The mid-infrared (MIR) spectral region is known as the “molecular fingerprint region”, and almost every kind of gas molecule shows a unique and strong absorption characteristic within that region, as shown in Figure 1a [1,2,3,4,5,6,7]. The MIR ultrashort pulsed lasers can be widely used in gas detection [7,8], cancer diagnosis [9,10,11], pollutant monitoring [12,13,14,15,16], food quality control [17,18,19,20,21,22,23], and other aspects [24,25,26,27,28], since they own much broader spectral ranges than ultrafast lasers in visible and near-infrared region. In addition, they also play essential roles in strong-field physics [29,30,31,32,33,34], laser surgery [35,36,37,38,39,40,41], attosecond science [42,43], and molecule dynamics [44,45,46] (as shown in Figure 1b), as they feature the ultrashort time domain and have the potential to achieve peak powers. Therefore, there is an essential and broad application value in MIR ultrashort pulse generation research.

MIR application is driven by the development of laser technology in related wavelengths. For example, the emergence of MIR laser frequency combs has facilitated the research process of highly sensitive (up to the order of several parts per billion (ppb)) molecular spectroscopy detection techniques [2]. Therefore, the timely promotion of the research on MIR coherent light sources is crucial to the development of science, technology, national defense, industry, medicine, and other applications. In the last decade, continuous breakthroughs and improvements have been made in the performance of MIR laser technology with regard to power, efficiency, wavelength range, time/frequency domain, beam quality, and stability [1,2,3,4,5]. The designs of the lasers have become increasingly compact, practical, portable, and miniature. However, due to the lack of suitable gain materials, the output of conventional solid-state lasers and fiber lasers cannot cover most of the MIR region, especially the long-wavelength MIR band of 8–20 µm. The application and development of spectroscopy in this band have been limited by the low laser power, narrow tunable wavelength range, and unstable operation [1,2,3,4,5]. At present, the main technologies for long-wavelength MIR laser generation include stimulated radiation methods (e.g., free-electron lasers, lead-salt semiconductor lasers, and quantum cascade lasers (QCLs)), and the nonlinear frequency conversion methods. Free-electron lasers are large, expensive, and usually used for X-ray pulse generation [47,48]. Lead-salt semiconductor lasers have low output power (about 0.1 mW), and require low-temperature cooling to achieve regular state operation, which are exceptionally sensitive to changes in temperature [8]. Due to the limited wavelength-tunable ranges of QCLs, it is usually required to use several QCLs with different working wavelengths, which is inconvenient and has high costs. Furthermore, mode-locking is hard to be realized in QCLs due to their picosecond-level gain recovery time. The picosecond-level laser pulse at 8 µm was only recently, in 2020, achieved in QCLs [49], and femtosecond pulses around 8 µm were generated via spectral filtering, gain modulation-induced spectral broadening, and external pulse compression in 2021 [50].

The nonlinear frequency conversion methods, including optical parametric oscillation (OPO), optical parametric amplification (OPA), difference frequency generation (DFG), and supercontinuum generation (SCG), have become the primary techniques to build up MIR ultrafast lasers above 8 µm. The principles of OPO, OPA, and DFG are shown in Figure 2 [51]. The OPO setup consists of a nonlinear crystal and a resonant cavity, which allows the signal or idler light or both of them to oscillate in the cavity to obtain the desired wavelength. In nonlinear crystals, the pump, signal and idler lights are overlapped, and the interaction between them leads to the corresponding gain obtained by the signal and idler light. The physical processes of DFG and OPA are basically the same and both require two beams of light (pump and signal lights) to interact in the nonlinear crystal. OPA aims to amplify the original signal light or to obtain high-energy idler light, while DFG aims to obtain high-average-power idler light. DFG is suitable for both continuous-wave and pulsed operation, while OPA is only suitable for pulsed operation. In addition, the signal light in the general OPA process is very weak compared to the pump light in power, while the signal light in the DFG process is stronger.

SCG is another effective nonlinear frequency conversion method for MIR generation. A supercontinuum is mainly generated from high-nonlinearity fibers, waveguides, or bulk crystals. In SCG, several nonlinear physical processes usually work together, such as self-phase/cross-phase modulation (SPM/XPM), parametric four-wave mixing, Raman scattering, soliton generation, soliton fission, soliton self-frequency shift, dispersive wave, and so on [5]. The current commercial product has already been available in the 2–5 µm of the MIR region [52]. However, the spanning range of supercontinuum greatly depends on the nonlinear material especially for MIR wavelengths longer than 5 µm or even 8 µm. Silicate glass is one of the most common and typical materials for SCG shorter than 2.5 µm. ZrF₄-BaF₂-LaF₃-AlF₃-NaF (ZBLAN) glass can extend the SCG wavelength to 10 µm. Chalcogenide glass, such as sulfur (S), selenium (Se), and tellurium (Te), is considered to be the most potential material for longer wavelength MIR generation [5,52,53].

The main drawback of OPO is the additional complexity brought by the resonant cavity and the relatively limited oscillation spectral range. It usually works in the region of 3–5 µm. Recent efforts show that the OPO-based MIR source could span over 8 µm by employing nonlinear crystals such as orientation-patterned gallium phosphide (OP-GaP) and silver gallium sulfide (AGS), owing to their long-wavelength transparency [54,55,56]. OPA usually works at a low repetition rate (from one to several hundreds of kHz level) but with high pulse energy (from one to several hundreds of μJ level), aiming to pursue high peak power or pulse energy in the MIR region [57,58,59,60,61,62,63,64,65,66,67,68,69,70,71]. However, there is still not much work regarding OPA working under an MHz repetition rate of over 10 μm because of the thermal effects. Many research efforts have shown that wavelengths generated by SCG could span over 8 µm or even 10 µm with the power of tens of milliwatts under well-engineered material structure [72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94]. However, long-wavelength (ranging from 2 to 9 µm) femtosecond pump sources with relatively high power (at least hundreds of milliwatts) and high-nonlinearity material are necessary for long-wavelength MIR generation by SCG, regardless of the pumping scheme (direct pumping or cascaded pumping). This makes the whole process more complicated since all the components mentioned above are not commercially available.

Nowadays, DFG is one of the most important techniques to achieve high average power MIR with a high repetition rate and has become an effective tool to generate optical frequency combs [51]. When the repetition rate is locked, the pump and signal pulse sequences come from the same oscillator and share the same carrier-envelope offset (CEO). This offset is automatically eliminated during the DFG process, resulting in a stable optical frequency comb. DFG has not only been applied to MIR but also to THz generation as the two wavelengths participating in the process are controllable. Thus, there is no doubt that DFG is still the mainstream method for generating MIR pulses.