Você está na página 1de 2

LT 21

Charge

Disorder

Proceedings of the 21st International Conference on Low Temperature Physics


Prague, August 8-14, 1996
Part $2 - Superconductivity 1: LTS - Theory

in Granular

Arrays

A.D.Zaikin ",b, and S.V.Panyukov b


a Institut ffir Theoretische Festkhrperphysik, Universit~it Karlsruhe,
76128 Karlsruhe, Germany
b I.E.Tamm Department of Theoretical Physics, P.N.Lebedev Physics Institute,
Leninskii prospect 53, 117924 Moscow, Russia
It is demonstrated that already weakly disordered distribution of offset charges destroys the KosterlitzThouless-Berezinskii charge-unbinding phase transition in two dimensional normal and superconducting granular arrays and films. Under these conditions the array conductance shows a purely thermally activated behavior with the effective Coulomb gap being decreased due to charge disorder. Our results may account for
recent experimental findings in two dimensional arrays of tunnel junctions.
Coulomb interaction in mesoscopic metallic systems gives rise to a variety of novel physical phenomena. One of them has been discussed by Mooij
et hi. [1, 2] who argued that two dimensional (2D)
granular arrays can exhibit a Kosterlitz-ThoulessBerezinskii (KTB) phase transition for charges with
the corresponding critical temperature TKTB of order Ec = e2/2C, where C is the typical intergrain
capacitance. Physically it implies that at T > TKTB
there is a nonzero concentration of free charges in the
system and its conductance remains finite G > 0,
whereas at T < TKTB all charges are bound in
charge-anticharge pairs and the system linear conductance drops to zero G = 0. The above conclusion
applies to both normal and superconducting granular arrays with an obvious substitution e --~ 2e for
the latter case [1, 2].
Several experiments [3, 4, 5] were performed to
study charging effects in both normal and superconducting 2D granular arrays and none of them
indicated the presence of a KTB phase transition
for charges. E.g. no specific KTB dependence
G(T) (x e x p ( - A / x / T K T B - T) in the vicinity of
TKTB "" E c / 4 z has been found. In contrast for
a wide temperature region the array conductance
was reported to follow a purely activation behavior
G(T) (x e x p ( - a E c / T ) , where for most of the samples the parameter a varied between 0.23 and 0.27
[1, 3, 4, 5]. Possible reasons for that might be finite size and the selfcapacitance effects (the selfcapacitance of the grain needs to b e equal to zero for
Czechoslovak Journal of Physics, Vol. 46 (1996), Suppl. $2

the KTB charge unbinding phase transition to occur [1, 2]) which turn a charge-KTB transition into
a crossover. Nevertheless for the sample parameters
[1, 3, 4, 5] such a crossover could be expected to be
sufficiently sharp to be distinguished from a purely
activation behavior.
Here we argue that another possible reason for
the absence of the KTB phase transition for charges
could be the presence of randomly distributed noninteger offset charges in the array and that the experimental results [1, 3, 4, 5] acquire a natural physical explanation provided the effect of random offset
charges is taken into account.
Let us consider a square tunnel junction array
with the junction capacitance C and assume that
the selfcapacitance of the islands is very small. If, to
begin with, we neglect intergrain electron tunneling
we can write the array Hamiltonian simply as
1
=

(1)

27~ t

where ~ is the charge operator for the grain x,


C~-~, is the inverse capacitance matrix and Q~ are
quenched offset charges randomly distributed over
the system, so that Qx = 0 and
e2
"Q=Q=, = ~--~2g(x - x').

(2)

Depending on the physical situation different types


of disorder can be considered. E.g. the values Q= on
629

each island can be fixed by strong local potentials ill


the substrate being completely independent of each
other. Then we have g ( x - x ' ) = g,,6(x - x'). Alternatively one can assume that Coulomb interaction
between offset charges on different islands dominates
over the local potentials. In this case we have
gk = 4 [sin2(k,/2) + sin2(k2/2)] g~ = 4A(k)g~, (3)
gk is the Fourier component of g ( x - z') and ttle
components kt,2 of the wave vector k are normalized
by tile inversed lattice spacing constant 1/a.
Let us consider the case (3). Making use of the
replica trick one can average over disorder [6] and
derive the scaling equations

dlny
(
Ec
geE~ ~
d l = 2 - 2 r T + --~Th- ,I '

d(T/Ec)
7r 2
-~
- -2 Y '

and dgddl = 0. These equations have the same form


as those derived and investigated earlier in Ref. [7]
for a 2d XY model with random phase shifts.
For ge > 1/327r the fugacity y monotonously increases during the scaling procedure. Therefore the
density of free charges in the system remains finite
at any T and the KTB phase transition for charges
never occurs. Note that already for a relatively weak
charge disorder the condition ge > 1/32r is well satisfied. E.g. if one assumes that correlated charges on
grains are uniformly distributed between - e / 2 and
e/2 one gets the value gc = r2/12 >> 1/327r. Furthermore one can argue that at nonzero T [8] and/or
for a relatively large fugacity [6] (the latter is actually the case) the KTB phase transition is smeared
out even for smaller values of go. Without entering
a more detailed discussion of the system behavior
at extremely small g~ we just conclude that already
very weak disorder of correlated charges completely
destroys the KTB phase transition for charges. For
uncorrelated random charges the condition for this
phase transition is even more stringent [6].
Let us now turn to a calculation of the system
conductance in the presence of the charge disorder.
Describing intergrain electron tunneling perturbatively in the junction conductance 1/Rt we obtain
the expression for the free energy of a junction F in
the array for a given distribution of the offset charges
Q(k). Then, proceeding along the same lines with [9],
we perform an analytic continuation of F, calculate
its imaginary part and obtain the junction conductance for a given Q(k). In the absence of the offset
charges Q(k) = 0 the conductance of each junction in
the array reads G = (2Rt) -1 e x p ( - E c / 2 T ) . As the

630

total conductance of a 2D square array is Gtot = 2G


we conclude that the effective Coulomb gap of such
an array is equal to E c / 2 (cf. [3]).
For nonzero Q(k) one should find tim average
value of the conductance integrating over all disorder
configurations. Assuming thermodynamic distribution of offset charges [6] we find

G(T) = (2Rt) -1 e x p [ - ( 1 / 2 - 1/r)Ec/T}.

(4)

Provided the disorder in the distribution of local


conductances is not very large the array conductance Gto~ can be evaluated within the mean field
approximation which yields Gtot(T) " G(T) cx
e x p ( - a E c / T ) with a = 1 / 2 - 1/Tr ~_ 0.2. This value
of a is close to that found experimentally [1, 3, 4, 5]
especially if we take into account the finite size effect which would make the Coulomb gap somewhat
larger than that for an infinite array.
Thus disorder in the distribution of the offset
charges decreases the Coulomb gap of 2D tunnel
junction arrays. Interpretation of this result is transparent. Indeed in the presence of randomly distributed offset charges Q= the effective charge of a
half of the tunnel junctions in the array is positive.
Thus the Coulomb gap for electron tunneling across
such junctions in one direction becomes smaller than
that without an external charge Ec/2. At sufficiently low T these junctions yield a dominant contribution to the array conductance Gtot for the corresponding direction of the current. For the current of
the opposite sign the same argument applies for another half of the junctions. Hence, in the presence of
a charge disorder one always has a < 1/2. The value
a may depend on the disorder type and temperature.
The same effect takes place in superconducting arrays. At sufficiently low temperatures the
number of quasiparticles above the superconducting
gap Ao is exponentially small and we get Gtot cx
exp[-(aEc + Ao)/T]. This result is also in a good
agreement with the experimental findings [3, 5].
REFERENCES

liI

P ys vt

R. Fazio, G. SchSn, Phys.Rev.B43,5307(1991).


T.S. Tighe el al., Phys.Rev.B47,1145(1993).
P. Delsing el aL, Phys.Rev.B50,3959(1994).
H.S.J. van der Zant et al., preprint (1995).
S.V.Panyukov, A.D.Zaikin, e-print condmat/9510159.
[7] M.Rubinstein, B.Shraiman, and D.R.Nelson,
Phys.Rev.B 271800(1983).
[~] T.Nattermannetal, J. de Phys. I 5,565(1995);
G.Schhn, A.D.Zaikin, Phys.R.ep.198,237(1990).

Czech. J. Phys. 46 (1996), Suppl. $2

Você também pode gostar