Letter pubs.acs.org/NanoLett
Ultrafast Spectral Migration of Photoluminescence in Graphene Oxide Annemarie L. Exarhos, Michael E. Turk, and James M. Kikkawa* Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, United States S Supporting Information *
ABSTRACT: We use subpicosecond time-resolved photoluminescence measurements to study the nature of photoluminescence in graphene oxide and reduced graphene oxide. Our data indicate that, in contrast to prior suggestions, the photoluminescence spectra of graphene oxide and reduced graphene oxide are inhomogeneously broadened. We observe substantial energy redistribution and relaxation among the emitting states within the first few picoseconds, leading to a progressive red shift of the emission spectrum. Blue shifts that arise in time-integrated spectra upon photothermal reduction are easily understood within this dynamical context without invoking a modified distribution of dipole-coupled states. Rather, reduction increases the nonradiative electron−hole recombination rate and curtails the red-shifting process, which is consistent with an increase in quenching through the introduction of larger and/or more numerous sp2 clusters. Polarization memory measurements show energetic signatures of electron−hole correlations, established on a subpicosecond time scale and developing little thereafter. KEYWORDS: Graphene, graphene oxide, ultrafast dynamics, photoluminescence
T
systems. Specifically, shifts in time-integrated spectra are not related to changes in the energetic distribution of dipolecoupled states (such as energy gap changes or the addition/ removal of transitions related to electronic states localized on functional groups) but instead reflect changes in ultrafast kinetics. Using subpicosecond time-resolved PL, we find that that emission from GO involves picosecond spectral migration, and we observe that photothermal reduction leads to changes in nonradiative relaxation rates that curtail this spectral evolution and thereby blue shift the observed time-integrated PL spectra. Techniques to reduce GO employ hydrazine,26−28 hydrogen plasma,29 electrochemical,30−32 photocatalytic,33 photothermal,25,34,35 and thermal29,36−39 treatments. These processes all vary the quantity of oxygen functional groups (alternatively thought of as varying the sp2 to sp3 ratio of carbon bonds) that decorate the two-dimensional graphene sheet so as to obtain reduced graphene oxide (rGO), though different reduction methods affect the quantities and types of oxide groups which are removed. In this work, photothermal reduction via continuous wave xenon (Xe) lamp exposure25 provides a simple and convenient way to study the reduction of GO without the need for chemical reducing agents or thermal annealing. Aqueous suspensions of GO are exposed to a continuous wave Xe lamp for varying amounts of time in order to achieve differing levels of photothermal reduction. The
he unique characteristics of graphene have attracted increasing attention since single-layer graphene was isolated.1 Much work has focused on the transport properties of this novel semimetal,1−6 but more recently there has been a steady rise in publications discussing its optical properties.7−9 Adding excitement to this emerging field are observations of photoluminescence (PL) from graphene oxide (GO),10,11 which, combined with analogous results from fluorinated graphene,12 suggests the possibility of lateral band gap engineering of graphene and avenues for opto-electronic integration. At the same time, the physical basis for PL emission from these materials is unclear with numerous fundamental questions as yet unresolved. Various mechanisms for generating electron volt (eV)-scale (optical) gaps in graphene have been proposed,10,13−16 and optical transitions involving functional groups have also been suggested.17,18 Additional intrigue stems from differences among reported spectra and their interpretation. PL from GO has been reported to peak in the red10,11,19 and blue20,21 portions of the visible spectrum depending on preparation, and some groups claim the spectrum is homogeneous,19,22 or associated with few emissive species,19,22−24 whereas others favor an inhomogeneous picture related to disorder.10,11,20,25 This Letter illuminates this discussion by demonstrating that GO emission is inhomogeneously broadened and that energy redistribution within an ensemble of emitting states is important to consider in most prior studies. By comparing samples at different stages of reduction we are able to show how the dynamics of energy redistribution inform the discussion on the origins of red and blue PL from these © 2013 American Chemical Society
Received: July 16, 2012 Revised: January 7, 2013 Published: January 22, 2013 344
dx.doi.org/10.1021/nl302624p | Nano Lett. 2013, 13, 344−349
Nano Letters
Letter
efficacy of this approach has previously been demonstrated through X-ray photoelectron spectroscopy (XPS)25 and is confirmed here by our own XPS measurements (Supporting Information Figure S1). For transient measurements, we drop cast the resulting aqueous suspensions onto cleaned and polished UV fused silica substrates, yielding layered dried films of GO (or rGO) flakes with normals orthogonal to the substrate. Time-resolved photoluminescence (TRPL) measurements are performed utilizing an optical Kerr gate (OKG) method40 with 3.1 eV (400 nm) pulses, 120 fs in duration, exciting the sample at normal incidence. The configuration of our experiment is shown in Supporting Information Figure S2 and is essentially that described in a prior publication41 but with added control over the polarization of the excitation. The ultimate pulse cross correlation of our system is as low as 250 fs41 but to reduce the background for the experiments presented here we use a slightly oblique gate beam angle resulting in a cross correlation of 400 fs. The OKG method enables the collection of broadband PL spectra at every delay position and is thus well suited to study the dynamics of the broadband emission in GO. For TRPL measurements, samples are held in vacuum in an optical cryostat. To minimize damage and photoreduction due to laser exposure, the excitation pulse fluence is kept below 18 μJ/cm2, and at each delay point the sample plane is translated over a set path that is identical for each exposure. The latter action not only reduces the exposure at any given point on the sample but also reduces systematic errors from spatial variations in film properties. Furthermore, TRPL scans are taken in both increasing and decreasing time delay directions to verify the reproducibility of the decay curves and to ensure that sample degradation is not responsible for the signal decay. Sample numbers are indicated on TRPL maps. Time-integrated spectra for drop cast samples are collected within the TRPL system by uncrossing the OKG polarizers, thus eliminating the optical gate. The reproducibility of our TRPL measurements and sample preparation is demonstrated for other samples in Supporting Information Figure S3, along with control measurements on a bare substrate. All data are taken at 300 K unless otherwise noted. Sample temperature does not significantly alter the dynamics; TRPL captured at 10 K shows only marginal changes relative to 300 K (Supporting Information Figure S4). Time-integrated PL spectra from drop cast samples of GO in various stages of reduction are shown in Figure 1. As Xe lamp exposure time is increased, the PL shifts from predominately red emission to blue at long reduction times, which is in agreement with the observations of ref 25. We note that prior to drop casting, the PL spectra of aqueous suspensions actually shows an initial small red shift for the shortest Xe lamp exposure times (Supporting Information Figure S5), not reported by ref 25. However, because this curious feature does not remain in drop cast samples necessary for our ultrafast measurements, further study is beyond the scope of this manuscript. In both the drop cast solid samples and in the aqueous suspensions, we find that the PL blue shift is accompanied by a significant quenching of the emission (Supporting Information Figure S6). This drop in emission intensity is characteristic of all our photothermally reduced GO samples and is correlated with an increase in absorption and a visible color change from brownish-orange to dark gray as the Xe lamp exposure time is increased.
Figure 1. Unpolarized time-integrated PL spectra for GO exposed to a continuous wave Xe lamp for varying times and then drop cast onto a substrate. All spectra are scaled to unity.
The time-integrated spectral changes in Figure 1 can be understood in terms of dynamical changes that occur upon reduction. Figure 2 contrasts the TRPL during the first 10 ps following excitation for GO with that of the three other reduced samples exposed for 5, 10, and 15 min to the Xe lamp. Unreduced GO has a notably longer lifetime than the reduced samples and shows a progressive migration to lower energies. (Note that the gap in data between 1.85 and 2.05 eV is due to a large emission background from the Kerr medium, which overwhelms the TRPL signal at these energies. See Supporting Information.) The dynamical red shift in unreduced GO is elaborated in Figure 3, which shows the PL spectrum at different time delays. In contrast, reduced samples have significantly shorter PL lifetimes and display little or no dynamical red shift (Figure 2b−d). However, despite these dynamical differences and markedly different time-integrated spectra (Figure 1), all samples show remarkably similar spectra just after the pump event (Figure 4a). The instantaneous PL spectrum for all samples is peaked in the blue and, moreover, this spectrum is essentially the same as the blue time-integrated spectrum for the 15 min exposed rGO sample (Figure 4b). This correspondence suggests a simple explanation for observations of time-integrated blue PL from rGO. In particular, the quenching of rGO PL on time scales faster than spectral migration strongly suggests that abbreviated carrier lifetimes in rGO suffice to explain its time-integrated blue PL simply because spectral relaxation and red shifting do not have sufficient time to occur. The importance of these observations can be understood through their contrast with prior work. For example, ref 19 reported an absence of spectral migration in TRPL of oxidized single layer graphene and concluded that GO PL was the result of a single, homogeneously broadened emission line. However, that work focused on the longer time scale dynamics (hundreds of picoseconds to nanoseconds) and thus did not observe the ultrafast spectral migration reported here. Not only do we observe migration typical of heterogeneous broadening, but we also observe a truncation of the emission spectrum above the excitation energy that also indicates a distribution of emitters (Supporting Information Figure S7). The same authors also reported an absence of spectral hole burning as evidence for 345
dx.doi.org/10.1021/nl302624p | Nano Lett. 2013, 13, 344−349
Nano Letters
Letter
Figure 2. Unpolarized TRPL maps for drop cast GO samples exposed to a continuous wave Xe lamp for (a) 0 (GO), (b) 5, (c) 10, and (d) 15 min. Increasing Xe lamp exposure corresponds to reduction of the GO. All maps are scaled to unity. Quenched emission characteristic of reduced samples leads to a decrease in S/N in (b−d).
Figure 4. (a) Zero delay PL spectra for samples shown in Figure 2. (b) Comparison between time-integrated PL spectra for GO and the 15 min rGO sample and the TRPL zero delay spectrum for GO. Figure 3. Progressive red shift of PL spectra at various delay times for unexposed GO. Spectra are linecuts of Figure 2a. For 0, 0.5, 1.5, and 5 ps, data are averaged over a 0.25, 0.25, 1, and 2 ps ranges, respectively. All spectra are scaled to unity.
horizontally polarized pump. Resolving the excitation polarization dependence of the emission, however, can give insights into the nature of excited electron−hole pairs in the system. A largely heterogeneous system of emitters that draws its population from a higher energy pump event typically shows an energy-dependent memory of the optical polarization of the pump.43−45 This characteristic behavior can arise, for example, from excitons that lose polarization memory during the process of kinetic relaxation. In this case, the likelihood of polarization memory loss is higher for electron−hole pairs excited further (in energy) from the zero momentum exciton since more scattering (depolarizing) events can occur. This type of electronic polarization memory reflects electron−hole correlations and is significantly different from structural polarization memory, where a polarization dependence is imposed by anisotropy. In materials exhibiting structural polarization memory, a projection of the incident polarization is created for each absorption/emission event depending on the orientation of the anisotropy axes. Carbon nanotubes, for example, act as linear absorbers and emitters and display polarization memory simply because of geometric constraints on the electron−hole pairs.46,47 Structural polarization memory
homogeneous broadening, which our data also explains. In particular, spectral migration in these systems occurs so quickly that it would erase a spectral hole within picoseconds and would thus be missed by spectroscopy lacking sufficient time resolution. In unreduced GO, time-integrated spectra indicate that spectral relaxation is incomplete within the time window probed here. Experiments show that emission from unreduced GO may actually be dominated by a slower (nanosecond) component.19,20,25,42 The OKG method used here focuses on picosecond dynamics at the expense of long lifetime signals, which may contribute significantly to the time-integrated signal but are often within the OKG noise due to their smaller timedomain signal. The unpolarized PL data (both time-integrated and timeresolved) discussed thus far are constructed by averaging the collected spectra when exciting with a vertically and a 346
dx.doi.org/10.1021/nl302624p | Nano Lett. 2013, 13, 344−349
Nano Letters
Letter
Figure 5. (a,b) Polarized TRPL maps of unreduced GO for co- and cross-polarized excitation and collection combinations (VV and HV respectively). TRPL maps for rGO appear in Supporting Information Figure S9. Maps are scaled to unity to emphasize the dominant emission energies for each polarization combination. (c−f) Polarization memory versus energy loss ΔE = Eexcitation − Eemission for the samples shown in Figure 2 showing the time-integrated and zero delay polarization memory. Zero delay polarization memory is calculated using the polarized TRPL spectra integrated around zero delay (−0.5 to 0.5 ps).
structural anisotropy is almost certain, and the energy dependence we observe here is not obviously implied. But continuing with this model for the moment, some authors have suggested PL emission from GO may involve a small number of impurity species characterized by a range of lifetimes.24 Within this context, the polarization evolution seen here could represent an evolution of oscillator strength from one emissive species to another, each characterized by a different degree of P. What must be understood within this picture is that our data would then indicate the emissive impurity levels have a large degree of inhomogeneous broadening and that they exchange energy from 2 to 10 ps without losing polarization memory. In many ways, electron−hole correlations associated with disorder of in-plane states provide a natural explanation for the data presented here, both polarization-resolved and otherwise. Within this context, disorder in the oxidation profile establishes a spatially varying band gap, resulting in broad band emission and a nearly continuous distribution of emitting states. The fact that the PL line shape is truncated above the excitation energy whenever the latter falls within the emission band (Supporting Information Figure S7) is then understood as reflecting the intrinsically heterogeneous nature of the spectrum. The dynamics of TRPL spectra, excited above the emission band, would then imply a sequence of relaxation events in a disordered band model. First, ultrafast kinetic and polarization relaxation occurs on a time scale faster than our instrumental resolution of 400 fs. Once the carriers have reached local minima, some recombine whereas others migrate to lower
does not generically show the type of energy dependence associated with electronic polarization memory (see Supporting Information). The layered orientation of GO flakes in drop cast samples permits us to excite and probe orthogonal in-plane polarizations, denoted vertical (V) and horizontal (H). Figure 5a,b compares the co-polarized (VV) and cross-polarized (HV) TRPL maps for unreduced GO. At short delay times, VV is blue shifted with respect to HV, indicating nonzero polarization memory. We quantify the in-plane polarization memory in terms of PL intensities, Iij, as P = (IVV − IHV)/(IVV + IHV), where the first and second subscripts correspond to the incident and collected polarizations respectively. Figure 5c−f shows a comparison between the time-integrated and instantaneous time-resolved polarization spectra for different GO and rGO samples. All samples exhibit a similar decrease in P for larger energy loss values, ΔE = Eexcitation − Eemission. Furthermore, a striking correspondence between the instantaneous and the time-integrated polarization memory for each sample indicates that the degree of polarization memory is largely established on ultrafast time scales (
