Shortwave quantum cascade laser frequency comb for multi-heterodyne spectroscopy

Optical frequency combs¹ emitting a spectrum of discrete, evenly spaced narrow lines with a stable phase relation serve as unique optical rulers in frequency domains. The relation of radio frequency (RF) to the optical domain establishes a phase coherent link and provides a “clockwork” that enables counting of optical frequencies at unprecedented precisions.² In the mid-infrared (mid-IR) range, intense development of optical frequency combs is mainly driven by their applications in laser spectroscopy.³ Since a vast number of molecular species have their absorption fingerprints in this wavelength range, the mid-IR comb is an appealing technology which enables high-speed and accurate detection of various molecules.^4,5

The recently developed quantum cascade laser (QCL) frequency combs^6,7 are especially attractive to mid-IR spectroscopy because of their tailorable emission wavelengths, intrinsically narrow spectral linewidth,⁸ and great potential for high-power high-efficiency operation.^9,10 When the QCL dispersion is engineered, the large third-order nonlinearity originated from QCL intersubband transitions is able to lock the dispersive modes into equally spaced comb modes via four wave mixing. Because the gain recovery time (<0.5 ps) is much shorter than the photon roundtrip time in the cavity (∼60 ps for a 3-mm cavity), ultrashort optical pulses cannot be developed by typical mid-IR QCLs unless active mode locking techniques are applied.^11–13 In other words, QCL combs are do not emit a train of ultrashort optical pulses but an array of comb lines with a stable phase relation resembling frequency modulated lasers.¹⁴ Since QCL combs are direct emitting comb sources without using external pumping sources, the advantage is that they are monolithic and chip-based comb sources and have potential for mass production using standard semiconductor technologies. Shortly after their demonstrations, dual comb spectroscopy based on QCL frequency combs^15,16 with high resolution and high speed is proven to be promising for real spectroscopy applications.

Although significant advancements have been achieved, yet most of the demonstrated mid-IR QCL combs are concentrated in the λ ∼ 7–9 μm range.^{6,7,9,10,17,18} The advantage of this spectral range is that InP has negative material dispersions, while the active region material and waveguide have positive dispersions. This makes the overall dispersion close to zero, which favors stable comb generation. Figure 1(a) shows the calculated group velocity dispersion (GVD) distribution for a typical QCL waveguide, i.e. a 5–μm × 2–μm active region surrounded by 4–5 μm cladding and cap layers. The total dispersions of a cold QCL waveguide without considering the gain are the summation of the material dispersions, weighted by the modal confinement in each material, and the waveguide dispersion. Clearly, beyond the low-dispersion wavelength regime, the high dispersions (GVD > 500 fs²/mm) for QCLs in the shorter and longer wavelength ranges need to be effectively compensated to achieve efficient locking for comb operation. In the mid-IR range, the refractive index of QCL materials strongly depends on the free carrier concentration and the doping level. It is therefore possible to engineer a negative material dispersion by using a heavily doped plasma enhanced waveguide to compensate the positive active region and waveguide dispersions, as has been done for QCL combs at λ ∼ 8 μm.^10,15 However, for QCL wavelengths shorter than 6 μm, the dispersion compensation becomes difficult because of the increased free-carrier absorption and the maximum doping limits for the contact layers.

The dispersive Gires–Tournois interferometer (GTI) mirror provides another option for dispersion compensation.¹⁹ However, because of the relatively thick mirror required for sufficient dispersion engineering, the mirror materials should be carefully chosen to avoid excessive optical absorption. SiO₂/Al₂O₃ GTI mirror works for QCLs with limited output power because of the relatively high optical absorptions of SiO₂ and Al₂O₃.²⁰ TiO₂/Y₂O₃ based GTI mirror,¹⁷ on the other hand, has low optical absorptions and is promising for dispersion compensation for QCL combs in shorter wavelength ranges. Recently, a harmonic frequency comb at wavelength λ ∼ 4.5 μm was demonstrated based on self-starting harmonic generation in a QCL.²¹ Frequency comb operation with a sub-THz repetition rate was observed by using an asymmetric cavity. Nevertheless, the fundamental comb operation is only observed in a narrow current range due to the high dispersion inside the laser cavity.

Here, we address the prominent dispersion issue for shortwave QCLs and demonstrate 1-W dispersion compensated shortwave QCL frequency combs at λ ∼ 5.0 μm. The dispersion of the QCL frequency comb is compensated using a GTI mirror with an optimized waveguide design. A very narrow intermode beating linewidth of 500 Hz and a broad lasing spectral coverage of 100 cm⁻¹ were observed. The multi-heterodyne spectrum from the beating of two of these QCL combs exhibits 95 comb lines, which corresponds to a 30 cm⁻¹ overlapping spectral range.

In this work, the QCL structure is based on a dual-stack active region structure with strain-balanced emitter designs at λ ∼ 4.5 and 5.5 μm. As pointed out in the previous works,²² the dual-stack design helps to increase the lasing spectral bandwidth and decrease the gain-induced dispersion. The gain-induced dispersion in the short wavelength range is generally smaller than the long wavelength range. This is because the device in the short wavelength range generally has a wider gain due to the larger transition broadening and lower threshold current density due to reduced waveguide loss and free-carrier absorption. The gain induced dispersion in this work is estimated to be around 150 fs²/mm, which is compared with 350 fs²/mm in Ref. 9. The QCL structure presented in this work is based on the strain-balanced Al_0.64In_0.36As/Ga_0.31In_0.69As/Ga0_.47In_0.53As material system with AlAs, Ga_0.47In_0.53As Al_0.48In_0.52As inserts²³ grown by gas-source molecular beam epitaxy (MBE) on an n-InP substrate. A standard waveguide structure is adopted except that the doping level of the cap layer is increased from 1 × 10¹⁹ cm⁻³ to 2 × 10¹⁹ cm⁻³ and the cladding layer thickness is reduced from 3 μm to 2.5 μm. This change is able to reduce the dispersion for about 100 fs²/mm while introducing a limited additional optical loss of <0.2 cm⁻¹. A 5-mm long QCL device was directly coated with an Y₂O₃/TiO₂ based GTI mirror on the back facet using an ion-beam deposition (IBD) system, as shown in Fig. 1(b). The GTI coating was evaporated with a 120-nm Au layer. The layer thicknesses of the multi-layer dielectrics are 940/230/940/230/940/230/940. Y₂O₃ layers are in normal font, and TiO₂ layers are in bold font. The unit is nm. The GTI layer thicknesses were chosen to align the negative dispersion of the GTI mirror with the lasing spectrum to cancel out the dispersion of the QCL device in the lasing range, as shown in Fig. 1(c). The GTI dispersion and reflectivity were obtained following the description in Ref. 17. A theoretical calculation of the group delay dispersion (GDD) spectrum based on a transfer matrix method was performed and matches well with the experimental data. Owing to the excellent material quality and low optical absorptions,²⁴ the dispersive mirror possesses a reflectivity higher than 95% in the lasing wavelength range, which enables high-power operation.

A comparative study between a GTI-coated device and the standard high-reflection (HR) coated device with similar geometry was first investigated, as shown in Fig. 2. Both devices are 5-mm long and epi-down mounted on diamond submounts for characterization. Figure 2(a) shows the optical power-current-voltage (P-I-V) characterization under CW operation at 293 K. The GTI-coated device emits a CW power up to 1 W with a threshold current density of 2.1 kA/cm², while the HR-coated device emits to 1.25 W with a threshold current density of 1.8 kA/cm². The slope efficiency is 2.45 W/A, which is slightly higher than 2.38 W/A for the GTI-coated device. More importantly, the GTI-coated device exhibits no break down, and comb operation over a wide dynamic current range is observed.

To assess the dispersion of the HR-coated device and the dispersion compensation via the GTI mirror, measurements of subthreshold amplified spontaneous emissions were carried out using a Bruker Fourier transform infrared (FTIR IFS 66v/s) spectrometer. The dispersion was extracted from the interferogram following the description in Ref. 17. The HR-coated device exhibits a high group delay dispersion (GDD) of ∼7500 fs² which corresponds to a GVD of 750 fs²/mm in the lasing spectral range, while the dispersion of the GTI-coated device was largely reduced to GDD < 1000 fs² (GVD < 100 fs²/mm) as shown in Fig. 2(b). For the spectral range below 1800 cm⁻¹ and above 2100 cm⁻¹ in Fig. 2(b) where dispersion induced by GTI mirror is close to zero as shown in Fig. 1(c), the measured dispersion of ∼5000 fs² is close to the dispersion of the device without GTI modulation. Given the dispersion of ∼1000 fs² after compensation highlighted in the grey area in Fig. 2(b), the amount of the reduced dispersion is estimated to be about 4000 fs². This is close to the introduced dispersion of about −3500 fs² by GTI mirror. The change in threshold current densities for HR- and GTI-coated devices is also responsible for an additional GDD change in Fig. 2(b). This is because that the gain gets flatter as the current increases, and the increased carriers due to the increased current and internal temperature also decrease the material dispersions.

The substantial dispersion engineering significantly changes the lasing spectra and beatnote spectra at the round-trip frequencies for the two devices, as shown in Figs. 2(c) and 2(d). The HR-coated device exhibits a high phase noise regime at high operating currents with a wide beatnote (taken at a high current of 850 mA) as shown in Fig. 2(c). However, the GTI-coated device has a narrow beatnote width of <500 Hz at a similar current of 820 mA, which indicates stable frequency comb operation. Here, the beatnote was measured with an electrical technique, and the current source is passed from the dc to the ac + dc port of a high-frequency bias tee (SHF BT 45 B) to the QCL, so that current modulation induced by any intracavity intensity modulation can be recorded using a spectrum analyzer connected to the ac port of the bias tee. The different baselines of noise floors shown in Fig. 2(c) are due to different resolution bandwidths (RBWs) used in the beatnote measurements. To capture the narrow beatnote spectrum for the GTI-coated device, the spectral averaging is set to be 1 to reduce the temperature and current drifting effects and 50 for the HR-coated device since the drifting effects are not critical for the broad beatnote measurement. The rest of the beatnote spectra in the following paragraphs were taken with an average of 10. The spectrum for the GTI QCL comb spans ∼90 cm⁻¹ at 820 mA and contains about 285 modes. A pronounced modulation of the spectral intensity with a period of 160 GHz was observed, which is attributed to parametric gain suppression caused by temporal population pulsations.²⁵ Compared with the HR-coated device, the spectral bandwidth of the GTI-coated device is truncated somehow due to the modulations of the dispersion and the reflectivity of the GTI mirror as shown in Fig. 1(c).

Since an optical comb is characterized by its equally spaced modes, analyzing the mode spacing of the QCL cavity modes is another way to assess the comb quality. However, the limitation of this technique is that the resolution of the FTIR spectrometer (0.12 cm⁻¹, 3.6 GHz) is not high enough to measure the mode spacing precisely, as discussed in Ref. 26. To overcome this issue, we perform numerical treatment to the acquired spectrum and generate more points between the peaks by using spline interpolation. As a result, more information can be extracted without requiring a higher FTIR resolution for the measurement. The mode spacing is calculated as a function of the mode number and then subtracted from its linear fitting. This generates the frequency residuals of the mode spacing as a function of the mode number, as shown in Fig. 2(e), which serves well as an indicator of dispersion of the laser spectra. For the GTI-coated device with a narrow beatnote linewidth, the frequency residual curve oscillates around zero and the fitted line is almost flat. For the HR-coated device, the frequency residuals extracted from the lasing spectrum in Fig. 2(d) and the calculated frequency residuals based on GDD spectrum in Fig. 2(b) are compared in Fig. 2(e). Both plots show clear dispersions in the mode spacing distributions, which indicates the large dispersion inside the cavity induces a broad beatnote and high phase noise operation.

Another advantage of dispersion compensation of the QCL comb is the improved frequency comb operation in a wide current range. Figure 3 plots the lasing spectra and their corresponding beatnote at different currents. The spectral range spans over 100 cm⁻¹ at a high current of I ≥ 890 mA. A narrow beatnote with a signal-to-noise ratio (SNR) greater than 23 dB was measured up to roll-over currents, and no high-phase noise regime was observed, as shown in Fig. 3(b). The intermode beat frequency shifts as a function of current, with a tuning rate of ∼100 kHz/mA. This is because both the repetition frequency f_rep and the carrier-envelope offset frequency f_ceo are related to the group index n_g which is susceptible to the current induced temperature change. Narrow beatnote linewidths were also observed for the HR-coated QCL. However, they were only observed in a very narrow current range (∼30 mA) near the onset of frequency comb operation. The reason behind this phenomenon is that the four-wave mixing process is indispensable for the mode locking of adjacent laser modes with a stable phase relationship, and this is only effective when the phase mismatch and dispersion between involved the modes is as low as possible.

To demonstrate the capabilities of the GTI-coated QCL combs for dual-comb spectroscopy applications, we measured the multi-heterodyne beat signals by beating the two combs onto a high-bandwidth HgCdTe detector (Vigo PV-4TE-10.6, 1 GHz of 3 dB bandwidth), as shown schematically in Fig. 4(a). A neutral density filter with transmission <1% is used to attenuate the power focused on to the detector and to avoid any possible saturation of the detector. In parallel, light from the two combs is sent to a FTIR for spectral characterization. The heterodyne signal is acquired using a high bandwidth oscilloscope with a 1 GHz analogue bandwidth and a 5.0 GSample/s sampling frequency. The acquired waveform is then converted using a Fourier transform technique to generate the heterodyne spectrum. Both combs are driven with low noise current drivers (Wavelength electronics QCL 2000) with a specified average current noise density of 4 nA/$Hz$⁠. In the experiment, both the repetition frequency difference (Δf_rep) and the offset frequency difference (Δf_ceo) have to be optimized by controlling temperature and currents so that the heterodyne frequencies are within the detection limit of the detector. In the experiment, comb 1 is biased at a current of 820 mA and stabilized at a temperature of 288 K, while the comb is set to 850 mA and 293 K, in order to target the desired Δf_ceo and Δf_rep. As a result, a repetition frequency difference of Δf_rep ∼ 5.1 MHz is measured, as shown in Fig. 4(b). At the same time, both comb offset frequencies were optimized to be close to each other which is determined by monitoring the emission spectrum acquired by FTIR. Figure 4(d) shows that 95 modes are observed on the multi-heterodyne spectrum, corresponding to an optical bandwidth of 30 cm⁻¹, as the comb repetition frequency is 0.31 cm⁻¹ (9.165 GHz). Note that the dual comb sources were from different sections of the same wafer, leading to incomplete spectral overlap. More beating modes can likely be observed by using adjacent comb devices on the wafer. Moreover, the multi-heterodyne beating between the combs with Δf_ceo < 400 MHz can be readily reduced by slightly adjusting the temperature and current for the two combs.

The next step of our research will focus on dual comb spectroscopy based on the high power QCL combs. The high-power characteristics will enable fast detection of the unknown molecules with high SNRs. To further enhance the capability of this system detecting multiple species of molecules, the comb bandwidth will be increased continuously using a balanced-gain heterogeneous active structure design.²⁷ To further expand the bandwidth of the dispersion compensation, off-resonance dispersion of GTI designs is favorable²⁸ and will be investigated in future works.

In conclusion, we demonstrate high-power shortwave dispersion compensated QCL frequency combs for dual-comb spectroscopy at λ ∼ 5 μm. This dispersion compensated frequency comb emits up to 1 W at room temperature with a minimal dispersion in the emitting spectral range. A very narrow intermode beating linewidth of 500 Hz with a broad spectral coverage of 100 cm⁻¹ was observed. The multi-heterodyne beating from two dispersion compensated QCL combs exhibits 95 comb lines which corresponds to a 30 cm⁻¹ overlapping spectral range. The demonstrated monolithic, high-power shortwave QCL frequency comb source paves the way for the realization of chip-based spectrometers covering the entire mid-infrared wavelength range for rapid detection of complex chemical mixtures.

This work was partially supported by the National Science Foundation (Grant Nos. ECCS-1505409 and ECCS-1607838). The published material represents the position of the author(s) and not necessarily that of the National Science Foundation. The authors would also like to acknowledge the encouragement and support of all the involved program managers.