Anda di halaman 1dari 14

J. Phys. Chem.

C 2007, 111, 4431-4444 4431

Analyzing Molecular Current-Voltage Characteristics with the Simmons Tunneling Model:


Scaling and Linearization

Ayelet Vilan†
Ilse Katz Center for Nanotechnology, Ben Gurion UniVersity, Israel
ReceiVed: October 18, 2006; In Final Form: December 31, 2006

Use of the Simmons model for analyzing tunneling transport across molecular junctions is reviewed, and its
inherent limitations are examined, specifically for cases where there are no molecular length-dependent data
(to extract a decay parameter), be it for experimental reasons or because of changes of the molecular energetics
or packing with molecular length. The potential barrier across a molecular junction is shown to be strongly
bias-dependent, much more so than is assumed in the commonly used version of the Simmons model. The
means to distinguish true tunneling from conduction via pinholes (or hot spots) are also considered. Power
expansion to the Simmons model shows that I/V vs V2 plots should be linear over that range, thus providing
a simple and standardized parameter extraction. From such plots, we can extract values for the equilibrium
conductance and for the “shape factor”, a complementary parameter that describes the shape of the I-V
relations. The applicability of these two parameters for describing actual transport is illustrated by analyzing
data for three different types of molecular junctions. The linearity of the I/V vs V2 plots can be used to
evaluate if, and if so, at which bias (and if at all) direct tunneling occurs and under which conditions this is
not the case. The extracted equilibrium conductance and the “shape factor” provide an empirical method for
quantifying electronic transport across practical molecular junctions where the exact packing of the molecules
is rather uncertain. As such, the analysis can be used to weigh and classify the effect of chemical modifications
to molecular junctions or compare contacting methods, so as to allow a deeper understanding of transport via
molecular junctions.

1. Introduction nominal molecular structures, ignoring possible structural


uncertainties (predominantly pinholes).
Molecular electronics is a fascinating scientific frontier, which
already has demonstrated proof of concept for such phenomena Indeed, although the example, given above, suggests that
as rectification,1 negative differential resistance,2 and others.3 chemical control over the exact binding site can have far-
Yet, basic understanding of electronic charge transport across reaching implications for molecular electronics, it is still an
molecular systems is still rudimentary. One basic question is elusive goal. We would like to devise and perform sets of
that of structure-function relations,4 viz., which of the junction’s experiments to measure current-voltage (I-V) characteristics
parameters (e.g., conjugation, position, density and symmetry that respond in a controllable and reproducible manner to a
of energy levels, bonding to the electrodes) affect the current controlled change in conditions. However, even if this is
and to what extent? achieved, the experimentalist still faces the problem of how to
quantify the observations and assign them to specific functional
An extreme example for a seemingly minor structural change
properties such as energy level positions (which dictate the
corresponding to a huge functional effect was given by Tomfohr
barrier height), carrier delocalization (or coupling between
and Sankey5 for alkyl thiols on gold. On the basis of complex
energy levels), or, for molecular monolayers, defect density,
band structure calculations, they showed that varying the
something that may well affect not only the energy levels but
adsorption position from a hollow lattice site to an on-top Au
also the contact area for transport.
atom site shifts the HOMO level of the adsorbed molecule by
∼1 eV, which is expected to exponentially affect the molecular The limited knowledge on the exact molecular arrangement
conductance.5 in actual molecular junctions impedes the use of accurate
Although this is only one example of what is nowadays a theories for quantification of experimental measurements.
very significant endeavor,6-9 transport modeling and simulations Furthermore, reducing detailed theoretical models into few
generally rely on extremely well-defined atomic spatial posi- averaged characteristic parameters (e.g., barrier height or
tions, far beyond our experimental accuracy. This is important effective mass) enables decomposing experimentally measured
because, although it is accepted that charge transport across current-voltage (I-V) curves into a few basic parameters.
molecular junctions is dictated by both the spatial structure Therefore, approximate, effective models are at present a
(thickness, uniformity) and the energetics (energy levels and necessary step for quantifying the increasing amount of mo-
effective mass or coupling), analyses of molecular current- lecular transport data.
voltage (I-V) curves most commonly rely on schematic, Within the largely accepted simplifications of nonresonant
tunneling and that of the WKB (semiclassical) approximation
† Present address: Department of Chemical Research Support, Weizmann to nonresonant tunneling, one of the most ubiquitous of such
Institute of Science, Rehovot, 76100 Israel. parameters is the decay parameter (β)10-12 (in Å-1, see equations
10.1021/jp066846s CCC: $37.00 © 2007 American Chemical Society
Published on Web 02/28/2007
4432 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan

a1 and a2 of the Appendix).13 This parameter describes the The shape of the I-V curve is characterized by a parameter
exponential decay of the current, I, with the length of the which we will call the “shape factor” (F), the ratio between the
molecule (l): I ∼ exp(-βl). A major disadvantage of the decay width and the height of the tunneling barrier. These para-
concept is that it requires a set of data with systematically meters reflect the product of the height and the width of the
varying tunnel barrier width (molecular length), while keeping tunnel barrier, which dictates the current magnitude, G0,
all other molecular parameters the same. Practically, such a (see section 2.2.3, below) and the ratio between these two, (F),
strategy is limited mostly to simple alkyl chains and related which dictates the shape of the I-V curve (see section 2.2.2).
molecules, because, if functional groups are introduced (e.g., Note that G0 and F are functional parameters within which are
conjugated moieties, polar groups, redox centers), some variation contained several physical parameters, such as the junction’s
of energy levels or change in packing with molecular length is barrier height, width, and contact area. In this manuscript, we
nearly inevitable. will occasionally refer to them as lumped parameters so as to
Alternatively, measuring current as a function of temperature contrast them to the more familiar parameters that describe the
can be very informative. It can tell if transport is thermally junction.
activated (pure tunneling is temperature-independent) and, if Although the combination of these two parameters enables
so, what is the activation energy.14,15 This is a factor that needs us to unravel the known coupling between barrier height and
to be considered in the choice of experimental measurement effective mass,18 we show that this is possible only if the contact
setup, as several types (e.g., Hg drop and electrochemical area is known or if data with systematic length variation are
junctions) allow only very limited temperature variation. available (i.e., at least one spatial parameter is required). Another
Furthermore, monolayers can undergo a liquid-solid phase basic limitation of the Simmons model is the fundamental
transition16 which can naturally influence their charge transport assumption that the potential profile across the molecule(s)
characteristics. Other molecules, such as proteins, are mostly between the electrodes is linear (section 2.1). We show, for three
incompatible with vacuum or dry conditions required for quite different experimental data sets, that this so-called “linear
cooling, again limiting the range of temperature variation model” is not applicable to real systems (4.1). We suggest that
experimentally. this problem can be overcome by limiting the analysis to low
The question that we will address here is the following: Can bias voltages, because at those voltages such a linear potential
we extract from experimental current-voltage characteristics profile is mostly a reasonable approximation (2.3.1). This
functional parameters that allow meaningful comparisons approach has as further benefit that a much simpler polynomial
between experiments, in a way that is independent of tunnel I-V relations result (2.3.2) along the lines originally suggested
barrier thickness, i.e., without the need for experimental by Simmons24 and used later by Brinkman et al.28
current-thickness relations? Both the scaled analysis and the polynomial approximation
In molecular electronics8,9,14,17-21 Simmons’ inelastic tun- converge to the same characteristic magnitude and shape
neling model22-24 (see section 2 for a summary) is used often, parameters (2.3.3 and 2.3.4). The proposed analysis is tested
due to its simplicity, even though, in most cases it is likely for analysis of I-V data measured across junctions of alkyl thiols
oversimplified as it ignores important contributions such as at three very different contacting configurations (3). Alkyl chains
super-exchange17,21 or hopping.9,25 Fitting the Simmons model are by far the cleanest and most reproducible of all molecular
to I-V data (instead of I-l, current-thickness data) was found junctions and thus serve as a basic model system for molecular
to be useful for evaluating the contribution of chemical electronics studies. In addition, tunneling is expected to dominate
modifications at bilayer interfaces17 and was shown to produce charge transport across alkyl chains.14 Extracting characteristic
parameters in good agreement with I-l extracted parameters.14 parameters from actual data is tested (4.2) and compared to
Wang et al. also reported on negligible temperature dependence traditional numerical fits (4.3). Finally, it is shown that a model
of the measured current, indicating transport by tunneling rather which allows for a slightly nonlinear potential profile provides
than hopping14,15 (on tunneling vs hopping see also ref 25). an excellent fit over the full bias range (4.4).
Aswal et al. distinguished between three different bias ranges The discussion section reviews the possible merits in using
(low, medium, and very high), using simplifications for each the lumped parameters under typical experimental conditions
of these ranges.20 The present work focuses on the intermediate where the junction area29-31 and width are not well-known. In
bias range. cases of a roughly known contact area (within an order of
magnitude), the product β0l can be extracted from G0, and the
However, as noted by Cui et al.18 and Wang et al.,14,15 fitting
ratio (β0l/F) provides the barrier height. Alternatively, if length-
over different voltage intervals yields quite different results. A
dependent data (I-V-l) are available, we show that the intercept
discussion on the problematic interdependence between the
of ln(G0·F) vs F plot can be used to extract experimentally the
different parameters controlling the Simmons relations was given
contact area, A, of a molecular junction.
in ref 26. In addition, our own experience27 shows that a visually
reasonable fit to the Simmons model can be reached fairly easily,
2. Revisiting the Simmons Model: Power Expansion,
but the resulting extracted parameters are extremely unstable
Linear Presentation, and Scaling
and sensitive to the initial choice of starting values. Such
parameters can hardly be regarded as “characteristic”. This section starts with the formal definitions of the Simmons
This is the reason for reconsidering the limitations inherent model and its characteristic equilibrium parameters (2.1). Then
in, and the possibilities for using, the Simmons approach (2.1). the full model is reduced to its commonly used linear form (2.2),
We can understand the problem of multiple fits of the Simmons and, finally, a polynomial approximation to the Simmons model
relations to a data set, within which no length-dependent data is presented (2.3).
are available, by using a “scaled” analysis (2.2). The idea is to 2.1. Basic Underlying Assumptions. The Simmons model
separate between the magnitude of the current (2.2.3) and the describes inelastic, nonresonant tunneling through an insulating
shape of the I-V curve (2.2.2). The magnitude is character- barrier and was formulated by J.G. Simmons in three successive
ized by the equilibrium conductance, G0 (in Ω-1), rather than papers in 1963.22-24 This model has found wide acceptance and
by an absolute current value (in Ampere, at a specific bias). use in the analysis of oxide tunnel barriers as well as barriers
Analyzing Molecular Current-Voltage Characteristics J. Phys. Chem. C, Vol. 111, No. 11, 2007 4433

expression of asymmetry in the barrier height. The price is a


significantly more complicated mathematical formulation.
2.1.2. Linearity of the Simmons Relations. The widely used
form of the Simmons relations (equation a3 in the Appendix)
represents only a special case of the general Simmons model.22
This form is based on two very stringent assumptions:
1. The potential profile is spatially linear, meaning that it
has either a rectangular or a trapezoidal shape, for symmetric
(Figure 1a) and asymmetric (Figure 1b) electrodes respectively.
2. The applied bias adds linearly to the potential profile.
The formal form of the linearity requirement is written as
follows:

φleft + φright (Wleft + Wright)/e + V V


h)
ψ ) ) φ0 + (1)
2 2 2
where ψ h is the averaged potential, φ is the potential difference
(barrier height) at each electrode (left and right), W is the work
function, e is the electron charge, and the subscript 0 emphasizes
that this is an equilibrium (V ≡ 0) property.
Thus, the so-called Simmons model really is the “linear
Simmons model” as it is based on the assumption that the
potential profile is linear in space and responds linearly to
applied bias. A nonlinear potential profile was originally
included by Simmons for describing the effect of the image
Figure 1. Schematic illustration for different barrier types and force.22,23 Recently Hansen and Brandbyge used a parabolic
characteristic parameters. Barrier within the linear Simmons model potential profile,34 as shown schematically in Figure 1c, to
include the symmetric rectangular barrier (a) and the asymmetric, improve modeling of experimental data. The first assumption
trapezoidal barrier (b) where the variation of potential profile is linear of averaging could be wrong for cases where traps contribute
following eq 1 (dark and light gray for equilibrium and biased potential
dominantly to the transport, as considered by Probst35 and
profiles). Equation 1 is incorrect for nonlinear cases such as image
force lowering (c) or traps within the barrier (d). The averaged potential schematically shown in Figure 1d.
is given by φ, and its equilibrium value (φ0) for the linear cases (a and Despite these limitations, the Simmons model has the great
b) equals to the average of the difference in work function at the two merit of simplicity, whereas the alternatives mentioned above
interfaces (eq 1). Panel e illustrates the characteristics parameters of a have generally much more complicated mathematical forms.
rectangular barrier: dimensionless thickness (βl) and the equilibrium Keeping this in mind, we will show how we can use the
barrier height (φ0). High and low shape factors (F ) β0l/φ0) are Simmons model to analyze I-V characteristics to yield unique
demonstrated on the two rows of panel e (constant φ0 at each column).
The width ratio (R ≡ L/l) is shown on the top-right barrier: L is the parameters, while at the same time maintaining simplicity. To
width of the potential barrier above the Fermi level, which can be achieve this, we will explore two independent approaches, that
narrower than the molecular length (l) due to packing tilt or more of scaling (2.2) and linearization (2.3).
fundamentally due to, e.g., image force lowering (c) or contact 2.2. Scaled Nature of the Simmons Relations.
resistance. 2.2.1. Modified Simmons Relations. The relation describing
the tunneling current (I) as a function of the applied bias (V)
formed by molecular films, since the very early days of this originally defined by Simmons22,23 (see equation a3, Appendix)
field32,33 and also recently.14,17,18 The underlying assumptions includes five unknown barrier characteristics, namely:
of the model are essential for defining the range of conditions • equilibrium barrier height (φ0)
over which it can be applied and its limitations, viz. • nominal barrier width (l)
• the potential is spatially averaged (2.1.1) and • contact area (A)
• the potential varies linearly with space and applied voltage • effective mass (m*)
(2.1.2). • ratio of the barrier width over the molecular length (R)
We discuss both these assumptions next. However, mathematically there are only three distinguishable
2.1.1. AVeraged Potential. The Simmons model is based on parameters. One of these is obviously the barrier height (φ0),
averaging the potential-thickness profile of the tunnel barrier, which scales the voltage. Several alternatives can be conceived
ψ(V,x) resulting in a single characteristic parameter, which is of to define the other two parameters. As will be justified below,
the averaged barrier potential, ψ h (V). The “Simmons view” of a we suggest rewriting the original Simmons relations (see
potential barrier for tunneling is shown schematically in Figure equations a3 to a5 in the Appendix) in the following manner:
1a,b where the dark and light gray depict the barrier at
equilibrium (V ≡ 0) and under arbitrary applied bias, respec-
tively. The averaged potentials for the two cases (φ0 and φ(V), I)
2G0φ0
Fφ0 - 2 {(
‚ 1-
V
2φ0 ) [ ( x )]
exp Fφ0 1 - 1-
V
2φ0
-

) [ ( x )]}
respectively) are also shown. One consequence of the averaged
potential assumption is that asymmetry between the height of
the barrier at the two ends (Figure 1b) almost does not play a
role (as long as the bias does not exceed the smaller of the two
( 1+
V
2φ0
exp Fφ0 1 - 1+
V
2φ0
(2)

barriers). To account for asymmetry, Brinkman et al. revised for |V | e φ0, where the three independent parameters are
the Simmons model by integrating over transmission prob- • φ0, the averaged, equilibrium (V ≡ 0) barrier height
abilities rather than over the potential profile,28 allowing a direct (eq 1 above)
4434 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan

• F, the equilibrium shape factor, defined as follows:

x x
4π 2m*m0e m*
F) Rl ) 1.025Rl (3)
h φ0 φ0

which (see equation a1 in the appendix) implies F ≡ β0l/φ0.


• G0 the equilibrium conductance, defined as follows:

e2 A 4π
G0 ) (
4πh R2l2 h
Rl x2m*m0eφ0 - 2 × )

exp -
h x(
Rl 2m*m0eφ0 (4a) )
A
G0 ) 310 (Fφ0 - 2) exp(-Fφ0) (4b)
R2l2

(Fφ0 - 2)
G0 ) 325Am* exp(-Fφ0) ≈
F2φ0
Am*
325 exp(-Fφ0) (4c)
F

where h is Planck’s constant and e, m0, and m* are the electron


charge, mass, and effective mass, respectively. R is the ratio
between the true junction width (L) and the experimentally
known molecular length (l), as schematically shown in the top
right of Figure 1e (see section 2.2.4). The numerical coefficients
in the short forms have the following units: 310 in Ω-1 [Å/µm2],
325 in [Ω-1 V-1 µm-2], and 1.025 (eq 3) in [Å-1 V-0.5].
Note that both F and G0 (eqs 3 and 4) are by definition
equilibrium parameters as no bias term is included. Because,
by definition, Fφ0 ≡ β0l, eq 2 is equivalent to equation a4 in
the Appendix, the two products will be occasionally exchanged,
as (β0l) is a more familiar expression than (Fφ0).
2.2.2. Shape Factor, F. The importance of the shape factor,
F,36 is demonstrated schematically in Figure 1e, as the ratio Figure 2. Illustration of the scaled nature of the Simmons I-V relation.
between the width and the height of the barrier.37 Figure 2a (a) The effect of the shape factor, F on the shape of Simmons’ tunneling
illustrates the effect of varying F on simulated I-V curves for I-V curves is demonstrated on simulated I-V curves of identical
junctions with constant β0l and G0. The dotted line (labeled G0) dimensionless thickness (β0l ) 30) and varying shape factor, F and,
reflects the simple Ohmic relation (I ) G0V), which provides accordingly, varying barrier height, φ0 (numbers in brackets), i.e., F
the current “magnitude”, and all curves converge to this curve (φ0): (I) 7.5 (4); (II) 10 (3); (III) 15 (2); and (IV) 20 (1.5). For all
curves, a constant equilibrium conductance was used, G0 ) 10-9 Ω-1
at very low bias. At higher bias, the tunneling current rises faster (meaning that the nominal contact area varied between 250 and 700
than the Ohmic one, and the higher F is, the steeper the curve µm2). The dotted line marked as G0 gives the Ohmic relation (I ) G0V).
is. The problem of using β0l and φ0 as parameters is illustrated The inset shows the strong similarity between curves for systems with
in the inset of Figure 2a, where we see a blown-up section of constant shape factor (F ) 15) and varying barrier height (β0l). Shown
three I-V curves for junctions with constant F and G0. Hence, are a blow-up of curve III from the main panel and curves for two
the inset shows that for a set of curves where β0l and φ0 differ systems with identical F and varying φ0, i.e., φ0 (β0l): (III) 2 (30); (V)
1(15); and (VI) 3 (45). In this case, the contact area varied between
but their ratio (F) is constant, the I-V curves cannot be 170 nm2 to 17 cm2 to be able to keep the same G0 ) 10-9 Ω-1 (b and
distinguished visually, even at strong magnification (a range of c) Maps showing the sensitivity of numerical fitting process to the
0.1 V is shown, at intermediate bias). choice of free variables. The gray level of a pixel represents the MSE
Varying F has a minor effect on the total G0, but this is (mean square error, eq a9) between a “target” I-V curve and the I-V
compensated for by having the contact area (A) vary between curve, based on the parameters given in the X-Y coordinates for that
250 and 700 µm2 (Figure 2a main panel) However, keeping pixel. The third parameter is either the correct G0 (b) or the correct
A/L2 (c). The varying fitting parameters were the barrier height
F constant and varying β0l (φ0) affects G0 exponentially. Thus,
(Y coordinate) and either the shape factor or dimensionless thickness
the three curves in the inset to Figure 2a represent 13 orders of (X coordinate) for panels b and c, respectively. The choice of specific
magnitude change in contact area, A (this is equivalent to coordinates is explained in the text. Maps of other combinations of
keeping A constant and normalizing the curves by a 13 orders coordinates appear in the Suppl. Mater. (S2).50 The “correct” minimal
of magnitude change in G0). point is the central pixel. The gray levels are linear with log(MSE)
Figure 2a also allows us to illustrate visually the effect of and normalized to the span of MSE, with white and black equal to
the term in braces of eq 2. That effect is expressed as the ratio maximal and minimal MSE respectively. The fitting is derived by the
strength of the gradient along the variables’ coordinates. Thus, a
between any of curves I-IV and the dotted line, the asymptotic direction of low gradient (constant gray level) means large tolerance
Ohmic relation (G0). This behavior differs from that of the for variation in this parameter. “Target” I-V data were calculated
common version of the Simmons relations (equations a3 and from equation a4 (Simmons’ model) using φ0 ) 2 V, β0l ) 15, and
a4 in the appendix) where the term in braces includes also a A/L2 ) 106.
Analyzing Molecular Current-Voltage Characteristics J. Phys. Chem. C, Vol. 111, No. 11, 2007 4435

strong contribution from the current magnitude, because the diagonal valley (cf. orthogonal one in Figure 2b,c), which is an
argument of the exponent in eq 2 will be between 0 and 1, in indication for correlation between the two fitting parameters.
contrast to that of eq a4 which is close to β0l (and similarly for To summarize, Figure 2b shows that, although F is a very
a3). Thus, the name “shape factor” reflects both the “aspect robust characteristic of the Simmons I-V relations (for known
ratio” of the barrier (cf. Figure 1e) and the shape of the I-V G0), φ0 becomes well defined only if we know the contact area
curve (cf. Figure 2.a). Furthermore, considering that to a first (A). Naturally, the question that remains is how well we know
approximation only the potential barrier is affected by the the true contact area (see section 5.3).29-31
applied bias and the kinetic energy of the electrons is rather 2.2.4. Width Ratio, R, and EffectiVe Mass, m*. Historically,
constant, it becomes clear that F, instead of the more common there was a direct translation from β0 (or F in our present
β0 parameter, is the natural parameter for describing the bias- terminology) to the barrier height, as β0 included only universal
dependent current transport. In contrast, β0 is the characteristic constants and the barrier height. However, attempts to fit
length-dependent parameter, dictating primarily the magnitude molecular I-V data with the Simmons model always encoun-
of the current. tered the problem that the experimentally measured currents
2.2.3. Equilibrium Conductance, G0. The equilibrium con- were larger than those expected theoretically.17,18,21 As can be
ductance (G0) is the first derivative of the current with respect seen from substituting eq 4 into eq 2, the current scales
to voltage near 0 V (G0 ) dI/dV|V f0, see section 2.3.2). As exponentially as well as inversely with the product Fφ0 (≡β0l),
such it includes information on the current magnitude, as so that the high observed currents correspond to a reduced value
demonstrated by the dotted asymptotic line in Figure 2a for β0l. However, φ0 cannot vary much as it scales the bias
(simulated I-V’s) and in Figure 3 (experimentally measured (see eq 2) or dictates the shape of the curve.
I-V’s, see section 4 for details). To compensate between these contradicting requirements to
The use of G0 as a characteristic parameter is fairly com- fit to the experimental magnitude and shape of the I-V curve,
mon,12,18,21,26,38 though it was not related to analyses within the further parameters were introduced in β0 (F). Holmlin et al. used
Simmons model. The uniqueness of the G0 concept within the an asymmetry parameter (R) to yield a reduced decay param-
Simmons model is that, in principle, it allows reducing the eter.17 However, the work of Simmons22,23 does not provide
number of numerically fitted parameters, by extracting G0 (from justification for such asymmetry effect (see eq 1 above). An
a linear fit of the low-bias range) and then using G0 as a fixed effective mass, smaller than the free electron mass that appears
parameter in a subsequent numerical fit over the full bias in the original Simmons model, was introduced by Cui et al.18
range. They extracted a value of m* ∼ 0.16, in reasonable agreement
This idea was tested by simulations shown in Figure 2b,c, with theoretical calculations that were based on the spectrum
which are maps illustrating numerical curve fitting to the of the electronic states of the molecular wire.40
Simmons model (eqs 2 or a4). First, a fictitious I-V data set Another possible cause for a reduced F value is the uncertainty
was constructed from eqs 2 or a4 with arbitrary junction regarding the actual width of the potential profile (L, shown
parameters. We then calculated the current (for the same voltage schematically in Figure 1c,e). Although L is the relevant
array) for a large matrix of φ0-F values (with eq 2, assuming tunneling property, it can differ from the experimentally known
that G0 is known, Figure 2b) or of φ0-βl values (with eq a4, parameter, which is the molecular length (l), possibly modified
assuming that the contact area, A, is known Figure 2c), and the by the tilt (Figure 1e), if that value is actually known. This is
mean squared error (MSE; see appendix A4) between the why we have interpreted R as a ratio of widths, R ) L/l. Such
new current array and the target current array was calculated. reduction in effective width can be due to image force narrowing
These MSE values are shown as gray levels, where the φ0-F of the potential22,23 (Figure 1c), to high contact resistance, which
(φ0-βl) matrices are used as X-Y coordinates. considerably alters the shape of the potential profile,7 or to
Figure 2b shows that for a fixed G0 the numerical fit is very significant molecule-electrode orbital hybridization at the
sensitive to the shape factor parameter (the dark valley along F interfaces.41 Through-space, rather than through-bond, transport
) 7.5 in Figure 2b) and rather insensitive to the barrier height can also reduce the actual width (L), compared to the nominal
(the weak vertical contrast along the F ) 7.5 valley in Figure molecular length (l) for tilted molecules10 (Figure 1e). The
2b; vertical contrast reflects variation in φ0. presence of oxide layer on the electrode20 or of residual water
Figure 2c illustrates the common fitting approach,14,17,18,21 (a distinct possibility if the top contact is evaporated on a cooled
where the contact area (A) is used as a known quantity. In that substrate, in the absence of a separate cold finger in the
case, the convergence of the numerical search (Figure 2c) is evaporation chamber that is at a temperature, well below that
much sharper than for fixed G0 (Figure 2b), with a shallow of the cooled sample) could lead to an actual thickness larger
valley along a constant β0l value.39 The reason for this difference than the nominal one.
between parts b and c of Figure 2 is that in Figure 2c we Because of their different physical meanings, both R and m*
optimize two properties: the current magnitude (dictated by β0l) are used here, but it is clear that they are coupled and lumped
and the shape of the curve (dictated by φ0). In contrast, in Figure together in F. The two parameters contribute differently only
2b, the magnitude of the current is almost completely accounted to the pre-exponent in G0 (eq 4b vs eq 4c), which is expected
for by the known G0, and the numerical search looks mostly to be negligible compared to the exponential term. Thus, whether
for the shape of the curve (up to the very fine differences as in to interpret the data in terms of R or m* is a matter of choice,
the inset to Figure 2a). As can be seen, F is by far the dominant as they cannot be distinguished based on analyses of experi-
factor, and for each φ0, a complementary A value can be found mental I-V data in the frame of the Simmons model.
to give the same (constant) G0 (see eqs 4b and 4c, A varies In summary, introducing the G0 and F as parameters allows
exponentially with φ0, if F is fixed). to fit experimental I-V data in such a way that one can sepa-
The choice of the independent parameter on the X axis for rate the magnitude of the current from the shape of the
the two maps (Figure 2b, c) comes from this insight, and was current-voltage curve. The importance of this parameter couple
also verified by constructing maps for other possible permuta- emerges also from the linearized analysis, as is discussed
tions (see S2 of the Supporting Information), producing a next.
4436 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan
-1 -2 -3
F4 β0l/15 - (β0l) - 3(β0l) - 3(β0l)
2.3. Linearization.
2.3.1. Assuming a Linear Potential Profile near Equilibrium P) ‚ ≈ 0.3C2
2048 β 0l - 2
is Justified. Section 2.1.2 above emphasized that the common (7b)
“Simmons model” refers only to linear potential profiles (eq 1,
Figure 1a,b). Similarly, using β2-V plots (for β extracted from However, practically, it was very difficult to extract a robust P
thickness dependent data, see, e.g., ref 18 and equation a2) relies coefficient even from simulations. First, it is almost a direct
on the same linearity assumption (eq 1). It is, however, very function of C (see approximated term in eq 7b, right-hand side),
reasonable to assume that molecules (especially incompletely and second, at small bias, the fifth order contribution is very
conjugated molecules, the ones of most interest for analysis with small, and at large bias, it is incorrect to approximate the original
tunneling models) develop nonlinear potential profiles across expression (eq 2) by a power series (eq 5).
them (such as those of, e.g., Figure 1c,d), if only because of The approximated relations (right-hand side of eqs 7) follow
their varying atomic composition or the spatial localization of Simmons24 and Brinkman et al.,28 where it was assumed that
the molecular orbitals. β0l . 2, which allows one to neglect all powers of β0l apart
Nevertheless, even nonlinear profiles can be approximated from the highest one. Similarly, with this approximation also
with the linear model over small intervals, such as near the (Fφ0 - 2) term in the G0 expression (eq 4) reduces to Fφ0.
equilibrium. Namely, we assume the potential varies linearly The β0l . 2 approximation is questionable for molecules, where
with the applied bias (eq 1) and with negligible bias effect on ∼0.5 < β0 < ∼1 Å-1 and ∼15 < l < ∼30 Å are typical values,
the net shape of the profile. Thus, linear (common) Simmons which yield a β0l product that is not much larger than 2.
model should properly be used only near equilibrium where, Nevertheless, the β0l . 2 approximation has a major merit that
for the small biases, quasiequilibrium conditions can be as- it allows to extract F without knowing β0l (i.e., directly from
sumed, instead of forcing it over a large bias range, where its C), something that requires knowing the contact area, A (see
applicability will, to say the least, be questionable. Accepting eqs a6 and a7). We will show later (section 4.2) that practically
this logic, we can, with the small signal assumption, replace the β0l . 2 approximation does not introduce large errors.
the highly nonlinear eq 2 with its power expansion.24,28 Equation 5 can be solved analytically, in contrast to eqs 2 or
2.3.2. Polynomial Approximation to the Linear Simmons a3-a5 that can be solved only numerically. However, the
Model. In a paper24 lesser known than his later ones, Simmons accuracy of the approximation (eq 5) deteriorates as V f φ0,
showed that eq a3 can be rewritten using a polynomial expansion and the fifth order parameter (P) is rather useless. Therefore,
(with V/2φ0 as the expansion variable), to give from here onward, we shall consider only the graphical
extraction of G0 and C from plots following eq 6. The
I = G0(V + CV3 + PV5 + ‚‚‚) (5) polynomial coefficients can then be translated into junction
parameters (e.g., barrier height and effective mass or width ratio)
by combining eqs 4, 7a, and 7b (for a full glossary see S1 of
Equation 5 can be further rearranged (ignoring the fifth order
the Supporting Information).
term) to get a linearized presentation of experimental I-V data
2.3.4. Linearized Approach Combined with Thickness Varia-
tion. Even though the main purpose of our analysis is to find
I
= G0 (1 + CV2) (6) ways to analyze I-V data from experiments without thickness-
V dependent information, we will now briefly consider how the
linearized analyses discussed here can also be useful for data
Thus, plotting I/V (the integral conductance) vs the square of sets that include length information.
the bias will yield two functional parameters, G0 and C (see The terms for G0 can be related to plots of G0 vs length
eq 8.b below). Equation 6 is very similar to the parabolic (l).18,21,38 Substituting for Fφ0 ≡ β0l in eq 4, neglecting the “-2”
expression developed by Brinkman et al.28 but with two term and rearranging we get
significant differences, namely that eq 6 gives the integral
conductance (in contrast to the differential one in ref 28) and
the absence of a linear term in eq 6.42
The I/V vs V2 relations (eq 6) are expected to be linear, as
(
ln(G0l) ≈ ln 310

R2 )
- βl (8a)

long as (1) the error in the power expansion is small; (2) the ln(G0F) ≈ ln(325Am*) - φ0F (8b)
averaged potential decreases as V/2 (linearity in bias, eq 1); (3)
the (low) bias has a negligible effect on the width of the barrier where, as noted above, the units of the numerical factor 310 is
(linear profile, see section 4.4); and (4) the asymmetry between [Ω-1(Å/µm)2] and that of 325 is [Ω-1 V-1 µm-2].
the barriers at the two sides of the junction is negligible.28 Thus, for a system of varying barrier thickness, the decay
2.3.3. Relating Polynomial Coefficients to Barrier Param- parameter (β0) can be extracted from the slope of a plot of ln-
eters. The coefficients of eq 5 are directly related to those of (G0l) vs the molecular length, l, as is well-known. Yet eq 8
eq 2. The equilibrium conductance, G0, is identical (eqs 4). The provides also an interpretation for the pre-exponential term (or
cubic coefficient (C) is closely related to the shape factor, F the intercept) of the semilogarithmic fit, which can be used to
extract the effective electrical contact area for transport, A scaled

( )
β0l + 3 by either β0/R2 or m*.
F2 F2
C) ‚ 1- ≈ (7a) In eq 8b, the independent parameter is now the extracted one,
96 β0l(β0l - 2) 96 F, rather than the assumed length, l. A linear ln(G0F) vs F plot
indicates a constant barrier height that can be extracted from
where, for convenience, the dimensionless thickness, β0l, is used, the slope. The effective electrical contact area can be extracted
instead of Fφ0. from the intercept as above. If the plot is not linear, it could
The expansion was continued until the fifth order (P) indicate that the barrier height varies between the different
coefficient, attempting to extract more independent information junctions. Clearly eq 8b could be useful for cases where there
(see section 4.3) is uncertainty about the exact barrier width (Rl). It also allows
Analyzing Molecular Current-Voltage Characteristics J. Phys. Chem. C, Vol. 111, No. 11, 2007 4437

test case since these molecular junctions are well established


and the results of our test analyses can then be compared with
those obtained in a more conventional manner, from the length
dependence.
The first type of molecular junction was formed by a single
dithioloctane (HS-(CH2)8-SH), bound chemically on one side
to a Au substrate and on the other side to a Au nanoparticle.
The molecule is embedded in a monolayer of monothioloctane,
i.e., all of the other molecules are bound only to the Au substrate.
Transport was measured with a conductive probe atomic force
microscope (CP-AFM; Figure 2a of ref 18).
The second type of junction contained a monolayer of
a several thousands of monothiol dodecane molecules,
(HS(CH2)12H), chemically bound to a Au electrode, in a
nanopore configuration, with no chemical bond to the other Au
electrode, that was evaporated onto the nanopore structure
(Figure 5 of ref 14).
The last junction was formed by a bilayer of ∼108 monothiol
docosane molecules (HS(CH2)22H), with the bottom layer bound
chemically to lithographically patterned gold electrodes and the
top layer bound chemically to a micrometer-sized floated gold
flake.27 Further details on that system are given in the Supporting
Information (S4) and will also be part of a separate publication.
Graphical fitting, numerical searches, and simulations were
done using MATLAB software. The specific functions employed
are given in Appendix A4. The voltage array for simulated I-V
data ranged from -φ0 to +φ0, with steps of 1 and 10 mV below
and above φ0/10, respectively.

Figure 3. Curve fitting to typical molecular I-V data (a) single 4. Extracting Characteristic Tunneling Parameters with
alkyldithiol;18 (b) alkylthiol monolayer;14 and alkylthiol bilayer.27 Dots Revised Simmons Model
represent measured data (diluted for clarity) and curves are different
fits to the Simmons model, including (I) full-range numerical fitting We will now review how we can extract characteristic
with two free variables (gray line) (eq 2); (II) quasi-equilibrium tunneling parameters from experimental current-voltage curves
approximation (dotted line, eq 6; see Figure 4); (III) bias dependent of molecular junctions. We start by describing the problems,
barrier width (green line, eq 9; see Figure 5); and (0) linear (Ohmic)
fit over 10 mV (dashed line). Because the measured interval for (b) inherent in how the Simmons model is commonly used (4.1).
was 20 mV, the linear fit was done over the (50 mV range. The inset We then introduce an alternative way to extract parameters from
to each panel shows a blow-up of the (30 mV range, except for (b) near-equilibrium data (4.2). This approach can be expanded so
where the inset covers the (200 mV range. The inset tables show the as to be applicable also to data, farther away from equilibrium,
fitting parameters for each curve: equilibrium conductance (G0 in nS); either by finding bias-independent parameters that are valid over
barrier height (BH in V); shape factor (SF, in V-1); and decay parameter the full measurement range (4.3) or by calculating bias-
(beta in Å-1). The width ratio (R) was unity for curves I and II, and
equals the following cubic polynomial for curve III: (a) R ) 1.00 -
dependent parameters (4.4).
0.03|V| + 0.23V2; (b) R ) 1.00 - 0.05|V| + 0.23V2; (c) RV>0 ) 1.00 4.1. Fitting Experimental I-V’s Depends on Voltage
- 0.03V + 0.31V2, and RV<0 ) 1.00 - 0.03|V| + 0.27V2. The nominal Range. Curve fits over part or all of the data ranges are shown
thickness and contact area for the different junctions are in Figure 3. In the three panels, we reproduce the experimental
(a) l ) 17.6 Å, A ) 22 Å2; (b) l ) 18.2 Å, A ) 1600 nm2; (c) l ) 53.2 I-V curves, which were measured for the three types of
Å, A ) 50 µm2. molecular junctions, viz. single embedded molecule with two
chemicontacts18 (Figure 3a; “single molecule”), a small area
a direct extraction of the equilibrium barrier height (φ0), without monolayer with one chemicontact (∼104 molecules per junction;
plotting β2 vs V (eq a2). A complementary test is a plot of F vs “monolayer”)14 (Figure 3.b), and a large area (∼108 molecules)
l, which should be linear through the origin (eq 3). In the results bilayer junction, with chemicontacts to the electrodes but not
section (4), the expressions that have been derived here are tested between the monolayers27 (Figure 3c; “bilayer”).
on three experimental data sets. The dots in Figure 3 are the experimental data, and the lines
represent fits to these data. The common numerical fit over the
3. Experimental Section
full bias range is labeled “I” (gray solid line, see section 4.3
The use of the proposed analysis approach is tested on the for details on the fitting). The insets to each panel are
well-studied Au/alkylthiol/Au system, prepared with different magnifications of the low signal range, showing a sever
contacting methods. Two of these junctions refer to published disagreement between this numerical fit (I) and the data at this
data.14,18 The third set was measured using a “ready-made” low bias range (as well as at very high bias, except for 3b).
contacting method of gold flakes floated over pre-patterned gold Limiting the fitting range to a (0.1 V (II, dotted line, parameters
electrodes.27 It was added here to verify the generality of the were extracted from the fitting of Figure 4 discussed next, yet
analysis. similar fits were received also by common numerical search)
Generally, alkyl molecular junctions are not expected to produces a good fit over that range, which strongly deviates at
benefit much from the proposed analysis approach, as their higher bias. This problem was noticed already by Wang et al.14
length can be easily varied. However, they were chosen as a and by Cui et al.18
4438 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan

Clearly the graphical differences between fits to different bias


ranges translate into considerably different tunneling parameters
corresponding to these fits, which are given in the Tables that
appear as insets to Figure 3. Such parameters are thus not unique
and change with the choice of bias range for the fits. In the
following sections, we present alternative ways to extract
characteristic parameters.
4.2. Extracting Equilibrium Parameters using Power
Expansion. This is a practical implementation of sections
2.3.2 and 2.3.3. Generally one can use eq 5 with increasing
degree of complexity to extract more information from
the data. Fitting to only one coefficient, i.e., imposing linearity
on the I-V relation (I ≈ G0V) is the well-known equilibrium
conductance approach.12 This approach is presented by the
dashed lines (“0” numbered) in the main part of Figure 3.
Clearly, G0 is the most robust parameter (cf., G0 values in the
tables that appear as insets to Figure 3 and Tables s2-s4 in the
Supporting Information). Unfortunately, it contains only infor-
mation on β0l (≡Fφ0; eq a7 of appendix). If more information
is needed, we need to find ways to extract more parameters
from the data.
Imposing cubic relations (eqs 5 and 6) on the I-V’s pro-
vides G0 and C, and with the knowledge of contact area
(A, for which an approximate value will suffice) allows
extracting two parameters: β0l and F (eqs a6 and a8, respec-
tively). The barrier height (φ0) is the ratio between these
parameters (β0l/F) and can be extracted without knowing the
exact barrier width (l). Note that the translation of G0 into β0l
relies on knowing the contact area (A). We checked if it is
possible to use the fifth order parameter (P of eq 5) to provide
one more experimental parameter, which would allow extracting
β0l independently of A. Yet, as explained in section 2.3.3 above,
and confirmed by simulations, mathematically this is not
possible.
The cubic relations can be presented in a linearized manner
by plotting I/V (the integral conductance) vs V2 (eq 6). Figure
4 shows a linearized presentation of the data (symbols) from
Figure 3, where the lines are linear fits to the data. This
linearization shows clearly the relevant fitting ranges, or
“quasiequilibrium” regions,43,44 which can vary from junction
to junction. For example, the monolayer (Figure 4b) shows some
asymmetry in the I-V since the forward bias (black dots) and
reverse bias (gray dots) data are slightly shifted. For a first
approximation, these differences were ignored and the linear Figure 4. Linearized presentation of molecular I-V data (a) single
fit considers all data. For the bilayer (Figure 4c), the asymmetry alkyldithiol;18 (b) alkylthiol monolayer;14 and (c) alkylthiol bilayer.27
was fairly large, and in this case, the reverse bias data were As emerges from the power expansion (eq 6), a plot of the (integral)
omitted for clarity. Furthermore, this junction shows a clear conductance (I/V) against the squared bias (V2) should be linear for
low enough bias. The black and gray dots represent data measured in
change of slope at ∼175 mV (0.03 V2). The definition of the forward and reverse bias, respectively, and lines are linear fits. In panels
proper fitting range was somewhat questionable also for the a and c, two slopes are evident and expressed by two fitting lines.
single molecule data (Figure 4a), where there is a fairly steep The linear coefficients are: (a/I) G0 ) 9.33 × 10-10; C ) 35.4; (a/II)
slope at very low bias (<80 mV), which becomes more G0 ) 9.74 × 10-10; C ) 11.0; (b) G0 ) 4.35 × 10-9; C ) 10.7; (c/I)
moderate at higher biases (80 < V < 250 mV). G0 ) 1.26 × 10-7; C ) 15.3; (c/II) G0 ) 1.75 × 10-7; C ) 1.35.
The coefficients C and G0 of the fits to Figure 4 were
translated into barrier parameters (using eqs a6 and a8) by junction the extracted β0 ()0.84) reproduces fairly well the
assuming a potential width that equals the molecular length published results,14 whereas for the other two junctions
(i.e., R ≡ 1, and literature values for molecular length) and (Figure 3a,c), β0 was ∼0.4, much lower than the published
the nominal (geometric) contact area (A, for alternatives see value18 and also well below the values for alkyl chains,
Table s1 of the Supporting Information). The corresponding commonly extracted from length-dependent data. Interestingly,
barrier parameters were inserted into eq 2 to produce the dotted notwithstanding the very different contacting environments of
curves (labeled II) in Figure 3. the single molecule and bilayer junctions, the linearized
The parameters extracted from the linearized presentation approach yields similar β0 values for these two types of
(Figure 4, eq 6) can be compared to those extracted using a junctions.
numerical fit, to check their reproducibility and robustness. As The quasi-equilibrium shape factors (F), extracted from the
can be seen from the inset to Figure 3b, for the monolayer data, were 38, 34, and 40 V-1 for the single molecule (line II
Analyzing Molecular Current-Voltage Characteristics J. Phys. Chem. C, Vol. 111, No. 11, 2007 4439

of Figure 4a), monolayer and bilayer (line I of Figure 4c) which we consider as a reasonable accuracy. However, for the
junctions, respectively. These values could be up to 10 times monolayer junction, the extracted contact area was ∼3 nm2
larger than the F, calculated from numerically fitting of data to (cf. nominal value of 1600 nm2), which could be taken as an
eq 2 over the full measurement range (curves I, see below). indication for defect-mediated transport. For the bilayer junction,
The shape factor, F, can be extracted independently of an the numerical search completely diverged. We take this result
assumed contact area, by using the β0l . 2 approximation. This to reflect the inability of this fitting procedure, in this case, to
procedure generally yields only a few percent decrease in match the shape of the curve, if the current magnitude is dictated
F compared to the values given above, thus confirming the by the value of G0, derived from the low bias fit. Interestingly,
validity of the extracted numbers and showing the scaled nature despite this huge divergence, it was still possible to extract a
of the analysis. reasonable value for the shape factor, which again emerges as
The extracted barrier heights (φ0) were 0.18, 0.46, and 0.52 a robust parameter. However, we find that, by limiting the bias
V, and the extracted effective mass values (m*) were 0.83, 1.49, range over which we fit the data, we can extract a set of
and 0.27 for single molecule, monolayer, and bilayer junctions, parameters with a contact area of the order of a few defects in
respectively. The resulting φ0 values are rather low compared the layer.
to what is commonly accepted (1-2 eV). We note, though, that Low A values are accompanied by low β0 and φ0 values,
a 50% reduction in barrier height was also observed by Dorneles because G0 cannot change much since it reflects the current
et al., after accounting for the effective electrical contact area, magnitude (see eq 4). Considering which of the various possible
compared to the barrier height that was extracted, using the extracted parameters is the most characteristic one, it clearly
nominal contact area.31 The extracted effective mass values are emerges that G0 is the most robust one. The variation in F
reasonable, except for the monolayer, as m* > 1 for molecular increases if high bias ranges are included, but considerably
junctions is unusual, though recently m* ) 1.6 was reported improves if only low bias ranges are taken. The reproducibility
for Hg/alkyl thiols/GaAs junctions.45 Note that m* > 1 can also in β0 was similar to that in G0, but the conventional characteristic
be interpreted as R > 1, meaning a thicker barrier than what junction parameters (i.e., the nonlumped parameters φ0, m*, A)
was assumed in the analysis (although a narrower barrier than showed much poorer reproducibility.
what is assumed from physical measurements and modeling is As noted before, extending the voltage range toward
the more common scenario). high bias could produce erroneous parameters simply be-
To overcome the need to know the contact area, we can try cause the linearity assumption does not hold there. The next
to extract more information from the data (i.e., extract three section considers the possibility of recovering the bias
parameters), using a numerical fit to the original eq 2, as dependence of the potential profile from experimental I-V
described next. characteristics.
4.4. Potential Profile across a Molecular Junction is
4.3. Equilibrium Approach, Compared to Numerical Fits
Nonlinear. Information on the bias variation of the barrier
over a Large Bias Range. The gray curves (I) of Figure 3
characteristics is of major interest18 and commonly expressed
represent the common fitting approach,14,17 based on a (presum-
as bias variation of the decay parameter, β (see eq a2, in con-
ably) known contact area (A), with φ0 and β0l (or m*)39 as two
trast to the equilibrium β0 of eq a1). Engelkes et al. found a
free fitting parameters. This procedure also uses the full
bias-independent β parameter but a strongly bias-dependent
measurement range for fitting.44 It gives a β0 value that is rather
residual resistance (the reciprocal of the pre-exponential term
similar to the one that is extracted at quasiequilibrium, reflecting
in eq 4).26 As described in section 2.1, the commonly used
mainly the magnitude of the current. However, the barrier height,
version of the Simmons model assumes that the shape of the
extracted from fitting the data over the large bias range, is larger
barrier is linear, instead of a more realistic parabolic shape22,23,34
than that extracted from quasi equilibrium data, expressing the
(see Figure 1, panels a and b vs c and d). In such a case, both
much slower increase in current at high bias (gray curves, I)
the averaged potential (ψ h ) and the barrier thickness (R(V)l) are
than that expected, based on the quasiequilibrium fit (dotted
expected to vary with the applied bias (see eq a5). This variation
curves, II).
will be different from the linear dependence of ψ h predicted by
The small change in β0 between the low and full measured eq 1.
bias ranges suggests that β0 can be considered as a constant, We will show now that accounting for the bias variation of
dictated by the current magnitude (G0, provided that A is the barrier width (R(V)l) can produce a good fit to the data over
known). Thus, the fitting is mainly one of the barrier height, the full measured bias range, as shown by the green curves in
and, instead of letting β0 vary freely it can be fixed to its value Figure 3 (an equivalent approach is to use ψ h (V) as a bias-
extracted from G0 (extracted graphically from the data in a dependent parameter). The potential width-bias (R-V) relations
preceding step). We tested such fitting procedure (see Tables are extracted from the I-V data by assuming that all of the
s2-s4 of the Supporting Information) and found only minor other parameters of eq 2 are independent of bias and equal their
differences between the parameters, extracted by optimizing equilibrium values as extracted from Figure 4. This is done by
solely on φ0 or on both φ0 and β0l. We conclude that there is defining a bias-dependent shape factor F(V) ) R(V)F(0),
no need for a numerical fitting approach to find β0 as it is a where,for simplicity, we assume R(0) ≡ 1. Substituting F(V)
direct function of G0 and A (cf. eqs a6 and a7).39 for F into eq 2 yields
Alternatively, a known G0 can be introduced into the fitting
search, combined with φ0 and A as free fitting parameters (where
β0l ) f(G0, φ0, A)), to extract the contact area from the I-V I)
2G0φ0
R(V) (Fφ0 - 2)
2 {(
‚ 1-
V
2φ0
×)
[ ( )] (
data. However, as shown in Figure 2b, in such case, the fitting
becomes very insensitive to the barrier height, which translates
into changes of orders of magnitude in the (now) extracted exp Fφ0 1 - R(V)
x 1-
V
2φ0
- 1+
V
2φ0
×)
[ ( )]}
contact area. For the single molecule junction such procedure

x
yielded a contact area of only 4 Å2 (cf. nominal alkyl thiol V
exp Fφ0 1 - R(V) 1+ (9)
footprint of 22 Å2, see Table s3 of the Supporting Information), 2φ0
4440 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan

Similar parabolic R-V curves were found for the other two
data sets, confirming the general behavior of a bias-dependent
potential profile. The green curves (III) in Figure 3 were
calculated from eq 9 using G0 and F values of curve II. However,
the low barrier height extracted for curves II could not be used
for fitting the whole bias range (note that both eqs 2 and 9 are
limited to V e φ0). In view of the insensitivity to the exact
barrier height, shown in Figure 5a, we arbitrarily set φ0 ) 1V
to allow fitting the whole bias range.
Figure 5 is thus a phenomenological proof that fitting
molecular I-V data over large bias ranges44 must consider
nonlinear potential profiles. Naturally, a meaningful parameter
extraction should rely on a physical model rather than use the
empirical approach employed in eq 9. A physical model could
be the detailed image force model by Simmons22,23 or the
parabolic model by Hansen and Brandbyge.34 Adding G0 as a
fitting constraint (see section 4.3, above) can be useful in
decreasing the large number of fitting variables in these
advanced models.
We will now discuss how we can use the quasiequilibrium
Simmons model, rather than the commonly used version, for
analyzing molecular I-V characteristics.

5. Discussion
5.1. Why Use the Simmons Model? We conclude from the
results presented in section 4 that formally the (linear) Simmons
model cannot describe real molecular current-voltage charac-
teristics. As reasons we note the lack of robustness of the
extracted barrier parameters and the inability to find bias-
independent parameters to cover all bias ranges, the low, near-
equilibrium, and the high bias one, far from equilibrium44
(Figure 3). This bias dependence (Figure 5) explains the
Figure 5. Bias variation of the width ratio (R) plotted for the monolayer inadequacy of the linear Simmons model in describing charge
junction14 for different equilibrium barrier height (φ0 ) 1, 1.6, and 2
eV, see curve labels) and constant shape factor (F ) 10.1) (a); or transfer via molecular junctions. This conclusion is a fairly
constant dimensionless thickness (β0l ) 16.16) (b). At each I-V point, general one, as it was found for three very different molecular
the width ratio was extracted by numerically solving eq 9 using junctions.
G0 ) 4.67 nS, and other parameters specified in the figure. Because, at present, there is no convenient adequate alterna-
tive, we need to find ways to use the model, while being aware
of its limitation, rather than forego using it altogether and we
Thus, at each I-V data point, R(V) can be extracted by solving will do so in the following sections:
(numerically) eq 9 for R(V). The results for such calculation • The deviations from linearity at high bias can be taken care
for the monolayer junction (Figure 3b) are shown in Figure 5. of largely by limiting the range of fitting as will be discussed
The extraction procedure was repeated for several sets of in the next section (5.2).
equilibrium parameters (reflecting a possible uncertainty in • The critical issue of contact area will be addressed in the
β0l due to the fact that the actual contact area is not known, see following section (5.3).
eq a7). • In the last two sections of this chapter, we suggest ways to
Figure 5 compares the shape of the R-V relations obtained analyze the experimental transport data within the linear
for three different barrier height values and either the same shape Simmons model by using G0 and F as phenomenological
factor (Figure 5a) or the same dimensionless thickness (Figure parameters (5.4 and 5.5).
5b). The figure shows that the R parameter varies by ∼6% with 5.2. Simplifying Parameter Extraction by Linear Presen-
the applied bias, which can be fit very well to a quadratic tation of the Experimental Data. Presenting molecular I-V
parabola (on each side of |V| ) 0). Interestingly, this 6% data in the form of Figure 4, rather than as numerical fits to
variation is equivalent to the error that Simmons predicted in eq 2 (cf. curves I in Figure 3), enables a simple visible evaluation
his model (eq a5) if the integration correction factor (termed of the relevance of the model to a given experimental data set.
there β, yet equivalent to R in the symbols used here and This is in clear contrast to the multiple possible solutions to
nowadays) is set equal to 1.22 Varying φ0 with a constant F, eq 2 shown in Figure 3. Hence, presenting molecular I-V data
changes only the scaling of the curve but not its shape (Figure as I/V vs V2 allows determining the bias range where the
5a). In contrast, varying φ0 for fixed β0l drastically changes Simmons model is relevant or identifying a change in conduction
the shape of the R-V curve (Figure 5b). The strong increase of behavior, as was evident for the bilayer junction (Figure 4.c).
R away from equilibrium (for φ0 ) 1; Figure 5b), which reflects Recently, Beebe et al. used a linearized presentation for Fowler-
a significant widening of the barrier, appears to be an artifact Nordheim field emission transport to evaluate the onset of this
due to an incorrect choice of β0l and φ0. This type of sensitivity transport mechanism.43 The linear I-V presentation, suggested
to F (≡β0l/φ0) further demonstrates the importance of F as the here, completes this approach by covering the intermediate
characteristic parameter in bias-dependent analyses. range, between the linear behavior at very low bias and the
Analyzing Molecular Current-Voltage Characteristics J. Phys. Chem. C, Vol. 111, No. 11, 2007 4441

Fowler-Nordheim relations at very high bias, as used earlier (increase) together if G0 is kept constant, as explained in section
by Aswal et al.20 A transition to Fowler-Nordheim tunneling 2.2.2 and inset to Figure 2a).
will appear as curving down of the linear trend in a presentation This suggests that the result obtained, using the contact area
such as that shown in Figure 4. Applying this presentation to as a free parameter, has physical value. Nevertheless, our
the experimental data-sets tested here, reveals no transition to confidence in the “pinholes” results is not better than in the
field emission for the monolayer14 and bilayer27 junctions. The “nominal contact area” results, as there is not much difference
single molecule junction18 does show a possible turning point in the accuracy of the two numerical fitting approaches. This is
at V ∼ 0.4 V, which is just outside the scale of Figure 4a in contrast to the work of Dorneles et al., on Al/Al2O3/Al
(V2 ∼ 0.16). Indeed the data of Figure 4a start deviating below junctions where the accuracy of the fit improved drastically if
the linear trend line at ∼0.06 V2. the contact area was allowed to vary (i.e., three free fitting
The mathematical approximation (eq 5) is scientifically variables).31 The difference in achieving a good fit could be
justified as it follows directly from the assumption of a linear explained by that the behavior of the inorganic insulator fits
potential profile (eq 1). If we limit the voltage range that we the Simmons model closer than does that of molecular barriers.
use to extract parameters to low bias quasiequilibrium condi- Possible deviations from the Simmons model are the major
tions,44 we can compare consistently and in a standardized problem, if one wants to distinguish between transport via
fashion between different experiments. However, such limitation pinholes and tunneling across all of the area occupied by the
leads to loss of important information regarding the bias molecules, solely based on a single I-V data set. Thus, it is
variation of the potential barrier. At the same time, extending critical to have an independent determination of the actual
the linear assumption to bias ranges where the barrier depends contact area, which can be done in cases were also length-
nonlinearly on the bias merely increases the inaccuracy of dependent data are available, as described in section 5.5 below.
extracted parameters (even if it appears to provide a visually Even in cases where we have direct evidence for tunneling
good fit; cf. curves I in Figure 3, see main figures vs insets). from temperature-independent currents or inelastic electron
Although the analysis of section 4.4 and Figure 5 is a tunneling spectroscopy (IETS) that shows characteristic mo-
qualitative one, it shows clearly that the potential profile over lecular vibronic modes involved in the charge transport,15,25,47
a molecule or over molecular layers is nonlinear, and not only we cannot assume that the geometric (nominal) contact area is
the barrier height but also the barrier shape varies with applied the effective one for electronic transport. This was demonstrated
bias. Therefore, analysis over a bias range beyond the near- clearly for Al/Al2O3/Al junctions where the effective area for
equilibrium range44 must account for nonlinearity, by either electronic transport was found to be 5 orders of magnitude
smaller than the nominal contact area. This was explained by
using existing models or developing new ones. Such models
arguing that the current flux was limited to “hot-spots” in the
are naturally more complicated than the one expressed by eq 2
insulator film. Transport across a “hot-spots” is still by tunneling
and require extraction of the parameters numerically. The
but the tunneling length is now much shorter, resulting in
concept of separating the magnitude issue (using G0) from the
exponentially higher conductance, and thus, most of the current
shape of the curve (using numerical search), as suggested here
will flow there.31
can simplify the application of these models.
5.4. Use of G0 and F as Phenomenological Parameters.
5.3. Do We Really Know the Contact Area? The ambiguity Figure 2, panels a and b, show that G0 and F, rather than β and
regarding whether the charge transport is attenuated by the φ0, are the fundamental, independent properties characterizing
length of the molecules or by the reduced contact area for the Simmons tunneling model in terms of scaled relations
transport (e.g., via defects in the monolayer) has been a worry (magnitude and shape). Moreover, these parameters have a
for molecular (or monolayer) electronics since its early days29 natural presentation of data (Figure 4) allowing their linear
and is relevant also for other fields that involve charge tunneling, (analytical) extraction, without knowing any other parameters.
such as spintronics.30,31 Remarkably, though, the issue is not Extraction of the approximated F (i.e., F ) x96C, from eq a4,
considered in most molecular electronics publications, which with C defined in eqs 5-7) bypasses the contact area ambiguity,
use the nominal (geometrical) area of the electrical contact. discussed in the previous section (as it is independent of
Recently, Engelkes et al. tried to identify the actual contact area β0l, which necessarily depends on A).
in conducting AFM measurements of Au/decanethiol/Au junc- Wold and Frisbie12 already suggested G0 as a phenomeno-
tions. They found that the measured zero resistance (equivalent logical characteristic parameter for comparing net current
to reciprocal G0, see eq 4) was insensitive to the diameter of magnitude. We propose that the shape factor, F can be used
the AFM tip, suggesting that the current is confined to the similarly for comparative purposes, to indicate how fast the
narrowest path between a rough protrusion of the substrate current increases away from equilibrium. The physical meaning
electrode and the very end of the tip.46 of the equilibrium conductance (G0) is straightforward as it
The critical effect of the contact area on the extracted barrier scales exponentially with the effective tunneling length or
parameters is clear from Figure 2c, which shows the sharp dimensionless thickness (β0l). The historical origin for allowing
minimum if the data are fitted, using a known contact area, in F36 as a free parameter (equivalent to R or m* as a free
contrast to the shallow valley that we get if we fit, using a known parameter) instead of being a fixed function of φ0 and l is that
equilibrium conductance (Figure 2b). Allowing the contact area the measured currents were higher than expected or the barrier
to be a free fitting parameter for the monolayer and bilayer was narrower than expected.14,17,18
(Tables s3 and s4 of the Supporting Information) results mostly Clearly, transport across a barrier is affected by both the
in values that are much smaller than the nominal one. The view barrier’s height and width (see Figure 1e). The “width” of the
of a junction that is controlled by pinholes/defects is consistent barrier can be related to its geometrical width (l) possibly
with the low barrier height (approximately tens of mV) that we narrowed by image forces22,23 or wave function hybridization41
extract for fits that yield small contact areas, as such small (R) and to coupling between the molecular orbitals and the
barriers are reasonable for two metals at very near proximity electrode (m*). The barrier width scales the current exponen-
(note, however, that the magnitude of φ0 and A decrease tially, with only minor bias dependence (slightly stronger
4442 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan

dependence if a nonlinear potential barrier is assumed, see if linearity is not achieved, such a presentation can be very
Figure 5 and eq 9). This is in contrast to the effect of the barrier informative about qualitative differences between different
height which directly varies with bias (eq 1). F reflects the ratio junctions.
of these two contributions, in contrast to the decay parameter,
β0 which is related to their product. Thus, the “heavier” the 6. Summary
charge carrier (m*) is or the wider the barrier (Rl) is, the larger The possibility for extracting information on transport
F is and the steeper the I-V curve is away from equilibrium characteristics from analyses of molecular current-voltage (cf.
(Figure 2a). Increasing the height of the tunneling barrier will current-length) data was considered. For this purpose, the
have the opposite effect (smaller F and shallower, more linear accepted Simmons model was reconsidered. Major failures are
I-V). (1) the strong dependence of extracted parameters on the
This can serve as a partial answer to the introductory example assumed contact area for transport, and (2) the strong deviation
of how to interpret a set of systematic changes in a molecular of experimental current-voltage data from the assumptions that
junction. If a systematic structural change leads only to an underlie the Simmons model, viz., the barrier profile is linear
increase in barrier height, it will be expressed as a decrease in with distance, and the barrier profile varies linearly with bias.
both F and G0 (increase in β0). However, if either the net The seemingly good graphical correlation between raw data
junction width or the effective mass increases, then F is expected and numerical fits to the Simmons model probably results from
to increase while G0 should decrease (β0l increases). Naturally, the scaled nature of the model, without proving its validity or
a combination of these effects will be more difficult to interpret. accuracy. The applicability of the Simmons tunneling model to
If, by comparing a series of junctions, we find a constant a specific I-V should be tested using the linearized version (I/V
F (i.e., β0l is also constant) and a changing G0, this can indicate vs V2) rather than using the direct form (I vs V).
that pinholes dominate conduction and that their density changes. The relevant range for the Simmons model is near equilib-
In addition, the linearized presentation of Figure 4 is a straight rium, and based on this, an alternative parametrization procedure
forward tool for evaluating the bias range where the (linear) was suggested. This procedure emphasizes the equilibrium
Simmons model is applicable and for a possible change in conductance and shape factor as robust characteristics of the
mechanism as indicated by the kink in Figure 4c. junctions, independent of any further information about the
The numerical values extracted here and presented in the junction. These two parameters can be related to more common
insert Tables to Figure 3 (and in Tables s2-s4 in the Supporting barrier characteristics, such as the decay parameter and barrier
Information) cannot provide an answer to the question what height but only if the contact area is known or from data where
are the junction parameters. Rather, they stress the ambiguity the molecular length (barrier width) is varied systematically.
in the extraction of these parameters. Nevertheless, we can Thus, the proposed approach is complementary to the common
characterize junctions in an acceptable manner as described next. thickness-dependent analysis, to which it can be added to allow
5.5. Suggested Routine for Parameter Extraction. In view separating the contribution of barrier height from that of an
of the discussion in the previous sections, I suggest the following effective mass parameter.
guidelines for analyzing molecular I-V characteristics: The quasiequilibrium I-V analysis will be especially useful
(a) Predetermine the contact area by testing the specific for analyses of chemically unique molecular junctions, the
contacting method on a known molecular system of varying thickness of which cannot be varied. Despite the gross over-
length (e.g., alkyls), using eq 8a. simplification of the Simmons model, the justification for using
(b) Measure as many data at low bias (eφ0/3) as possible. it originates from the need for quantifying imperfect molecular
Logarithmic bias intervals are convenient for this.18 Naturally, junctions. Such quantification is crucial for achieving more
this requires a low-level meter capable of measuring the controlled, reliable, and reproducible experiments, as well as
expected low currents. acquiring a predictive power toward a more realistic description
(c) Use the presentation of Figure 4 and eq 6 to extract the of transport across molecular junctions and more adequate
characteristic G0 and C parameters and to validate if the interpretation for extracted averaged characteristic parameters.
Simmons model is at all relevant for analyzing the data. A
complementary linearized presentation (e.g., Fowler-Nordheim, Acknowledgment. I thank the Kreitman foundation for a
see ref 43) can be added to map the high bias range of the fellowship at Ben Gurion University of the Negev; M. Reed
measured data. and S. Lindsay for kindly providing their I-V data; Y. Selzer
(d) In the case of deviation from the model (nonlinear I/V vs and A. Solomon for discussions; and N. Kalisman for helpful
V2 presentation), only G0 can be regarded as a characteristic remarks on numerical analyses. I am grateful to D. Cahen for
parameter and can be translated to β0 ) f(G0, l, A) (eq a7). numerous discussions, especially on contact area issues, and
Only if the I/V vs V2 plots are linear, can one use the following critical readings of, and constructive remarks on, the manuscript.
recommendations:
(e) Extract approximate F value from C (eq a8). Appendix
(f) In the case of high confidence in the effective electrical A1. Glossary of Terms
contact area (see point a above), translate G0 into β0l (using C A ) contact area [µm2]
and A, eq a6) and calculate F more accurately (F ) f(C, β0l), C ) cubic polynomial coefficient [V-2]
eq a8). I ) current [A]
(g) Combining F and β0l (eqs 3 and a1) yields φ0 ) β0l/F G0 ) equilibrium conductance [Ω-1]
and lRxm* ) 1.025(Fβ0l)1/2, where the last product can be L ) (true) junction width [Å]
further interpreted depending on the experimental framework P ) fifth order polynomial coefficient [V-4]
(e.g., systematic change in molecular length, conjugation or V ) bias [V]
electrode work function). W ) work function [eV]
(h) Achieving a linear plot in any of the versions of eq 8 is e ) electron charge [C]
required for extracting believable absolute numbers. Still, even h ) Plank’s constant [Js]
Analyzing Molecular Current-Voltage Characteristics J. Phys. Chem. C, Vol. 111, No. 11, 2007 4443

l ) molecular length [Å] A3. Conversion Equations.


The two quasi-equilibrium coefficients can be combined to
m0 ) free electron mass [kg]
extract β0l, under a priori known effective mass, m*, and contact
m* ) electron effective mass area, A, by combining eqs 4c and 7a
R ) width ratio ()L/l)
β0 ) decay parameter [Å-1]
1
β0l - ln 1 -
2 ( 5
+
3
+
6
β0l β 2l2 β 3l3
) )
( )
0 0
φ0 ) barrier height [V] G0x6C
F ) shape factor [V-1] ln(2πm0e3/h3) - ln (a6)
Am*
ψ ) potential [V] Note that second term on the left ∼0 and can be neglected to
A2. Other Forms of the Simmons Model. get an explicit expression for β0l. In addition, any uncertainty
in the effective mass, m*, has little effect on β0l, as these
a. Equilibrium (V ≡ 0) expression for the decay parameter, parameters are logarithmically related.
β0 In case of extracting G0 only, it can be translated to β0l, if
the junction width (Rl) and real contact area, A, are known by
4π rearranging eq 4b and substituting β0l ≡ Fφ
h x
β0 ) R 2m*m0eφ0 ) 1.025Rxm*φ0 (a1)
β0l - ln(β0l - 2) ) ln(e2/4πh) - ln(G0R2l2/A) (a7)
Although the logarithmic term on the left cannot be ignored
The numerical constant is valid for β0 in [Å-1], φ0 in [V], and (∼3), it is still possible, though, to calculate β0l without
dimensionless R and m* values (1.025 is in units of [Å-1 V-0.5]). considering that term, to arrive at an initial guess.
b. Nonequilibrium (V * 0) expression for the decay Rearranging eq 7a provides an explicit term for the shape
parameter, β0 factor, F, or the barrier height, φ0, as a function of the cubic
coefficient, C, and the dimensionless thickness, β0l, extracted
from G0 (eq a6)

h x
β) R 2m*m0e(φ0 - V/2) ) β0x1 - V/2φ0 (a2)

x
β0l 1 - 2/β0l
F) ) 4‚ 6C ≈ 4x6C (a8)
φ0 1 - 3/β0l - 3/β02l2
c. The widely accepted form of the linear Simmons relations

x 1 (β0l) - 3/β0l - 3
2
is 1 β 0l
φ0 ) ≈ (a9)
4 6C 1 - 2/β0l 4x6C

I)
e2 A
2πh (Rl)2 {(V
φ0 - ×
2 )
where the approximated equality refers to the condition of β0l
. 2.

[ x ( )] (
A4. Mathematical Tools.
4πRl V V Mean square error (MSE) fitting was done by a MATLAB
exp -
h
2m*m0e φ0 -
2
- φ0 + ×
2 ) function named “fminsearch” which uses the Nelder-Mead

exp - [ x ( )]}
4πRl
h
2m*m0e φ0 +
V
2
(|V| e φ0) (a3)
simplex48 direct searching algorithm to find a parameter array,
p, that minimizes the MSE defined as49
N
1
The same relations, expressed in terms of the equilibrium decay
MSE ) ∑ [f (Vi, p) - Ii]2
N i)1
(a10)
parameter, β0, are
where p ) (G0, F, φ0), [Vi] and [Ii] are N long arrays of

{( ) ( x )
voltage and current, and f represents eq 2 (or any variation of
e2 Aφ0 V V it).
I) 1- exp - β0l 1- - Implicit equations (e.g., eq a6) were solved using the
2πh (Rl)2 2φ0 2φ0

( ) ( x )}
MATLAB function “fzero” or “fminsearch”.
V V
1+ exp - β0l 1+ (a4)
2φ0 2φ0 Supporting Information Available: Glossary of translation
functions from polynomial coefficients into tunneling parameters
(S1); additional maps of mean square error as function of fitting
d. The general Simmons relation including a bias-dependent variables (S2); summary tables comparing the tunneling pa-
potential, ψ(V), and profile (width ratio), R(V), indicated by a rameters extracted by various fitting procedures (S3); and a
subscript V detailed experimental section for the bilayer junction (S4). This
material is available free of charge via the Internet at http://
pubs.acs.org.
e2 A 4π
I)( ‚ψ [
h exp -lRV
2πh R 2l2 V
V
(
h x
hV -
2m*m0eψ ) References and Notes
(1) Collier, C. P.; Mattersteig, G.; Wong, E. W.; Luo, Y.; Beverly, K.;

(ψ (
h V + |V|) exp -lRV
h x
h V + |V|)
2m*m0e(ψ )] (a5) Sampaio, J.; Raymo, F. M.; Stoddart, J. F.; Heath, J. R. Science 2000, 289,
1172.
4444 J. Phys. Chem. C, Vol. 111, No. 11, 2007 Vilan

(2) Chen, J.; Reed, M. A.; Rawlett, A. M.; Tour, J. M. Science 1999, (28) Brinkman, W. F.; Dynes, R. C.; Rowell, J. M. J. Appl. Phys. 1970,
286, 1550. 41, 1915.
(3) For a comprehensive review of the field, see: Kagan and Ratner,4 (29) Miles, J. L.; McMahon, H. O. J. Appl. Phys. 1961, 32, 1176.
and other articles in this special issue of MRS bulletin. (30) Zhang, Z.-S.; Rabson, D. A. J. Appl. Phys. 2004, 95, 199.
(4) Kagan, C. R.; Ratner, M. A. MRS Bull. 2004, June, 376. (31) Dorneles, L. S.; Schaefer, D. M.; Carara, M.; Schelp, L. F. Appl.
(5) Tomfohr, J. K.; Sankey, O. F. Phys. Rev. B 2002, 65, 245105. Phys. Lett. 2003, 82, 2832.
(6) Magoga, M.; Joachim, C. Phys. ReV. B 1997, 56, 4722. (32) Mann, B.; Khun, H. J. Appl. Phys. 1971, 42, 4398.
(7) Galperin, M.; Nitzan, A.; Sek, S.; Majda, M. J. Electroanal. Chem. (33) Polymeropoulos, E. E.; Sagiv, J. J. Chem. Phys. 1978, 69,
2003, 550, 337. 1836.
(8) Zahid, F.; Paulsson, M.; Polizzi, E.; Ghosh, A. W.; Siddiqui, L.; (34) Hansen, K.; Brandbyge, M. J. Appl. Phys. 2004, 95, 3582.
Datta, S. J. Chem. Phys. 2005, 123, 064707. (35) Probst, O. M. Am. J. Phys. 2002, 70, 1110.
(9) Berlin, Y. A.; Ratner, M. A. Radiat. Phys. Chem. 2005, 74, 124. (36) A very similar expression (up to a factor of 4/π) was used by
(10) Slowinski, K.; Chamberlain, R. V.; Miller, C. J.; Majda, M. J. Am. Hansen et al.34 as the exponential coefficient in the WKB transmission
Chem. Soc. 1997, 119, 11910. function.
(11) York, R. L.; Nguyen, P. T.; Slowinski, K. J. Am. Chem. Soc. 2003, (37) The shape factor can be viewed in analogy to the Reynolds number
125, 5948. in fluid mechanics (the ratio of inertial forces to viscous forces, see: http://
(12) Wold, D. J.; Frisbie, C. D. J. Am. Chem. Soc. 2001, 123, 5549. en.wikipedia.org/wiki/Reynolds_number) which determines the flow pattern
(13) Definition of the decay length, β varies. Most formally β should (i.e., laminar or turbulent) regardless of the absolute flow speed. Similarly,
scale the width of the potential, L; however as this property is hard to the shape factor dictates the shape of the tunneling I-V characteristics,
measure experimentally, β is defined clearer as scaling of the molecular regardless of the absolute current magnitude.
length, l. Note that in latter case, β includes an additional contribution from (38) Wold, D. J.; Haag, R.; Rampi, M. A.; Frisbie, C. D. J. Phys. Chem.
the ratio R ) L/l. It is also common to express β in terms of methylene B 2002, 106, 2813.
units or per carbon atom. Here we shall follow the 1/molecular length (Å-1)
(39) Fitting with β0l as a free parameter is equivalent to use of m* or
convention.
R, but β0l is the general lumped parameter, regardless of any assumptions.
(14) Wang, W.; Lee, T.; Reed, M. A. Phys. ReV. B 2003, 68, 035416.
(15) Wang, W.; Lee, T.; Reed, M. A. Rep. Prog. Phys. 2005, 68, 523- (40) Joachim, C.; Magoga, M. Chem. Phys. 2002, 281, 347-352.
544. (41) Segev, L.; Salomon, A.; Natan, A.; Cahen, D.; Kronik, L.; Ami,
(16) Parikh, A. N.; Allara, D. L.; Azouz, I. B.; Rondelez, F. J. Phys. F.; Chan, C. K.; Kahn, A. Phys. ReV. B 2006, 74, 165323.
Chem. 1994, 98, 7577. (42) The linear term in the analysis of Brinkman et al.28 expresses
(17) Holmlin, R. E.; Ismagilov, R. F.; Haag, R.; Mujica, V.; Ratner, M. asymmetry between the barrier heights at the two electrodes. It emerges
A.; Rampi, M. A.; Whitesides, G. M. Angew. Chem.-Int. Ed. 2001, 40, from integration over probabilities in contrast to the integration over potential
2316. used by Simmons22-24 and followed here.
(18) Cui, X. D.; Primak, A.; Zarate, X.; Tomfohr, J.; Sankey, O.; Moore, (43) Beebe, J. M.; Kim, B.; Gadzuk, J. W.; Frisbie, C. D.; Kushmerick,
A. L.; Moore, T. A.; Gust, D.; Nagahara, L. A.; Lindsay, S. M. J. Phys. J. G. Phys. ReV. Lett. 2006, 97, 026801.
Chem. B 2002, 106, 8609. (44) Full or high bias range refers to V < φ0, which is the high limit on
(19) Adams, D. M.; Brus, L.; Chidsey, C. E. D.; Creager, S.; Creutz, eq 2. For V . φ0 the tunneling can be described by the Fowler-Nordheim
C.; Kagan, C. R.; Kamat, P. V.; Lieberman, M.; Lindsay, S.; Marcus, R. relations.22,23,43 By the term near- or quasiequilibrium, we refer to the range
A.; Metzger, R. M.; Michel-Beyerle, M. E.; Miller, J. R.; Newton, M. D.; where the I/V vs V2 plot is linear.
Rolison, D. R.; Sankey, O.; Schanze, K. S.; Yardley, J.; Zhu, X. Y. J. Phys. (45) Nesher, G.; Vilan, A.; Cohen, H.; Cahen, D.; Amy, F.; Chan, C.;
Chem. B 2003, 107, 6668. Hwang, J.; Kahn, A. J. Phys. Chem. B 2006, 110, 14363.
(20) Aswal, D. K.; Lenfant, S.; Guerin, D.; Yakhmi, J. V.; Vuillaume, (46) Engelkes, V. B.; Beebe, J. M.; Frisbie, C. D. J. Phys. Chem. B
D. Small 2005, 1, 725. 2005, 109, 16801.
(21) York, R. L.; Slowinski, K. J. Electroanal. Chem. 2003, 550, 327. (47) Kushmerick, J. G.; Lazorcik, J.; Patterson, C. H.; Shashidhar, R.;
(22) Simmons, J. G. J. Appl. Phys. 1963, 34, 1793. Seferos, D. S.; Bazan, G. C. Nano Lett. 2004, 4, 639.
(23) Simmons, J. G. J. Appl. Phys. 1963, 34, 2581. (48) Lagarias, J. C.; Reeds, J. A.; Wright, M. H.; Wright, P. E. SIAM J.
(24) Simmons, J. G. J. Appl. Phys. 1963, 34, 238. Optim. 1998, 9, 112.
(25) Selzer, Y.; Cai, L.; Cabassi, M. A.; Yao, Y.; Tour, J. M.; Mayer, (49) In comparison, the Microcal Origin nonlinear search algorithm relies
T. S.; Allara, D. L. Nano Lett. 2005, 5, 61. on derivatives rather than on direct search. Yet the MSE function is
(26) Engelkes, V. B.; Beebe, J. M.; Frisbie, C. D. J. Am. Chem. Soc. similar to the χ2 function of Microcal Origin for a single independent
2004, 126, 14287. variable (Vi).
(27) Vilan, A.; Hikmet, R. A. M. Manuscript in preparation 2007. On (50) Similar maps were produced also for the experimental data
preparation of gold flakes, see: patent No. WO 2005085389; Hikmet, R. providing fairly irregular contours, thus not presented. A comparable
A. M., Koninklijke Philips Electronics N.V., The Netherlands, 2005 example can be found in Figure 6 of ref 15.

Anda mungkin juga menyukai