Simple single-section diode frequency combs

Optical frequency combs^1,2 initially attracted attention for their impact on the field of precision measurement where they revolutionized optical frequency metrology^3,4 and synthesis⁵ and enabled optical atomic clocks.⁶ These applications leverage the ability of combs to coherently link disparate optical frequencies and directly connect optical and radio frequencies. A second wave of interest was triggered by the development of dual-comb spectroscopy and related methods,^7–11 which rely on the mutual coherence of two combs to realize optical spectroscopy with a combination of spectral resolution, acquisition time, and size that cannot be achieved using conventional methods. The development of methods to measure and control the frequency comb generated by mode-locked lasers^1,12 was key to these applications.

In parallel with the broadening of applications based on highly controlled combs, new sources of combs were being developed. One prominent current approach is based on nonlinearities in micro-resonators.¹³ This approach features a chip-scale resonator with a large free-spectral range, although the required high power continuous-wave pump presented a challenge to realizing fully chip-scale systems, which has only recently been achieved.^5,14,15 Another important approach was launched by the discovery that quantum cascade lasers (QCLs) could produce frequency combs.¹⁶ While QCLs are compact and electrically pumped, their development was important as they provide a simple source for mid-infrared (IR) to terahertz frequency combs, although the long operating wavelengths brought with them a need for cryogenic cooling.

Directly generating combs from inter-band diode lasers in the near-IR or visible spectral regions is certainly attractive as they are electrically pumped, are compact, and do not require cryogenic cooling. While there have been scattered reports of comb generation in single section devices,^17,18 most effort on engineering diode lasers to produce combs has focused on traditional mode-locking, which is designed to produce a train of ultrashort pulses. However, the deleterious gain dynamics in semiconductors have made this challenging.¹⁹ Consequently, diode-laser combs have lagged behind other source. For example, dual comb radio frequency (RF) spectra have only recently been demonstrated.^20,21 Success in developing an inexpensive, miniature, portable, simple, and low power comb source, whether a diode laser or otherwise, would greatly extend the reach of dual-comb spectroscopy to provide a path to ubiquitous spectroscopy for a large variety of applications such as atmospheric monitoring, pollution monitoring, chemical sensing, or breath analysis. One could even imagine them being incorporated into mobile phones, as accelerometers are today.

To this end, we develop and characterize the electric field output of electrically pumped, single section diode lasers outputting frequency combs.²² These devices are monolithic and cheap to manufacture. Each diode’s volume is ∼0.01 mm³, allowing several dozen devices to fit on a 0.5 × 2 × 1 mm³ chip. We demonstrate tuning of individual comb teeth over multiple free-spectral ranges (FSRs) by adjusting pump temperature or current, fulfilling a key requirement for combs used in precision metrology. Furthermore, we demonstrate dual comb spectroscopy of molecular transitions²⁰ and power our chip scale combs with AA batteries. These results show that diode frequency combs (DFCs) have sufficient coherence to be used in dual comb spectroscopy and thus realize the vision for ubiquitous and portable comb-based spectroscopy. The epitaxial growth of our diode laser layers was performed by a commercial foundry, enhancing the possibility for them to be manufactured inexpensively, using materials standardized by the telecommunications industry, an important distinction from the recent demonstration of dual comb spectroscopy at 2 μm.²¹ 

Microscopic ring resonators are the current state of the art in near-IR compact frequency comb sources. The resonators are typically rings of silicon nitride or fused silica pumped externally by high-power erbium-doped fiber amplifiers, although multi-section single-comb, battery powered sources have recently been developed.^14,15 Entering the most useful low-noise single-soliton comb states requires a complex algorithm of parametric pump detuning.^23,24

An alternative route has been the development of miniaturized semiconductor comb sources. Due to their broader gain bandwidth through inhomogeneous broadening, quantum dash frequency combs^25,26 and quantum-dot frequency combs^27,28 are more common than quantum well based devices^17,18,20,29 and related quantum well cascade devices.^30,31 Quantum well devices, however, have the advantages of higher gain coefficients and ease of production. Furthermore, due to the higher gain coefficients, quantum-well devices are more flexible because the repetition rate can be easily tuned over a wide range (∼10 GHz–100 GHz) by varying the quantum well number and device length and offer a path toward convenient and compact comb generation in the visible and near-infrared spectral regions.³² The gain bandwidth of quantum-well devices can be increased by using multiple quantum wells with varying width.²⁰ 

In a coherent optical frequency comb, the frequency of every output mode is determined by two radio frequencies: the repetition rate (f_rep), or the inverse of the cavity round-trip time, and the offset frequency (f_o), which is set by the dispersion of the cavity. The frequency of the nth mode of a comb’s output is f_n = nf_rep + f_o. This relationship undergirds the fields of precision optical frequency measurement and generation.

Two key properties are required for a laser to generate a coherent frequency comb: consistent gain for multiple modes simultaneously and a mechanism to force coherence between adjacent cavity modes. For DFCs and other semiconductor devices that rely on frequency-modulated (FM) mode locking,²² spatial hole burning (SHB) supports consistent gain between multiple modes, satisfying the first requirement. We ensure the presence of SHB by choosing the operating wavelength to be shorter than the ambipolar diffusion length of carriers²⁰ because each standing-wave cavity-mode “burns” a spatial hole in the gain distribution, but gain will still be available between field maxima. In order to allow comparable gain among multiple modes and ensure coherent output, four-wave-mixing mediated by the Kerr effect between modes enforces that the phase of each mode satisfies Φ = 2ψ₀ − ψ₊ − ψ₋ = ±π, 0, where ψ₀ is the phase of an arbitrary lasing mode and ψ_± are the phases of the adjacent modes.²² The solution Φ = 0 (corresponding to a pulse train) requires a saturable absorber, but the solutions Φ = ±π force the electric fields out of phase, maintaining the SHB-induced gain profile. The net result is that DFCs output a frequency- and amplitude-modulated comb.

Frequency combs are generated in p–n InGaAsP laser diodes with four quantum wells providing gain.²⁰ The active regions are composed of a quaternary InGaAsP quantum well structure, with a ridge waveguide providing lateral mode confinement. Device construction details can be found in the supplementary material and Ref. 20. Chips, shown in Fig. 1(a), are typically 1 mm–2 mm long and 2 mm–4 mm wide, containing up to two dozen diodes. DFC comb generation, shown in Fig. 1(b), initiates once the pump current is above the lasing threshold. The laser output spectrum stabilizes only when comb generation commences (typically observed for currents ∼10%–20% above the lasing threshold). The diodes were operated between 10 °C and 23 °C, with injection currents of between 110 mA and 230 mA (although currents > 300 mA were achieved). The diodes had a series resistance of ∼1.5 Ω, leading to a power dissipation of ∼0.3 W at an injection current of ∼200 mA. Tungsten probes contacted the diode’s top face, and a brass plate was used as a current return from the bottom face.

Operating wavelength, power, and repetition rate are coarsely set by device design and fabrication. A typical spectral width (−5 dB from the peak) is 0.8 THz (6 nm) at 1550 nm, while the repetition rate (roughly set by the cavity length) was in the range of 19 GHz–25 GHz. Although the FSR of these combs was on the order of 20 GHz, one to two orders of magnitude larger than desired for atmospheric gas sensing, the FSR could be adjusted by simply lengthening the cavity. Devices even ten times longer than our current combs would still have a compact form factor (∼2.5 cm in length). Fine control of f_rep and f_o is achieved by adjusting pump current and diode temperature, tuning individual comb teeth over multiple FSR, as seen in Figs. 1(c) and 1(d), and is dominantly linear. Although the changes in f_rep and f_o are likely due to the interplay between different physical effects, the ability to tune over multiple FSRs distinguishes these devices from micro-resonator-based comb generation. It should be noted that the tuning range presented was limited by technical considerations, not by fundamental device physics. Typical tooth widths were on the order of 200 kHz, likely determined by a number of technical and fundamental details to be addressed by future work. The output spectrum can be broadened by the addition of quantum wells of other widths,²⁰ though the limits of this broadening are as of yet unknown. The output power (typically 1 mW) can be increased by adding quantum wells.

The basic property upon which all frequency comb applications rely is the mutual coherence between the comb teeth. Mutual coherence is typically difficult to prove and requires measuring the spectral phase profile of the comb spectrum. It is not enough to simply show a comb-like power spectrum to confirm that a light source is a coherent frequency comb. To show that the DFC output fulfills this requirement, we performed a cross-correlation dual comb (XCDC) measurement, shown in Figs. 2(a) and 2(b). This technique derives from traditional dual comb techniques and spectral interferometry techniques^7,33,34 but is broadly applicable to rapidly characterizing the electric field output of frequency combs. Given the quasi-CW nature of the DFC output, more common techniques such as frequency-resolved optical grating measurements would be impossible to implement here because they require a nonlinear step. We use it to confirm the prediction that the observed combs are due to frequency modulation mode-locking, as occurs in QCLs,^16,22,35,36 and has been hypothesized for diode lasers.²² In addition, the XCDC measurement confirms the overall phase coherence of the DFC output.

To realize XCDC, we combined the DFC output (f_rep = 25.0012 GHz) with that of a dispersion-compensated frequency comb with f_rep,r = 250 MHz. Although our coupling efficiency into the optical fiber was 50%–60% for the DFC, there was still plenty of power to characterize the DFC field. In the time domain, this corresponds to a cross correlation between the pulse trains of the DFC and reference comb. In the radio frequency domain, equivalent image spectra occur every f_rep,r/2. We acquired 350 000 samples at 0500 MS/s, taking the data from 125 MHz to 250 MHz to be the XCDC signal (the frequency interval from f_rep,r/2 to f_rep,r was chosen for technical reasons). Since DFCs were stable over seconds, we converted between the RF and optical domains by fitting the RF spectra to traces taken on an optical spectrum analyzer (OSA). After coherent averaging of the time-domain data to remove the effect of crosstalk in the acquisition electronics (see the supplementary material), we unwrapped the spectral phase in the frequency domain by monitoring point-to-point changes in the spectral phase and adding the appropriate value of either ±2π when the discontinuity exceeded ±π radians. As seen by the good qualitative fit of the data in Fig. 2(c), the spectral phase was parabolic in good agreement with the theory presented in Refs. 22 and 32. This fit allowed us to extract the estimated group delay dispersion of −4.3 ps², which corresponds to the lowest frequency portion of the comb arriving at the detector first. For details on the phase convention used in our analysis, see the supplementary material. The device output [Fig. 2(d)] is periodic, with strong amplitude and frequency modulated components. We note that the reconstructed quasi-linear chirp in Fig. 2(d) is similar to that observed from quantum cascade lasers.^35,36

To establish mutual phase coherence between two comb sources, it is not enough to simply demonstrate a comb structure in their RF beat spectrum. One must also demonstrate that the comb teeth in the spectrum are coherent with each other. This requirement is more challenging and requires that the phase noise in the RF spectra be correctible. To demonstrate this, we independently powered two DFCs on the same chip with injection currents of ∼210 mA and ∼195 mA and an operating platform temperature of 13.3 °C. We then combined their outputs and measured RF dual comb spectra (DCS) before and after a 300 Torr H¹³C¹⁴N calibration cell along with an attendant OSA trace.^7,37–39 This high-pressure calibration cell was chosen because it has transition linewidths of roughly ∼25 GHz, well-matched to the repetition rate of our devices. Figure 3(a) presents a schematic of this experiment. The DCS data were collected over 10 µs at 2 GS/s and noise-corrected (see the supplementary material for details).^40–42 This process produced a pair of well-corrected RF combs, from which the amplitude of each comb tooth was used to calculate the absorption profile using Beer’s law, comparing the result to a typical spectrum taken using a broad-band light source. Figure 3(b) shows a typical DCS absorption measurement. This measurement is similar to that in Ref. 37 using a mode-locked laser and Ref. 38 using a microresonator. It demonstrates the ease of taking useful multi-comb spectra with even free-running, miniaturized DFC devices on the same chip at ambient operating conditions. Furthermore, due to the nature of DFCs, the repetition rate can be chosen during manufacture to be optimally matched to remote spectroscopy applications. Alternatively, greater resolution spectra could be acquired with even the high rep-rate devices by acquiring multiple DCS spectra over several sets of injection currents. Figure 3(c) shows the spectrum of a DFC device powered with AA batteries (two batteries provide roughly an hour’s operation at present efficiency), though it should be clarified that both the DCS and XCDC results were obtained using a traditional diode power supply to maintain precise injection currents.

We have shown a diode-laser frequency comb source that is a viable platform from which to launch future portable, comb-based spectroscopy, frequency metrology, time-transfer, or ranging applications. To be useful in such applications, chip-scale frequency comb generators must be (1) compact, (2) efficient, (3) self-coherent, (4) tunable, and (5) coherent with other such devices. We have shown that combs can be generated in diode lasers grown commercially, with two-dozen or so combs fitting on each device. We have shown that they are efficient enough to be powered with standard AA batteries. Our XCDC measurements show that DFCs output coherent, frequency modulated combs. We have measured the phase profile of the field in both the time and frequency domains, showing that the output is dominated by linear chirp, as expected.²² Furthermore, their output is tunable over multiple free-spectral ranges with both current and temperature, demonstrating that they are a potential source of stabilizable frequency comb spectra for precision time-transfer or frequency metrology applications. Finally, we demonstrated that our combs are coherent with other devices on the same chip by conducting dual-comb spectroscopy of HCN vapor.

DFCs fulfill the five requirements outlined above necessary for ubiquitous and portable application. Although we did not stabilize their spectra carefully in this work, future work could focus on simple injection locking schemes to increase control over the comb spectra and decrease tooth linewidths. As is, teeth are roughly on the order of 100 kHz wide.²⁰ The comb generation, however, is incredibly simple and was sufficiently stable to enable us to characterize the coherence and operation of these light sources. These results are not an advance in the state of the art for DCS; rather, they provide a route to making DCS, and combs more broadly, ubiquitous with even greater potential impact.