning spectroscopy, atomic physics, and quantum science.