Optical frequency combs (OFC) operating in the mid-infrared (MIR) spectral range are a powerful spectroscopic tool [1]. Measuring the molecular fingerprint in the MIR region allows the identification of chemical species the determination of their concentration. MIR frequency comb sources based on the nonlinear conversion of near-infrared mode-locked lasers [2,3] and microresonators [4] have reached a high level of maturity featuring octave spanning spectra [5]. Semiconductor laser OFCs [6] are advantageous in applications that require compactness and low power consumption. Quantum cascade laser (QCL) frequency combs [7,8] are among the most investigated technologies. However, gain bandwidth and dispersion [9,10] have so far limited their spectral bandwidth on the order of $100{\mathrm{cm}}^{-1}$. One way to overcome this issue is spectral broadening in an external nonlinear fiber or waveguide. Recent results [11] have revealed, however, that the temporal output of QCLs is strongly chirped accompanied by the suppression of amplitude modulation. In fact, the ultrafast gain dynamics of QCLs are believed to be highly unfavorable for the formation of light pulses [12]. Previous attempts of mode-locking in monolithic QCLs were limited to cryogenic temperatures and low peak powers [13]. This makes nonlinear techniques for spectral broadening very inefficient. Further possible applications of short and intense mid-infrared pulses are numerous and promising, including nonlinear spectroscopy as well as optical ranging [14,15]. First results to enter the mid-infrared from shorter wavelengths were demonstrated using passively mode-locked GaSb-based type-I cascade diode lasers with 10 ps pulse duration around 3.25 μm [16].

Type-II interband cascade lasers (ICL) are an interesting alternative that already cover a major part of the mid-infrared up to 6 μm [17–19]. They combine the carrier injection and extraction scheme of QCLs with the advantages of an interband lasing transition. Hence, the upper-state lifetime of the optical transition in ICLs is on the order of 1 ns and thus significantly longer than the cavity round-trip time [18]. This enables low dissipation operation and has important consequences for mode-locking of ICLs. While the short upper-state lifetime of QCLs prevents the formation of short pulses, this issue is not present in ICLs. Furthermore, the ICL active material can be switched to absorption at the laser wavelength, which was shown by using ICLs as photodetectors at zero-bias [20]. Together with the fast carrier injection scheme, this allows the realization of efficient, high-speed modulators with cutoff frequencies of several gigahertz [21]. Hence, ICLs exhibit all required properties for efficient active mode-locking via modulation of the gain at the cavity round-trip frequency $f_{\mathrm{rep}}$ [22]. Recent efforts have been aimed at the generation of OFCs via passive mode-locking of ICLs. However, neither the interferometric autocorrelation nor multiheterodyne beating experiments showed the formation of pulses [23]. Instead, such passive ICL frequency combs are characterized by a continuous output intensity with a strong frequency modulation [21], similar to what was found in QCLs [8,11].

In this Letter, we report on the generation of picosecond pulses in two-section Fabry–Perot ICLs. The dry etched laser ridges are 6 μm wide and split into a 3520 μm long gain section and a 480 μm long modulation section [Fig. 1(a)]. The modulation section was designed to minimize parasitic capacitance for efficient RF injection. A 1.5 μm thick ${\mathrm{Si}}_3{\mathrm{N}}_4$ passivation layer was used for the modulation section while keeping its top contact area as small as possible. The passivation layer of the gain section is thinner (250 nm) to improve the thermal performance of the laser. The back facet of the device was high-reflection coated using ${\mathrm{Si}}_3{\mathrm{N}}_4$ and gold, while the front facet was left uncoated. The active region is comprised of six stages and operates at 3.85 μm ($2600{\mathrm{cm}}^{-1}$). At room temperature, the ICL emits up to 4.2 mW of optical power in continuous wave operation when both sections are biased homogeneously [Fig. 1(b)]. When the bias of the modulation section is set to 2.5 V additional loss is added to the cavity causing the threshold current density to increase and the maximum output power decreases to 1.9 mW. The injection of an RF signal at $f_{\mathrm{rep}}$ into the modulation section reduces the threshold by about 20%, showing that the laser is strongly influenced by the active modulation.

 

Fig. 1. (a) Scanning electron microscope picture of a 320 μm long modulation section of the ICL. The modulation section (left) and a ground contact (right) are optimized for RF injection via RF tips (Supplement 1, Fig. 1). A three-dimensional sketch of the entire device is illustrated in Supplement 1, Fig. 2. (b) Light-current-voltage (L-I-V) characteristics of the ICL at 15°C for a homogeneously biased laser (blue line) as well as for 2.5 V absorber bias with (dotted red line) and without (solid red line) RF modulation at $f_{\mathrm{rep}}\approx 10.15\mathrm{GHz}$. The RF power is 31 dBm.

Download Full Size | PDF

The characterization of the temporal output intensity of mid-infrared semiconductor lasers is challenging. Due to the high repetition rate and the relatively low average power, the peak power is expected to be too low for established nonlinear pulse characterization techniques [24]. Instead, we employ a linear phase-sensitive autocorrelation technique called “SWIFTS” [25]. This method uses a Fourier transform infrared (FTIR) spectrometer and a fast quantum well infrared photodetector (QWIP) to measure the amplitudes and phases of the beatings between adjacent laser modes (details in [25]). Hence, SWIFTS allows the reconstruction of both the temporal intensity and instantaneous frequency of the ICL frequency comb. Furthermore, recent experiments with a passively mode-locked quantum dot laser proved that SWIFTS and conventional intensity autocorrelation in a nonlinear crystal retrieve the same pulse width [26].

At 2.5 V bias of the modulation section, the laser generates a narrow beat note at the cavity round-trip frequency. This beat note results from the beating of adjacent cavity modes. Its narrow linewidth on the kHz level indicates that the cavity modes are phase-locked. To provide a stable reference for SWIFTS, we inject a weak RF signal at $-2\mathrm{dBm}$ into the modulation section. Previous experiments showed that such a weak modulation is able to lock the frequency of the beat note and leaves the spectral phases of the free-running OFC unchanged [27]. The SWIFTS analysis of the ICL in this state [Fig. 2(a)] shows a $28{\mathrm{cm}}^{-1}$ wide intensity spectrum, which consists of several lobes. The SWIFTS spectrum has the same shape as the intensity spectrum [blue dots in Fig. 2(a)] over its entire span, which proves full phase-coherence and thus frequency comb operation. The intermodal difference phases $\mathrm{\Delta }\varphi$ retrieved from the SWIFTS data decrease linearly over a range of exactly $2\pi$. This particular frequency comb state was found in QCLs operating at 8 μm [11], ICLs at 4 μm [21], and quantum dot lasers at 1.25 μm [26], and appears to be universal in semiconductor laser OFCs. Recent theoretical work attributes its origin to the interplay of dispersion and the Kerr effect in lasers with spatial hole burning [29]. Both SWIFTS interferograms [Fig. 2(a) right] have a local minimum at zero-path difference, which indicates the suppression of amplitude modulation [8]. Indeed, the reconstructed intensity [Fig. 2(b) top] does not show isolated pulses. In contrast, the instantaneous wavenumber is strongly modulated and linearly chirps through the entire spectrum within a cavity round-trip period.

 

Fig. 2. (a) SWIFTS characterization of the ICL at $-2\mathrm{dBm}$ injected RF power. The gain section is operated at $770\mathrm{A}/{\mathrm{cm}}^2$ and the modulation section at 2.5 V. The modulation section bias was optimized to obtain the narrowest beat note linewidth [28]. Blue line: intensity spectrum. Red line: SWIFTS spectrum. Blue dots: geometric average of neighboring modes of the intensity spectrum, which should be equal to the SWIFTS amplitudes in case of full phase coherence. Green dots: intermodal difference phases of adjacent comb lines. The right part shows a zoom-in on the center burst of the intensity and SWIFTS interferograms. (b) Reconstructed intensity and instantaneous wavenumber of the ICL frequency comb.

Download Full Size | PDF

In the following, we will investigate the influence of an increased modulation strength on the ICL frequency comb dynamics. Figure 3 shows the detailed SWIFTS analysis of the ICL for injection power levels from 6 to 36 dBm. The modulation frequency is altered slightly around $f_{\mathrm{rep}}$ to account for a small detuning of the laser round-trip frequency with temperature due to the high injection power. At 6 dBm [Fig. 3(a)], the ICL still operates in a similar OFC state as in Fig. 2(a) and the reconstructed time signal [Fig. 3(b)] does not show isolated pulses. When the injected power is further increased to 18 dBm [Fig. 3(c)], the spectral bandwidth grows to $33{\mathrm{cm}}^{-1}$ and the multiple lobes of the spectrum start to disappear. Interestingly, the SWIFTS spectrum shows that the ICL is not fully phase-locked in this transition state to the actively mode-locked regime. The intermodal difference phases still decrease linearly, but only cover the range of $0.53·2\pi$. Since $\mathrm{\Delta }\varphi$ is directly proportional to the group delay with $2\pi$ corresponding to one cavity round trip, this means that the ICL emits pulses with roughly $0.53·2\pi /2\pi \approx 53\%$ duty cycle. Indeed, the reconstructed time signal [Fig. 3(d)] shows broad and linearly chirped pulses. However, one should be careful when interpreting the reconstructed time signal of this partially coherent state because there could be an additional incoherent background. At 30 dBm, the spectrum consists of a single lobe spanning over $36{\mathrm{cm}}^{-1}$ [Fig. 3(e)]. The slope of $\mathrm{\Delta }\varphi$ decreases, which corresponds to a decrease of the group delay dispersion (GDD) to $-5.4{\mathrm{ps}}^2$ at 30 dBm compared to $-19{\mathrm{ps}}^2$ at 6 dBm. The reconstructed time signal [Fig. 3(f)] shows a train of isolated pulses with 27 ps full width at half maximum (FWHM). When the injected power is increased by another 6 dB, only a minor change in the shape of the spectrum and the GDD is observed [Fig. 3(g)].

 

Fig. 3. SWIFTS analysis of the ICL frequency comb for 6 dBm (a), 18 dBm (c), 30 dBm (e), and 36 dBm (g) injected power. The wavenumber in the top right corner indicates the bandwidth of the spectrum. The gain section was operated at $800\mathrm{A}/{\mathrm{cm}}^2$ and the modulation section bias was 2.5 V. The GDD is deduced from the slope of the intermodal difference phases (red dotted line). (b), (d), (f), and (h) The corresponding reconstructed temporal intensity and instantaneous wavenumber. The shaded grey area indicates the phase range of $\mathrm{\Delta }\varphi$ and the duty cycle of the pulses. The time signal in (d) only shows the coherent part of the ICL output.

Download Full Size | PDF

Detuning the modulation frequency away from the round-trip frequency $f_{\mathrm{rep}}$ of the free-running laser is another knob to influence the temporal output of the laser. Figure 4(a) shows the SWIFTS characterization at 32 dBm injected power, while the modulation frequency is 15 MHz higher than $f_{\mathrm{rep}}$. The spectrum consists of a single lobe spanning over $45{\mathrm{cm}}^{-1}$ and is fully coherent. Indeed, the intermodal difference phases do not show a linear chirp as in Figs. 3(a)–3(h) and occupy the phase range of only $0.13·2\pi$. The reconstructed time signal [Fig. 4(b)] displays much shorter pulses than in Fig. 3(g) with a FWHM of 3.2 ps. The peak power is 114 mW, which is an enhancement of more than 40 with respect to the average power of 2.7 mW and proves that the energy can be efficiently stored by the gain medium over a round trip. The inset in Fig. 4(b) shows that the pulse is not transformation-limited with several smaller pulses arriving shortly after the first intense pulse. This is because the intermodal difference phases in Fig. 4(a) do not synchronize perfectly.

 

Fig. 4. (a) SWIFTS characterization for $920\mathrm{A}/{\mathrm{cm}}^2$ gain section current density and 3 V modulation section bias. The modulation power is 32 dBm and the modulation frequency is roughly 15 MHz higher than $f_{\mathrm{rep}}$. In contrast to Fig. 2(a), both SWIFTS interferograms now have local maxima at zero delay of the FTIR mirrors, indicating strong amplitude modulation of the laser intensity. (b) Reconstructed time domain signal. Inset: Zoom on a single pulse, revealing a FWHM of 3.2 ps.

Download Full Size | PDF

In conclusion, our results show that ICLs enable the generation of MIR pulses within a single semiconductor laser ridge. The weakly modulated ICL frequency comb operates in a regime characterized by a strong linear chirp, which was also found in QCLs and quantum dot lasers. When increasing the modulation amplitude, the laser spectrum broadens accompanied by the formation of broad and chirped pulses. By carefully tuning the modulation frequency away from the round-trip frequency of the free-running laser, the pulse duration decreases to 3.2 ps with a peak power enhancement of over 40. This illustrates the large potential of ICLs as a compact source for mid-infrared pulses. From the spectral bandwidth, one can assume that further optimization of the laser dispersion [30] (see Supplement 1, Fig. 3) and modulation section length will enable the generation of subpicosecond pulses [31].