1. Introduction

A large variety of time-resolved optical spectroscopy methods allow the investigation of non-equilibrium dynamics of complex materials by monitoring the transient response of relevant optical transitions. Particularly through the advent of ultrafast pulsed lasers, time-resolved spectroscopy has been used in numerous studies in physical, chemical, and biological fields, achieving time resolutions down to the femtosecond and attosecond range [1–6]. Such measurements are often performed in a pump-probe framework, in which a pump pulse excites a sample that was initially in equilibrium, thus creating a time-dependent evolution of states of the system as it relaxes back to equilibrium. Following the pump pulse, a probe pulse interacts with the sample and yields a measurement of the evolving states by probing their optical signatures [7].

Depending on the probe source, its interaction with the sample and the type of detection, different categories of pump-probe techniques exist that can probe transient optical phenomena [7]. Among them, transient absorption (TA) spectroscopy is a simple yet powerful all-optical technique for measuring population dynamics, where the transmission of the probe through the sample is measured with and without the presence of the pump excitation, yielding a pump-induced differential absorption (or transmission) response of the sample with a tunable time delay between the pump and probe beams. Because the temporal resolution, in this case, is limited by the involved pulse durations, ultrashort laser pulses achieve time resolutions that can vary from sub-femtoseconds to nanoseconds. Ultrashort laser pulses also enable powerful wavelength conversion methods through various non-linear optical processes. In most conventional setups, the arrival time of either pump or probe pulses is controlled using the finite speed of light. Varying the path length to the sample using a mechanical translation stage allows precisely changing the delay between pump and probe. However, generally due to space constraints as well as beam divergence and walk-off issues, the length of the delay line is practically limited to a few meters, which corresponds to time delays of usually less than ten nanoseconds in time.

To extend the measurement time window, a variety of methods have been developed, for instance, laser flash photolysis [8]. Here, the time resolution is commonly limited to a few nanoseconds, as determined either by the pump pulse duration, the detector instrument response, or electronic synchronization. Flash photolysis requires considerable pump energies in the mJ range to achieve sufficient excitation densities. Other methods rely on two or more synchronized ultrafast lasers. Two laser oscillators with locked repetition rate offset can quickly acquire long scans using asynchronous optical sampling (ASOPS) or as a dual-frequency comb if the carrier-envelope-phase is stabilized [9–11]. Another approach is to employ two independent synchronized femtosecond lasers consisting of two oscillators that each separately seed two amplifiers, and the delay between two pulses is controlled electronically [12]. In a slight modification, one oscillator with two amplifiers was employed with an electronically inserted delay for coarse delay control and a spatially separated translation stage for fine adjustments [13]. Another approach makes use of a pulsed laser for excitation and a pulse train of broadband light as the probe that is not synchronized with the pump laser. By recording the probe spectra of each pulse at random pump-probe delays repeatedly, the whole time delay trace can be reconstructed [14]. One simple approach is to employ a pump pulse and a monochromatic continuous-wave (CW) laser as a probe and gather the transient probe transmission with a high-speed photodetector and pump-triggered digitizer [15–20]. The time resolution, in this case, is determined by the rise time and fall time of the detection instruments, e.g. photodetector and amplifier, as well as the bandwidth of the recording device, i.e. oscilloscope or data acquisition card. There are many variations of the techniques featured in this paragraph, mostly using different lasers and electronic trigger schemes [21–29]. However, most of these methods either demand complicated and expensive equipment, e.g. multiple synchronized lasers and complex electronic triggering circuits, or the probe does not provide a sufficiently broad spectrum or limited wavelength tunability, as is the case for monochromatic CW lasers.

In this paper, we introduce an approach that employs a wavelength-tunable femtosecond pump pulse and an incoherent CW white-light (WL) probe that provides broadband spectral coverage from 0.4 to 2.0 µm. The WL transmitted through the sample is passed through a monochromator and detected with a fast photodiode from which the transient current is then amplified and recorded with an oscilloscope. For each pump pulse, a full pump-probe transient is recorded for a specific probe wavelength and consecutive transients are averaged. As opposed to traditional TA techniques, the time resolution here is not determined by the pulse durations of pump and probe, but by the electronic bandwidth of the detection and digitization devices. The setup generally achieves a time resolution of around one nanosecond with scan durations up to milliseconds.

2. Experimental setup

As the pump, we employ an ultrafast Yb laser (Amplitude Tangerine SP) with tunable repetition rate emitting about 150-fs long pulses at 1030 nm. Wavelength tuning of the pump laser is achieved either through a non-collinear optical parametric amplifier (NOPA rainbow) or second harmonic generation (SH, 515 nm) and third harmonic generation (TH, 343 nm) in BBO crystals. As the probe, we use a broadband, laser-driven Xenon plasma light source (XWS-30, ISTEQ BV). As opposed to conventional Xenon arc lamps, where the emitting plasma is generated through high voltage electrodes in a discharge arc, here an infrared CW laser is tightly focused into the Xenon gas, thus generating the emitting plasma with high brightness (up to 45 mW/(mm²·nm·sr)) and within a smaller volume compared to conventional filaments - around 150-200 µm in diameter. The confined emitter volume allows incoupling into a multimode fiber with high power throughput and imaging of the probe beam to small spot sizes on the sample.

In the setup, the emitted WL from the filament is coupled into a 200 µm multimode fiber with a numerical aperture (NA) of 0.22, and the output power is ∼65 mW. The exiting fiber-coupled WL is collimated using a reflective collimator (RC02FC-P01, Thorlabs Inc.) in the setup, for which the schematic is shown in Fig. 1-(a). A variable linear band pass spectral filter (Delta Optical Thin Film) mounted on a translation stage is placed after the collimator, the purpose of which is explained later. Due to the broadband spectrum of the WL, only reflective optical components and mostly protected silver mirrors are used to avoid chromatic aberrations. The collimated light is deflected with two flat mirrors (M1 and M2) and focused on the sample with a curved spherical mirror with a focal length (FL) of 37.5 mm (M3). The WL spot size at the focus on the sample is 290 µm in diameter (measured by a camera in the focal plane), which is about an order of magnitude smaller than common flash photolysis systems, hence allowing to employ much lower pump energies for excitation. Since the sample is housed in a closed-cycle helium cryostat (ARS-10HW), in which the temperature can be as low as 4 K, it is not possible to focus the WL tighter and the spot size is determined by the fiber diameter and NA. The pump beam is focused (using Lens L1 with a FL of 1250mm) on the sample at a small angle with respect to the WL to a spot size slightly larger than the WL, around 350 µm in diameter. The WL transmitted through the sample is recollimated with a parabolic mirror (M4 with a FL of 50 mm), deflected (M5 and M6), and focused on the input slit of a monochromator (Cornerstone 130 1/8m, Newport Corporation) with a parabolic mirror (M7 with a FL of 50 mm). The monochromatic output light is finally focused on a fast photodiode (PD) with an ellipsoidal mirror (M11 - 25.4 mm Sq., 2X, Protected Aluminum Off-Axis Ellipsoidal Mirror, Edmund Optics) for detection.

 

Fig. 1. a) Schematic of the implemented setup. FC demonstrates the fiber coupling of the WL, C1 a collimator, F1 a variable linear band pass spectral filter mounted on a translation stage and L1 a lens. M1, M2, M5, M6, M7, M8, and M10 are right-angle flat prism mirrors, M3 and M9 are spherical mirrors, M4 and M7 are parabolic mirrors, and M11 is an ellipsoidal mirror. The monochromatic light is detected with a photodiode (PD) and the output signal recorded with an oscilloscope after an amplification stage (Amp.). b) Schematic of the operation principle and signal acquisition with the oscilloscope.

Download Full Size | PDF

Depending on the signal level, spectral range and desired time resolution, various PDs or avalanche photodiodes (APDs) made of silicon or InGaAs are employed for detection. APDs have an initial amplification of the photon-generated electron-hole pairs due to impact ionization and are more suitable for low-intensity light detection. The PDs and APDs used to collect the light in different experiments are as follows: DET025AL (Si-PD with a bandwidth of 2 Ghz, Thorlabs Inc.), UPD-70-UVIR-P (InGaAs-PD with a bandwidth of 5 Ghz, ALPHALAS GmbH), G12182-103K, G8931-20 (InGaAs-PD (extended) and APD, with a bandwidth of 140 and 900 MHz respectively, Hamamatsu Photonics) and SARP500 (Si-APD with a rise time of 450 ps, Laser Components GmbH). The spectral coverage of the aforementioned PDs (or APDs) is from 400 to 1050 nm for silicon and from 800 to 1700 nm and 2070 nm for InGaAs and Extended InGaAs detectors, respectively. The output signal of the PD is amplified with a high-bandwidth electronic amplifier. Depending on the situation, different amplifiers from FEMTO Messtechnik are used, namely HSA-Y-2-40 (100x voltage gain), HSA-Y-1-60 (1000x voltage gain) and HCA-400M-5K-C (5000x transimpedance gain). Their bandwidths are 10 kHz-2 GHz, 10 kHz-1 GHz, and DC-400 MHz, respectively. The amplified signal is acquired with an oscilloscope (4GHz HDO9404, Teledyne LeCroy) that is triggered by residuals of the pump pulse detected by a fast PD.

The operational principle of the setup is shown in Fig. 1-(b). In this approach, with a single pump shot, a full delay-time trace of the signal is captured at a specified wavelength, contrary to most common TA techniques that acquire a full probe spectrum at a specific time delay. The time resolution of the measurement is determined by the lowest bandwidth in the PD-amplifier-oscilloscope acquisition chain, which in our case is either the PD or the amplifier. The maximum measurement duration is defined by the oscilloscope (buffer size etc.) and can in principle be extended up to the time separation of two pump pulses as defined by the laser repetition rate, i.e. at 1 kHz one can scan up to 1 ms, which is also the maximum scan duration that we chose here.

To increase the signal-to-noise ratio (SNR), the signal is averaged in the oscilloscope while acquiring the data and then stored on the computer. All hardware control and data read-out were performed with a homemade program written in LabVIEW. In order to reduce the effect of electromagnetic interference (EMI) on the signal, the cables are designed to be as short as possible, and all electronic components and cables are shielded with EMI shielding wire mesh, foils, and boxes. Transient signals are recorded at each wavelength, and the whole spectral range is scanned by rotating the grating in the monochromator. The spectral resolution is defined by the input and output slit size, which are usually adjusted to have a spectral resolution of around 2 nm. The final measured spectrum of the WL after the monochromator is shown in Fig. 2, which is stitched together from measurements with Si APD, InGaAs APD, and extended InGaAs PD with appropriate rescaling.

 

Fig. 2. Spectrum of the white-light (WL) source after the monochromator, measured with Si APD, InGaAs APD, and extended InGaAs PD and stitched together.

Download Full Size | PDF

3. Results and discussion

To benchmark our setup, we measured the optical dynamics of nitrogen-vacancy (NV) color centers in diamond. The NV center is a point defect in diamond that features a particular combination of favorable optical and spin properties that make them promising candidates as solid-state spin qubits for quantum information processing [30,31], as well as room temperature nanoscale sensors for magnetic fields [32–34]. The photo-excitation dynamics that ultimately determine important features of NV centers reach from picoseconds to microseconds and beyond, with distinct spectral features covering the visible to the near-infrared, thus making NV centers a perfect testbed for our method [35–37]. The sample is a type 1b diamond substrate grown with the high-pressure, high-temperature technique, having substitutional nitrogen with a concentration of ∼100 ppm. Vacancies were introduced by irradiation of 1 MeV electrons and subsequent annealing in vacuum, yielding a NV concentration of roughly 1.5ppm. The absorption spectrum of the sample is shown in Fig. 3-(a).

 

Fig. 3. a) The absorption spectrum of the sample. The electronic structure of NV centers is shown in the inset. b) Time and spectrally resolved photoluminescence (PL) of NV centers in diamond at room temperature. Early time PL trace is shown in the inset of top graph that shows a rise time of 0.28 ns. The impulse response of the system is shown on the top-right graph.

Download Full Size | PDF

The electronic structure of NV centers is shown in the inset of Fig. 3-(a). It consists of ground and excited spin triplet states (³A₂ and ³E), and two singlet intermediate states (¹A₁ and ¹E) within the bandgap of diamond. It has a zero-phonon line (ZPL) transition at 637 nm (1.945 eV) between the triplets, and an intermediate infrared transition at 1042 nm (1.190 eV) between the singlets. The latter pathway can be accessed from the triplet states after optical excitation via intersystem crossing (ISC) [38]. In our measurements, we pump the sample with the laser SH at 515 nm to excite electrons from ³A₂ to ³E. The excited electrons either relax directly to ³A₂ by radiative relaxation or via the singlet states through a series of largely non-radiative decay channels. The lifetimes of the photoluminescence (PL) and ISC from ³E to ¹A₁ are both around 10 ns. Population from ¹A₁ decays to ¹E within about 100ps [35]. The lower singlet level (¹E) has a lifetime of ∼200 ns at room temperature.

Without using the WL, our setup allows us to measure the time- and spectrally-resolved PL of a sample, shown for the NV sample at room temperature in Fig. 3-(b). The time resolution of this measurement is given by rise time and fall time of the detector that are around 300 ps each, calculated with the impulse response of the system shown in top-right panel in Fig. 3-(b). The impulse response is recorded by detecting the scattered light of the pump pulse with the photodetector. By fitting the PL of the sample with an exponential function, we found the lifetime to be around 8.5 ns which is in good agreement with previous publications [39].

Introducing the WL, we can measure the pump-induced differential transmission (ΔT) between photoexcited (T_p) and non-photoexcited sample (T) at each set probe wavelength. Normalization by T corrects for detection efficiency and throughput variations and yields the differential transmission ΔT/T. Since T_p can also contain PL (as shown above), we have to acquire a reference measurement without WL that will be subtracted from T_p. To do so, a motorized shutter in front of the WL fiber-incoupling is used to block the WL at each wavelength to perform measurements with and without probe consecutively. After having obtained ΔT/T, it is converted to ΔA which is the differential absorption of the sample.

A typical time- and spectrally-resolved ΔA result of the sample is shown in Fig. 4. The measurement shown in Fig. 4-(a) covers the whole spectrum from 400 to 1100 nm except the pump region (508-530 nm) with a measurement duration of 1 µs, a time resolution of ∼1 ns and a spectral resolution of ∼2 nm. Spectrally-resolved ΔA at different times and time-resolved ΔA at different wavelengths are shown in Fig. 4-(b) and Fig. 4-(c), respectively. Also, a longer measurement, up to 1 ms measured within the region of interest, featuring long-persisting ground state bleaching from 530 to 580 nm, is shown in Fig. 4-(d). The energy density of the pump pulses with a central wavelength of 515 nm on the sample in this measurement is 8.7 mJ/cm² with a pulse duration of ∼150fs, a repetition rate of 10 kHz for the shorter measurement and 1 kHz for the longer one.

 

Fig. 4. a) Time- and spectrally-resolved transient absorption results of NV centers in diamond at room temperature. b) Spectrally resolved ΔA at different times. c) Time resolved ΔA at different wavelengths. d) 2D time and spectrally resolved data for longer time delays up to 1 ms showing long persisting ground-state bleaching. A time trace at 560 nm is shown in the inset.

Download Full Size | PDF

We now want to briefly highlight the observed dynamics. A more detailed discussion will follow in a separate publication. In Fig. 4-(c), we plot the dynamics of four distinct spectral regions: 450 nm (excited state absorption, ESA), 550 nm (ground state bleaching, GSB), 700 nm (stimulated emission, SE) and 1042 nm (singlet transition, ST). The ESA and SE signal decays within 8.5 ns, in line with the PL lifetime measurement. The ST absorption signal rises within 9 ns, indicative of the upper ISC time, and subsequently decays within 200 ns, reflecting the lower ISC time. The GSB signal initially decays bi-exponentially with time constants of 8.5 ns and 200 ns, respectively, reflecting the direct radiative relaxation channel and the ISC relaxation channel. The measured lifetimes are in agreement with the reported values [39,40]. There is however still a non-vanishing GSB signal present after 1 µs which originates from photoionization of the negatively-charged NV centers that lead to charge conversion to the neutral charge state. The following decay, shown in Fig. 4-(d), thus monitors the dynamics of ionized NV centers converting back to their original negative charge state. This decay is bi-exponential, with the two time constants being 8 and 300 µs, respectively.

The 1-µs measurement, shown in Fig. 4 a-c, was performed with three different detectors. From 400 to 508 nm, the Si APD with a gain of 50 was used and amplified with HSA-Y-2-40. The PL emission is large in the range between 570 and 800 nm, at its peak reaching transient intensity levels almost 100 times larger than the WL. As a consequence, the SNR is degraded in this spectral range in the first 10s of nanoseconds, as can be seen in Fig. 4-(b). Furthermore, the silicon APD shows a slightly non-linear response in this range, presumably arising from small gain changes as a consequence of variations of the CW light illumination (in two measurements: one with and the other without the WL). This gain modulation is negligible in other regions. However, due to the strong transient PL signal in this spectral range, the final ΔA transient contains artifacts in the first 60 ns for the Si APD. For this reason, in the 530 to 800 nm range, the silicon PD amplified with HSA-Y-2-60 was employed, which does not show any transient artifacts. From 800 to 1100 nm, the InGaAs APD (with a gain of 10) amplified with HSA-Y-2-40 was used. Interestingly, this detector does not show any non-linear response like the Si APD. In this measurement, the signal was averaged for 20000 and 5000 times for PD and APDs, respectively. The whole spectrum measurement took 4 hours to finish. In addition, to have a smoother spectral response, the spectral scan was performed with 0.33 nm or 0.5 nm steps for the PD and APDs, respectively, and then smoothed with moving average in the spectral domain. In the longer measurement, shown in Fig. 4-(d), the Si APD amplified with HCA-400M-5K-C was employed. The signal was averaged 4000 times and scanned with 1 nm spectral steps, and the whole measurement took 45 minutes.

The total measurement time depends on various parameters: repetition rate of the pump laser, measurement duration, sampling rate of the oscilloscope and processing time, spectral scanning steps, and number of averaging at each wavelength. For shorter measurement durations, the repetition rate of the pump laser can be increased, which decreases the oscilloscope acquisition time. However, the required time to capture a transient trace with the oscilloscope at each wavelength is determined by not only the acquisition time, but also the computational time e.g. data averaging and storing. Therefore, the measurement duration and sampling rate of the oscilloscope, which can be up to 40 Giga-samples per second, directly affects the computational time, which in some cases can be several times more than the acquisition time. Conceivably, the computational time could be shortened by employing alternative data acquisition devices that allow faster processing on an FPGA and/or in parallel to the measurement process. Since the sampling rate also limits the maximum frequency of the captured signal, for slower dynamics that require longer measurement time, time resolution can be compromised by decreasing the sampling rate in order to decrease the computational load and achieve faster acquisition. The required time to acquire data with 2000 averaging for various sampling rates with 1 kHz pump repetition rate is shown in Table 1. In the table, computational time is the time consumed to average the signal and total time is the required time to acquire and average the signal.

Table 1. Time required to acquire and process the signal with 2000 averaging for various sampling rates (indicated in Giga-samples per second) and 1 kHz pump repetition rate.

View Table | View all tables in this article

In order to compare the noise of the output signal for various photodetectors with varying amount of averaging, three time traces for each photodetector at a specified wavelength are shown in Fig. 5. It is noticeable that the signal of the PD even with 10 times more averaging is noisier than that of the APDs. To have a better comparison of noise for various configurations, the standard deviations of noise in ΔA time traces, shown in Fig. 5, were calculated and shown in Table 2. The standard deviation was calculated in ΔA traces before time zero, which gives the noise floor (in the absence of any transient absorption signal). As can be seen, the PD has an order of magnitude larger noise floor than the APDs for. However, due to aforementioned non-linearity of the silicon APD caused by strong PL, we resort to using the PD in this spectral range.

 

Fig. 5. Time traces of ΔA at specified wavelengths for multiple averaging times for various photodetectors: a) Silicon PD, b) Silicon APD, and c) InGaAs APD. Note that the number of averaging for PD is 10 times more than for APDs.

Download Full Size | PDF

Table 2. Standard deviation of the noise in a time trace of ΔA for various photodetectors and different amounts of averaging.

View Table | View all tables in this article

We noticed some slight differences in the PL intensity with and without applied WL, presumably due to photo-excitation of NV centers by the broadband WL. In order to solve this issue, we filtered the WL before interacting with the sample by using a variable linear bandpass spectral filter (Delta Optical Thin Film) that is mounted on a motorized translation stage. In this way, only a narrow-bandwidth part of the WL (40 nm bandwidth) matching the selected monochromator wavelength is transmitted onto the sample. Another possible approach to deal with this issue is to use the monochromator before the sample.

A few other drawbacks need to be mentioned. First, the measurements can be slower than other TA techniques depending on the configuration and measurement duration and may take up to a few hours, thus requiring a stable pump laser beam. Second, in case of having a strong PL, which was the case with the investigated sample, the silicon APD shows non-linear response. Therefore, in this spectral region, a PD must be used that requires more averaging to obtain a SNR comparable to APDs. Furthermore, the transient absorption data shows higher fluctuations in the PL range and needs to be averaged more than other spectral parts.

4. Conclusion

To conclude, we proposed a new approach for transient absorption spectroscopy, covering multiple timescales with a broadband spectral probe coverage from 0.4 to 2 µm with a spectral resolution of <2 nm. We achieved time resolutions of around 1 ns with measurement durations on the order of microseconds that can easily be extended up to milliseconds and beyond. Furthermore, we are able to measure time- and spectrally-resolved PL besides the transient absorption of the sample. As a proof-of-principle experiment, we measured the dynamics of NV centers in diamond up to 1 ms in the visible and NIR range. Compared to existing approaches, our setup is simpler and requires less equipment.

Even though the spectral coverage of the current setup ranges from 0.4 to 2 µm, due to the simple concept of using a single-element detector, it may also be implemented in other spectral regions (e.g. deep-UV and mid-IR) if a sufficiently high brightness light source, monochromator and suitable detectors are available. For instance, a modified version of a similar laser-driven plasma light source as employed here as a probe using a diamond exit window has been shown to provide broadband radiation in the mid-IR with high brightness [41].