NANO LETTERS

Voltage-Induced Nonlinear Characteristics of Arrays of Metallic Quantum Dots

F. Remacle, and R. D. Levine*,,
The Fritz Haber Research Center for Molecular Dynamics, The Hebrew UniVersity of Jerusalem, Jerusalem 91904, Israel, Departement de Chimie, B6c, UniVersite de Liege, ` B4000 Liege, Belgium, and Department of Chemistry and Biochemistry, ` The UniVersity of California Los Angeles, Los Angeles, California 90095
Received April 7, 2002; Revised Manuscript Received May 16, 2002

2002 Vol. 2, No. 7 697-701

ABSTRACT
We show by computations that even low voltages can significantly modify the electronic states of a small compressed array of metallic quantum dots. This is because the voltage counteracts the effects of inherent disorder on the wave function. The low-lying excitations can be thermally probed. At lower voltages the array breaks into islands that are superexchange coupled. At higher voltages the response is ohmic, then it cuts out due to formation of a junction layer.

The current interest in quantum dots (QDs) is motivated in part by their potential use in quantum electronic devices and in optoelectronics. This is made possible by the development of chemical methods that enable the production of nanoparticles with very narrow size distributions.1 Even when there is some variability in the sizes of the individual dots, they can self-assemble into ordered 2D or 3D lattices with domain sizes on the order of tens of unit cells.2 The control of the properties of such solids is made possible not only through the composition and size of the dots and the nature of the ligands but also by the application of external fields, including mechanical compression, that tune the coupling of the dots3 (see ref 4 for a perspective). Here we specifically discuss the control of arrays of metallic dots using an external static electrical field, as revealed by low-temperature measurements of conductance5 and of surface potential.6 In this letter we discuss dots of relatively small size (2-8 nm in diameter). By now the quantum confinement phenomena in such individual dots is well understood.7 To discuss the coupling of dots in a compressed array one can use the model where individual dots are treated as atoms with discrete electronic states, and we here take account only of the valence electrons. This model is not always sufficient8 but it has the advantage that the physical picture can be related to the computations. We offer the additional justification that the parameters in the Hamiltonian that we use have been able to account for a number of different electronic
* Corresponding author. E-mail [email protected]. Fax 972-2-6513742. Universite de Liege. ` Hebrew University of Jerusalem. UCLA. 10.1021/nl025571z CCC: $22.00 Published on Web 06/08/2002 2002 American Chemical Society

properties of arrays of Ag nanodots as measured by Heath and co-workers, see ref 4 for a review. We are specifically interested in fairly compressed, small ordered domains of quantum dots, where the wave functions of the valence electron on neighboring dots do overlap to a non-negligible extent.9 The dominant coupling is then by exchange and not the classical-like dipole-dipole terms that are appropriate for larger dots and/or an expanded lattice.10 Indeed we take it that the energetic cost () charging energy) for moving the highermost (valence) electron from one dot to an adjacent one (which already has one electron on it) can be small as compared to the strength of the exchange coupling between dots. This coupling strength is maximal at contact and exponentially decreasing with the distance D between adjacent dots. In units of the dot radius R, it is usually the case that for D/2R < 1.3-1.5 (depending on the size of the dot), the exchange coupling already is larger than the charging energy I. So fairly compressed lattices, which are discussed in this paper, are on the left of the Mott11 insulator to metal transition. Yet such arrays are not metallic because of the variation in size of adjacent dots. The size, through the confinement, determines the energy level R of the highermost electron and the fluctuations in R are at best above 5%, which for metallic dots means that R can vary by R (0.125 eV. Electron transfer between adjacent dots is therefore not necessarily resonant, and it takes a further compression of the array before the exchange coupling can completely bridge the energy gaps between neighboring dots. This is the Anderson transition12 to a delocalized electronic state. For weaker coupling we argue

that it is still possible for two dots, which are not near neighbors but which have nearly comparable energies, to be effectively strongly coupled via nonresonant exchange coupling mediated by a sequence of dots that bridge between them. This bridge-mediated electron transfer is an analogue of what is known as superexchange in molecules.13 The difference is that the coupling of the bridge to its two anchoring dots is nonresonant because of size fluctuations. In this letter we discuss the electrical response in the intermediate to strong coupling regime so that the charging energy is small14 and the dominant effects are the fluctuation in site energies due to the variations in size, the exchange coupling, and the externally applied DC field. All three are subject to experimental control. The computations are carried out assuming that charge migration in a single ordered domain of quantum dots is coherent. The Hamiltonian in the tight binding approximation for one valence electron per dot has the form H ) iRiaai i + i,jaaj. Here i is an index of a site and the prime on i,j i the second summation restricts it to the case where i,j are neighboring sites. The essential point is that we include both size variations and the Stark shift of the voltage in the site energies, Ri ) R0(1 + Ri + eVi). R R0R is the range of variation in the site energies due to the fluctuations in the size of the dots. Ri is R times a random number between -0.5 and 0.5; R0eVi is the external potential at the position of the ith dot, computed from the geometry of the array and the applied voltage V, and e is the charge of the electron. Without an external potential the energies of different dots are uncorrelated. When a potential is applied, there is a systematic shift in the site energy but for the voltages that are experimentally accessible without destroying the sample, this shift is small compared to the energy R, Ri > eVi. It will require a much narrower size distribution or a much higher voltage drop per dot to be able to experimentally probe the other side of this inequality. When the diagonal elements of the Hamiltonian are arranged by their magnitude, they span a range of up to R so that, if there are N sites in a domain, two adjacent entries on this list differ on the average by R/N. This is a (sizedependent) rather small gap because for the hexagonal domain used in our computations, with 55 dots per side, there are N ) 8911 dots. The small gaps need not be between two dots that are physically adjacent. Say that they are separated by n other dots. Then the effective coupling between the two dots is of the order of (/R)n, and this need to be compared with R/N+cnV/ N. Here cnV/ N is the voltage drop between the two dots, c; 0 < c < 1 is a geometric factor that depends on the orientation of the two dots with respect to the electrodes; and the factor N is because the distance between the electrodes scales with the side length of a 2D array. Effective coupling is thereby possible for n >1. This is analogous to electron transfer through a nonresonant bridge,13 except that here the bridge is due to the local variation in site energies. It follows that there can be localized patterns spanning the range of n dots15 and geometrically arranged in the direction of the voltage gradient. In the limiting case V > R, the sorting of diagonal
698

elements of the Hamiltonian arranges the dots by a pattern dependent on the voltage at a given dot. The strength and range of the exchange coupling is that fitted to the second-harmonic spectral response and the frequency-dependent dielectric constant of domains of Ag quantum dots.9,16 The exchange coupling is sensitive both to size variations of the dots and to local fluctuations in the separation of neighboring dots. Both effects are taken into account17 in the functional form of . The experiments we refer to5,6 used dots of 7 nm in diameter and probed samples with different disorder (14-6%) at variable but not low compressions, D/2R <1.4. The surface potential (SP) measures the potential difference between the surface and a tip probe that is manipulated after it is moved to a new position, to remain at a fixed elevation from the surface. SP probes the charge on the surface due to the imposed voltage gradient. We compute SP as the local charge density of the unoccupied molecular orbitals, the contribution of each orbital being weighted by the temperature. Explicitly, the local value of the surface potential is determined by unocc exp(- E/kT)|Ci|2 where Ci is the amplitude of orbital at the ith site. So it is primarily the lowest lying unoccupied orbitals that are important at low temperatures. Figure 1 shows contour plots of the surface potential across the array, exhibiting the changes in the coupling with increasing voltage, for dots with moderate disorder. Weak coupling results in (Anderson-like) purely localized states. When there is effective coupling leading to domain localized states, the corresponding image is that of islands oriented along the direction of the voltage gradient. At a higher voltage the behavior is ohmic with contours of equisurface potential oriented perpendicular to the gradient. When the range of size fluctuations is more limited, the onset of the ohmic regime occurs for a lower voltage. The ohmic regime requires delocalized states. In the combined presence of disorder and voltage, the transition to delocalization depends on both. The criterion g R is valid when the voltage is low and the domain is large. When < R, ohmic behavior already sets in at a voltage that satisfies V g . This is because the voltage skews the weights of the sites in the direction of the gradient. When overcomes both the effects of size variation and the voltage, the delocalization of the states is essentially uniform across most of the array and a gradient in the surface potential is seen only in the immediate vicinity of the electrodes, at the two ends of the array. There is then a junction layer of several dots deep, Figure 2. When V g > R the behavior is ohmic, except for a junction layer. The drop in the surface potential along the potential gradient is lower for lower voltages. The computation of the conductance is in two parts. First, the transmission function is computed as a function of energy. This is done by the usual scattering formalism18,19 but with the voltage being part of the Hamiltonian of the array. In the scattering formalism, the transmission is given by the matrix elements of the transition operator, T, between the two electrodes as shown explicitly in eq 1 below. We sum over all initial sites of the array that are directly coupled
Nano Lett., Vol. 2, No. 7, 2002

Figure 2. The line charge above a row of dots vs site location for intermediate (solid curve, R ) 0.1 eV) and high (dashed curve, R ) 0.25 eV) disorder for a voltage drop of 0.11 V. Note the directional character and the appearance of junction region when V g R.

Figure 1. Contours of surface potential of a 2D array for a voltage drop along the abscissa. (a) Voltage drop of 110-4 V per dot for 5% site and packing disorder. For a lower voltage the pattern is localized and random. (b) Voltage drop of 110-3 V per dot for 2% site disorder (same amount of packing disorder).

to the electrode. The transmission to the final sites (that are coupled to the other electrode) can be added coherently or incoherently. The coherent sum is larger by a factor of about 2. The domain is not conducting as a sum over independent channels. The dominant nonlinear feature in the transmission of the array is the decrease at higher voltages, Figure 3. This is because the transmission is a weighted density of states. Explicitly, the th orbital of the domain contributes as Tj, j ) j|UGU|j j|UGU|j = i

Figure 3. The transmission of a 2D hexagonal array, with 55 dots per side, vs energy at three different values of the voltage as indicated. The other parameters of the Hamiltonian are kept constant for the three computations: D/2R ) 1.23 and a 5% site and packing disorder. The break at higher voltages seen in Figure 4 is due to the cutoff seen here and is a manifestation of the junction layer seen in Figures 1 and 2.

j|U|kk|(E - E)|kk|U|j k k

(1)

where j and j are states of the electrodes, k and k are sites of the array coupled by U to the electrodes, and the effect of the voltage enters in computing the Greens function G
Nano Lett., Vol. 2, No. 7, 2002

of the Hamiltonian and in the amplitudes Ck k| and |k of the sites on opposite sides of the array of the same orbital . It is these correlated (by the index ) amplitudes that are responsible for the shrinking of the band of conducting states at higher voltages. As already discussed (and experimentally imaged by the surface potential6), when the voltage is increased there is a gradual decrease of the weights along the voltage gradient. The decrease is particularly noticeable along the junction layer, Figure 2. In eq 1, the sites k and k are at the two opposite extremes of the
699

two junction layers. For low disorder, the difference between their zero-order energies is therefore about equal to the strength of the applied bias V. Lets say that k is on the positive bias side and k on the negative bias site, Rk > Rk. When V becomes larger than the interdot coupling strength , the amplitude Ck of a MO, , of higher energy will be large on site k but very small on site k and vice versa for a MO of low energy. Since it is the product of the amplitudes that weighs the density of conducting states in eq 1, there will be a cut off of the transmission at both high and low energy when V > . The cut off on the high energy side is clearly seen in Figure 3 where the three transmission functions are computed for increasing strengths of the bias, V, while the strengths of the interdot coupling (D/2R ) 1.23, ) 0.027 eV) and the amount of disorder (5%) are kept constant. As can also be seen in Figure 3, the transmission strongly varies from one MO to the next. This is because of the resonance condition that appears in eq 1. The orbital energies, E, are obtained by diagonalization of the Hamiltonian of the array and depend on the strength of the exchange interdot coupling, on the amount of packing and site disorder and on the strength of the bias, V. The only difference between the three transmission functions reported in Figure 3 is the strength of the applied bias. Figure 3 shows that in addition to the overall effect of the applied bias discussed above, the external voltage also affects individual orbital energies E so that the resonance condition occurs at different energies for different strengths of the external potential. The second part of computing the observed conductance is a double thermal average. One thermal average is on the Fermi distribution of the one-electron states in the electrodes. We do not approximate the difference between the Fermi functions for the two electrodes as linear in the voltage.19 Unlike a single molecule, the empty orbitals of a domain of quantum dots are rather close above the highest occupied orbital. This obvious point means that more states of the array are contributing and, if the Fermi energy of the electrode is resonant with the array, the computed conductance is significantly higher than when the linear approximation is used. The second thermal average is over the empty orbitals of the array so that, as in the computations of the surface potential, higher states can be accessed at even low temperatures. This second thermal averaging tends to reduce the range of states that are available and so counteracts the broadening of the range of states that are possible due to the voltage effect of the Fermi functions. Therefore, at lower temperatures, the net conductance is seemingly linear but it is really due to a cancellation of two effects in the regime where V is larger than the spacings of orbitals available to accept an electron from the electrode. Again, this need not be the case for a molecule but is the typical regime for any but a tiny domain. A typical current-voltage (I-V) curve is shown in Figure 4. At low voltages and low temperatures the domain is not conducting because the Fermi level of the electrodes is a shade below the lowest unoccupied orbital. The trivial type of nonlinearity is that increasing the voltage brings the
700

Figure 4. Current (quantum units)-voltage curves at four temperatures for a fairly compressed array (D/2R )1.21, 5% in site and packing disorder). The increase in conductance with temperature is due to excited states of the array becoming available. The conduction gap at low voltages and low temperatures is due to the energy difference between the Fermi energy of the electrode and the lowest unoccupied orbital. It can be closed by increasing the voltage and/or the temperature. The break at higher voltages is due to junction effects as discussed in Figures 2 and 3 and the text.

electrode into resonance. The first break in the I-V curve, where the current is increasing, is the transition from domain localized (Figure 1a) to ohmic behavior. The breakdown is where the transmission of the array is cut out at the upper band edge, Figure 3, due to the junction layer at both electrodes. Due to the small differences in the energies of quantum dots and the weak exchange coupling when the lattice is not very compressed ( ) 0.04 eV for Figure 4), even a low voltage can significantly perturb the electronic structure and thereby lead to rather nonlinear current-voltage dependence. Furthermore, the I-V curves are not symmetric for negative and positive bias. This is unlike what is expected in the absence of disorder. It comes about because we do not make the linear approximation for the effect of the external voltage and include in the Hamiltonian the Stark effect on site energies due to the external potential. Since the random fluctuation affecting the packing and the site energies has no spatial correlation, a positive or a negative bias leads to different orbitals , which result in differences in the I-V curves for positive and negative bias. Note however that despite the absence of reflection symmetry, the overall behavior as a function of temperature is the same. In conclusion, the DC electrical response of mesoscopic arrays of exchange-coupled metallic quantum dots is computed and discussed. Both local (surface potential) and collective (conductance) measures are reported at low temperatures. We consider compressed and ordered arrays of nearly identical dots, but the computations allow for a limited variation in the size of the dots and for packing disorder. The electrical response is nonlinear because the
Nano Lett., Vol. 2, No. 7, 2002

application of even a low voltage significantly modifies the electronic states of the array. This is because we incorporate the effects of the DC field in the Hamiltonian of the array. (We also do not linearize the role of the voltage in the density of states of the electrodes). There are several effects. First, we see clear evidence for a junction layer in the array near the electrodes. This reduces the conductivity at higher, (but still quite low) voltages. In general, the voltage counteracts the effects of inherent disorder on the wave function of the array and it broadens the range of states of the array that are resonant with the electrodes. The response is ohmic when the voltage is comparable to the exchange coupling of the dots. At lower voltages the array breaks into finite islands that are superexchange coupled surrounded by an insulating background as can be probed by the surface potential. Since the electronic structure of the lower excited states is significantly perturbed by the inherent disorder of the array, thermal effects at low voltages are particularly important and offer a probe of very low lying excitations. Acknowledgment. F.R. is a Matre de Recherches, FNRS, Belgium. We thank Prof. J. R. Heath and his group, K. C. Beverly in particular, for many stimulating discussions. This work was supported by the James Franck program and used the computational facilities provided by SFB 377 (Hebrew University) and NIC (Liege University). ` References
(1) Lin, X. M.; Jaeger, H. M.; Sorensen, C. M.; Klabunde, K. J. J. Phys. Chem. B 2001, 105, 3353. Peng, Z. A.; Peng, X. G. J. Am. Chem. Soc. 2001, 123, 183. (2) Andres, R. P.; Bein, T.; Dorogi, M.; Feng, S.; Henderson, J. I.; Kubiak, C. P.; Mahoney, W.; Osifchin, R. G.; Reifenberger, R.

(3) (4) (5) (6) (7)

(8) (9)

(10) (11) (12) (13) (14)

(15) (16) (17) (18)

(19)

Science 1996, 272, 1323. Murray, C. B.; Kagan, C. R.; Bawendi, M. G. Annu. ReV. Mater. Sci. 2000, 30. Collier, C. P.; Saykally, R. J.; Shiang, J. J.; Henrichs, S. E.; Heath, J. R. Science 1997, 277, 1978. Beverly, K. C.; Sample, J. L.; Sampaio, J. F.; Remacle, F.; Heath, J. R.; Levine, R. D. Proc. Natl. Acad. Sci. U.S.A. 2002, 99, 6456. Beverly, K. C.; Sampaio, J. F.; Heath, J. R. J. Phys. Chem. 2002, 106, 2131. Sample, J. L.; Beverly, K. C.; Chaudhari, P. R.; Remacle, F.; Heath, J. R.; Levine, R. D. AdV. Mater. 2002, 114, 124. Bawendi, M. G.; Steigerwald, M. L.; Brus, L. E. Annu. ReV. Phys. Chem. 1990, 41, 477. Alivisatos, A. P. Science 1996, 271, 933. Ashoori, R. C. Nature 1996, 379, 413. Smith, B. B.; Nozik, A. J. Nano Lett. 2001, 1, 36. In Remacle, F.; Collier, C. P.; Markovitch, G.; Heath, J. R.; Banin, U.; Levine, R. D. J. Phys. Chem. B 1998, 102, 7727, we show that the overlap of the tails of the wave function of the valence electrons of two adjacent dots, as estimated from the effective mass and binding energy of the highermost orbital, is consistent with our length scale for the exchange coupling. Shiang, J. J.; Heath, J. R.; Collier, C. P.; Saykally, R. J. J. Phys. Chem. B 1997, 102, 3425. Mott, N. F. Metal-Insulator Transitions; Taylor & Francis: London, 1990. Zallen, R. The Physics of Amorphous Solids; Wiley: New York, 1983. McConnell, H. M. J. Chem. Phys. 1961, 35, 508. As seen by comparing with computations based on a Hubbard model Hamiltonian; Remacle, F.; Levine, R. D. J. Am. Chem. Soc. 2000, 122, 4084. Remacle, F.; Levine, R. D. J. Phys. Chem. B 2001, 105, 2153. Remacle, F.; Levine, R. D. J. Am. Chem. Soc. 2000, 122, 4084. Remacle, F.; Levine, R. D. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 553. Imry, Y.; Landauer, R. ReV. Mod. Phys. 1999, 71, S306. Yaliraki, S. N.; Ratner, M. A. J. Chem. Phys. 1998, 109, 5036. Hall, L. E.; Reimers, J. R.; Hush, N. S.; Silverbrook, K. J. Chem. Phys 2000, 112, 1510. Datta, S. Electronic Transport in Mesoscopic Systems; Cambridge University Press: Cambridge, 1995.

NL025571Z

Nano Lett., Vol. 2, No. 7, 2002

701