Achieving ultra-high sensitivity is one of the major goals and challenges of any detection scheme. In the case of chemical sensors aiming at single molecule detection, advanced schemes based on the measurement of the photothermal shift [11], the evanescent wave amplitude in nanowires [12], the plasmonically enhanced scattering [13,14], fluorescent behaviors [9,15] and the mode splitting at the exceptional point [16,17] have been proposed. However, detection of individual gas molecules using integrated photonic devices is still a challenge because of the stringent requirements in terms of spectral resolution and ultralow noise. Dual-comb spectroscopy based on Kerr or Raman solitons [18] naturally has narrow linewidth and high frequency stability and could thus be the ideal tool to achieve individual molecule detection on microscale photonic devices and on-chip. However, the inert nature of the materials (silica, silicon nitride or metal fluorides) that are typically used for soliton microcomb devices inhibits gas adsorption and sensing applications. In this context, chemical functionalization could significantly expand the capability of dual-comb devices for sensing applications. Here we demonstrate that functionalization of an over-modal microresonator with a single layer of graphene allows to realize a dual-comb spectrometer with unprecedented chemically sensitivity. Leveraging such graphene functionalized dual-comb microresonator, we successfully achieve individual gas molecule detection at high-speed. Such graphene-microcavity device, which has been used for optoelectronic tuning of frequency comb generation [19,20,21,22], was never used in comb spectroscopy to date. Our results stem from the precise spatial positioning of the graphene flake on the microcavity to create a unique ‘reference-probe’ dual-comb system. In particular, by depositing graphene 30o away from the equator of the microcavity we generate one Stokes soliton interacting with graphene (probe) and one Stokes soliton from the pristine microcavity (reference). Molecule adsorption on graphene changes its Fermi level [19,23] and thus modifies the free-spectrum-range of the probe comb. Finally, taking advantage of an advanced optoelectronic heterodyne detection scheme, we trace the frequency shift of the dual-soliton beat-note with uncertainty < 0.2 Hz to detect single molecule dynamics.
Fig. 1a shows the conceptual design of the graphene based micro dual-comb device (GMDC). A silica microsphere with diameter ≈ 600 μm and typical intrinsic Q factor ≈ 3×108 is used for the Kerr and Stokes soliton generation. Thanks to its large mode volume, such a microsphere supports multiple transverse co-oscillating intracavity modes, driven by one single pump laser. This enables different soliton frequencies to be generated simultaneously either in a low-order mode (blue arrow) or in a high-order mode (yellow). In this architecture, a mechanically exfoliated graphene flake is deposited on the surface of the microsphere by deterministic dry-transfer [24]. The position of graphene is carefully controlled to be 30o above the equatorial plane, to ensure the overlap only with the high order modes (with wider energy distribution) of the microcavity. On the other hand, the fundamental mode, which distributes tightly along the equator, will not be affected by the presence of graphene. Such scheme will provide both the “probe” (graphene functionalized) and “reference” (pristine) combs, as we show in the following. Moreover, positioning the graphene away from the equator also prevents its damage and heating when the intracavity power is high (up to tens of watts). Fig. 1b shows a top-view optical microscopy picture of our device and a scanning electron microscopy image where the atomically smooth surface of the silica microresonator and the 80 × 30 μm2 graphene layer are clearly visible. More details about the transfer method and characterization of the device are available in the Supplementary Notes S3 & S4.
Fig. 1c illustrates the measured intracavity intensity evolution of the Kerr and Stokes combs, which are traced synchronously by using a C+L/U band wavelength division multiplexer. When red scanning the pump laser (at fixed power 200 mW) from 1549 nm to 1550 nm with a scanning speed of 500 MHz/ms, a Kerr comb begins to form and creates Raman amplification. To observe both Kerr and Stokes solitons, the following four conditions must be satisfied: 1) the Stokes soliton lines must lie within the Raman gain spectrum generated by the Kerr soliton; 2) the FSR of the Stokes solitons must be close to that of the Kerr soliton; 3) the mode family of Kerr soliton and Stokes solitons overlap efficiently in both space and time; 4) the pump power reaches the Raman threshold. In this way, the Stokes solitons rely on the existence of the Kerr soliton due to the spatial-temporal overlap for Kerr effect trapping and Raman amplification. In Fig.1c one can also observe that the Kerr soliton step appears earlier than the Stokes soliton step. Further theoretical analysis is available in Supplementary Notes S1 and S2. Fig. 1d plots the optical spectrum of the co-generated Kerr and Stokes solitons. The Kerr soliton spans from 1500 nm to 1600 nm, while the multiple Stokes solitons are generated in the band from 1650 nm to 1700 nm. Although their central wavelengths are different, the Stokes solitons shows FSRs similar to those of the Kerr soliton. In the zoomed-in panel, we also observe that the excited Stokes solitons have distinct comb envelopes, since they belong to different mode families. This means that one Kerr soliton can trap many Stokes solitons thanks to the over-modal nature of the microresonator. As a consequence, these Stokes solitons can beat with each other, offering a powerful tool for dual-comb spectroscopy in the electrical domain.
Fig. 2a maps the frequency-resolved auto-correlation traces of the Kerr soliton and the Stokes solitons based on second-harmonic generation (SHG). First, in the C band (1550 nm), the pulse structure with 10.24 ps interval clearly suggests the existence of a single Kerr soliton. Based on the autocorrelation trace, the measured pulse duration of the Kerr soliton is 350 fs, as expected from the 3 dB spectral range of 0.93 THz. On the other hand, in the U band (1670 nm), the signal to noise ratio of the auto-correlation trace is lower because the Stokes solitons contain pulse trains with different repetition rates. These Stokes solitons with different repetition rates can interfere leading to a frequency down-conversion in the radio frequency domain. Fig. 2b plots the beating signal of a pair of dual Stokes solitons. This beat note can’t be due to the Kerr-Stokes interaction, as they don’t overlap in frequency (photon energy). The dual Stokes soliton beating provides an electrical comb with 7.514 MHz spacing, such a frequency difference is more than 4 orders of magnitude smaller than the soliton FSR. (in Supplementary Note S5 we also show that the dual soliton modes with 7.514 MHz FSR difference have orthogonal polarizations.). The zoomed-in panel shows more details on the dual Stokes soliton beating. The signal-to-noise ratio (SNR) of the first beating line is > 55 dB and its spectral linewidth is < 10 Hz. Moreover, Fig. 2c shows the measured single-sideband (SSB) phase noise of this beat note and reveals that the phase noise is < -130 dBc/Hz at 10 kHz, and < -140 dBc/Hz at 1 MHz. This result could be further optimized by using active feedback [22,25,26]. In Fig. 2d, we measure the long-term stability of the 7.514 MHz beating signal. For a continuous measurement of 2 hours at room temperature the frequency shift is < 2.5 Hz and the intensity variation is < ±0.1 dB. Such a high stability with uncertainty at the Hz level offers a unique platform for gas sensing applications.
Fig. 3a shows a schematic of our gas sensing device for individual molecule detection. In the microresonator, the reference Stokes soliton at the equator has a FSR fR, while the probe Stokes soliton interacting with the graphene flake has a FSR fP, and their beating down conversion frequency is Δf = fR - fP. Once gas molecules are adsorbed on the graphene flake the refractive index experienced by the probe soliton will change, leading to a change of fP. As a result, we will detect a shift of the beating frequency Δf, that we can measure in the electrical domain to obtain the gas dynamics with individual molecule sensitivity. Fig. 3b explains the optical refractive index change induced by adsorption of individual gas molecules on graphene. The Fermi level of graphene is related to the carrier density by |EF| ≈ ℏ|vF|(πN)−1/2 [27], where N is the carrier density, vF ≈ 106 m/s is the Fermi velocity, and ℏ is the reduced Plank’s constant. In this equation, N includes both the intrinsic doping and the external doping, determined in our case by the gas adsorption. Tuning of the Fermi level affects the refractive index of graphene, as explained in details in Refs [19,28] and in the Supplementary Note S6. In the right panels of Fig. 3b, we show the calculated real and imaginary refractive index spectra of the graphene (Re{ng} and Im{ng}) for different values of the Fermi level and wavelengths in the range 1500 nm – 1700 nm. For our application, at the wavelength ≈ 1670 nm an increment of the Fermi level from 0 eV to 0.31 eV induces a change of the Re{ng} from 2.75 to 3.25, thus approximately 1.61/eV. In this work, we use ammonia (NH3) molecules as a sensing target [23,29]. The adsorption of a single NH3 molecule provides 2 additional electrons to graphene. Considering that the area of the (p-doped) graphene flake is 1.6×10-9 m2, the adsorption of a single NH3 molecule will induce a |EF| reduction of 1.7×10-4 eV. As a result, Re{ng} will decrease by 2.74×10-4 inducing an increase of the probe comb mode FSR and a spectral shift of the reference-probe comb beat note.
Fig. 3c shows the measured shifts of the reference-probe comb beat note for different values of NH3 concentration in the pM/L range. For this experiment, we put our GMDC device in a vacuum chamber with volume of 8 L (the setup is shown in the Supplementary Note S7) and we inject pure NH3 gas in the vacuum chamber, controlling the minimum concentration of the NH3 in steps of 0.5 pM/L. When the NH3 concentration is 0.5 pM/L, 1 pM/L, 2 pM/L, 4 pM/L, and 8 pM/L, we record dual-comb beat note shifts of 85 Hz, 159 Hz, 203 Hz, 230 Hz, and 248 Hz respectively. Fig. 3d summarizes the performances of our sensing device. The maximum sensitivity reaches 170 Hz/(pM/L) in the 0 ~ 0.5 pM/L region, while a higher in-chamber gas concentration gives lower sensitivity due to saturation of the gas adsorption. In repeated measurements, the gas molecules attached on graphene can be > 99% released simply by heating the device via an electrical heater. Moreover, after repeated use the sensitivity of the dual-comb sensor does not deteriorate.
To achieve single molecule detection, we further implement a heterodyne lock-in amplification scheme, as shown schematically in Fig. 4a. We use a signal generator to produce a RF line with a stable frequency of 7.464 MHz to beat the 7.514 MHz signal (Δf), thus forming a new 50 kHz frequency (ΔfD) that falls within the bandwidth of the lock-in amplifier (125 kHz, Stanford Research SR 810) used for our experiments. Any shift of the dual-comb beat note will induce a shift of ΔfD. We then used a narrow filter (bandwidth 0.3 Hz) on the lock-in amplifier to lock and amplify only the amplitude at 55 Hz. Since the 50 kHz ΔfD beat note has a linewidth of 10 Hz (defined by the spectral linewidth of the dual Stokes soliton beating) with an electrical intensity 1 mV, the frequency shift dependent amplitude change reaches 0.1 mV/Hz in sech2 approximation. Such intensity variation could be further amplified by 30 dB. Considering that the resolution of the oscilloscope used for our measurements (see Supplementary Fig. S7) is 0.01 mV, we obtain a theoretically maximum resolution of 10-4 Hz. More details on the experimental implementation are shown in Supplementary Fig. S7.
Finally, we inject 0.08 pM NH3 into the chamber (corresponding to concentration of 0.01 pM/L in chamber) and we measure the dual-comb response (Fig. 4b). Following interaction between the graphene flake and NH3, the lock-in amplified intensity increases from 0 to 34 mV on a time-scale of approximately 100 ms, defined by the gas diffusion process in vacuum. When we zoom-in the gas response trace of Fig. 4b, we observe clear steps caused by individual molecule adsorption/desorption (Fig. 4c). Before injecting the NH3 gas, the trace is uniformly flat and there is no evidence of any molecular on/off case (state i). Once the NH3 gas is injected in the chamber, we observe that the intensity curve increases in small steps, suggesting that adsorption of individual molecules occurs (state ii). When the interaction between the graphene and the NH3 gas reaches the dynamic balance, the intensity curve becomes flat again, although we can still observe on/off steps due to microscopic molecular thermal motion. Fig. 4d shows that the height of all the observed steps are multiple integers of 0.2 mV, which is the smallest number corresponding to individual molecule adsorption. This is another strong evidence that individual molecule dynamics can be detected with our device. We also count the molecular on-off cases occurring within 200 ms after the injection of the gas in the chamber. Gas adsorption dominates during the state ii, and the on-off cases are balanced in the state iii. Moreover, when the GMDC is exposed to NH3, the large steps are rare (such as > 2 molecular on/off events), whereas unit steps are dominant. These statistical results obey a power-law distribution, which is also a sign of individual molecule adsorption events [10].
To conclude, dual-comb spectroscopy is demonstrated in a graphene functionalized over-modal microresonator. Stokes soliton combs with ≈ 1670 nm central wavelength in distinct mode families are co-generated and trapped by the Kerr soliton in the communication band. By placing graphene away from the equator of the microresonator, we managed to simultaneously generate one Stokes soliton interacting with graphene and another Stokes soliton interacting only with the pristine cavity. The former is used as a probe and the latter as a reference for gas sensing. To this end, we measured the probe-reference intermode dual-comb beating signal in the in the RF spectral region and thanks to a heterodyne lock-in scheme we achieved sub Hz spectral resolution and individual molecule sensitivity. This scheme offers a label-free optical tool to realize individual gas molecule detection. Such a compact device not only demonstrates a unique potential for chemical sensing, but also paves the way to design novel microcomb devices for applications ranging from radio signal generators, frequency modulators and to spatial rangefinders.