Future technology for information process needs higher density of integration and higher speed. Surface plasmon polaritons (SPPs) [1], propagating bound oscillations of free electrons and light at a metal surface, are promising candidates for next-generation highly integrated nanophotonic devices and are widely investigated in recent years [2–5]. However active plasmonics with short response time are not quite easy to realize. One possible route is to engage the ultrafast electron relaxation process in the metal, since the existence of excited electrons will change the dielectric properties of the metal film which can result in an active action [6]. Thus, ultrafast relaxation dynamics of excited electrons in metal nanostructures must be well investigated, which requires a technique with both spatial and temporal high resolutions. Using femtosecond-SNOM, Imura K et al. have tried to investigate the ultrafast excited electron dynamics in gold nanorod [7]. However, due to the one-color pump-probe configuration and both the pump and probe exciting the sample at the plasmon resonance wavelength of the nanorod, the resulting pump-probe signal is dominated by the plasmon contribution and does not correspond to the distribution of excited electrons [8]. We propose that, to detect the spatial distribution of excited electrons, the probe light must ensure local detection and keep away from plasmon resonance which represents a collective contribution of electrons. In the letter, by employing a two-color pump-probe femtosecond-SNOM and choosing the probe wavelength to locate at the interband transition wavelength of the metal which has no evident SPP effect, we successfully detect ultrahigh spatiotemporal resolved excited electron dynamics near a gold nano-slit and directly observe the ultrafast spatial diffusion process of excited electrons.

By combining femtosecond ultrafast optical spectroscopy and scanning near-field optical microscope (SNOM), the femtosecond-SNOM can obtain ultrahigh spatial and temporal optical resolutions simultaneously [9,10]. And it has been successfully applied to study the ultrafast carrier relaxation dynamics in semiconductor nanostructures [11], quantum dots [12] and so on. Figure 1 schematically shows the experimental setup of our two-color pump-probe femtosecond-SNOM (commercial SNOM, Omicron TwinSNOM). Laser pulses from a femtosecond Ti:Al₂O₃ laser (120 fs, 1000 nm, 76 MHz, Mira 900F, Coherent) are split into two parts. One passes through the optical delay line and is used as the pump light, and the other is frequency doubled to 500nm to match the interband transition wavelength of the gold and used as the probe light. The pump and probe beams are recombined at a dichroic mirror and then focused onto the sample by a 4 × objective. The transmitted signal is collected in the near field above the sample by a gold-coated fiber tip (chemical etched) with aperture diameter of about 200 nm. The output from the fiber then pass through a color filter to eliminate the 1000 nm pump light and only the 500 nm probe light is detected by a PMT. A mechanical light chopper is engaged to modulate the pump light at 1.2 kHz and the transient transmission of the probe light (that is the pump-probe signal) is retrieved by a lock-in amplifier at the modulating frequency.

 

Fig. 1 Schematic of the two-color pump-probe femtosecond-SNOM system.

Download Full Size | PDF

For sample preparation, a 20nm-thick gold film is first evaporated on a glass substrate by electron-beam evaporator. AFM measurement (by NTEGRA Spectra AFM, NT-MDT) gives a rms roughness of about 1nm, which means a quite flat sample surface. A 200 nm wide and 30μm long nano-slit is then fabricated on the gold film by focused-ion-beam (FIB). When the 1000 nm pump pulse with electric field perpendicular to the nano-slit illuminates the sample, SPPs are excited by the nano-slit and propagate along the two surfaces of the gold film. The spot size of the pump light is about 15 micron, which is bigger than our scan range (10 micron), so SPPs will interfere with the direct transmitted pump light (schematically shown in Fig. 2(a) ). Numerical simulations by finite element method (FEM) using Comsol Multiphysics show that the SPP on the gold/substrate interface is much stronger than the SPP on the gold/air interface (not shown here). So the electric field in the gold film is dominated by the interference between the SPP on the gold/substrate interface and the direct transmitted pump light, and the interference period should be equal to the SPP wavelength on the gold/substrate interface. This is well demonstrated by Fig. 2(b) which displays FEM simulation results of the electric field intensity in the middle of the gold film, since the electric field mainly possesses an interference period of about 620nm which corresponds well to the SPP wavelength on the gold/substrate interface. This periodically distributed electric field then excites electrons via intraband absorption in the pump pulse duration, and the resulting initial distribution of excited electrons corresponds to the distribution of the excitation electric field. After that, the excited electrons undergo relaxation processes in both space and time. Since the existence of the excited electrons changes the dielectric property of the gold film, this will lead to the modulation on the transmittance of the probe light [13]. Because the probe light locates at the interband transition wavelength of the gold without evident SPP effect, the probe light collected by the near-field tip only locally interacts with excited electrons just below the tip. So the spatiotemporal resolved transient transmission of the probe light detected by femtosecond-SNOM gives a direct measurement on the spatiotemporal evolution of the excited electrons.

 

Fig. 2 (a) Schematic view of SPP generation and subsequent interference with the direct transmitted pump light. (b) FEM simulation results of the electric field intensity in the middle of the gold film (with the nano-slit locates at sample position of 0 nm).

Download Full Size | PDF

Figure 3a shows pump-probe signals across the nano-slit with different time delays. The transverse axis indicates sample positions while the vertical axis represents time delays. Such image is obtained by interpolation to original data which are obtained by repeating scanning the fiber tip across the nano-slit at ten time delays from −333 fs to 2.67 ps with equal interval of 333 fs. The simultaneously obtained topography signal is displayed in Fig. 3b with the nano-slit clearly seen in the middle. Totally speaking, the near-field pump-probe signal in Fig. 3a shows a fast rise and a slow decay which consists well with previous far-field pump-probe results on gold film with no structures [13,14]. This verifies that the near-field pump-probe signal also reflects the states of excited electrons as in the far-field case, and the fast rise and slow decay correspond to electron-electron and electron-phonon scattering processes respectively. Well, besides the total electron relaxation process, the near-field results present more detailed spatial evolvement information. As noted in Fig. 3a, the periodical distribution of the pump-probe signal tends to disappear as time delay increases. In order to clarify such process, near-field pump-probe signals at four typical time delays of −330 fs, 0 fs, 660 fs and 2 ps are extracted from Fig. 3a and shown in Fig. 3c. It is noted that, at −330 fs (that is before excitation), the pump-probe signal shows no periodical structure and it is a background signal coming from longlived thermal effect. At 0 fs, the periodical structure appears with a period of about 610 nm, which is approximately equal to the wavelength of the SPP on the metal/substrate interface. This periodical structure agrees well with the FEM simulating electric field in the gold film displayed in Fig. 2(b), since such excitation electric field distribution gives the initial distribution of excited electrons in the metal film. At 660 fs, the contrast of the periodical pump-probe signal is reduced although the signal is even stronger than at 0fs. At 2 ps, while the pump-probe signal is still much higher than at −330fs, the periodical structure can hardly be identified. To obtain the decay time, the contrast between the peak at position 2000 nm and the valley at position 2300 nm is extracted from Fig. 3(a) and plotted in Fig. 3(d) as a typical example, and single exponential fit gives an ultrafast decay time of 676 ± 135 fs.

 

Fig. 3 (a) Pump-probe signals across the nano-slit with time delays from −333 fs to 2.67 ps. (b) Simultaneously measured topography. (c) Pump-probe signals (without offset) at four different time delays of −330 fs, 0 fs, 660 fs and 2 ps, respectively. (d) Extracted contrast between the peak at 2000 nm and the valley at 2300 nm from Fig. 3(a) and corresponding fitting results by single exponential decay.

Download Full Size | PDF

According to references [13] and [14], the excited electrons can be roughly divided into thermal electrons and nonthermal electrons. After being excited, the electrons are nonthermal first and then they will thermalize through electron-electron scattering on a time scale of about 500fs. Since we probe at 500nm, the pump-probe signals are mainly contributed by the thermal electrons [13]. So the measured near-field pump-probe signal approximately gives the spatial distribution of the thermal electrons. In the experiment, the initial nonthermal electron distribution is nonuniform due to the nonuniform excitation field, so the nonthermal electrons not only relax to thermal electrons locally but also diffuse to other sample positions and relax to thermal electrons there. Suppose there is no significant diffusion process of nonthermal electrons and only electron-electron and electron-phonon scattering exist, the thermal electrons and the corresponding pump-probe signals at different sample positions will undergo same relaxation dynamics and the contrast of the periodical pump-probe signals will keep unchanged. Well, our experimental results are opposite to such hypothesis. Moreover, the ultrafast diffusion of the nonthermal electrons can just account for the ultrafast decay of the contrast in Fig. 3, since previous studies indicate that the energy transport induced by diffusion of nonthermal electron has an effective velocity of about 10⁶m/s [15] and the corresponding diffusion time through a distance of 300nm (peak to valley distance) is just about a few hundreds femtoseconds. Other mechanisms are not on the same time scale and can be excluded. For example, the diffusion of thermal electrons has an effective velocity two orders of magnitude smaller than that of nonthermal electrons, which is negligible in the above ultrafast decay process.

The ultrafast electron diffusion process can also be identified from the different relaxation times at different sample positions. Figures 4 displays typical time-domain measured pump-probe curves by scanning the time delay while fixing the near-field fiber tip at specific sample positions. The green symbols and the red symbols are acquired at sample positions of 2000 nm and 2300 nm in Fig. 3, respectively, corresponding to two adjacent peak and valley of the pump-probe signals. Figure 4(a) shows the original data and Fig. 4(b) shows the normalized results with respect to the peak intensities. The relaxation dynamics are obviously different and double exponential fit shows that the pump-probe signal at position 2000 nm possesses both a faster rise time and a faster decay time (rise time of 266 fs and decay time of 1446 fs) compared with those at position 2300 nm (rise time of 413 fs and decay time of 1563 fs). Considering that the nonthermal electron density after excitation at position 2000 nm is larger than that at position 2300 nm, they will diffuse from position 2000 nm to position 2300 nm. So that the relaxation process at position 2000 nm is speeded up (in contrary to the pump power dependent behavior in reference [14], where higher pump intensity gives slower dynamics), which gives faster rise and decay times. On the contrary, due to the additional nonthermal electrons coming from its neighborhood, the relaxation process at position 2300 nm is slowed down. Thus the different relaxation processes indicated by the pump-probe curves at different sample positions also give indications of the ultrafast electron diffusion process in the metal nanostructure.

 

Fig. 4 Green symbols and red symbols show pump-probe curves acquired at sample positions of 2000 nm and 2300 nm in Fig. 3, respectively. (a) Original data and (b) normalized results with respect to the peak intensities. Lines correspond to double exponential fitting results.

Download Full Size | PDF

For comparison, the experiment with one-color pump-probe configuration is also carried out. Since the pump and probe lights cannot be distinguished by color filter, we employ the equal pulse correlation (EPC) measurement [16]. Due to the lower signal to noise ratio, the experiment is performed at 830 nm wavelength with higher pump and probe intensities. The resulting pump-probe signals obtained by scanning the fiber tip across the same nano-slit at different time delays are presented in Fig. 5 . The signals possess periods of about 800 nm corresponding to the SPP wavelength on the gold/air interface. Considering that the probe light changes to the intraband transition wavelength of 830 nm, the near-field detected probe signal is no longer a local detected signal as in the two-color pump-probe configuration, but mainly comes from the interference of the direct transmitted probe light and the propagating SPP on the gold/air interface excited by the probe light, and has a period of about 800 nm. Consequently, not only the excited electrons at the probe position but also the excited electrons on the propagating path of the SPP can change the detected probe signal and give contributions to the pump-probe signal. Thus the pump-probe signal is much more complicated and cannot be directly related to the excited electron distribution as in the previous two-color pump-probe case. This is well demonstrated by Fig. 5, since the one-color pump-probe signals always show the periodical structure (even at 6ps after excitation). The signal variation is quite different from that in Fig. 3c and the electron diffusion process is hidden.

 

Fig. 5 Pump-probe signals at time delays of 0 fs, 660 fs, 2 ps and 6 ps obtained by scanning the fiber tip across the same nano-slit as in Fig. 3 with one-color pump-probe configuration. The signals are amplified and offset for clarification.

Download Full Size | PDF

In conclusion, we have successfully observed the spatiotemporal evolvement of excited electrons in a gold nanostructure by employing two-color pump-probe femtosecond-SNOM and probing at the interband transition wavelength of the gold. Ultrafast nonthermal electron diffusion process is successfully identified on a time scale of about 700 fs, while such phenomenon is not observed in the one-color pump-probe configuration in the comparison experiment. Due to the simultaneously obtained ultrahigh spatial and temporal resolution, the two-color pump-probe femtosecond-SNOM can reveal more detailed information on the ultrafast electron relaxation processes in metal nanostructures and even in ultrafast active plasmonic devices.