Report

Slow Electron Cooling in Colloidal Quantum Dots

See allHide authors and affiliations

Science  07 Nov 2008:
Vol. 322, Issue 5903, pp. 929-932
DOI: 10.1126/science.1159832

Abstract

Hot electrons in semiconductors lose their energy very quickly (within picoseconds) to lattice vibrations. Slowing this energy loss could prove useful for more efficient photovoltaic or infrared devices. With their well-separated electronic states, quantum dots should display slow relaxation, but other mechanisms have made it difficult to observe. We report slow intraband relaxation (>1 nanosecond) in colloidal quantum dots. The small cadmium selenide (CdSe) dots, with an intraband energy separation of ∼0.25 electron volts, are capped by an epitaxial zinc selenide (ZnSe) shell. The shell is terminated by a CdSe passivating layer to remove electron traps and is covered by ligands of low infrared absorbance (alkane thiols) at the intraband energy. We found that relaxation is markedly slowed with increasing ZnSe shell thickness.

When semiconductors are formed into quantum dots (QDs), discrete electronic states arise through confinement by the boundary. These states can be exploited in optical applications, where size-tunable narrow fluorescence emissions with lifetimes in the nanosecond range are useful (1). However, such narrow fluorescence emission entails the dissipation of any initially absorbed excess energy as heat within hundreds of femtoseconds (2, 3). This rapid energy loss (“electron cooling”) is not useful in electronics applications such as photovoltaics or infrared (IR) devices. Slower dissipation might give time to extract the energy of hot carriers for more efficient photovoltaics (4), and it would also enable the detection and emission of IR radiation via the intraband transitions of quantum dots.

In small QDs, such as colloidal CdSe, with a low electron mass, slow dissipation is in fact predicted on the basis that the lattice vibrations (30 meV) cannot couple widely separated electronic states (300 meV) (58) (Fig. 1B). The striking absence of such a “phonon bottleneck” in photoexcited QDs has been explained by fast excitonic cooling, where the electron transfers its energy to the much larger density of states of the more massive hole (911) (Fig. 1C). Nonetheless, efforts to decouple the electron and hole led to marginally longer picosecond relaxation times (1214), raising the possibility of other mechanisms involving local intermediate states, generically called traps (15, 16), and high-frequency local modes (17) or molecular vibrations (14).

Fig. 1.

(A) In bulk semiconductors, hot carriers cool by sequential emission of phonons. (B) However, in a small dot, electronic states are too far apart for phonon emission. (C) An electron can cool by transferring its energy to the hole with higher state density that then relaxes via phonons. (D) With a reduced coupling to the hole, shown here as surface-trapped, the electron can cool via intermediate trap states. (E) The electron can transfer its energy to a resonant high-frequency vibration.

In our experiments, electron cooling between the two lowest conduction band states of small colloidal CdSe dots, 1Se and 1Pe, is slowed to longer than 1 ns, which is more than three orders of magnitude slower than the relaxation time between the 1S and 1 Pexciton (2). This slower cooling is achieved by using a thick ZnSe shell to separate electrons and holes and to increase the distance of the electronic states from the ligands. We observed three complementary mechanisms shown schematically in Fig. 1. With an exposed ZnSe surface, electron trapping takes place on a time scale of 10 to 30 ps (Fig. 1D). Capping the ZnSe shell with one monolayer (1 ML) of CdS reduces electron trapping, but it can also prevent hole trapping and allows fast excitonic cooling (<6 ps) (Fig. 1C). A CdSe capping monolayer reduces electron trapping as well, but with hole-trapping ligands, electron cooling slows from 10 ps to >1 ns, in agreement with a mechanism of energy transfer to vibrations via dipole coupling (Fig. 1E).

In our experiments, the electron in the conduction band state of the dots is photogenerated, which leaves a hole initially in the valence band. Reducing electron-hole coupling requires a shell that preserves electron confinement and extracts the hole to a remote state, as shown in Fig. 2A. To avoid intermediate local electronic states or vibrational modes that could arise at defects, the shell must be epitaxial, and its surface should have no states that would trap the electron. To limit coupling to ligand vibrations, the shell should be thick and its outer surface capped by ligands of low IR absorbance.

Fig. 2.

(A) Schematic of the electron-confining and hole-extracting colloidal structure used to probe electron relaxation from 1Pe to 1Se. (B) TEM pictures of CdSe/ZnS1ML/ZnSe9ML/CdSe1ML. Scale bar, 5 nm (inset, 20 nm). The CdSe core is ∼3 nm in diameter. (C) Absorption spectra at various stages of shell growth.

To satisfy these criteria, we used a combination of shell materials resulting in a CdSe/ZnS1ML/ ZnSexML/CdSe1ML structure with the core CdSe in the zincblende structure (18). We used alkane thiolate ligands, which act as hole traps on chalcogenide QDs. They also have no hydrogen bonding mid-IR background, and their IR absorption is smallest between the CH stretches and CH bends, between 0.3 and 0.23 eV (14). This energy window determines where we set the 1Se-1Pe intraband resonance and hence the core size. For the electron-confining/hole-extracting shell, we chose ZnSe because it has a weak type II alignment with CdSe, with a conduction band offset of ∼1 eV (19). However, only thin ZnSe shells (<1 nm thick) have been grown previously (20). To grow a thick epitaxial and spherical ZnSe shell up to ∼12 ML (each monolayer is ½ lattice constant, i.e., 0.28 nm), a high temperature (280°C) was required. To retain the monodispersed core size at that temperature, we found that an initial ZnS layer (∼1 ML) was most helpful. Finally, with QDs terminated by ZnSe, the electrons were also easily trapped; this problem was solved by growing a final monolayer of CdSe.

The samples were characterized by optical spectroscopy and transmission electron microscopy (TEM) at all stages of the growth. The TEM images in Fig. 2B show the spherical and epitaxial shell growth. The standard deviation of the number of ZnSe monolayers (estimated from sizes) varies from <1 for thin shells to ∼2 for the thickest shells. The optical spectra taken at various stages of growth are shown in Fig. 2C. With increasing ZnSe shell thickness, the weakly shifting lower-energy exciton peak (called the S-exciton) confirms the retention of monodispersed cores and the good electron-confining potential of ZnSe. The disappearance of several higher-energy features (e.g., the light hole to 1Se transition seen at 510 nm in Fig. 2C) that arise from the discrete valence states is consistent with the increasingly delocalized hole in ZnSe (19). For the transient measurements, the cleaned QDs are dissolved in C2Cl4 and placed in a 1-mm-thick cell.

The electron cooling time was measured by a sequence of three optical pulses designed to (i) prepare the dots with one electron in 1Se, (ii) resonantly pump the electron from 1Se to 1Pe, and (iii) probe 1Se recovery and 1Pe depopulation (13). The setup and procedure have been described for CdSe cores (14). To create a population of dots with one electron in 1Se, we use a green pulse (0.532 μm) above the S-exciton. Electrons and holes are created, and after multicarrier recombination (occurring within ∼100 ps) and cooling, the dots are left with a single electron and a hole. The hole is not detectable in absorption, whether it is trapped or delocalized, because of weak overlap or high degeneracy, respectively. The electron in 1Se can be clearly identified by the appearance of the strongly allowed 1Se-1Pe intraband transition, as well as by the bleach of the S-exciton, because 1Se is only doubly degenerate. Figure 3A shows a typical bleach probed by a tunable visible pulse 600 ps after the green pulse. The bleach has an optical density of 0.04, which corresponds to ∼15% of the sample optical density and implies that ∼30% of the dots have one electron in 1Se. The electron remains in 1Se until radiative and nonradiative recombination with the hole or trapping. As such, recombination/trapping of the 1Se electron takes place over ∼1 ns for ZnSe-terminated surfaces and much longer (≫1 ns) with a CdS and CdSe outer monolayer.

Fig. 3.

(A) Absorption spectrum (red line) and transient bleaches of a CdSe/ZnS1ML/ZnSe5ML/CdSe sample taken 600 ps after the green pulse (blue circles) and 10 ps after the IR pump (green circles). The left schematic shows how a 1Se electron mostly weakens the S-exciton absorption (from black to red), whereas a 1Pe electron mostly red-shifts the S-exciton. The bleach after the green pump is due to state filling by the 1Se electron. The bleach after the IR pump reflects state filling by 1Se electrons and redshift caused by 1Pe electrons, which is most visible at A1. (B) Transient bleach Δα/α at B1 (red dots) and A1 (blue dots) for the three different samples discussed in the text. The solid lines are biexponential fits (left and right graphs) and a Gaussian (center graph).

At a varying delay after the green pump, an IR pump pulse (∼4 μm, 20 μJ) resonantly excites the electron from 1Se to 1Pe. This excitation reduces the bleach in Fig. 3A to α – Δα, where α denotes the bleach without the IR pump. The probe for the 1Se population is at the bleach peak, at position B1 in Fig. 3A. [The notation B1 was introduced in (2) for CdSe cores.] At B1, Δα mostly detects the change of the 1Se occupation and Δα/α gives the fraction of dots for which the IR pump removed the electron from 1Se. This is ∼45% in Fig. 3A. The probe for the 1Pe population is at the position noted A1 (2) on the red edge where α is smaller. Δα at A1 is caused by both the reduced bleach that results from the loss of the 1Se electron and the absorption by the redshifted exciton (3), as shown in the schematic in Fig. 3A. The redshift is explained by a stabilizing polarization interaction of the S-exciton with the electron charge in 1Pe. The factor of 5 to 10 difference in the magnitude of α between A1 and B1 is normalized out by the ratio Δα/α. If the electron relaxes from 1Pe directly to 1Se, A1 and B1 follow identical kinetics but Δα/α is larger at A1. If the electron leaves 1Pe by recombination/trapping, Δα reflects only the loss of the 1Se electron; hence, it is a fractional change of α, and Δα/α has identical values at the B1 and A1 positions.

Figure 3B shows the very different results obtained with different outer surfaces. With an outer ZnSe surface capped with amine and carboxylate ligands, Δα/α at A1 and at B1 merge to a constant value after ∼20 ps. This result shows that the electrons leave 1Pe in 20 ps but that most, ∼80%, do not relax to 1Se because B1 does not recover. The response may be interpreted as being inhomogeneous with different dynamics for different dots, where 80% of 1Pe electrons undergo rapid recombination/trapping and 20% relax to 1Se within 20 ps. Alternatively, the data can be fit to a three-level system model with decay from 1Pe via recombination/trapping (23 ps) and intraband relaxation (74 ps) (21). The fast recombination/trapping of the electron in 1Pe is consistent with its poor stability in 1Se for this ZnSe-terminated surface.

In contrast, with an outer monolayer of CdS terminated with amine ligands, the 1Se electron is much more stable but IR pumping only shows a very weak and prompt Δα/α, ∼7% maximum for B1 and within the noise at A1. This result shows that the 1Pe electron is short-lived—shorter than the laser pulse width of 6 ps—and it also shows that 1Pe completely relaxes to 1Se. For this system, the low-valence band of CdS confines the hole into the core + shell, and the hole is not trapped by amine ligands. This lack of trapping reduces the decoupling between electron and holes and leads to bright photoluminescence, at least for moderate shell thicknesses. The very fast relaxation confirms the importance of decoupling electrons and holes to observe slow electron cooling.

Our focus here is on the QDs that exhibit slow electron cooling. These are capped by a monolayer of CdSe treated with dodecanethiol (DDT) hole-trapping ligands. For this system, the fluorescence is quenched, as should be the case with well-decoupled electrons and holes, but the intraband spectrum and bleach α are also strong, indicating that CdSe termination, like CdS, removes electron traps. As shown in Fig. 3B, for this system, Δα/α at B1 is large, typically 50%, and long-lived. A1 and B1 have similar kinetics, but the magnitude for A1 is much greater. This difference shows that the electron stays in 1Pe for a long time and relaxes to 1Se with no intermediate state. We note that in all samples the relaxation is not a single exponential. In Fig. 3B, 60% decays with a time constant of 1.7 ± 0.3 ns, and 40% decays in 30 ± 5 ps. This range of kinetics might be caused by sample inhomogeneity because the response is so sensitive to the outer surface. We think that it could also be caused by the inhomogeneous width of the intraband absorption, across which the sample molecular IR absorption varies by one or two orders of magnitude.

We now discuss the role of ligand vibrations. Consider an intraband transition with a dipole p and an IR-absorbing surface. A dipole-coupling model gives an energy transfer time constant Tnr, Embedded Image(1) (14), where R is the outer radius of the QD, λ is the wavelength, σ is the infrared cross section at the intraband transition caused by the surface ligands, ϵ0 is the medium dielectric constant, ϵ1 is the material dielectric constant, and Embedded Image is Planck's constant divided by 2π. This model predicts a linear relation between relaxation rate and IR absorption σ. To date, alkane thiols and alkyl amines are the ligands with the lowest IR absorbance that we have used, and they always show the slowest relaxation times. Phosphonic or carboxylic acid ligands always lead to increased IR absorption (from broad hydrogen-bonding vibrations) and faster relaxation. Even mild exposure of a slow sample to these acid ligands leads to faster relaxation (Fig. 4A). Starting with a thiol-capped sample where A1 and B1 follow biexponential relaxation with a slow relaxation time of ∼180 ps, a room-temperature exposure to stearic acid increases by a factor of ∼7 the IR absorbance of the sample in the vicinity of the intraband transition. Correspondingly, the electron now relaxes from 1Pe to 1Se in ∼12 ps. The low magnitude of the long tail seen in the A1 and B1 response also shows that it mostly returns to 1Se, indicating small trapping. This correlation between increased ligand absorbance and faster relaxation supports the role of vibrations in the cooling dynamics.

Fig. 4.

Sensitivity of the intraband relaxation dynamics to the sample IR absorption and shell thickness. (A) CdSe/ZnS/ZnSe8ML/CdSe/DDT. A1 and B1 show the 1Pe decay with a slow time constant of 180 ± 40 ps. After exposure to stearic acid, A1 and B1 show faster recovery. The intraband spectrum of the sample (linear scale) is shown with solid dots and Gaussian line fit. The infrared absorption of the sample (log scale) is shown before stearic acid exposure (dotted line) and after exposure (solid line). (B) Slow relaxation time constant plotted against the outer radius for CdSe/ZnS/ZnSe/CdSe/thiolate of different ZnSe shell thicknesses. The line is an R4 fit.

Equation 1 further predicts that if the electron-confining shell is grown with similar ligand coverage, σ will increase as R2, such that Tnr will increase as R4. Figure 4B shows data consistent with this predicted scaling. The slow time constant increases from ∼10 ps to ∼1.7 ns for samples with similar intraband energy (kept within 27 ± 3 meV for all the samples) but with shell thickness increasing from 0 to ∼3.5 nm. Finally, the absolute magnitudes of lifetimes also agree with predictions from the equation. Using p = 0.3eR0 (where R0 is the core radius) and σ ∼2.10–18 cm2, measured in the 0.28-eV region for thiol-passivated CdSe QDs with R = R0 = 2.25 nm, yields Tnr ∼23 ps. A shell 3.25 nm thick, for a total radius of 5.5 nm, would give Tnr ∼ 1 ps, in agreement with the data in Fig. 3B. Thus, it is likely that the vibrational coupling mechanism is now dominating and that even longer lifetimes will be achieved by using ligands with lower IR activity, more complete exchange procedures, or fully inorganic matrices.

Slow electron cooling had long been expected in quantum dots, but it had remained elusive in colloidal dots as well as in those grown by molecular beam epitaxy (2224). Colloidal quantum dots allow a strong confinement, which should provide a good decoupling between the electronic states and the lattice vibrations. However, to observe slow (>1 ns) electron cooling, we had to use core/shell quantum dots designed to reduce several other ways by which energy is dissipated: via coupling to holes, intermediate trap states, and molecular vibrations. The cooling time could be tuned from less than 6 ps to longer than 1 ns with identical cores and different shell compositions and thicknesses. For emitting or detecting IR radiation, the increase in cooling time by three orders of magnitude opens the prospect of using the intraband transitions of colloidal dots. Proposals to attempt to extract the energy of hot carriers from quantum dots to improve photovoltaic yield are also better validated.

Supporting Online Material

www.sciencemag.org/cgi/content/full/322/5903/929/DC1

Materials and Methods

Figs. S1 to S6

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article