1. Introduction

Ultrafast spectroscopy provides insight into molecular dynamics on the femtosecond timescale. Coherent two-dimensional (2D) spectroscopy [1] is a generalized version of transient absorption spectroscopy with frequency resolution for both the pump and the probe step. The method detects the third-order nonlinear response of the system under investigation and is often carried out on liquid-phase solutions, though it has been demonstrated also for other sample types. Examples for investigated systems are molecules [2, 3], quantum dots [4], nitrogen-vacancy centers [5], thin molecular films [6–8], nanodots [9–12], nanocrystals [13], or carbon nanotubes [14]. Most geometries detect a coherently emitted four-wave-mixing signal following three-pulse excitation, but incoherent population-based observables can also be used as has been demonstrated for fluorescence [15–19], mass spectroscopy [20], or electron currents [21–24]. The latter offers the possibility for nanometer spatial resolution when measured via photoemission electron microscopy (PEEM) [21,23]. With all-optical methods, a spatial resolution of about half of the wavelength of the exciting electromagnetic field can be achieved according to the diffraction limit [25]. Ultrafast spectroscopy in the focus of microscope objectives has been successfully demonstrated [26–29], even down to the limit of single molecules in linear [30–33] and in nonlinear studies [34–37].

In general, 2D spectroscopy can be conducted in different geometries as reviewed recently [38, 39]. In box geometry, the three exciting laser pulses are overlapped on the sample from different directions and the nonlinear signal is collected in a fourth direction, given by the phase matching of the excitation wave vectors and allowing background-free detection. Especially for existing transient absorption experiments, a more convenient possibility is using the pump-probe geometry and splitting the pump pulse into two collinear pulses via a pulse shaper [40]. Another option is a fully collinear geometry, where all excitation pulses share a common direction, often used in experiments where the response is probed by incoherent observables [15–24]. In collinear geometry, the different linear and nonlinear response contributions can no longer be separated spatially by phase matching because they are not emitted in distingushiable directions from the sample, and anyway, for incoherent observables phase matching does not exist. Instead, phase cycling can be utilized, where the difference phases between the individual laser pulses of a sequence are modulated [41] and the nonlinear contributions (e.g., the rephasing and the non-rephasing signal) are retrieved by linear superposition of differently phase-modulated raw data [16,18–20]. Pulse shapers used for pulse sequence generation are inherently phase-stable, which is an advantage for these geometries.

In this work, we combine fluorescence-detected 2D specroscopy [19] and high-numerical-aperture (NA = 1.4) optical microscopy [42,43] to establish the novel method of coherent 2D fluorescence micro-spectroscopy. A collinear pulse-shaper-generated pulse sequence is directed to chromophore systems in the diffraction-limited focus of a microscope and the resulting fluorescence is collected as a function of inter-pulse time delays and phases. With this all-optical approach, spatial variations of the nonlinear third-order response function can be detected. Such variations might be due to heterogeneities in the surface morphology or mixtures of different chromophores. This makes the method ideal for the investigation of material systems used in opto-electronic devices such as solar cells or light-emitting diodes.

2. Experiment

An overview of the experimental setup is shown in Fig. 1(a), initially developed for linear spectral-interference microscopy [42,43] and used for the investigation of propagation effects in nanoplasmonic systems [44–46]. Modifications on the previously reported design have been made for the present new application of nonlinear microscopy, in particular to deal with sample degradation in an optimal way and to implement fluorescence detection as discussed now. The spectral bandwidth of the femtosecond oscillator (VENTEON Laser Technologies GmbH, Pulse One PE) ranging from 650 nm to 950 nm wavelength is confined by hard apertures in the Fourier plane of a 4f-based pulse shaper [Fig. 1(a)], in front of the liquid-crystal display (LCD, JENOPTIK Optical Systems GmbH, SLM-S640d). This way, the apertures act as a long-pass (LP) filter at 671 nm and a short-pass (SP) filter at 828 nm wavelength. In combination with a Schott KG5 color filter, this results in the spectrum of Fig. 1(b). A smooth shape ensures the absence of pronounced side peaks and other irregularities in the temporal pulse profile, which is crucial to avoid artifacts in spectroscopic measurements. The spectrum in Fig. 1(b) is measured between the dichroic beam splitter (DBS, AHF Analysentechnik, F48-810) and the microscope objective (Nikon Plan Apo, 100 × /1.40). For temporal pulse compression, the fixed phase correction of a pair of chirped mirrors (VENTEON Laser Technologies GmbH, DCM7) is combined with pulse-shaper phase correction. A two-photon photodiode (TPPD) is placed in the focus of the microscope objective to generate a nonlinear feedback of the laser pulse peak intensity. By utilizing the PRISM algorithm [43,47], the peak intensity is maximized, leading to a transform-limited laser pulse. To measure the pulse duration, a pair of (compressed) pulses is created by the pulse shaper and the normalized signal from the TPPD is recorded in dependence on the inter-pulse delay τ. Figure 1(c) shows the resulting interferometric autocorrelation. From the Fourier-filtered trace, the pulse duration at the sample position can be determined to be 12.1 fs (FWHM) assuming the experimental spectrum of Fig. 1(b). The measured result is close to the theoretical limit of 11.8 fs calculated using the experimental spectrum with a flat phase.

 

Fig. 1 Experimental basics. (a) Overview of setup for 2D fluorescence micro-spectroscopy. Chirped mirrors and a spatial light modulator are used for temporal pulse compression, the latter additionally for generation of pulse sequences. The laser spectrum is confined by longpass (LP) and shortpass (SP) apertures in front of the liquid-crystal display (LCD) and by a Schott KG5 color filter in front of the dichroic beam splitter (DBS). Excitation of the sample and collection of the fluorescence is conducted by a high-numerical-aperture (NA) microscope objective. Separated from scattered light by an additional LP filter, the fluorescence is detected either by an avalanche photodiode (APD) or a spectrometer. (b) Laser spectrum measured after reflection from the DBS. (c) Collinear interferometric autocorrelation (blue) recorded by placing a two-photon photodiode at the sample position. From the Fourier-filtered curve (red), the duration of a single pulse is determined to be 12.1 fs (FWHM). (d) Four-pulse sequence. The inter-pulse delays are defined by τ, T and t. Each pulse has an adjustable phase Φ_i (i = 1, . . ., 4).

Download Full Size | PDF

All measurements using ultrafast laser pulses presented in this manuscript were done with 12.1 fs pulse duration by applying the PRISM phase to the pulse shaper in addition to the mask pattern required for the pulse train. In the case of four-pulse sequences, the inter-pulse delays t, T and τ of the compressed pulses were controlled by the pulse shaper as well as the phases Φ_i (i = 1, . . ., 4) [Fig. 1(d)]. The overall absolute carrier-envelope phase (CEP) of the pulses, i.e., the zero-order coefficient of the Taylor-expanded temporal phase, is not controlled. Manipulating the relative (CEP) inter-pulse phases Φ_i (i = 1, . . ., 4) by the pulse shaper is sufficient. Applying a phase mask to the LCD-based pulse shaper takes about 1.2 s, mostly because of the physical rotation of the liquid crystal of the LCD array itself. On the other hand, acquisition times on an avalanche photodiode (APD) as used in our work can be set as short as 50 ms, still offering an acceptable signal-to-background ratio for samples with high fluorescence yield. This would correspond to a duty cycle of only 4 %, resulting in a high degree of sample photobleaching due to unnecessarily long exposure times. We therefore implemented a circular step-motor-based shutter [Fig. 1(a)] with a response time below 20 ms, increasing the duty cycle by a factor of 7 in this example and thus greatly reducing the photobleaching for a fixed acquisition time.

The laser focus in the microscope is mapped by a piezo scanning stage (P-517.3CL, PI, Germany). The reflected light is collected by the same objective, transmitted through a dichroic beam splitter (DBS), and detected by either an APD (Perkin Elmer, SPCM-CD 2801) or a spectrometer [spectrograph (Princeton Instruments, Acton SpectraPro 2500i) and charge-coupled device (e2v, CCD42-10 in Princeton Instruments, Acton Pixis2kB)]. The reflection intensity maps [Fig. 2(a)] provide a first optical characterization of the sample. By adding an additional emission filter (EF, AHF Analysentechnik, F76-832), a map of the sample fluorescence is generated in the same manner [Fig. 2(b)]. By these scans of the sample plane, sub-micron-sized features of geometrical and material composition can be distinguished at a lateral resolution of 260 nm according to the Abbe diffraction limit [λ / (2 × NA)]. The absorption of the thin film samples is separately measured with a commercial spectrometer (Jasco Deutschland GmbH, V670).

 

Fig. 2 Microscopic characterization of the in-plane structured F₁₆ZnPc pillar array. (a) Reflection intensity map, showing low reflectivity in regions where the regular hexagonal pattern is perturbed. (b) Fluorescence map of the same area as in (a). The regions of low reflectivity show the highest fluorescence yield due to the thick chromophore layer coverage. (c) AFM map from a homogeneously structured region of the same sample (but not necessarily the same area), indicating the defects in the lateral pattern of chromophore pillars. Scalebar 1000 nm.

Download Full Size | PDF

3. Sample

As reference system we chose the fluorinated zinc phthalocyanine (F₁₆ZnPc) dye to demonstrate the capability of the newly developed method of 2D fluorescence micro-spectroscopy. The dye is directly evaporated on glass slides, resulting in a homogeneous layer growth of F₁₆ZnPc [48]. Previous nonlinear measurements of an approximately 300 nm thick F₁₆ZnPc layer (not shown) confirmed the onset of a nonlinear response at about 1 GW/cm² peak excitation intensity in the center of the focus during pulse duration (averaged over the temporal FWHM). Furthermore, the thin films have proven to be sufficiently photostable for measurements lasting up to 40 h at a peak excitation intensity on the order of 100 GW/cm², corresponding to an average intensity on the order of 100 kW/cm² at the center of the focus. While the nonlinear response of the sample scales with the peak intensity, i.e., the inverse pulse length, additional measurements with uncompressed pulses have shown that photostability is mainly dependent on the average intensity. This indicates heating of the sample through absorption as the dominant bleaching mechanism rather than ultrafast photodamage.

As can be seen in the reflection image of Fig. 2(a), a regularly structured sample surface was obtained by evaporating the dye on a monolayer of spincoated polystyrene nanospheres of 500 nm diameter (Polysciences Europe GmbH) and subsequent lift-off of the spheres [49]. Thus, triangular pyramids of F₁₆ZnPc of an approximate size of 200 nm remain on the glass slide and form a hexagonal ordered lattice indicating the positions initially not covered by the spheres. Defects in the sphere monolayer led to fissures in the regular F₁₆ZnPc nanopillar arrangement, recognized by the reduced reflection in this region [Fig. 2(a)]. Looking at the fluorescence of the same sample regions [Fig. 2(b)], the locally constricted dye coverage is verified by the existence of intensity hotspots. Thus, one can choose between a more confined dye pyramid inside the regular pattern or a brighter hotspot offering a better signal-to-background ratio. The latter makes it possible to conduct nonlinear measurements with lower excitation power, decreasing photobleaching. The surface structure was confirmed by AFM measurements [Fig. 2(c)]. More detailed information on the preparation and characterization of equivalent thin films is provided by Kolb et al. [49].

Figure 3(a) shows the laser spectrum (blue) in comparison to the absorption of plane and nanostructured F₁₆ZnPc films. The origin of the absorption features of a homogeneous film (black) has been reported before [48] as resulting from two coexisting crystalline phases at room temperature. The laterally structured thin film of Fig. 2 shows a pronounced absorption around a wavelength of 600 nm [red in Fig. 3(a)]. However, this cannot directly be attributed to the crystalline structure of the microscopic pyramids since the absorption measurement averages over several mm² consisting of several domains. The surface contains areas with homogeneously ordered dye pillars and areas still covered by nanospheres. The latter remain after the top layer is removed by liftoff in areas where a multilayer of spheres formed in the course of spincoating. We attribute the broadband contribution over the whole shown spectral range to absorption and scattering from these remaining nanospheres. The absorption of the dye itself shows only slight modulation in the range of the laser spectrum.

 

Fig. 3 (a) Absorption of a nanostructured (red) and a homogeneous (black) F₁₆ZnPc thin film in comparison to the laser spectrum (blue). (b) Excitation-power-dependent fluorescence of the F₁₆ZnPc nanopillar array, showing an onset of the non-linear response above approximately 1 GW/cm² peak excitation intensity, deviating from the linear trend (black line). Each color corresponds to an individual measurement with a different set of neutral density filters where a half-waveplate is rotated in front of a linear polarizer to scan the attenuation.

Download Full Size | PDF

The nonlinear response needs to be above a certain level to allow reliable separation from the linear counterparts and the noise floor by phase cycling, since they are not geometrically separated when using a fully collinear measurement. To determine this level, we conducted a series of excitation power scans by rotating an achromatic half-waveplate (B. Halle Nachfl. GmbH) in front of a fixed linear polarizer. Figure 3(b) shows the resulting fluorescence response of a diffraction-limited hotspot on the previously described nanopatterned F₁₆ZnPc surface in dependence on the peak excitation intensity. We observe a linear relation for peak intensities up to the order of 1 GW/cm², similar to the case of homogeneous layers. In such low-power measurements, one would expect to see no nonlinear signal contributions above the noise floor after phase-cycling in 2D measurements. On the other hand, this means that if there are signals in the experimental data after phase cycling, they can possibly be identified as measurement artifacts, e.g., due to imperfections in the pulse-sequence generation. We use such a procedure to remove artifacts from the nonlinear response maps in high-power measurements, as shown in Section 4.

The 2D measurements are performed by scanning coherence time τ and signal time t in 2 fs steps from 0 fs up to 60 fs, keeping the population time T fixed at 30 fs. Several factors influence the chosen total interval of the coherence time axes. The controllable time window with the Fourier-domain pulse shaper is proportional to the number of SLM pixels, in our case ca. 2 ps. For a scan with (60 fs, 30 fs, 60 fs) for the three time delays we stay significantly below that limit. Longer time scans are possible, however, if the sample under investigation shows dynamics exceeding the chosen intervals or if a higher frequency resolution is desirable. For the system at hand, the parameters have been optimized to balance the requirements between frequency and time resolution, signal-to-noise ratio, and overall exposure time to reduce photobleaching.

The measurement was conducted in the partially rotating frame, where the observed frequencies ω^rot = ω^lab − Δω are downshifted from the frequencies in the laboratory frame ω^lab [1]. We use a shifting frequency Δω = 0.7 ω₀, where ω₀ = 2.59 rad/fs is the fundamental frequency of the laser spectrum used for excitation. The frequency shift is introduced to avoid undersampling when using time steps of 2 fs. Measuring in the fully rotating frame (Δω = ω₀) would in principle also be possible. In that case, however, the desired nonlinear signal would appear at an effective frequency of zero and would overlap with other, unwanted, contributions. Thus it is advantageous to choose a partially rotating frame. In our case we selected a shifting frequency of Δω = 0.7 ω₀. This introduces a spectral offset large enough to separate the whole range of the excitation spectrum from stationary contributions avoiding artefacts. The spectral separation is crucial for both identification and isolation of the nonlinear part of the measured data.

A 27-step (3 × 3 × 3) phase-cycling scheme is used, allowing the extraction of the absorptive spectrum from the rephasing and non-rephasing nonlinear contributions [1,41]. For each time step, the 27 different phase variations are measured as direct sequence with an averaging time of 1 s on the APD, followed by a reference measurement consisting of a single excitation pulse. This way, the advance of photobleaching is determined for each time step and accounted for by using the inverse intensity of the single pulse as a normalisation factor for the 27 preceding four-pulse sequences. This correction is valid as long as a reasonable fraction of the chromophores in the center of the focus, i.e., in the region of the highest optical fields and therefore nonlinear response, is still intact. For further advance of the bleaching (approximately below 20 % of the initial fluorescence yield), the nonlinear response can no longer be reliably extracted.

According to the power scans in Fig. 3(b), the nonlinear scans should be measured above approximately 1 GW/cm² peak intensity. It would be desirable to use much higher powers, but the sample photostability sets an upper practical limit. For peak intensities on the order of 100 GW/cm², corresponding to roughly 100 kW/cm² average intensity when using a single laser pulse, we observe a slow exponential decay of the fluorescence yield on the time scale of multiple hours which is accounted for in our nonlinear measurements by intensity correction according to the reference pulses. When conducting nonlinear scans, we use a peak excitation intensity of 65 GW/cm² (251 kW/cm² average intensity, both values for temporally separated sub-pulses in a four-pulse measurement), offering a good compromise between nonlinear signal-to-noise ratio and photobleaching stability of the sample at hand.

4. Results and discussion

We performed a four-pulse fluorescence measurement on a single diffraction-limited hotspot on the laterally structured F₁₆ZnPc film at a peak intensity of 65 GW/cm² (for temporally separated sub-pulses). The time-domain data was normalized to the recorded fluorescence intensity without pulse overlap, i.e., using regions from all 27 maps where both τ ≥ 40 fs and t ≥ 40 fs. The phase-cycled and Fourier-transformed 2D spectrum (absolute value) is shown for the rephasing [Fig. 4(a)] and the non-rephasing [Fig. 4(b)] contributions. The spectra are shown in the partially rotating frame as function of the observed frequency ω^rot downshifted by Δω = 0.7 ω₀. The major contribution is located along the anti-diagonal for the rephasing contribution [Fig. 4(a)] and along the diagonal for the non-rephasing contribution [Fig. 4(b)] spanning over the full laser spectrum. Additional features are seen along the diagonal lines and along a vertical line for ${\omega }_{\tau }^{\text{rot}}\approx 0$. When repeating the measurement at a low peak excitation power of 3.5 GW/cm² in the dominantly linear regime, as described in Section 3, these features remain and can thus be identified as linear artifacts both in the rephasing [Fig. 4(c)] and the non-rephasing [Fig. 4(d)] map.

 

Fig. 4 Zero-padded rephasing (upper row) and non-rephasing (lower row) 2D spectra before the final processing step. The high-power measurements (a,b) contain the expected nonlinear signals in the lower right (a) and lower left (b) quadrants. Additional artifacts occur in other positions. For low-power measurements (c,d), only the linear artifacts remain, allowing for an artifact correction. The correction is done by subtracting the measured low-power from the high-power maps in the time domain (not shown). The corrected 2D maps are then retrieved by Fourier transformation, from which the absolute values are shown (e,f).

Download Full Size | PDF

The artifacts are primarily caused by pulse-shaper imperfections, both inherent to the LCD-type shaper design as well as the voltage-to-phase calibration of the LCD array. More precisely, transparent gaps between the LCD pixels result in a fraction of the light transmitted through the pulse shaper without control over the applied phase. Furthermore, especially for neighboring pixels with large differences in the applied voltages, crosstalk shifts the actually applied phases for those pixels. The voltage crosstalk can be reduced, but not fully avoided, by setting the working range to low voltages. Additionally, the design of the LCD array results in reflections at the interfaces between the glass, the transparent electrodes and the liquid crystal layers because of differences in the refractive indices. Since the exact optical pathlength of the resulting Fabry-Pérot cavity varies for every applied phase and thereby, for every single pixel depending on the applied pulse shape, the resulting deviations in shaped phase and amplitude cannot easily be compensated for by the pulse shaper calibration. Therefore, we scanned the voltages of one LCD layer in intervals around the values for maximal and minimal transmission found by the initial phase calibration for different absolute phase values of the other LCD layer. Systematic deviations between expected and measured voltage-to-phase dependencies could thereby be individually corrected for each pixel, minimizing the average deviation. A full correction would include iterative optimization for every possible phase combination and thereby pulse shape with measurement of the complex electric fields, which would be impractical because of the immense time and computational resource consumption. To sum up, by using a low-voltage working range to reduce pixel crosstalk and an additional calibration step to minimize the average phase deviation individually for each LCD pixel, the pulse shaper imperfections can be significantly reduced with reasonable effort.

Note that the described challenges are in principle present in many pulse-shaping applications. However, when dealing with conventional pump-probe-type or quantum control experiments, these artifacts are in general not so critical and thus often remain unnoticed because tiny deviations from the desired pulse shape will not have a significant influence on the observed dynamics. Here, on the other hand, we rely on extracting a small nonlinear component from a strong linear fluorescence background [compare Fig. 3(b)], and thus even small deviations from the target pulse shape must be suitably addressed. For the measurements shown in Fig. 4, we have already implemented the described pulse-shaper calibration procedure, and the features seen in Fig. 4(c) and 4(d) are what remains.

Now the amplitude of the artifacts is low enough that the residual linear contributions can be removed. The measured time-domain maps have been normalized to the laser-power, therefore linear contributions can be removed from the individual time-domain maps by subtraction of the low-excitation-power map from the corresponding high-power map.

Note that, as opposed to four-wave-mixing-detected 2D spectroscopy using a local oscillator for phase recovery, we do not require data analysis of spectral interferograms via a Fourier-transformation procedure. Thus no Fourier windowing effects need to be considered. We rather use the full time-integrated and spectrally integrated fluorescence. After subsequent phase cycling of the 27 time maps and two-dimensional Fourier transformation, we obtain the corrected rephasing and non-rephasing 2D spectra shown in Fig. 4(e) and 4(f).

Detailed views of the real parts of the relevant quadrants of the corrected 2D maps of Fig. 4(e) and 4(f) are shown in Fig. 5(a, rephasing) and 5(c, non-rephasing), respectively, and the corresponding imaginary parts are plotted in Fig. 5(b) and 5(d). The axes have been converted from the partially rotating frame frequencies ω^rot used above to the laboratory frame frequencies ω^lab by adding Δω = 0.7 ω₀. Adding up the rephasing and the non-rephasing real-valued contributions, the absorptive 2D spectrum is retrieved [Fig. 5(e)]. This spectrum allows for comparison with 2D measurements in non-collinear box or pump-probe geometry. However, the latter are lacking the spatial resolution needed for investigation of structures heterogeneous on the sub-micron scale. To verify the integrity of the obtained 2D spectra, we integrate the absorptive spectrum along the ω_τ axis [Fig. 5(f, black)]. For comparison, we measured the linear fluorescence of the in-plane structured F₁₆ZnPc pillar array with a two-pulse delay scan (τ scanned from −100 fs to 100 fs in 1 fs steps, Δω = 0.7 ω₀) and subsequent Fourier transformation [Fig. 5(f, red)]. The two spectra show the same spectral bandwidth and ratio of the two distinct peaks, with the local minimum between the two peaks being more pronounced in the spectrum obtained by the two-pulse scan because of the better signal-to-noise ratio compared to the full 2D spectrum. The satisfactory agreement between the two curves is a validation of the new method.

 

Fig. 5 Final 2D spectrum of a hotspot in the nanostructured F₁₆ZnPc pillar array. The complex-valued maps are shown for the rephasing (a) real and (b) imaginary part as well as the non-rephasing (d) real and (e) imaginary part. The real-valued maps are added to obtain (e) the absorptive spectrum. (f) Projection of the 2D absorptive spectrum onto the ω_t axis (black) compared to a separately measured Fourier-transformed linear two-pulse scan (red).

Download Full Size | PDF

5. Conclusion

We have developed the novel method of coherent two-dimensional (2D) fluorescence micro-spectroscopy which makes it possible to probe the nonlinear response of fluorescent laterally heterogeneous thin films with diffraction-limited spatial resolution (NA = 1.4, in our case corrsponding to a resolution of 260 nm). We exemplified the method on a nanostructured array of F₁₆ZnPc pillars. The rephasing, non-rephasing, and absorptive 2D spectra were extracted from the collinearly measured data by 27-step phase cycling.

This scheme can be expanded to probe molecular dynamics in more detail by additional scanning of the population time T. Ansiotropy measurements of crystalline domains with different orientation can be carried out by rotating the polarization of the exciting light pulses. With the help of a second microscope objective, light emitted from the sample in transmission direction can be collected, thus enabling additional capture of absorptive (rather than fluorescence-detected) 2D spectra. The rephasing, non-rephasing and absorptive contributions from both observables, incoherent fluorescence and coherent absorption, could even be collected simultaneously in a single measurement.

Most importantly, already with the toolkit presented here, spatial variations of the nonlinear response in chromophore systems and couplings to other systems in their vicinity can be measured all-optically with sub-micron resolution.