1. Introduction

Raman spectroscopy is a noninvasive technique that has found wide chemical and biological applications by providing rich information about vibrational levels of molecules and material structures [1–3]. Coherent anti-Stokes Raman scattering (CARS) provides an enhancement of the spontaneous Raman signals by generation of the vibrational coherence, offering high temporal resolution, faster acquisition speed, and avoids the one-photon fluorescence [2–5]. CARS is a third-order nonlinear process in which the pump E(ω_p) and Stokes E(ω_s) fields generate coherence at Ω_R = ω_p − ω_s, probed by a third field E(ω_pr) to generate a blue-shifted CARS signal at ω_sig = ω_pr + Ω_R. Typically there is also a nonresonant four-wave mixing (FWM) background at ω_sig which masks and distorts the CARS signal. Various methods were developed to eliminate this background, including epi-detection [6], polarization control [7], heterodyne CARS [8, 9], nonlinear interferometric vibrational imaging [10, 11], Fourier transform CARS [12], and femtosecond adaptive spectroscopic techniques for CARS (FAST CARS) [13, 14]. Heterodyne CARS mixes a local oscillator (LO) field with the CARS signal and provides an improved shot-noise-limited detection in low-concentration measurements [15]. Most of these CARS schemes require splitting and recombination of ultrashort laser pulses, which is an experimental challenge in spatial and temporal alignment.

Single-beam CARS techniques eliminate the need for complicated alignment, generating CARS signals via coherent control of a broadband femtosecond pulse in a pulse shaper with the aid of a spatial light modulator (SLM) [16]. Other single-beam CARS schemes include polarization [17,18], time-resolved two color [19], binary phase [20], and spectral notch shaping [21,22]. Also, advancement in technology has allowed fast modulation speed and broadband excitation to address a wide range of Raman modes [23–25]. Incorporation of the conventional CARS technique into the single-beam regime has been achieved for heterodyne detection and spectral focusing with improved performance [26,27]. Heterodyne single-beam multiplex CARS was also demonstrated by dividing the femtosecond pulse into a broadband low-frequency pump/Stokes, a narrowband probe, and a high-frequency LO, and the CARS signal was retrieved via double quadrature spectral interferometry (DQSI) method [28].

Previously we demonstrated a single-beam FAST CARS spectroscopic technique, where the narrowband probe was delayed relative to the broadband pump/Stokes field to achieve the non-resonant background suppression at an optimal probe delay [29]. Instead of using the SLM, we translated the probe folding mirror using motorized actuator to achieve a large probe delay. The FWM background generated by the pump/Stokes field was used as the LO, which guaranteed stability [30]. However, the phase relation between the FAST CARS signal and the LO was not precisely controlled, leading to line-shape distortions. Also the background acquisition required long-range shift of the probe mirror actuator, which was time-consuming and not suitable for imaging.

Here we implemented a single-beam heterodyne FAST CARS microscopy by exploiting the phase relation between the signal and the LO. The energy level diagram of the heterodyne FAST CARS scheme is shown in Fig. 1(a), and the spectral intensities of the probe and pump/Stokes fields are plotted in Fig. 1(b). In addition to the large-scale delay controlled by the motorized actuator, we added a piezo actuator for fine tuning the delay. By modulating the probe mirror position locally within a fraction of the wavelength using the piezo actuator, we switched the probe delay between two close values τ₀ + δτ and τ₀ − δτ, and obtained two spectra I₊(ω) and I₋(ω), respectively. Here τ₀ denotes the central delay value and δτ is the modulation amplitude. By carefully choosing τ₀ and δτ, I₊(ω) and I₋(ω) will correspond to the constructive and destructive interference between the FAST CARS signal and the LO, as shown by the simulation data in Fig. 1(c). Then by using a modified DQSI method [18]:

 

Fig. 1 Schematic of the single-beam heterodyne FAST CARS. (a) Energy diagram of the FAST CARS scheme. (b) Spectral intensity of the excitation pulse with the probe and pump/Stokes parts marked with red and black lines, respectively. (c) Anti-Stokes signals I₊(ω) (red) and I₋(ω) (blue) simulated corresponding to delays τ₀ + δτ and τ₀ − δτ, respectively, with τ₀ ≈ 0. Inset shows the constructive (red) and destructive (blue) interferences between the FAST CARS signal and the FWM LO. (d) FAST CARS signal retrieved from the spectra in (c) using Eq. (1).

Download Full Size | PDF

2. Results and discussion

Results obtained from 1,1,2,2- tetrachloroethane (TeCA) are shown in Fig. 2, demonstrating the probe-delay effect on the retrieved Raman peaks. Figure 2(a) shows the spectrogram of TeCA acquired for different probe-delay times, in which we can see the tails of various Raman modes due to their long-lasting coherences. Nonresonant background was reduced at all delay values τ₀, and we plot three spectra at delays of 0 ps, 0.6 ps, and 1.2 ps in Fig. 2(b), and they are offset for visibility. For comparison, we also plot the zero probe delay spectrum from our previous work [29] in Fig. 2(c), in which the resonant signal is buried in the large nonresonant background. As the delay τ₀ was increased from 0 ps to τ₀ ≈ 0.6 ps, the estimated height of the main peak around 353 cm⁻¹ has increased by around 30%, and the full width at half maximum decreased from 32 cm⁻¹ to 21 cm⁻¹ compared to the zero probe delay signal. Other peaks showed similar behavior, and the effect is similar to the π-step-phased probe case in previous literature [31]. Also noticeable is the lineshape distortion to the retrieved peaks, with dispersive shapes forming on both sides of the 353 cm⁻¹ peak at 0.6 ps delay. When the delay was further increased to τ₀ ≈ 1.2 ps, the distortion became so large that the peak height information was lost, but the central peak became more distinct. One interesting point to note is that besides the main peak at 353 cm⁻¹, there is also a spectral structure appearing at 370 cm⁻¹. Its origin could be the 366 cm⁻¹ vibrational mode of TeCA, whose spontaneous Raman spectrum is shown in Fig. 2(d). We note that at large τ₀ values, the delay induces a spectral phase across the probe spectrum and modifies its interaction with the nonlinear susceptibility, which can lead to the appearance of the doublet structure that would otherwise be hidden as in the zero probe delay case.

 

Fig. 2 Single-beam heterodyne FAST CARS of TeCA using the forward detection: (a) 2D plot of the heterodyne FAST CARS spectra at different probe delays. (b) Heterodyne FAST CARS signal at 0 (black), 0.6 (red), and 1.2 ps (blue) delay. (c) FAST CARS signal of TeCA at zero probe delay from [29]. (d) Spontaneous Raman spectrum of TeCA. Spectra in (b) are normalized to the same factor.

Download Full Size | PDF

Next we carried out measurements probing coherence lifetimes of the Raman modes using fused silica. It was previously shown that fused silica contains a broad vibrational band, denoted as ω₁, due to an internal continuous random network with a short decoherence time [32]. The retrieved FAST CARS signals of fused silica at different probe delay times are shown in Fig. 3. We identify the ω₁ band as the broad spectral feature in the range between 200 cm⁻¹ and 440 cm⁻¹. As the probe was further delayed relative to the pump/Stokes, this band intensity gradually decreased and only the narrow peak due to the four-fold ring structure at 495 cm⁻¹ [32] was preserved. The coherence lifetime of the ω₁ band is estimated to be shorter than 1 ps due to the convolution of the FAST CARS signal with the probe temporal shape, which limits the temporal resolution to ∼1 ps.

 

Fig. 3 FAST CARS spectra of fused silica using the forward detection at different probe delays: (a) 0 ps, (b) 0.2 ps, (c) 0.4 ps, (d) 0.6 ps, (e) 0.8 ps, and (f) 1 ps.

Download Full Size | PDF

We used carbon tetrachloride (CCl₄) as the test sample to verify the power and concentration dependence. Because the raw signal intensities I_±(ω), generated by third-order nonlinear process, are proportional to the third power of the total input laser intensity I_±(ω) ∝ (Input power)³, then Eq. (1) the retrieved signal has a 1.5th-power dependence on the input power:

 

Fig. 4 Input power and concentration dependence of the CCl₄ FAST CARS signal measured in the forward detection. (a) Retrieved FAST CARS spectrum from CCl₄ at zero probe delay. (b) Log-log plot of the intensity of the 459 cm⁻¹ peak as a function of the laser power. (c) Intensity of the 459 cm⁻¹ peak measured at different concentrations of CCl₄ in acetone.

Download Full Size | PDF

Finally we demonstrate the epi-detection capability of the setup by imaging microscopic structures of the gold-coated silicon (Si) and molybdenum disulfide (MoS₂) flakes. The samples were scanned using piezo-driven stages, and characteristic peak intensities using 1 ps probe delay value at each spatial location were collected to obtain the FAST CARS images. Figures 5(a) and 5(d) are the microscopic optical images of the Si and MoS₂ samples, respectively, with marked scan areas. Figures 5(b) and 5(e) compare two FAST CARS spectra acquired at different pixels from the two samples, with one pixel containing the target substance and the other not. The characteristic Raman peak values were used for imaging: for Si the 520 cm⁻¹ Raman mode was used, and for MoS₂ the peak corresponding to the E_2g transition around 385 cm⁻¹ was used. The retrieved FAST CARS images are shown in Figs. 5(c) and 5(f), which are in good agreement with the optical images for both samples.

 

Fig. 5 Epi-detection FAST CARS microscopy of Si and MoS₂ microstructures. Optical images of the structured Si (a) and MoS₂ flake (d) samples, with scale bars of 2μm. FAST CARS spectra of Si (b) and MoS₂ (e) plotted in black lines compared to the background in red lines, with corresponding FAST CARS images in (c) and (f), respectively, using the Si peak at 520 cm⁻¹ and the MoS₂ peak at 385 cm⁻¹. The two spectra in (b) were acquired at (6.4 μm, 6.4 μm) (black) and (4.0 μm, 4.0 μm) (red) corresponding to the image in (c), and the two spectra in (e) were acquired at (4.6 μm, 3.8 μm) (black) and (1.0 μm, 2.2 μm) (red) corresponding to (f).

Download Full Size | PDF

The appropriate power assignment to the probe and pump/Stokes field could contribute to an enhanced signal level. For the experiment on TeCA described earlier, the total power on the sample was 8.3 mW, with a probe to pump/Stokes power ratio ∼ 1:75. By placing an ND filter in front of the pump/Stokes mirror to solely attenuate this part, we modified the probe to pump/Stokes power ratio to ∼ 1:9.3, and the total power on the sample changed to 7.8 mW. We found that for TeCA the signal-to-noise ratio of the primary retrieved peak at 355 cm⁻¹ is doubled as compared to the previous case using 8.3 mW total power but with a smaller probe to pump/Stokes power ratio. Because the LO is also dependent on the pump/Stokes field, careful consideration is needed to reach the optimal ratio between the probe and pump/Stokes power for the FAST CARS signal retrieval.

The limited sensitivity and acquisition speed of the current setup can be improved from various perspectives. Higher pulse energy lasers such as cavity-dumping oscillator [31] or high-repetition-rate regenerative amplifier system [11] could lead to an order of magnitude signal enhancement and therefore less acquisition time with better sensitivity. Also, the piezo actuator controlling the probe mirror can be operated faster than 10 kHz, so higher detection speed can be achieved by adopting fast acquisition detector like a frame-transfer CCD, scientific CMOS camera, or the recently devised TAMP array [33]. Another limitation concerns the bandwidth, that is, given a broader spectral coverage, the full fingerprint region can be addressed, therefore leading to broader applications in chemistry and biology. Currently we are limited to multiplex detection in the deep fingerprint region, but the setup can easily be modified for broadband sources such as sub-10 fs oscillators [27,28], or oscillator-pumped supercontinuum [9,25].

3. Methods

3.1. Samples

TeCA and CCl₄ samples were commercial products (SigmaAldrich), and fused silica sample was from UV-grade window material (Newport). The Si sample was prepared using the e-beam lithography on Si with the PMMA patterned mask followed by coating with 50 nm of gold; the MoS₂ flakes were prepared by exfoliation of bulk MoS₂ using tape onto a Si substrate.

3.2. Experimental setup

The sketch of our experimental setup is shown in Fig. 6. Details of the pulse shaper have been previously described [29]. A Ti:Sapphire oscillator (KMLabs, TS kit) was pumped by a green laser (Millenia eV, Spectra Physics) to produce 25 fs pulses centered around 810 nm, with a repetition rate of 85 MHz. A variable ND filter was used for attenuation, and followed by a chirp mirror pair for dispersion compensation. Then the high-frequency part of the pulse that coincides with the CARS signal was blocked by a long-pass filter (LP02-785RU-25, Semrock), and the pulse was sent to the pulse shaper where its spectrum was dispersed across the Fourier plane [34]. A major part of the pulse on the low frequency side was reflected by the mirror M₁ that served as pump/Stokes, and a narrowband portion was reflected by the mirror M₂, serving as probe. A right angle mirror M₃ was placed closely in front of M₂ such that the edges of M₂ and M₃ determine the bandwidth of the probe, which was chosen as 30 cm⁻¹. M₂ was mounted on a linear stage with a motorized actuator for coarse tuning τ₀ and a piezo actuator for fine tuning τ₀ and the modulation δτ. The following procedure was used to determine the relative position between M₁ and M₂: distilled water was used as the sample, and a large range of probe delays was scanned using the motorized actuator, with spectra taken at each increment of the stepper motor. Then by plotting the intensity at a specific wavelength on the signal curve versus the probe delay we retrieved a fringe pattern similar to an autocorrelation trace, resulted from the interference between the LO and the nonresonant signal generated by the probe. The envelope of this pattern indicates the temporal shape of the probe pulse, and we can determine the zero delay position according to this envelope. Then the probe delay value τ₀ can be calculated according to the motorized actuator shift from the zero delay position. Also, optimal modulation amplitude for the piezo actuator was found to be ∼190 nm to maximize the retrieved Raman peaks, which corresponds to δτ ≈ 1.3 fs.

 

Fig. 6 Experimental setup: DC, chirp mirror pair for dispersion compensation; LPF, long-pass filter; G, grating; CM, concave mirror; M, mirror; SPF, short-pass filter; L, lens; Obj, microscope objective; S, sample on X–Y translation stage; spec, spectrometer. The excitation and FAST CARS beams are shown by red and green colors, respectively.

Download Full Size | PDF

The reflected beams from M₁ and M₂ were tilted so that they could be guided out collinearly and sent to the microscope, where both beams were focused on the sample by an objective (N40X-NIR, Nikon). For transparent samples, the transmitted light was collimated by another objective (LPlan50X, Nikon) and sent through a short-pass filter (TSP01-790-25x36, Semrock) before entering the spectrometer (SureSpectrum, Bruker Optics) for detection; for opaque samples, back-scattered light was collected by the same focusing objective and reflected back towards the pulse shaper. After reaching the Fourier plane, the blue-shifted signal was guided by M₃ to the telescope consisting of the concave mirror CM₂ and the lens L, and imaged using the CCD camera (Spec10, Princeton Instruments). The focal lengths of CM₂ and L were 500 mm and 200 mm, respectively. The same short-pass filter was placed inside the telescope for filtering the residual excitation light. Both the forward arm spectrometer and the CCD in the epi-detection arm were calibrated by a spectral lamp, and the Raman shifts were calculated from the frequency shift from the probe central wavelength which was measured by the forward arm spectrometer.

TeCA, fused silica, and CCl₄ were transparent samples and were measured in the forward direction; Si and MoS₂ samples were imaged using epi-detection. The laser power on the sample without attenuation by the ND filter was 50 mW, and typical laser powers used on each sample were 9.8, 8.3, 19.2, 42, and 16.7 mW for CCl₄, TeCA, fused silica, Si, and MoS₂, respectively, except for the power dependent measurement. The CCD exposure time for all samples was fixed at 40 ms, and CCl₄, TeCA, and silicon spectra were averaged 10 times, while fused silica and MoS₂ spectra were averaged 100 times due to their relatively weak CARS signals. During the acquisition the CCD received a trigger frequency of 14 Hz while the piezo was driven at 7 Hz such that the two spectra were collected at constructive and destructive interference positions, from which the FAST CARS signals were retrieved using Eq. (1). The image acquisition times for Si and for MoS₂ shown in Fig. 5 were 12 minutes and 2 hours, respectively. For comparison, spontaneous Raman spectra were acquired using a commercial Raman system (LabRAM HR Evolution, Horiba).

3.3. Theory

We denote E(ω) as the excitation laser field, which is separated into the probe field E_pr (ω) with a bandwidth of δω, and the pump/Stokes field E_ps (ω), as shown in Fig. 1(b). For the transform limited pulses, the electric fields are taken to have real values. The nonlinear polarization is given by [3]:

When a time delay τ relative to the pump/Stokes field is added to the probe, it is equivalent to adding a linear spectral phase to E_pr (ω) in the frequency domain, and Eq. (3) is modified as:

The above equation relates the detected CARS signal to the function f(ω). A closer look at this function in Eq. (7) reveals that it is essentially the susceptibility χ(Ω) gated by the probe with phase E_pr (ω − Ω)e^{−i(ω−Ω)τ₀} and a large smooth envelope function A(Ω). For ideal probe shape E_pr (ω) assumes the form of a delta function with infinitely small bandwidth so that f(ω) reproduces the shape of χ(Ω). In reality E_pr (ω) always has a finite bandwidth, leading to a compromised resolution. In the FAST CARS case here it is further complicated by the phase factor e^{−i(ω−Ω)τ₀}. Using Eqs. (7) and (11) we can express I_CARS as:

Next we consider the neglected imaginary part of the LO field $P_{\mathit{FWM}}^{\mathit{ps}}(\omega )$, which is due to the contribution from χ_R(Ω) in the expression for $P_{\mathit{FWM}}^{\mathit{ps}}(\omega )$. Consider the expression $P_{\mathit{FWM}}^{\mathit{ps}}(\omega )=\left|P_{\mathit{FWM}}^{\mathit{ps}}(\omega )\right|e^{i\theta (\omega )}$, where θ(ω) denotes the spectral phase of the LO field. Then the polarizations P₊(ω) and P₋(ω) change correspondingly, leading to a modification to the difference in Eq. (10):

The simulated curves in Figs. 1(c) and 1(d) were obtained by setting the Raman modes as Ω₁ = 300 cm⁻¹, Ω₂ = 650 cm⁻¹, Ω₃ = 1000 cm⁻¹, with Γ₁ = 15 cm⁻¹, Γ₂ = 10 cm⁻¹, and Γ₃ = 20 cm⁻¹. The probe pulse width was δω = 30.5 cm⁻¹, and the delay parameters τ₀ = 0.06 fs and δτ = 0.65 fs.

4. Conclusion

In conclusion, we demonstrated single-beam heterodyne FAST CARS microscopy using the temporal control of the probe delay inside the pulse shaper. Phase shifting between the signal and LO was achieved using real-time piezo modulation, and the nonresonant background was reduced using hetero-dyne detection, allowing for imaging applications as well as the study of the dynamic evolution of FAST CARS signals at various probe delays. We also improved the epi-detection design by incorporating spectral detection into the pulse shaper, which is more efficient and compact compared to previous work [22,29].