Principle concept and structural design
Figure 1 schematically illustrates the proposed waveguide structure for achieving UV to MIR (UV–MIR) SCG using a C-band pump laser. As shown in Fig. 1, the structure consists of three key sections (I, II, III) with varying waveguide widths and poling periods. Section I is for cascaded χ² SCG to achieve near-infrared (NIR) SCG, section II is for up-conversion to generate UV and visible light, while section III is for down-conversion to generate the MIR light. Here, in section I, efficient cascaded χ² SCG is first achieved for NIR SCG by using a uniform PPLT waveguide, whose poling period is designed for phase-matched SHG of the pump wavelengths. Meanwhile, near-zero anomalous dispersion and small group velocity mismatch (GVM) in NIR are also required. The up-conversion waveguide in section II is based on PPLT with spatially linearly decreasing poling periods, allowing the phase-matching wavelengths to vary from the visible to UV band along the propagation direction. In this section, SHG and SFG are the key mechanisms for expanding the spectra into the visible and UV ranges. In section III for down-conversion, the poling periods increase linearly to enable wide-band phase-matched DFG, further broadening the spectrum into the MIR region. Particularly, the width (W2) of the waveguide in section III is broadened to improve the mode confinement for the long wavelength in MIR, as analyzed in details in the supplementary material. On the other hand, for UV-visible light, such a broadened waveguide greatly reduces the power density, and thus the wavelength conversion efficiency is weakened so that the energy in the original short wavelength bands is preserved. As a consequence, the proposed waveguide structure comprising sections I/II/III facilitates the SCG from UV to MIR. Since the waveguide features a slowly varying structure with a fixed propagation direction, and its operation relies on the TE₀ mode only, the birefringence of LT does not limit the bandwidth of the generated supercontinuum.
Fig. 1: Schematic illustration of the proposed device.
The alternative text for this image may have been generated using AI.
The structure consists of three cascaded components with varying waveguide widths and poling periods: a uniformly poled, dispersive-engineered waveguide for SCG; an up-conversion waveguide for UV and visible light generation; and a down-conversion waveguide for MIR generation
Here, a LT-on-insulator (LTOI) wafer with a 600-nm-thick LT core layer and a 4.7-μm-thick buried-oxide layer is used, and the etch depth is chosen to be 300 nm. Figure 2a–c shows calculation results of the phase mismatch, dispersion, and group velocity of LTOI photonic waveguides with different widths of W = 500 nm, 1300 nm, 2100 nm, and 3000 nm, respectively. As observed, the waveguides exhibit anomalous dispersion at the pump wavelength of 1550 nm when the width is chosen as 1300 nm or 2100 nm. In particular, the 1300-nm-wide waveguide exhibits the smallest GVM of 1.027×10-10 s/m between the pump and SHG lights, resulting in enhanced effective self-phase modulation (SPM) due to cascaded second-order nonlinearity41. Moreover, smaller mode area contributes to higher effective nonlinear coefficients. As a consequence, we choose the width W1 as 1300 nm for the uniform PPLT waveguide in section I in our design. Here the etching depth was not specifically optimized, as they are designed to be compatible with our fabrication process. Dispersion calculations for different etching depths are included in supplementary material.
Fig. 2: The simulation of the proposed device.
The alternative text for this image may have been generated using AI.
a–c The phase mismatch, dispersion and group velocity of LT-on-insulator waveguides at different widths of W = 500, 1300, 2100 and 3000 nm, respectively. d On left bottom is the time evolution on the PPLT waveguides, and in the center is the frequency evolution on the PPLT waveguides. On the right is the poling period as a function of the propagation distance of this structure. In all the simulation it is slab waveguide with Height=600 nm and etch depth=300 nm
Figure 2d shows the simulated time- and frequency-domain evolution along the PPLT waveguide with the following parameters: W1 = 1300 nm, W2 = 3000 nm, L1 = 4 mm, L2 = 3 mm and L3 = 4 mm. In this simulation, we use100-fs pulse pump at 1550 nm. Considering both the final bandwidth performance of the SC source and the cost of the pump femtosecond laser, we choose a on-chip pulse energy of 100 pJ, which corresponds to an on-chip average power of 10 mW (see more details in supplementary material). Initially, the poling period is set to be 4.13 μm. The incident pulses undergo strong temporal compression, followed by soliton fission occurring at the propagation distance of z = ~2 mm, which leads to significant spectral broadening. SHG and THG are also observed around the wavelengths of 775 nm and 515 nm, respectively. At the propagation distance of z = 4 mm, the energy in the visible, especially in the UV range is still weak, mainly due to strong SPM phase mismatch occurring at these wavelengths. After propagating through the chirped PPLT waveguide for up-conversion in the distance from z = 4 mm to z = 7 mm, where the poling period decreases from 4.13 μm to 0.95 μm, the χ² phase-matching condition is sequentially satisfied for light in the wavelength range from NIR to the visible and eventually to UV. As shown in Fig. 2d, the short-wavelength edge even reaches 270 nm, covering the entire UVA/UVB bands and extending to the UVC band, indicating that the on-chip fifth harmonic generation (5th HG) of the C-band pump is achieved. Finally, for the down-conversion PPLT waveguide of section III (i.e., 7 mm < z < 10 mm), the poling period increases from 4.13 μm to 10 μm to satisfy the QPM condition for DFG, resulting in further spectral expansion and enhancement in the MIR region. As a result, the achieved SCG spectrum spans from 270 nm (UVC) to > 4000 nm (MIR).
Fabrication and measurement
The designed PPLT waveguide was fabricated using regular lithography and etching processes without any special requirement. Here our design introduces a gradual transition in the poling period from 0.95 µm to 10 µm, which presents some significant challenge for ferroelectric domains, particularly for submicron ferroelectric domains, which is a long-standing difficulty for poling techniques. To improve the poling quality, we employ three strategies. First, experimental observations revealed that, for a fixed electrode duty cycle, smaller periods yield wider inverted domains especially for submicron periods. To compensate for this, the duty cycle of the poling electrodes is increased from 10% to 50% for the case when the poling period increases from 0.95 µm to 4 µm, and maintained at 50% when the period is larger than 4 µm. Second, since the waveguide comprises three cascaded sections with different poling periods, the electrodes are divided into three independently patterned segments. This allows the fine-tuning of poling parameters for each section, thereby improving the poling quality separately. Finally, due to the non-uniformity of nucleation across the wafer, poling defects may occur in certain regions. To mitigate this, a high-voltage pre-poling pulse (700 V) opposite in polarity to the main poling voltage is applied prior to the formal poling process, which is found to greatly reduce poling defects. Figure 3a shows the false-color Piezoresponse Force Microscopy (PFM) image of the fabricated PPLT, demonstrating that the poling is uniform and exhibits an appropriate duty cycle, even at a submicron poling period of 0.95 μm. After poling, waveguides are defined using the electron-beam lithography (EBL) and etched using Ar⁺ inductively coupled plasma (ICP) etching of LT. Compared with wet etching, Ar⁺ dry etching often exhibits a much weaker dependence on the ferroelectric domain orientation31. The Scanning Electron Microscope (SEM) image of the fabricated waveguides with a poling period of approximately 1 µm is shown in Fig. 3b. As observed, the LT waveguide has high etching quality and uniform core-width, and there are no periodic surface features induced by domain polarization.
Fig. 3: UVC to MIR SCG.
The alternative text for this image may have been generated using AI.
a False-color PFM image of the PPLT with a poling period of 0.95 µm. b SEM image of fabricated waveguide. c Photograph of a glowing PPLT chip under test. d, e Measured and simulated SCG output spectrum from a chirped PPLT waveguide and unpoled waveguide. SHG, third-harmonic generation (THG), and fourth-harmonic generation (4th HG), are labeled in distinct colors. The green and pink traces represent measurements using optical spectrum analyzers AQ6373 (350–1200 nm) and AQ6375 (1200–2400 nm) from YOKOGAWA, respectively. f Measured UV SCG output from the chirped PPLT waveguide after the rotary fiber filter. g Measured SCG spectra after additional filtering by a 351-nm bandpass filter, a 343-nm bandpass filter, and a 310-nm short-pass filter. The corresponding fifth harmonic generation (5th HG) wavelengths of the 1550-nm pump are also indicated. All experiments were conducted using 1550-nm pump pulses with 100 MHz repetition rate and 100 fs pulse duration
In our experiment, we use a 100-fs pulse pump at 1550 nm with a pulse energy of 100 pJ on the chip, and the in-waveguide peak power is 1000 W. It should be noticed that LT exhibits a significantly higher optical damage threshold (1064 nm, 240 MW/cm²) compared to LN (1064 nm, 120 MW/cm²). Due to the insufficient power of the femtosecond laser available in our lab, we were unable to experimentally test the damage threshold for the PPLT waveguide. On the other hand, we found that the LT waveguide could safely handle the continuous-wave (CW) light at 1550 nm without damage even when the on-chip power is as high as 30 dBm (1 W). The pump pulses are butt-coupled into the waveguide using a lensed fiber, resulting in a coupling loss of approximately 10 dB for the pulse pump. We replaced the original output fiber (PM1550, approximately 1 m in length) of the femtosecond laser with a lensed fiber of the same type and length to eliminate the influence of the coupling fiber on the pulse parameters. The propagation loss of the fabricated 1.3-μm-wide LT photonic waveguides were measured with the cut-back method to be ~0.5 dB/cm at 1550 nm, ~0.6 dB/cm and 1064 nm and ~0.8 dB/cm at 400 nm. The output light was collected using a polarization-maintaining (PM) lensed fiber optimized for 1550 nm and directed to two optical spectrum analyzers including AQ6373 YOKOGAWA (350–1200 nm) and AQ6375 YOKOGAWA (1200–2400 nm). The resolution of spectrometer is set to be 2 nm. Figure 3c shows the photograph of a glowing PPLT chip under test, and Fig. 3d displays the measured and simulated spectra output from the chirped PPLT waveguides, covering a broad range from 350 nm to 2400 nm. The results show that the chirped PPLT waveguide enables continuous supercontinuum spectrum with a spectral bandwidth extending from below 350 nm to beyond 2400 nm. The measured total output power is -0.88 dBm. The measured conversion efficiency from the near-infrared pump to the ultraviolet–visible region of 350–760 nm is 19.3%. In the simulation, we modified our model by introducing a small phase mismatch Δβ = 7.3 (mm-1) in section I by considering the mismatch arising from the fabrication imperfection in the PPLT waveguide, such as variations in the film thickness, the waveguide width, and the etching depth. A detailed discussion of the impact of this phase mismatch on the supercontinuum generation results is provided in the Supplementary Materials. After incorporating this correction, the simulation results show good agreement with the experimental one. Note that such fabrication-induced mismatches can be effectively compensated through adaptive poling techniques42. For comparison, we also measured and simulated the spectrum generated with an unpoled LT waveguide at the same pump condition, as shown in Fig. 3e. It can be seen that the produced continuous spectrum ranges from ~510 nm to 2230 nm, which is considerably narrower and more irregular compared to the present SCG from the chirped PPLT waveguide. Furthermore, Fig. 3e also shows that the third-harmonic generation (THG) and fourth-harmonic generation signals are below the detection limit (i.e., ~ -70 dBm/nm) of the commercial spectrometer. Basically speaking, the lack of phase matching causes low conversion efficiency in the unpoled LT waveguide. In contrast, the chirped PPLT waveguides exhibit a much broader and flatter supercontinuum spectrum than the unpoled counterparts, particularly in the visible, ultraviolet, and MIR regions. Our laboratory currently lacks an optical spectrometer for the wavelength range beyond 2400 nm; therefore, the chip performance in this spectral region remains uncharacterized. In Supplementary Materials, we present the numerical simulation results to compare the complete waveguide comprising Sections I–III and a truncated waveguide only with Sections I–II. It is shown that the complete waveguide exhibits a significant enhancement relative to the truncated waveguide within the 3300–3800 nm wavelength range, confirming that Section III plays a key role for facilitating the down-conversion process to achieve spectral broadening in the long wavelength-band. And MIR via femtosecond-laser-driven DFG has already been experimentally demonstrated and theoretically analyzed in PPLN crystals43. The measured SNR at 750 nm is ≥37.6 dB. One should note that the SNR measurement is limited by the PD sensitivity rather than the SC waveguide itself. A higher-power pump source is required for measuring the true SNR of the chirped PPLT, but it is currently unavailable in our lab. Nevertheless, the SNR higher than 37.6 dB is usually sufficient for most biological and chemical sensing applications. Previous studies on PPLN have reported that carrier–envelope offset frequency (fceo) detection, also confirming the performance excellence of SNR in the spectral range from ultraviolet to visible30. And it should be noticed that the noise-generating processes associated with such massive spectral broadening involving both χ (2) and χ(3) nonlinearities remain an important topic for future investigation.
For UV spectral measurements, we used a UV single-mode fiber (SM300-SC) from FIBERCORE for output collection. A primary challenge is the crosstalk from longer wavelengths because the stray light in the visible and NIR ranges may be mistakenly detected as UV if not blocked. However, currently there is no bandpass filter available to transmit the UV light but block visible/NIR light. To address this, we wrapped the collection fiber around a cylinder with a diameter of 2 cm to induce a high bending loss for wavelengths longer than the fiber’s cutoff (430 nm), effectively functioning as a short-pass filter. Such a fiber filter has a transmission dropped to <0.001 at the wavelength of 500 nm in experiment, which is beyond the measurement limit of a high-performance CCD (ILX511), achieving high extinction ratios comparable to or better than many commercial spectral filters. The filtered spectrum is then analyzed using an Ocean Optics USB 2000+ optical spectrum analyzer. Figure 3f presents the measured output spectrum of the chirped PPLT waveguides after the fiber filter, demonstrating that continuous SCG extending beyond 270 nm is achieved, thus covering both the UVB and UVA bands. To further validate these results, we measured the spectrum after the fiber filter by introducing additional three separate optical filters, i.e., a 351 nm bandpass filter (FBH351-10, FWHM 10 nm, Thorlabs), a 343 nm bandpass filter (FBH343-10, FWHM 10 nm, Thorlabs), and a 310 nm short-pass filter (XUV0310, Asahi Spectra). These measurements confirm the presence of spectral components near 351 nm, 343 nm, and 300 nm, as shown in Fig. 3g, providing strong evidence that the supercontinuum (SC) spectrum extends to the fifth harmonic generation ( ~ 310 nm). Note that the shortest wavelength detected after filtering is approximately 270 nm, which aligns well with our simulation results, while the UV bandwidth limitation is likely due to the LT absorption which increases rapidly for wavelengths below 280 nm.
UV to infrared (IR) absorption spectroscopy
The supercontinuum is a promising broadband light source for optical absorption spectroscopy. Conventionally, a fiber-based supercontinuum source is often used in the spectroscopy system together with the cuvettes for sample absorption and a spectrometer for detection44,45. Since it’s important for the sensing waveguide to be compact and low-loss across a wide range of wavelengths, we introduce a compact and wide-band waveguide spiral on the chip to work as the sensing element. Compared to LN, whose anisotropy is as high as -0.0735, LT offers more than 10-fold reduced birefringence ( ~ 0.004)35, making it a more suitable platform for ultra-low-loss, densely integrated spiral waveguides for enhanced light–matter interaction. In the supplementary material, we compare the effective refractive indices of bending LN and LT photonic waveguides. It can be seen that the TE0 mode in the LN photonic waveguide exhibits notable mode hybridizations with the TM1 mode at around 900 nm and with the TM0 mode at 1350 nm when the bending radius is 100 µm. In contrast, there is no mode hybridization found for the LT photonic waveguide even at a small bending radius of 60 µm.
In this study, we present an on-chip UV-to-IR spectral sensing platform that integrates a chirped poled PPLT supercontinuum source with a sensing waveguide on the same chip. The platform enables sensing and spectroscopy of various liquid and gas samples across the UV, visible, and IR wavelength-bands. Figure 4a illustrates the experimental setup for the on-chip UV-to-IR spectral sensing. The pump pulse is butt-coupled into the PPLT waveguide, generating a broad-band SCG light. The following light-matter interaction occurs in the 2-cm long waveguide spiral, while the transmitted spectrum is measured using an optical spectrum analyzer, with data processing performed by a computer. Figure 4b shows a photograph of the on-chip UV-to-IR spectral sensing chip under test, while Fig. 4c presents an image of the spiral sensing waveguide on LTOI.
Fig. 4: UV-to-IR spectral sensing using the present chirp PPLT supercontinuum light source.
The alternative text for this image may have been generated using AI.
a Experimental setup for on-chip spectral sensing. b Photograph of the on-chip UV-to-IR spectral sensing chip under test. The scattered light from the on-chip supercontinuum sources is clearly visible, and the droplet on top of the chip is the test sample. c Microscope images of the fabricated waveguide spiral for sensing. d Measured UV-visible transmission spectra of air and aqueous methylene blue solutions with the concentrations of 0.75 mg/L and 1 mg/L. e UV-visible absorbance spectra of methylene blue solutions at different concentrations, obtained by normalizing each sample spectrum to the spectrum measured with no solution on the waveguide spiral. f Measured UV transmission spectra of cuvettes containing methylene blue solutions at concentrations of 0 (empty), 1, 2, and 10 mg/L. g Measured transmitted IR spectrum with or without the HCN gas cells, respectively. The insets display the wavelength differences of the measured absorption peaks of HCN relative to the HITRAN database.48 h Measured transmitted IR spectrum with or without the HF gas cells, respectively. The insets display the wavelength differences of the measured absorption peaks of HF relative to the HITRAN database48
First, we investigate the absorption spectroscopy of methylene blue (C₁₆H₁₈N₃ClS) aqueous solutions with different concentrations. As it is well known, methylene blue is widely used for staining tissues and bodily fluids before or during surgery and diagnostic procedures, as well as serving as an antiseptic, an internal wound-healing agent, and a dye in microscopic analysis46,47. Thus, measuring methylene blue concentration is highly relevant in the applications such as activated carbon performance evaluation and concentration analysis of polymeric amino acid compounds. As shown in Fig. 4d, we measured the transmitted spectra in the UV-visible-IR range for air, and methylene blue solutions with the concentration of 0.75 and 1 mg/L, respectively. It can be seen that, when the concentration increases, the absorption increases as well. Figure 4e presents the normalized absorption spectra with respective to the measured spectrum for the case without samples. A strong absorption peak appears around 660 nm, corresponding to a commonly used spectral signature for methylene blue detection. We also conducted UV-absorption measurements for the light from the PPLT waveguide, passed through a 2 cm-long cuvette containing either methylene blue solution or air, followed by a 310 nm short-pass filter and, finally, an optical spectrum analyzer. The results are shown in Fig. 4f. As it can be seen, the absorption increases as the concentration of methylene blue increases. As a consequence, our chirped poled PPLT supercontinuum source is capable of UV to visible liquid detection with both supercontinuum source and sensing spiral on chip.
Moreover, we also performed IR absorption spectroscopy for H13C14N (25 Torr, 55 mm cell path length, Technicasa) and HF (50 torr, 27 mm cell path length, Wavelength references) using our PPLT supercontinuum source. The light from the PPLT waveguide in the spectroscopy module is fiber-coupled into the gas cells and the transmitted spectrum are sent to an OSA for analysis. Figure 4g, h shows the measured transmitted IR spectra with and without the HCN and HF gas cells, respectively. The insets display the wavelength differences of the measured absorption peaks of HCN and HF relative to the HITRAN database48. It shows that the mean wavelength differences are −0.0329 nm and 0.045 nm, respectively, and the corresponding variances are 0.00040 and 0.00053. During the experiment, coupling between the optical fiber and the chip exhibited certain instability, which was a major source of error in gas absorption measurements. Currently, the measurement accuracy is primarily limited by the precision of the commercial OSA used (MS9740B, Anritsu). In the future, the implementation of an on-chip spectrometer based on high-Q micro-resonators49 on the same chip could significantly improve measurement accuracy and reduce coupling instability.