Metal nanostructures have been the focus of several recent studies due to their ability to generate high energy, non-equilibrium “hot” electrons for use in photochemical and photocatalytic applications. In particular, there is growing interest to understand how differences in the electronic structure and optical response of different metals may impact the behavior and utility of their hot electrons in chemical reactions. Using a continuous wave anti-Stokes Raman spectroscopy technique recently developed in our laboratory, in this study, we measured the temperature and lifetime of hot electrons in gold and copper nanostructures in order to understand how the choice of metal impacts hot electron dynamics during steady state illumination. We found that hot electrons in copper are more abundant and more reactive than those in gold, suggesting that copper nanostructures may be a more promising platform for performing hot electron photochemistry.
Plasmon resonances in metal nanostructures, the resonant collective oscillations of free charge carriers during optical excitation, generate “hot” electrons and holes when they decay. The decay process, resulting from either Landau damping or direct interband absorption, efficiently produces transiently excited electrons with kinetic energy greatly in excess of the thermal energy of the metal lattice. As established in ultrafast transient absorption (TA) studies, on a very short time scale, ∼fs, these electrons equilibrate via electron–electron scattering to produce a subpopulation of electrical carriers at an elevated thermal distribution with a characteristic hot electron temperature that is several thousands of K greater than the temperature of the lattice.1,2 These hot carriers eventually relax to the lattice temperature via coupling with phonons on a ∼ps time scale.3
These hot electrons have high kinetic energy, and several studies have explored if the significant photopotential can be utilized before relaxation, for example, for power conversion4 or photodetection.5,6 There is also significant growing interest to take advantage of hot electrons in mechanisms of solar-powered photocatalysis and photochemistry. To this end, plasmonic noble metals with known catalytic activity have been explored, including gold,7,8 silver,9,10 and copper.11 Remarkably, a wide variety of photochemical reactions have been demonstrated, albeit mostly with low efficiency to date. Recent reports indicate that CO2 reduction,12,13 water splitting,14,15 and ammonia generation,16 as well as more sophisticated chemistry using plasmonic antenna-reactor hybrid systems can be achieved during relatively low intensity optical excitation.17
When considering how plasmonic materials may be used in solar photochemistry, of particular interest are plasmonic structures made from copper. This is due in part to the low cost of copper compared to other plasmonic metals such as gold or silver, without sacrificing strong plasmonic resonances and tunability of the optical response by controlling size, shape, and geometry.18,19 In addition, copper is currently one of the best-performing electrocatalysts for CO2 reduction, suggesting that the high chemical potential of photogenerated hot carriers may be used to drive the same reduction chemistry.20 Moreover, a better understanding of differences in the behavior of plasmonically generated hot electrons in different noble metals, especially under conditions of steady state relatively low power continuous wave (CW) optical excitation that is more directly comparable to sunlight, may allow for better tailoring of future plasmonic solar applications.
Our report explores how the photothermalization dynamics of plasmonic hot electrons in copper are distinct from gold during CW optical excitation. We use the anti-Stokes Raman spectroscopy technique our laboratory recently developed that reports the lattice temperature, the electronic temperature, and the size of the hot electron sub-population in a metal during steady state illumination.21,22 Information about the hot electron lifetime as well as the rate of coupling to thermalization pathways and chemical reactions can also be deduced from the anti-Stokes signal. Our experiments provide an important point of comparison for understanding the dynamic behavior of plasmonic hot electrons that is distinct from the insights that can be learned during the high intensity pulse and decay dynamics observed in transient absorption experiments. In comparative studies of nanostructure arrays with equivalent dimensions made from both metals, we show that gold reaches a much greater lattice temperature and much greater hot electron temperature than copper under equivalent CW optical intensity. However, copper shows a larger steady state population of hot electrons compared with gold. Furthermore, the hot electrons in copper are more chemically reactive, which manifest as a relative decrease in hot electron lifetime when the metals are under different illumination conditions but equivalent electronic temperature. Thus, we outline how these differences in the photoresponse of the metals entail a trade-off in the desired photophysical response. Gold may be preferable if the goal is to achieve the highest possible hot electron temperature inside the nanostructure, whereas copper provides a larger population of non-equilibrium electrons for increased chemical reactivity in the environment around the nanostructure.
Fabrication: Nanostructures were prepared by thermally evaporating 5 nm of Cr as a sticking layer followed by 150 nm of Au or Cu (Lesker PVD e-beam evaporator) onto a Si substrate. A resist layer of PMMA/MMA 9% in ethyl lactate (MicroChem) followed by a layer of 2% 950 k PMMA in anisole (MicroChem) were spin coated onto the samples, followed by exposure with an electron beam for lithography using a Tescan FE-SEM instrument. A 100 nm Au or Cu layer was then evaporated on top, and the polymer resist was removed using acetone.
Spectra: Reflection spectra were taken using a Witec RA300 confocal microscope with a 100 × 0.9NA objective and white light source. These spectra were normalized to the source spectrum to find the absorptivity. Using the same microscope, anti-Stokes Raman spectra were collected using a vacuum stage (Linkam TS1500V) evacuated from atmosphere to a pressure of 0.010 mbar. Spectra were collected using a 532 nm CW Nd:Yag laser and a 20 × 0.4NA objective.
In order to explore differences in hot electron dynamics between gold and copper nanostructures, we used electron beam lithography to fabricate nanostructures of the same geometry made from both metals. The nanostructure consisted of nanopillars with a diameter of 250 nm, a height of 100 nm, and a pitch of 700 nm on a metal film of 150 nm as shown schematically in Fig. 1(a). Optical and SEM images of the nanostructures are displayed in Figs. 1(b)–1(e). Both the gold and copper nanostructures are more absorbing than their respective films; however, the copper nanostructure is approximately twice as absorbing as the gold nanostructure for this geometry. An additional benefit of studying nanostructures, rather than smooth films, is that nanostructures provide extremely large surface-enhanced Raman scattering (SERS) enhancement of the Raman signal.23
The photothermalization dynamics of samples were probed by an analysis of an anti-Stokes Raman signal that resulted during optical excitation (Fig. 2). The beam diameter of our laser exceeded the pitch of the nanostructure; thus, our measurements represent a spectral average of several cylinders. Several reports, including previous studies from our lab, have established that this broad anti-Stokes Raman signal reflects the energetic distribution of the electron gas in the metal. Currently, there is some debate whether this signal is due to photoluminescence or a coherent scattering process as in conventional Raman.24,25 In either case, the signal can be used to calculate both the lattice and the electronic temperature.22,26,27 Electrons that are equilibrated to the lattice temperature report the energetic distribution of the phonons and, thus, can be fitted to a Bose–Einstein distribution, whereas the electrons that are thermalized in the hot-electron sub-population display Fermi–Dirac statistics. The overall anti-Stokes Raman signal, I(Δω), can be modeled according to
The signal intensity is a function of the energy difference from the Rayleigh line in cm−1, as well as the lattice and electronic temperature, Tl and Te, respectively. In the steady state, there is some fraction of the population at the elevated electronic temperature, corresponding to χ. This equation also accounts for the optical density of states, D(Δω), which is approximated by a white light reflection spectrum. A constant scaling factor, C, accounts for the experimental collection efficiency and is calibrated for each measurement. A representative spectrum from the gold nanostructure and the corresponding fit to Eq. (1) is displayed in Fig. 2(a).
We show the power-dependent series of anti-Stokes Raman spectra for both gold and copper nanostructures in Figs. 2(b) and 2(c), respectively, over the same range of incident optical powers. At higher fluences in both metals, two peaks grow in at −1350 and −1585 cm−1. This signal is due to the formation of amorphous carbon on the surface and has been observed previously in tip enhanced Raman spectroscopy measurements and other SERS studies.21,26 In gold, these peaks are less pronounced, especially at lower powers; however, this signal is much more apparent from the copper samples. This is a qualitative indication that copper is reacting more readily with organic impurities in the vacuum atmosphere that have adsorbed onto the metal surface.
An additional feature in the copper signal is due to the formation of copper oxide at elevated temperatures during photothermalization. This peak can be seen growing in at −485 cm−1 and corresponds to Cu4O3.28 At higher laser powers, this peak disappears and is replaced by broader peaks at −60–220 cm−1 due to the formation of Cu2O replacing Cu4O3.28 This is more clearly visible on the Stokes side of the spectrum (see the supplementary material). This reduction in the oxidation state of copper oxide on the surface may well be the result of hot electron injection from the metal into the oxide at higher optical fluences. We note that measurements were performed in a vacuum atmosphere of 0.010 mbar in order to minimize the formation of these oxide peaks and to improve signal-to-noise for the Raman spectra.
For each nanostructure, we used Eq. (1) to calculate the dependence of Tl, Te, and χ. Because the samples exhibit a different absorptivity at the incident Raman laser wavelength of 532 nm, we plot the data in Fig. 3 as a function of the power absorbed. All of these fitted parameters are also plotted as a function of the absolute incident power in the supplementary material (Fig. 2), and there are no significant differences in the observed trends. Thermal degradation occurred in both samples when Tl exceeded 600 K, providing an upper limit for the optical power range that could be studied. We estimated that our fitted temperatures have an error range of ±7 K for Tl, ±275 K for Te, and ±0.11% for χ. For further discussion on how error is analyzed for these data, see the supplementary material.
For both nanostructures, there is a monotonic increase in both Tl and Te as the power absorbed increases. At the same power absorbed, the gold nanostructure reaches Tl that is 200 K in excess of the copper nanostructure. The gold nanostructure also reached a higher Te than the copper nanostructure. Due to the much lower heat capacity of the electron gas,3 the electronic temperatures are over an order of magnitude higher than the lattice temperatures, in agreement with other studies.21,22,26,29 As reported in our prior studies of gold nanostructures, there is an inverse relationship between temperature and the fraction of hot electrons in both samples.21,22 This inverse trend results because electron–phonon coupling is thermally activated above electronic temperatures of 2000 K.30 Therefore, the increase in the excitation rate of hot carriers at higher fluence is overwhelmed by the decrease in hot electron lifetime at elevated temperature. We observed that the copper nanostructure has a higher percentage of hot electrons compared to the gold nanostructure at the same powers.
For deeper insight into these differences between samples, we use a modified version of the two temperature model (TTM) that is standardly employed in TA studies of plasmonic metals.31 This analysis allows us to calculate a coupling constant, G, that describes the rate of energy transfer out of the hot electron population,21,22
The expression describes the heat transfer out of the hot electron sub-population in terms of the electronic heat capacity, Ce, and the optical power, Pabs, that enters the volume of metal participating in absorption. This absorption volume is calculated using the radius of our laser spot as well as the penetration depth into the metal. In the steady state, the time derivative goes to zero so that G is defined in terms of the fitted parameters reported in Fig. 3.
In the traditional TTM, G is interpreted specifically as an electron–phonon coupling constant that is an intrinsic property of the metal. In the limit that this experiment could be performed in ultra-high vacuum, the coupling constant, G, we calculate would correspond to the intrinsic electron–phonon coupling of the metal. This is because the only way a hot electron could lose its energy and return from Te to Tl would be through electron–phonon coupling, assuming negligible photoluminescence. We note that thermal energy transfer to a substrate could also impact the relaxation dynamics, if the nanostructure and substrate are strongly vibrationally coupled.32 However, at the relatively low vacuum pressure in our study (0.010 mbar), we understand that multiple additional relaxation pathways exist for the hot electrons. Specifically, the hot electrons can participate in chemical reactions with surface adsorbates and gas molecules. Indeed, in previous studies, we established that moving a gold nanostructure from vacuum to ambient atmospheric pressure increased the coupling constant, G, dramatically by 2 orders of magnitude at electronic temperatures of 1 × 104 K, due to the increased interaction with chemical adsorbates and gas molecules in the environment.21,22 Thus, trends in G, as we report here, also serve as an indication of trends in the chemical reactivity of hot electrons with surface absorbed species. This interpretation is further supported by the observation of the formation of amorphous carbon and copper oxide when samples are subjected to higher laser fluence.
The fitted values of G are plotted in Fig. 4(a) for both copper and gold as a function of the temperature difference between Te and Tl. The magnitude of G is thermally activated.30 Because gold and copper reached different temperatures when the same optical power was absorbed, we compare the trends in G when hot electrons have same degree of excitation compared with the metal lattice, i.e., the same hot electron temperature Te. Under these more equivalent conditions, we find that the coupling constant for copper is higher than that for gold. This implies that the hot electrons in copper are more reactive than those in gold and, thus, are more likely to participate in chemistry at the copper surface, even though both populations of hot electrons in the separate metals are at the same temperature. Again, this interpretation is supported by the larger amounts of deposited amorphous carbon and greater copper oxidation indicated by the Raman spectra. Given that the hot electrons in copper are using their energy to participate in chemistry, rather than increasing the temperature of the system, this also can explain why copper obtains a drastically lower lattice and electronic temperature than gold when the same optical power is absorbed.
To give further perspective upon this behavior, we also calculate the lifetime, τ, of the hot electrons in both systems based on the steady state Raman signal. Lifetime can be determined by analyzing the size of the population of hot electrons, assuming that every absorbed photon produces a hot electron,
Here, ρ is the electron density,33 V is the interaction volume of the metal with the light, N is the incident number of photons per second, and α is the measured absorptivity at 532 nm.
In Fig. 4(b), we present the fitted hot electron lifetimes as a function of the power absorbed by the nanostructure. For any given power, copper has a significantly longer lifetime than gold. This is at first surprising since trends in G suggest that the hot electrons in copper are more reactive and, thus, are expected to be shorter-lived. However, this apparent contradiction is resolved when we analyze lifetime in terms of the temperature difference between the hot electrons and the lattice, as seen in Fig. 4(c). As mentioned above, the rate of hot electron relaxation is thermally activated, and the large differences in lifetime between hot electrons in copper and gold reflect the significantly higher temperature gold reaches compared to copper when absorbing the same optical power. When we compare hot electrons at the same temperature, we see that hot electrons in copper are, indeed, shorter lived by approximately half an order of magnitude, consistent with the interpretation that the hot electrons in copper are more likely to relax by participating in surface chemical reactions. Future experiments with samples in ultrahigh vacuum can also be used to deduce how much of the difference in lifetime may be due to the intrinsic differences between the electron–phonon coupling in the metals, when there are not relaxation pathways provided by surface adsorbates.
Finally, we comment on the very long lifetimes of hot electrons we measure in both metals, up to 100 ns, in comparison with the ps lifetimes commonly measured in TA experiments. The lifetimes also increase monotonically as the absorbed optical power is decreased. Again, this trend partially reflects that hot electron relaxation is thermally activated. As the optical power is decreased, the hot electron temperature obtained is lower, so the hot electrons relax more slowly. We note that the lifetimes we observe at the highest optical powers we probe are comparable to lifetime values reported in TA studies at equivalent optical powers.34 However, we hypothesize that the long lifetimes may also indicate that multiple photon absorption events excite the hot electrons, effectively extending their lifetime, before they thermalize with the lattice. At the lowest optical power densities in this report, the time between photon absorption events per nanoscale hot spot is shorter than the ∼ps lifetime of hot electrons established in TA measurements. In the low power limit where the time between photon absorption events is much longer than the electron–phonon coupling rate established in TA experiments, the lifetime may decrease significantly because there would be a small chance for the dilute population of hot electrons to absorb a second photon before thermalizing with the lattice. We are working to improve signal-to-noise in our experiments so that we can probe this lower fluence range, in order to understand trends in behavior under conditions even more comparable to solar fluence, ∼103 W/m2.
In this study, we used an anti-Stokes Raman thermometry technique to analyze the power dependence of the lattice temperature and the electronic temperature, as well as the fractional population of the non-equilibrium hot electrons in nanostructures of identical geometry fabricated from gold or copper. Our analysis provides insight into the steady state photothermalization dynamics of hot electrons in these different metals, informing the choice of which metal may be more appropriate for developing applications of plasmonic hot electrons. We found that gold reaches higher lattice temperatures and hot electron temperatures at lower incident optical powers compared to copper. The greater temperature increase in gold due to photothermalization is expected based on the superior plasmonic behavior of gold, i.e., lower plasmon damping, which results in a consequent increase in the local optical field concentration at plasmonic hot spots. However, our data suggest that this behavior also entails that hot electrons in gold are less reactive in surface chemical reactions. Thus, gold may be more appropriate in applications such as thermoelectrics,35 where the highest possible temperature is the most important design parameter, and photodetection,5 where efficient optical-to-electrical coupling is desirable, as opposed to the chemical reactivity of hot electrons. However, copper nanostructures may be more promising for photochemical and photocatalytic applications due to the larger population of hot electrons under equivalent optical power compared with gold. We also observe increased reactivity manifested as greater coupling of the hot electrons with molecules adsorbed on the copper surface, as well as decreased hot electron lifetimes in copper. Both the plasmon damping and the rate of hot electron decay are larger in copper.
In these experiments, the optical excitation wavelength (532 nm) also served as the probe laser wavelength for the Raman analysis. However, we believe that the same anti-Stokes analysis technique can be extended to provide information during steady state pump-probe experiments, where the Raman laser is a non-perturbative probe, with a tunable pump wavelength for generating hot electrons. Additionally, it may be possible to use this spectroscopic technique in order to understand the non-thermal photoexcited electron distribution before electron–electron scattering creates a thermal distribution, as commonly described by the so-called extended TTM.36 In combination with quantitative Raman analysis of the transformation of chemical species on the metal surface, we believe that this anti-Stokes technique may be used to analyze a range of plasmonic metals and geometries, as well as to provide strategies for best optimizing plasmonic systems for target applications and chemical reactions, especially under CW conditions more comparable to solar fluence.
See the supplementary material for the Raman spectra of Cu2O, the fitted values in Fig. 3 plotted alternately as a function of incident optical power, and an in depth discussion of error as it relates to data in Fig. 3.
This work was funded by the Air Force Office of Scientific Research under Award No. FA9550-16-1-0154. M.S. also acknowledges support from the Welch Foundation (Grant No. A-1886) and the Gordon and Betty Moore Foundation through Grant No. GBMF6882.