Hybrid CO$\mathrm{_2}$-Ti:sapphire laser with tunable pulse duration for mid-infrared-pump terahertz-probe spectroscopy

Ultrafast optical excitation with intense mid-infrared and terahertz pulses has emerged as a new tool to control materials dynamically. As most experiments are performed with femtosecond pulse excitation, typical lifetimes of most light-induced phenomena in solids are of only few picoseconds. Yet, many scientific applications require longer drive pulses and lifetimes. Here, we describe a mid-infrared pump - terahertz-probe setup based on a CO$\mathrm{_2}$ laser seeded with 10.6 $\mathrm{\mu}$m wavelength pulses from an optical parametric amplifier, itself pumped by a Ti:Al$\mathrm{_2}$O$\mathrm{_3}$ laser. The output of the seeded CO$\mathrm{_2}$ laser produces high power pulses of nanosecond duration, which are synchronized to the femtosecond laser. Hence, these pulses can be tuned in pulse duration by slicing their front and back edges with semiconductor-plasma mirrors irradiated by replicas of the femtosecond seed laser pulses. Variable pulse lengths from 5 ps to 1.3 ns are achieved, and used in mid-infrared pump, terahertz-probe experiments with probe pulses generated and electro-optically sampled by the femtosecond laser.


INTRODUCTION
Intense mid-infrared (MIR) and terahertz light pulses are of strong interest for nonlinear ultrafast spectroscopy as well as for coherent manipulation of materials properties. Recently, such pulses have enabled the discovery of unconventional phenomena such as the formation of magnetic polarization in antiferromagnets 1 , ferroelectricity in paraelectrics 2,3 , new topological phases 4 , as well as non-equilibrium high T c superconductivity in cuprates 5,6 and molecular organic materials [7][8][9][10] . Typically, these experiments are performed using subpicosecond pulses with high peak electric field (∼ 5 -10 MV/cm). The lifetime of the induced transient state is therefore often limited to a few picoseconds. In the pursuit of longer-lived, functionally relevant photo-induced states, there is an increasing demand for laser sources producing mid-infrared and terahertz pulses with tunable duration from hundreds of femtoseconds to hundreds of picoseconds 10 .
Laser sources based on narrow band gap semiconductors, such as IV-VI compounds (lead salts) and wavelength tunable quantum cascade lasers readily produce radiation in the midinfrared region of the electromagnetic spectrum (∼ 2.5 -30 µm wavelength). However, these sources often require cryogenic cooling 11 , and their emitted power is not sufficient to reach sufficiently intense pulses, as field strengths in excess of few MV/cm are required for nonlinear excitation of materials. Typically, high peak power mid-infrared pulses are generated via optical parametric amplification and frequency mixing starting from ultra short light pulses generated by conventional solid state lasers (e.g. Ti:Al 2 O 3 or Yb:YAG). This approach yields pulses with energies of 10 -50 µJ and allows to reach MV/cm peak electric fields only for pulses with sub-picosecond durations. Although gas lasers based on CO (λ = 5 -6 µm) and CO 2 (λ = 9 -11 µm) can easily produce high energy mid-infrared pulses [12][13][14] , they are typically not mode-locked resulting in pulse-to-pulse time and intensity fluctuations that make pump-probe experiments with picosecond precision difficult.~1 Here we present a hybrid optical source based on a combination of ultrafast Ti:Al 2 O 3 and CO 2 lasers to perform pump-probe measurements using intense MIR pulses with tunable duration. Figure 1 displays a schematic of the experimental setup. A CO 2 laser chain is seeded with radiation generated by an ultrafast optical parametric amplifier to generate midinfrared pulses at 10.6 µm that are optically synchronized to the Ti:Al 2 O 3 source, enabling experiments with sub-picosecond time resolution.

GENERATION OF MODE-LOCKED MID-INFRARED PULSES WITH CO 2 LASERS
The gain spectrum of the CO 2 laser consists of multiple lines with a bandwidth of about 3.7 GHz and about 55 GHz spacing originating from the vibrational-rotational transitions in the laser gas (see Fig. 2b). This bandwidth is much larger than the typical spacing of cavity modes of about 100 MHz (see Fig. 2c) so that multiple longitudinal modes can be excited at the same time even if the gain is limited to a single rotational line (e.g. with a grating reflector). Figure 2a displays the typical single shot output of a commercial transversely excited CO 2 laser, revealing multi-mode operation arising from amplified spontaneous emission.
Whilst mode-locking techniques based on saturable absorbers, electro-or acousto-optic modulators facilitate the formation of reproducible high power output pulses 15-17 , they are not well suited to synchronize the generated MIR pulses to a probe beam. This limitation can be overcome by externally seeding the CO 2 laser cavity, both as a means of active mode locking and to facilitate synchronization 18,19 . For this purpose, we generated 150 fs long 10.6 µm wavelength pulses in a 1.5 mm thick GaSe crystal by difference frequency mixing of the signal (1490 nm) and idler (1730 nm) outputs of a home built optical parametric amplifier (OPA). The OPA was pumped with 3 mJ pulse energy, 100 fs long pulses from a commercial Ti:Al 2 O 3 regenerative amplifier (see Fig. 1). A fraction of these femtosecond pulses (∼ 60 pJ) was injected through the semi-transparent (R: 80 %, T: 20 %) front window into the cavity of the CO 2 laser. The bandwidth of the injected femtosecond pulses was spectrally filtered to match the narrower gain bandwidth of the CO 2 laser cavity (see Fig. 2b, c) by spatial selection of the first diffraction order of the intracavity grating reflector, which acted as a rear mirror only for the desired rotational line around 10.6 µm. The injection of these seed pulses induces an active mode locking resulting in a train of output pulses, synchronized to the femtosecond seed laser 20 . A typical time-intensity trace of the seeded output of the CO 2 oscillator is shown in Fig. 3b (blue curve). The peak intensity of the pulse train was about one order of magnitude higher than in the non-seeded case (red curve) due to the phase locking of all longitudinal cavity modes induced by the seed. The single pulses in the train were ∼ 1.3 ns long, limited by the available gain bandwidth, and showed intensity fluctuations of less than 1 %. These residual fluctuations can be attributed to small temporal variations of the electrically triggered plasma build-up, which translated into a time jitter of the gain curve. For the specific laser used in this work, the optimal timing for seed injection was found to be ∼ 700 ns after the plasma excitation is triggered. Under these conditions, the pulse build-up time was reduced by ∼ 150 ns as compared to unseeded operation and the output intensity was highest.
The most intense pulse in the train was selected by a custom-built pulse picker based on the combination of a CdTe Pockels-cell ( Fig. 1-5) and a wire-grid polarizer. To reach the required pulse energy for the envisioned experiments, this isolated pulse was further amplified from ∼ 0.7 mJ to ∼ 10 mJ in a subsequent CO 2 laser amplifier, built from a commercial laser that we modified to operate as a 10-pass amplifier ( Fig. 1-6) at a repetition rate of 18 Hz.

MID-INFRARED PULSE DURATION CONTROL
Although a pressure or electric field (through the AC Stark effect) broadened spectral gain can support the amplification of picosecond CO 2 laser pulses, it requires either an amplifier with a gas cell pressurized at 10 -15 bar 21-23 or multi-stage preamplification for the generation of laser fields with intensities of more than 5 GW/cm² 24 . As the midinfrared pulses produced in our setup are inherently synchronized to a femtosecond near infrared source, another viable option for tailoring their pulse duration is the use of electronhole plasmas in photoexcited semiconductors as ultrafast shutters. This technique, first described in detail in Ref. 25 , has been applied in different laser systems for external pulse shortening 23,26 as well as for the reflection of a single pulse out of a laser cavity [27][28][29] .
In our implementation, shown schematically in Fig. 4a, the p-polarized mid-infrared laser pulse was transmitted through an optically flat semiconducting plate set at Brewster's angle to suppress any reflection in the unpumped state. A short near-infrared laser pulse, derived from the Ti:Al 2 O 3 chain used to seed the CO 2 oscillator was used to excite a plasma of electron-hole pairs. Optical excitation made the semiconductor surface highly reflective for all frequencies lower than the electron-hole-density-dependent plasma frequency. After FIG. 4. Principle of semiconductor switching. a. Schematic of the setup for generation of continuously adjustable pulse lengths. The incoming MIR CO2 laser pulse (blue) impinges on a semiconductor at Brewster's angle and its leading edge is fully transmitted. A second short NIR pulse (800 nm wavelength, ∼ 100 fs duration, red) excites an electron-hole plasma at spatio-temporal overlap with the MIR pulse. Within few picoseconds, the carrier density rises, which renders the semiconductor highly reflective and slices the MIR pulse. A subsequent switch can cut the trailing edge so that precise pulse length tunability is possible by adjusting the optical delay between the two NIR slicing pulses. b. Cross correlation curves of the unswitched (red) CO2 laser beam as well as for single transmissive (violet), single reflective (green) and double switched pulses (blue). Upon switching, the semiconductor reflectivity changes within less than 5 ps creating sharp edges in the pulse.
optical excitation, the critical carrier density for full reflection can be obtained from: where ω l is the laser frequency and 0 , r , m * , and e denote the vacuum and semiconductor dielectric constants, the effective carrier mass and the electron charge, respectively. To create a reflective surface for CO 2 laser radiation at 10.6 µm wavelength, a carrier density in excess of 10 18 − 10 19 cm −3 is needed depending on the type of semiconductor. Si, Ge, CdTe, and GaAs all feature high mid-infrared transparency and critical carrier densities that can be achieved with modest (< 10 mJ/cm²) excitation fluences. The combination of a reflection and a transmission switching enabled the generation of short pulses of variable length that could be adjusted by appropriately delaying the control pulses. In order to compensate the optical path in the CO 2 lasers of ∼ 400 m, the nearinfrared excitation pulse for semiconductor switching was derived from a second Ti:Al 2 O 3 amplifier seeded with a later pulse from the master oscillator (see Fig. 1). The remaining optical delay of ∼ 2 m was compensated with an optical delay line to achieve accurate timing.
To perform pump-probe experiments using these pulses we required high switching efficiency, wide pulse length tuning range (10 ps to more than 500 ps), and a high background suppression of at least 1:1000. To achieve a high switching contrast, good mid-infrared transparency in the unexcited state and high reflectivity in the photoexcited state were required. Figure 4b displays typical time-traces of the resulting 10.6 µm pulses before and after switching using silicon both as a reflection and transmission switch. These measurements were performed by cross-correlation of the mid-infrared pulse with an ultrafast NIR pulse (100 fs pulse length, λ = 800 nm) via sum-frequency generation in a 2 mm thick GaSe crystal. Different semiconductors feature distinct decay dynamics 30-32 of the electron-hole plasma. Figure 5a shows a comparison of the time-dependent 10.6 µm reflectivity of a single semiconductor switch made of silicon, germanium and cadmium telluride. These curves were extracted from cross-correlation measurements shown exemplarily in Fig. 4b by calculating the time dependent ratio of the reflected intensity from the semiconductor to the intensity of the unswitched beam. The decay rate of the reflectivity could be fitted by a single exponential decay in all the three cases and was found to be the longest in Si (550 ps) and the shortest in CdTe (75 ps). When output pulses below 50 ps are desired, using CdTe as the first reflection switch provided the best background suppression ratio (> 1000:1) thanks to its fast reflectivity decay. The background suppression was measured through cross-correlation of the double sliced pulse shown in Fig. 5b by averaging over a time window of 50 ps on the pulse and 50 ps after the trailing edge of the pulse. On the other hand, for output pulse durations in excess of 50 ps, a Si reflection switch was preferred to maintain high reflectivity for a longer time. In this case, due to the large penetration depth of the near-infrared excitation pulse, thick plasma layers are formed so that transmission of the MIR radiation is negligible 30 . This makes Si a good candidate also for the second switch, both for the generation of short and long pulses. As this switch is used to suppress the trailing edge of the pulse, it is most effective if the radiation is to be blocked for as long as possible after photoexcitation.
The pulse lengths of double sliced pulses obtained with two silicon switches could be tuned over a wide range from ∼ 5 ps to ∼ 1 ns whilst for durations up to 300 ps almost flattop-intensity pulses were achieved (see time profiles in Fig. 5b). Although the minimum rise and fall time of the mid-infrared pulse is determined by the near infrared excitation pulse duration, in our setup the two pulses impinge on the semiconductors with a 13°noncollinear angle leading to a few picosecond temporal spread on the semiconductor surface. The damage threshold was found to be significantly material dependent being just above 5 mJ/cm² for Si, 3.9 mJ/cm² for CdTe, and 1.7 mJ/cm² for Ge. For fluence levels below damage threshold a maximum switching efficiency of 50 % was achieved for all materials.

MIR-PUMP THZ-PROBE EXPERIMENT ON INSB
To demonstrate the applicability of the setup for pump-probe experiments with adjustable excitation pulse-lengths, we measured the time-resolved reflection of THz radiation from an InSb wafer excited with 10.6 µm pulses with durations varying between 5 ps and 100 ps (see Fig. 6). These pulses were generated using a CdTe reflection and a Si transmission switch and their time-intensity profile is shown in Fig. 6b. THz pulses with bandwidth from ∼ 0.5 THz to ∼ 3 THz were generated in a commercial photoconductive antenna excited with 100 fs long pulses at 800 nm wavelength. These THz pulses were focused on the InSb sample, and their electric field was measured after reflection by means of electro-optic sampling in a 1 mm thick (110)-cut ZnTe crystal. Upon irradiation with the 10.6 µm pump pulses, the reflected THz intensity (the square of the THz electric field) increased linearly with the integral of the pump pulse intensity cross-correlation measurements, which we determined to be linear with the number of pump photons. After the 10.6 µm pump pulses hit the InSb wafer, the respective reflectivity levels stayed nearly constant for the observable measurement time of several hundred picoseconds. Therefore, the InSb response acts like an optical integrator of the pump photons. In summary, we demonstrated a hybrid optical device that generates high-power midinfrared pulses based on seeded CO 2 lasers, with tunable pulse durations achieved with two semiconductor-based plasma mirrors. This source allows the extension of ultrafast experiments from femtosecond drives to pulse durations of hundreds of picoseconds. As the mid infrared pulses were naturally synchronized in time to a femtosecond laser, this device made pump-probe experiments with sub-picosecond resolution possible.