Ultraviolet (UV) pulse sources are crucial in various applications, including medical diagnostics and treatments [1–3], as well as materials processing [4,5]. Common methods for generating UV pulses involve stimulated emission from excimers in their excited state, i.e., excimer lasers, and frequency upconversion of near-infrared pulses in nonlinear crystals. However, excimer lasers typically produce pulse widths only in the nanosecond range or longer [6], and the peak power of frequency upconversion is often limited by the inherent properties of crystal materials, such as laser-induced damage and thermal dephasing [7].

Recently, UV pulse generation via dispersive wave (DW) emission in gas-filled hollow-core fibers has attracted significant interest. This technique exploits the capability of hollow-core fibers to guide UV light [8] and their unique dispersion properties [9], enabling the conversion of energy from a longer pump wavelength to a UV wavelength [10]. This solarization-free, fiber-based UV source provides pulse durations as short as a few femtoseconds [11] and offers wavelength tunability [12].

Conversion efficiency is a crucial factor for the practical deployment of this new technique. Although a record-high UV conversion efficiency of 15% was achieved in a gas-filled large core capillary [13], the weak light guidance in the capillary necessitated keeping the few-meter-long waveguide straight. Additionally, the large core requires intense pump pulses with single-pulse energy in the hundreds of microjoules range. These requirements present significant challenges in many application scenarios with constraints on pump energy and space.

In contrast, tubular-type anti-resonant hollow core fibers (AR-HCFs)—first introduced in 2011 [14]—offer strong light guidance due to their microstructured cladding, which inhibits coupling between modes in the hollow core and cladding regions, thus preventing light leakage from the core [15–18]. This makes AR-HCFs a more viable alternative for generating UV pulses in a compact system, because they require only a few centimeters of fiber and pump pulse energy as low as tens of nanojoules [19,20].

However, demonstrations of UV pulse generation using AR-HCFs typically report much lower conversion efficiencies, ranging from 1.2% to 6% [19–21]. We attribute these poor yields mainly to the presence of structural resonances, i.e., resonant coupling between the fundamental core modes in the hollow core and those in the dielectric cladding walls, which cause light leakage from the guided modes of the core [22]. In addition, the presence of resonances limits the extent of spectral broadening, which in turn affects nonlinear optical processes, such as DW generation and pulse compression. These processes can be significantly enhanced by eliminating the resonances. For example, an AR-HCF with a cladding wall thickness of 0.5 μm was able to compress 2 μm pulses down to just 0.4 optical cycles, thanks to uninterrupted nonlinear spectral broadening made possible by the absence of resonances around 2 μm [23].

In this study, we present numerical and experimental investigations on the impact of these resonances on UV conversion efficiency. Furthermore, to the best of our knowledge, we demonstrate, for the first time, the generation of UV pulses in a resonance-free AR-HCF, achieving a conversion efficiency as high as 12%. We show how wet-etching AR-HCFs can shift structural resonances away from the spectral region of interest, significantly enhancing UV conversion and enabling energy scalability with increased nonlinearity.

To investigate the impact of resonances on UV emission in a gas-filled AR-HCF system, we conducted an experimental study, followed by a comparative numerical analysis.

The AR-HCF used in the experiment has a core diameter of 31 μm and cladding tubes that have a glass wall thickness of 470 nm. The transmission spectrum of the AR-HCF obtained using a xenon lamp, and the scanning electron micrograph of its cross section are shown in Figs. 1(a) and 1(b). The 470 nm cladding wall results in the first resonant band appearing at 300–350 THz, followed by bands at 580–630 THz, 860–910 THz, and 1110–1160 THz, corresponding to the second, third, and fourth resonant bands, respectively [24]. To generate DW in the UV range of 900–1100 THz, the 17-cm-long, 14-bar argon-filled AR-HCF is pumped by 37 fs pulses at an 800 nm wavelength emitted from a Ti:sapphire laser operating at a repetition rate of 1 kHz. The coupling efficiency into the fiber was 75%. The output spectra were collected using an integrating sphere to ensure that the entire output light was captured by an intensity-calibrated spectrometer.

The UV is generated in the form of DW emission, which arises due to nonlinear phase matching. The pump initially undergoes self-phase-modulation-induced spectral broadening. Once sufficient broadening is achieved, the spectral edge can seed phase-matched conversion of the pump to a UV wavelength as DW emission [25].

In a gas-filled AR-HCF, soliton-DW phase matching can arise from two distinct origins. The first type results from the general dispersion landscape in spectral regions far from resonances, which closely follows the dispersion of a simple dielectric capillary with the same core diameter [26]. This capillary-model-induced phase matching can easily be tuned by varying the refractive index of the filling gas, such as by adjusting the gas pressure. For our system, this phase-matching point is indicated by a black dashed line in Fig. 1.

The second type, which we refer to as resonance-induced phase matching, originates from rapid changes in dispersion around the resonances, producing a phase-matching point at the blue edge of each transmission band [27–29], as marked by red dashed lines in Fig. 1. Since the filling gas has little influence on the dispersion near the resonances, these phase-matching wavelengths cannot be tuned significantly without modifying the fiber geometry.

Figure 1(c) shows the output spectrum of the gas-filled AR-HCF when the pump energy is 1.4 μJ, corresponding to the soliton number N of 4.5. A distinct phase-matched emission is observed at 1070 THz, marked by a black dashed line, with a conversion efficiency of 2.7%. The efficiency was determined by calculating the energy fraction within the 900–1140 THz range of the measured output spectrum. This DW emission is attributed to capillary-model-induced phase matching. Generally, UV conversion can be enhanced by seeding more photons at the phase-matching wavelength. This can be achieved by, for instance, increasing the pump pulse energy, i.e., increasing the soliton number, to obtain a wider spectral broadening through self-phase modulation.

However, Figs. 1(d) and 1(e) show that even with increased pump energies, we observe reduced UV conversion efficiencies. This is because the broader spectral expansion also entails the onset of additional phase-matched conversions at 580 THz and 840 THz induced by the resonances. As these phase-matching points are spectrally closer to the pump, conversions to these wavelengths are favored over capillary-model-induced phase matching, which is farther out in the UV region. The presence of these competing phase-matching points is detrimental to efficient UV conversion. Figures 1(f)–1(h) present the UV portion from 870 to 1140 THz of the output spectra of Figs. 1(c)–1(e) on a linear scale. They clearly show that in the absence of competing conversion to resonance-induced phase-matching points, a distinct UV emission peak can be obtained. With an increasing soliton number, the UV spectral peak is reduced by more than half from Figs. 1(f)–1(g) and eventually reaches the noise level in Fig. 1(h).

For the numerical analysis, we simulated UV generation in two AR-HCFs with different cladding wall thicknesses by solving a uni-direction field propagation equation that includes the effect of photoionization [30]. The first system has a wall thickness of 470 nm, while the second has a thinner wall thickness of 115 nm. This reduction in thickness removes the cladding wall-induced resonances from the spectral region between the pump and UV. Other simulation parameters that may influence the UV conversion efficiency, such as the pump soliton number and target UV wavelength, are kept the same in both systems, i.e., N=4.5, 5.5, and 6.5, and the UV phase-matching point of 1000 THz. Note that the fiber length is set to 1.2 times the soliton fission length [31–33] in all cases to ensure the completion of the DW generation process.

Figures 2(a) and 2(b) present the dispersive wave phase-matching diagrams for gas-filled AR-HCFs with cladding wall thicknesses of 470 nm and 115 nm, respectively. The red dots indicate the phase-matching points for DW emission at 1000 THz. Figure 2(c) shows spectra with UV conversion to 1000 THz in the AR-HCF with a cladding wall thickness of 470 nm and pump soliton numbers of 4.5, 5.5, and 6.5, corresponding to pump energies of 1.6 μJ, 2.4 μJ, and 3.3 μJ, respectively. Similarly to the experimental results shown in Fig. 1(c), at N=4.5, the UV emission exhibits a distinct peak with a conversion efficiency of 3.4%. We also observed minor resonance-induced peaks around 600 THz and 880 THz. As N increases to 5.5 and 6.5, spectral peaks near the resonances develop, while the one at 1000 THz becomes less pronounced.

The influence of resonances is twofold. First, the resonances introduce considerable transmission loss—one at 300 THz stops the pump spectrum from advancing into longer wavelengths, while another at 600 THz effectively prevents further spectral broadening into the blue side. The latter limits the availability of seed photons in the far UV region. Second, the conversion to resonance-induced phase-matching points depletes the available seed photons at the UV wavelength.

In contrast, Fig. 2(d) shows that, in the absence of resonant bands, the broadening is uninterrupted, and the conversion efficiency increases consistently with increasing nonlinearity, rising from 6%, 8.5%, to 10.3% as the pump soliton number increases from 4.5, 5.5, to 6.5, respectively.

Fabricating AR-HCFs with sufficiently thin cladding walls to eliminate resonances in the spectral region between near-infrared and UV directly from thermal drawing is challenging. A cladding wall thickness of 132 nm was achieved directly from the fiber drawing tower with a core diameter of 17 μm [34]. Realizing a thinner cladding wall in thermal drawing requires highly expanding the tubes to overcome surface tension [15,35–37], increasing the risk of the tubes sticking to each other. Hence, a further reduction in wall thickness requires a fiber post-processing technique such as tapering. For example, an AR-HCF cladding wall of only 84 nm was achieved by fiber tapering, although the core diameter had also to be scaled-down to 5.8 μm [38]. The main issue with such small-core AR-HCFs in UV generation experiments is their restrictive transmission window at the long-wavelength edge [39] and limited power scalability [13]. Moreover, the tight mode area leading to high peak intensity may cause premature photoionization, which can negatively influence the UV conversion process [40].

One solution to reduce the cladding wall thickness without shrinking the core is wet-etching [41,42]. A notable achievement with this method is a UV generation experiment conducted in an AR-HCF etched to a cladding wall thickness of 220 nm [21], where only one resonant band lays between the pump and the UV wavelengths. This experiment reported a relatively good conversion efficiency of 6%. Further improvement requires pushing resonances completely out of the entire spectral region involved in the UV generation process with a ∼100nm cladding-wall thickness. However, pursuing such extreme thinness introduces a challenge in achieving an axially uniform wall thickness. This is an important factor in the nonlinear frequency up-conversion because an axially varying wall thickness broadens the loss bands and narrows the low-loss transmission windows. The issue becomes more pronounced in a thinner-wall AR-HCF, as the same absolute thickness variation leads to a more severe narrowing of the transmission bands. As shown in Fig. 3(a), the transmission bandwidth narrowing worsens dramatically with reduced wall thicknesses. For example, a wall thickness variation of 500–520 nm exhibits the first resonance band that is 10 THz wide, whereas the same 20 nm variation of 100–120 nm results in 20 times the bandwidth increase to a 200 THz-wide loss band.

We addressed the issue of wall thickness uniformity using the following wet-etching technique and achieved a uniform wall thickness of just 115 nm along a 17-cm-long AR-HCF. A schematic diagram of the setup is shown in Fig. 3(b).

We diluted commercial buffered oxide enchant (BOE) in water to create an HF solution with a concentration of 0.02%. The low concentration slows the etching rate, allowing for precise control over the process. BOE helps to address the uniformity issue. When the HF solution flows through the fiber, the initial segments will be etched more heavily than the latter due to the decreasing HF concentration along its flow path. This loss in concentration is compensated by the buffering reaction between ammonium fluoride (NH4F) and water that generates HF [43]. The use of BOE also improves the surface roughness of the etched fiber [44].

The flow rate of the HF solution within the fiber is another crucial parameter. A faster flow rate reduces the concentration gradient between the head and tail of the fiber. We set the air compressor to provide 8 bar pressure. Furthermore, the different dimensions of the fiber core and cladding rings can cause problems with varying HF solution flow rates in their differently sized channels. Therefore, we collapsed the cladding rings, allowing the HF solution to reach only the central hollow core. We used an air compressor to inject a stream of compressed air that continuously pushes the HF solution through the pretreated AR-HCF. This enables uniform etching along the fiber length of 20 cm, which is sufficiently long for our experiments. After a designated etching period, the AR-HCF was thoroughly flushed with water and dried in an oven to remove any residual chemicals.

Figure 4(a) presents the measured transmission spectra of the original fiber with a cladding wall thickness of 470 nm (Fiber A) and the etched fibers with thicknesses of 235 nm (Fiber B), 200 nm (Fiber C), and 115 nm (Fiber D). The first resonant band shifts from 350 THz to 1150 THz as the cladding wall thickness is reduced. In Fiber D, a continuous transmission window from its long wavelength limit set by the core size to 1150 THz (260 nm) is achieved, which is broad enough to stage resonance-free UV generation. Figures 4(b)–4(e) present the scanning electron micrographs of the corresponding fiber cross sections and enlarged views of their cladding tubes.

We measured the wall thickness of seven cladding tubes from scanning electron micrographs of cross sections taken at different axial positions along Fibers A and D at 1 cm, 4 cm, 7 cm, 10 cm, 13 cm, and 16 cm, as presented in Fig. 4(f). In Fiber A, the average cladding-wall thickness across the six cuts is 464 nm, with a standard deviation of 3.2% of the average thickness. In Fiber D, the average thickness is 102 nm, and the standard deviation increases to 10% of the average thickness.

The propagation loss calculated via the finite element method for Fiber D is 2 dB/m at the pump frequency. This is sufficiently low for the fiber length used in our experiments, with a total loss of only 0.3 dB over 17 cm—meaning over 90% of the launched pump energy remains available for nonlinear conversion. By further inflating the cladding tubes in Fiber D, a loss as low as 0.004 dB/m can be achieved.

To generate UV at 1000 THz from the pump at 375 THz, we launched N=6 solitons into the 17-cm-long etched AR-HCFs with different cladding wall thicknesses (Fibers B, C, and D). The difference in the dispersion among fibers leads to different fission lengths. Nevertheless, extending the fiber length beyond 1.2 times the soliton fission length does not significantly alter the main spectral features in the UV band. For practicality, we used the same fiber length in all our experiments. We chose 17 cm based on Fiber D, which exhibits the longest soliton fission length—14 cm—among the fibers used. The gas pressure is adjusted to ensure the same UV phase-matching wavelength for all three fibers, i.e., 8.4 bar, 8 bar, and 7.3 bar for Fibers B, C, and D, respectively. The pump pulse energy is also tuned to maintain the soliton number of 6, which corresponds to 1 μJ for Fibers B and C and 1.14 μJ for Fiber D.

Note that in Fibers B and C, there is still one resonant band in the spectral range between the pump and the UV. This resonant band is closer to the pump in Fiber B, whereas it is closer to the UV in Fiber C. Fiber D is completely free of any resonances up to the UV. The output spectra of these three fibers are plotted in Figs. 5(a)–5(c).

In Fiber B, the UV conversion efficiency is 4.6%. This increases slightly to 5.7% in Fiber C, accompanied by a significant drop in resonance-induced emission at the short-wavelength edge of the first transmission band. This is because the resonance-induced phase matching in Fiber C is farther from the pump than in Fiber B, leading to fewer seed photons at the resonance-induced phase-matching wavelength.

The UV conversion is greatly enhanced to 9.5% in Fiber D. Here, the absence of the resonant band between the pump and the UV completely removes the resonance-induced phase matching. Furthermore, the spectral broadening of the pump is not interrupted by the presence of resonances. These factors increase the available seed photons in the UV region, significantly improving the conversion efficiency. Note that in all cases presented in Fig. 5, the generated UV has a single distinctive spectral peak.

The DW bands have broader spectra than the simulated ones in Fig. 2(d). This is due to longitudinal variations in the fiber geometry—such as cladding-wall thickness and core diameter—introduced during etching. These axial variations lead to local changes in dispersion and phase-matching conditions along the fiber, resulting in DW emission over a broader spectral bandwidth.

We also characterized the spatial profile of the generated UV using the knife-edge method. We integrated the collected spectra over the range of 850–1200 THz to isolate the UV components and obtained their spatial energy distribution in both the horizontal and vertical directions. The resulting beam profile is shown in the inset of Fig. 5(c). The close agreement with the Gaussian fit confirms that the UV output predominantly maintains a fundamental core mode profile.

We emphasize that achieving a wider spectrum is key to improving the UV conversion efficiency. Apart from launching a pump with a higher soliton number, which will lead to a broader spectrum, this can also be achieved by using a shorter pump pulse. A shorter pump pulse can increase the nonlinearity by providing a higher peak intensity for the same soliton number.

In our experiment, we can shorten the pump pulse by compensating a residual chirp at the fiber input using a chirped mirror. The residual chirp was introduced due to the input transmissive optics, i.e., a half-wave plate, a coupling lens, and a gas cell window, needed for coupling the pump to the gas-filled AR-HCF. The chirp-compensated pulse has a duration of 27 fs, as characterized by the frequency-resolved optical gating technique and shown in the Fig. 6 inset. The time-bandwidth product of the chirp-compensated pulse is 0.33, indicating that the pulse is close to being transform-limited.

We launch the chirp-compensated pump pulse into the argon-filled Fiber D to generate UV at 1000 THz, keeping the identical pump soliton number of 6. The measured output spectrum is plotted in Fig. 6 together with a simulation result that is obtained by using the experimentally measured pulse as the numerical input. The wider main spectrum indicates a stronger nonlinearity experienced by the compressed pump pulse. As such, a UV conversion efficiency as high as 12% is achieved while also preserving a single distinctive spectral peak in the UV centered at 1000 THz.

The UV pulse immediately after its emission is as short as 7 fs in the simulation. Since the UV is generated in the normal dispersion regime of the gas-filled AR-HCF, the pulse experiences temporal broadening as it propagates through the rest of the fiber. By the time when the pulse reaches the end of the fiber, it broadens to 27 fs.

We observed no evidence of significant photoionization in any of the cases studied. The photoionization fraction remained extremely low in the simulations. Also, no typical photoionization-induced blue-shifting of the pump was observed in our experiments [45].

This study highlights the significant potential of etched AR-HCFs in enhancing UV generation. Our experimental and numerical results demonstrate that the removal of resonant bands between the pump and target UV wavelengths can significantly increase the conversion efficiency, up to 12%, while maintaining an excellent single distinct UV spectral peak.

Note that the HF etching process usually deteriorates the surface roughness, which, in principle, increases the transmission loss, especially in the UV region. The purity of the silica substrate governs this roughness effect. The surface roughness of fused silica remains constant in the sub-nanometer regime within 40 μm etching depths [46]. The use of fiber-drawing-grade silica inherently suppresses etching-induced roughness. Moreover, the roughness-induced loss can be effectively mitigated by strategically adjusting the fiber length to match the UV generation point. By optimizing the fiber length to ensure that the UV is coupled out immediately upon its generation, we effectively circumvented the scattering loss of post-processed AR-HCFs. The use of post-processed AR-HCF allows for UV generation with conversion efficiency comparable to that of simple capillary systems while also retaining the compactness and reduced pump energy requirements of AR-HCF-based systems.