Report

Semiconductor interfacial carrier dynamics via photoinduced electric fields

See allHide authors and affiliations

Science  27 Nov 2015:
Vol. 350, Issue 6264, pp. 1061-1065
DOI: 10.1126/science.aad3459

Charge separation viewed in reflection

When light strikes a semiconductor, excited electrons travel across the interface. Y. Yang et al. applied ultrafast reflection spectroscopy to probe this process in a gallium indium phosphide system used for hydrogen generation from water (see the Perspective by Hansen et al.). Platinum and titanium dioxide (TiO2) coatings enhanced charge separation of the excited electrons from the positive holes they left behind. TiO2, however, was more effective at suppressing the reverse process of unproductive recombination.

Science, this issue p. 1061; see also p. 1030

Abstract

Solar photoconversion in semiconductors is driven by charge separation at the interface of the semiconductor and contacting layers. Here we demonstrate that time-resolved photoinduced reflectance from a semiconductor captures interfacial carrier dynamics. We applied this transient photoreflectance method to study charge transfer at p-type gallium-indium phosphide (p-GaInP2) interfaces critically important to solar-driven water splitting. We monitored the formation and decay of transient electric fields that form upon photoexcitation within bare p-GaInP2, p-GaInP2/platinum (Pt), and p-GaInP2/amorphous titania (TiO2) interfaces. The data show that a field at both the p-GaInP2/Pt and p-GaInP2/TiO2 interfaces drives charge separation. Additionally, the charge recombination rate at the p-GaInP2/TiO2 interface is greatly reduced owing to its p-n nature, compared with the Schottky nature of the p-GaInP2/Pt interface.

Semiconductor photoelectrodes used in photoelectrochemical (PEC) cells directly convert sunlight into stored chemical potential (15). Junctions that form between a semiconductor and a contact layer are the key to charge separation that drives photoconversion processes. Equilibration of chemical potential at such junctions creates an internal electric field (referred to as the built-in field) and establishes a region where mobile charges are driven away known as the depletion region. Absorption of light within the depletion region results in charge separation by the built-in field. As a result, photogenerated electrons (holes) transfer across the interface to participate in a reduction (oxidation) reaction and holes (electrons) are transported to the counter electrode via the external circuit for the oxidation (reduction) reaction. However, the photocarriers can also recombine across the same interface, and such recombination reduces the energy conversion efficiency. Thus, the carrier dynamics—charge separation and recombination across junctions—represent a key determining factor in the PEC performance. Time-resolved spectroscopies are normally employed to study ultrafast carrier dynamics in semiconductors (68) and semiconductor nanostructures (912) by probing the kinetics of the spectral features of initial and/or final states. Time-resolved surface-sensitive spectroscopy also has been exploited to study the semiconductor surface dynamics (13). However, isolating spectral signatures and/or the carrier dynamics that are specific to junctions—often strongly affecting subsurface depths of tens of nanometers such as in photoelectrodes—is challenging.

Here, we introduce transient photoreflectance (TPR) spectroscopy to probe the dynamics of the transient electric field (ΔF) caused by charge separation and recombination at a junction of interest. In the TPR method, the change in reflectance (ΔR) of a broadband probe from a specific interface is monitored as a function of pump-probe delay (Fig. 1). We applied TPR to study the p-type gallium-indium-phosphide (p-GaInP2) photoelectrode system, a well-known photocathode for light-driven hydrogen evolution (3, 14). We compare TPR spectral and dynamical signatures of bare p-GaInP2 with p-GaInP2/platinum (Pt) and p-GaInP2/amorphous titania (TiO2) photoelectrodes. We demonstrate that TPR can extract the dynamics of ΔF upon photoexcitation, as well as the magnitude of the built-in field (F) in these junctions.

Fig. 1 Schematic illustration of TPR spectroscopy applied to p-GaInP2.

A broadband probe pulse spanning the semiconductor bandgap is reflected from an interface of interest. An above-bandgap monochromatic pump pulse modulates the reflectance, either via band filling due to the presence of free carriers or via surface field due to charge separation across the interface. For this experiment, the pump pulse frequency was tuned to ensure an optical penetration depth shallower than that of the depletion region so that all of the photoinduced carriers were separated.

TPR spectroscopy is a pump-probe technique (Fig. 1) and in our experiment, the temporal response was ~150 fs. The pump photon energy of 3.1 eV corresponded to a pump penetration depth of 29 nm according to the absorption coefficient (fig. S1). The probe was a white-light continuum with photon energy ħω between 1.55 and 2.40 eV, corresponding to effective detection depths between 18 to 10 nm, as approximated by λ/4πn (where λ is wavelength and n is refractive index). The photocarrier density was calculated from the absorption coefficient and pump photon flux. The pump and probe beams were overlapped at the electrode surface, and the reflected probe directed to a spectrometer for reflectance spectrum (R) detection. The photoelectrode consisted of a ~1-μm-thick top layer of p-GaInP2 epitaxially grown on a GaAs substrate. The absorption coefficient and refractive index of p-GaInP2 were determined from ellipsometry characterization (fig. S1). The doping concentration was measured to be 2.7 × 1017 cm−3 from Mott-Schottky analysis (fig. S2). The platinum and amorphous TiO2 layers were deposited by atomic layer deposition.

The normalized change in reflectance (ΔR/R) at different probe photon energies (ħω) was recorded as a function of pump-probe delay. Reflectance modulations could arise from band filling by photogenerated free-carriers (15) and/or transient field changes arising from electro-optic effects (16, 17). Pseudocolor images of ΔR/R as a function of probe photon energy and pump-probe delay are displayed for the three samples in Fig. 2, A, C, and E, and their representative spectra at the specified pump-probe delays are shown in Fig. 2, B, D, and F.

Fig. 2 TPR spectra.

Pseudocolor image and spectral snapshot of TPR spectra for (A and B) p-GaInP2, (C and D) p-GaInP2/Pt, and (E and F) p-GaInP2/TiO2. Intensities of red and blue in pseudocolor images represent the magnitude of the reduced and increased reflectance, respectively. The blue and red spectra in (B), (D), and (F) are snapshots from the image at 2 ps and 1 ns delays, indicated by the dashed blue and red lines in (A), (C), and (E). For the p-GaInP2 sample, the spectra evolve over time owing to diffusion and surface trapping effects. The oscillations at 2.3 eV [red dash-line boxes in (D) and (F)] are assigned to the transition from the valence band edge to the upper conduction band (fig. S3). The black-dash traces are simulations discussed in the text and detailed in the supplementary material (SM section 3).

For the bare p-GaInP2 surfaces (Fig. 2, A and B), the data exhibit a negative and a positive band below and above the bandgap (~1.8 eV), respectively, for delays less than ~10 ps (Fig. 2B, blue circles); such spectral features can be attributed to band filling (15). Because the effective mass of electrons is much lighter than that of holes (18), the spectra are dominated by filling of conduction band states (photogenerated electrons) (19). Thus, in the low-excitation range, ΔR/R is proportional to the surface carrier density, and surface-specific carrier dynamics can be extracted (20, 21). As carriers undergo surface trapping or diffuse away from the surface, the TPR spectra evolve over ~100 ps to a derivative-like shape near the bandgap (Fig. 2B, red triangles). For bare semiconductor surfaces with flat bands, little to no field is present before photoexcitation, but a surface field forms upon photoinduced surface trapping of minority carriers (22). Because of the low surface trapping density, ΔF can be considered to be in the low-field regime, and thus, ΔR/R should resemble the third-derivative of the refractive index with respect to photon energy, Embedded Image, according to the electro-reflectance effect (16, 23). Thus, in the absence of free carriers that occupy band-edge states, ΔR/R results from the transient surface electric field (24). We simulated typical TPR spectra at short and long delay times (2 ps and 1 ns; Fig. 2B, dashed traces) based on the band-filling model and low-field electro-reflectance model, respectively [details in supplementary materials (SM) section 3.2 and 3.3]. We find good agreement between the simulation and the experimental data.

The situation is quite different in the presence of a charge-separating junction. In the p-GaInP2/Pt case, a built-in electric field (F) is present in the dark owing to the Schottky junction and drives ultrafast photogenerated charge separation, which produces a ΔF and thus ΔR. For p-GaInP2/Pt (Fig. 2, C and D), the carrier-induced spectral features are not observed at early delay times (as they are for p-GaInP2 for delays <100 ps). The absence of features associated with band- filling suggests that the electron injection into Pt is too fast to be resolved (<150 fs). Instead, a set of oscillatory features emerges immediately after excitation and remains for the whole delay time (~5 ns), with little spectral evolution occurring other than a gradual decrease of the amplitude.

Under low-fluence conditions, F is the dominant field and ΔF can be considered as a small perturbation. Because of the large built-in field, R should exhibit periodic oscillations above the semiconductor bandgap, known as Franz-Keldysh oscillations (FKO) (24), and ΔR/R will represent a small perturbation to those oscillations. According to Franz-Keldysh theory, the periodicity is determined by F, whereas the amplitude is proportional to ΔF (25).Embedded Image(1)where A(F) contains Airy functions, exhibiting the oscillatory behavior. Thus, in TPR spectroscopy, the dynamics of ΔF can be probed by following the FKO kinetics and F can be determined (SM section 3.4). Similar FKOs have been reported in photoreflectance and electroreflectance studies of p-GaInP2 junctions (26, 27). The TPR spectrum near the p-GaInP2 bandgap can be simulated by Franz-Keldysh theory (Fig. 2D, dashed trace). In our simulations, F is determined to be 158 ± 4 kV cm−1, corresponding to a depletion region width (w) of ~42 nm (w = Fε/ρ, where ε and ρ are the dielectric constant and doping charge density). There is an additional oscillatory feature at 2.3 eV (Fig. 2, D and F, red box, not resolved in the spectra of bare p-GaInP2) that corresponds to the electro-optic effect of the optical transition connecting the upper conduction band and the bottom valence band (fig. S3) (23), which is not included in the simulation. The periodicity of the oscillation remains constant for the various delays, indicative of the pertubative nature of ΔF.

The pseudo color image of TPR spectra for p-GaInP2/TiO2 (TiO2 thickness is ~35 nm) is shown in Fig. 2E. Similar to p-GaInP2/Pt, the data display FKOs, which are the signature of the electric field modulation. At early delay times, the band-filling–induced features are also not observed, implying that the electron transfer rate is faster than our temporal resolution. The TPR spectra were simulated with Franz-Keldysh theory (Fig. 2F, black-dash trace), and F was determined to be 139 ± 2 kV cm−1 (corresponding to a depletion region of ~37 nm). Owing to the observation of the field in p-GaInP2/TiO2, we conclude that the junction between p-GaInP2 and TiO2 can best be described as a p-n heterojunction.

To correlate the relationship between TPR magnitude, ΔF, and carrier density (N), we carried out TPR measurements for different excitation intensities. For p-GaInP2, the magnitude of ΔR/R at short delay (2 ps) is proportional to N (Fig. 3A, blue circles and dashed fit), consistent with theoretical prediction (see SM section 5) for the band-filling effect. In contrast, the low-field TPR signal at a long delay time (5 ns) is nearly constant with increasing N (Fig. 3A, red squares). For p-GaInP2/Pt and p-GaInP2/TiO2, the FKO amplitude is plotted as function of carrier density (Fig. 3B) on a logarithm scale. In the low–carrier density region (N < 5 × 1016 cm–3), the FKO amplitude is proportional to log(N). In keeping with the photovoltaic effect, ΔF also should be proportional to log(N). Thus, the FKO amplitude linearly depends on ΔF, which can also be theoretically deduced from the Franz-Keldysh theory (Eq. 1). The carrier-density–dependent measurements further reveal the nature of the different spectral features. The FKO amplitude begins to saturate for both the Pt and TiO2 samples when N > 5 × 1016 cm–3. In that case, ΔF becomes significant and the perturbation approximation is not valid. The FKO amplitude is higher for the p-GaInP2/TiO2 at a given photon flux because of the antireflective nature of the TiO2 layer.

Fig. 3 Carrier density dependence and TPR kinetics.

(A) Band-filling–induced and low-field–modulated TPR signal as a function of carrier density for p-GaInP2. (B) FKO amplitude as a function of carrier density plotted on a logarithmic scale. For each pump intensity, the data points represent the peak-to-peak amplitude between the first positive and negative peaks in the spectra at 50 ps delay time. The black dashed lines show a linear dependence on log(N). (C) Kinetics of the band-filling–induced TPR in p-GaInP2 (recorded at 2.2 eV), which represent the dynamics of the surface carrier density. The black dashed trace is a diffusion model discussed in the text. (D) Kinetics of FKO in p-GaInP2/Pt and p-GaInP2/TiO2 (recorded at 1.82 eV), which represent the dynamics of the transient field. The black dashed lines represent a model discussed in the text. All of the kinetic traces are scaled by normalizing the maximum to 1. In (C) and (D), the data are plotted on a linear scale for delays less than 10 ps and on a logarithmic scale from 10 ps to 4 ns (indicated by the vertical line).

Because the detection depth is less than 20 nm, the kinetics of ΔR/R caused by band–filling (p-GaInP2) represent the dynamics of photogenerated electrons in the detection region (surface carriers). The kinetics (Fig. 3C) were recorded at 2.2 eV to avoid overlapping with the trapping-induced derivative-like features (Fig. 2B, red trace). The bulk lifetime in p-GaInP2 at low carrier density is ~3 ns (28); therefore, surface trapping and carrier diffusion into the bulk are responsible for the short surface lifetime. Furthermore, the surface carrier dynamics are independent of excitation density (fig. S6), suggesting that the fast decay kinetics are not affected by bulk recombination (21). We fit a diffusion model to the kinetic trace with the ambipolar diffusion coefficient and the surface recombination velocity (SRV) as best-fit parameters with values of 0.5 cm2 s–1 and 500 cm s–1, respectively. Because the electron diffusion coefficient is much higher than the determined ambipolar diffusion coefficient (28), carrier diffusion is limited by holes, not electrons. The low SRV also implies a low density of surface trapping states. The rise of the signal corresponds to carrier cooling as the carrier distribution narrows in k-space, and the cooling time is determined to be 0.30 ± 0.01 ps.

Because of the linear relationship between FKO amplitude and ΔF, the dynamics of ΔF for p-GaInP2/Pt and p-GaInP2/TiO2 can be directly probed by following the FKO kinetics (Fig. 3D). The increase in ΔF results from the interfacial charge separation, and the kinetics of both samples show biphasic behavior. Our modeling indicates a common fast component (0.3 ps) for both kinetic traces, which we attribute to carrier cooling. The time constant of the slow component for p-GaInP2/TiO2 (5.7 ± 0.3 ps) is longer than that for p-GaInP2/Pt (2.8 ± 0.3 ps). The formation of ΔF results from electron and hole separation, and the electron transfer time is shorter than the time resolution (~150 fs); therefore, the slow component is attributed to holes that drift from the surface to the bulk. The electron and hole drift time can be estimated from their respective mobilities, distance over which they travel, and F. The average distance for electrons was approximated as the pump penetration depth (29 nm); the average distance for holes was determined by the depletion region width (~43 nm), and F is determined by the FKO simulations discussed above. The mobilities in p-GaInP2 are on the order of 103 cm2 V–1 s–1 for electrons and 10 cm2 V–1 s–1 for holes, and the corresponding drift times are ~20 fs and ~3 ps for electrons and holes, respectively, which is consistent with our measurement. Thus, the slower growth of ΔF in p-GaInP2/TiO2 arises from the smaller built-in electric field. The decay of ΔF reflects the interfacial charge recombination. We determined transient field decay time constants of 104 ± 25 and 11.3 ± 0.5 ns for p-GaInP2/TiO2 and p-GaInP2/Pt, respectively. According to the relationship between Embedded Image and ΔR/R (Fig. 3B), the interfacial charge recombination time can also be extracted from the field dynamics (fig. S7). We summarize our findings for p-GaInP2, p-GaInP2/Pt, and p-GaInP2/TiO2, by schematic illustrations in Fig. 4.

Fig. 4 Energy band and charge flow diagram.

Band bending and carrier dynamics at the surface or interface for (A) p-GaInP2, (B) p-GaInP2/Pt, and (C) p-GaInP2/TiO2. The energy band positions are determined by XPS and UPS characterization and the optical bandgaps. The energy scale bar (left) is referenced to the p-GaInP2 vacuum level. (A) For p-GaInP2, the photocarriers are initially generated near the surface and are primarily depopulated by diffusion to the bulk and surface trapping. The trapped electron and the remaining hole form a weak transient electric field at the surface, which modulates the reflectance at longer delay times. In contrast, (B) the Schottky (p-GaInP2/Pt) and (C) p-n (p-GaInP2/TiO2) junctions establish a depletion region width larger than the pump penetration depth. These built-in fields accelerate electrons and holes toward Pt/TiO2 and the bulk, respectively. The electron transfers from the depletion region to the interfacial layer within the time resolution (~150 fs) of the experiment, whereas the hole requires several picoseconds to drift out of the depletion region. The charge separation screens the built-in field and forms the FKOs in TPR spectra. Though the charge separation process is similar, the recovery of ΔF arising from charge recombination at the Schottky (p-GaInP2/Pt) junction is faster (by about one order of magnitude) than that at the p-n (p-GaInP2/TiO2) junction.

Our results uncover key factors that enhance the photon conversion process in amorphous TiO2-coated electrodes. For the bare electrode, surface trapping leads to localized electrons at lower energy states and diffusion transports electrons away from the surface, both of which are harmful to efficient photoconversion. Thus, appropriate junctions at the electrode interface (formed with Pt and TiO2) produce surface fields that can drive electron-hole separation and promote high charge separation yields. However, the nature of the junction (Schottky or p-n) has a profound influence on the carrier dynamics following separation. Compared with Pt, the amorphous TiO2 greatly retards the interfacial charge recombination without appreciably sacrificing the charge separation rate. The barrier for electron-hole recombination at the Pt interface is determined by the Schottky barrier height, which from our data we estimate as ~0.3 eV (ϕ = F·w/2), derived solely from p-GaInP2 valence band bending (Fig. 4B). In contrast, both electrons and holes are driven away from the interface in p-GaInP2/TiO2 (Fig. 4C) by the built-in fields that exist on either side of the junction. We estimate that the charges become separated by ~72 nm (total depletion width in GaInP2 and TiO2), establishing a substantial kinetic barrier to recombination. In addition, x-ray photoelectron spectroscopy (XPS) and ultraviolet photoelectron spectroscopy (UPS) characterization also show that the thermodynamic barrier to electron (hole) injection into p-GaInP2 (TiO2) for bulk recombination is unlikely, owing to the large band offset (1.88 eV in valence band and 0.64 eV in conduction band, Fig. 4C), and thus the recombination should occur at the p-GaInP2/TiO2 interface. In the latter scenario, both an electron in TiO2 and a hole in p-GaInP2 must overcome barriers of >0.3 eV (measured by XPS and UPS; see SM section 8 for details) and 0.26 eV, respectively, and find one another likely assisted by an interfacial recombination center. Thus, the interfacial recombination probability is further reduced by the energy barrier and low density of interfacial defect sites. Therefore, we conclude that the observed slower decay of ΔF in p-GaInP2/TiO2 versus p-GaInP2/Pt can be attributed to both kinetic and thermodynamic barriers that result from the p-n nature of this junction.

We investigated the effect of the amorphous TiO2 layer thickness on the interfacial carrier dynamics to explore the limiting thickness required to observe the beneficial effects of the p-n junction. For TiO2 thicknesses from 0.5 to 5 nm, the TPR spectral shape and magnitude are similar to the low-field modulation features found in p-GaInP2 at long delays (fig. S9A). The modulation field arises from charge separation via electron transfer from p-GaInP2 to TiO2. As the TiO2 thickness increases from 10 to 35 nm, the TPR spectra of p-GaInP2/TiO2 exhibit oscillations similar to that for p-GaInP2/Pt, meaning that the surface field increases substantially when the TiO2 thickness increases from 0.5 to 35 nm. The formation and decay time constant of ΔF for these samples are extracted from the corresponding TPR kinetics (fig. S9B). Best-fit parameters are tabulated in table S1. Thicker TiO2 layers exhibit slightly faster field formation but slower decay, which is likely due to the larger built-in field that drives carriers apart and separates them at a greater distance, both of which lead to slower recombination. We find that the kinetics are effectively unperturbed once a sufficient amorphous TiO2 thickness has been reached, suggesting that thicker layers would not drastically influence the photoconversion performance from a charge dynamics perspective. A thick TiO2 layer may still be necessary for other reasons (such as elimination of pinholes) that affect stabilization against photocorrosion, as has been found for 140-nm-thick amorphous TiO2 layers on Si, GaAs, and GaP photoanodes (2).

Our results uncover key beneficial roles of amorphous TiO2 in the energy-conversion process that have come under intense investigation after several recent reports of TiO2-stablized photoelectrodes (2, 29, 30). The TPR technique developed here furthermore introduces a general method to understand charge transfer at semiconductor junctions.

Supplementary Materials

www.sciencemag.org/content/350/6264/1061/suppl/DC1

Materials and Methods

Figs. S1 to S9

Tables S1 to S4

References (31, 32)

References and Notes

  1. Acknowledgments: This work was supported by the Division of Chemical Sciences, Geosciences and Biosciences, Office of Basic Energy Sciences of the U.S. Department of Energy, through the Solar Photochemistry Program under contract no. DE-AC36-08GO28308 to the National Renewable Energy Laboratory. J.L.Y. acknowledges NSF Graduate Research Fellowship Grant no. DGE 1144083. The U.S. government retains— and the publisher, by accepting the article for publication, acknowledges that the U.S. government retains— a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. government purposes.
View Abstract

Navigate This Article