High temperature electrical resistivity and Seebeck coefficient of Ge2Sb2Te5 thin films

L. Adnane, F. Dirisaglik, A. Cywar, K. Cil, Y. Zhu, C. Lam, A. F. M. Anwar, A. Gokirmak, and H. Silva

Citation: Journal of Applied Physics 122, 125104 (2017); doi: 10.1063/1.4996218

View online: http://dx.doi.org/10.1063/1.4996218
View Table of Contents: http://aip.scitation.org/toc/jap/122/12
Published by the American Institute of Physics
 JOURNAL OF APPLIED PHYSICS 122, 125104 (2017)

High temperature electrical resistivity and Seebeck coefficient

of Ge2Sb2Te5 thin films
L. Adnane,1 F. Dirisaglik,2 A. Cywar,1 K. Cil,1 Y. Zhu,3 C. Lam,3 A. F. M. Anwar,1
A. Gokirmak,1 and H. Silva1
1
Department of Electrical and Computer Engineering, University of Connecticut, Storrs, Connecticut 06269, USA
2
Department of Electrical and Electronics Engineering, Eskisehir Osmangazi University, Eskisehir 26480, Turkey
3
IBM Watson Research Center, Yorktown Heights, New York 10598, USA
(Received 14 July 2017; accepted 10 September 2017; published online 25 September 2017)
High-temperature characterization of the thermoelectric properties of chalcogenide Ge2Sb2Te5 (GST)
is critical for phase change memory devices, which utilize self-heating to quickly switch between
amorphous and crystalline states and experience significant thermoelectric effects. In this work, the
electrical resistivity and Seebeck coefficient are measured simultaneously as a function of temperature,
from room temperature to 600  C, on 50 nm and 200 nm GST thin films deposited on silicon dioxide.
Multiple heating and cooling cycles with increasingly maximum temperature allow temperature-
dependent characterization of the material at each crystalline state; this is in contrast to continuous
measurements which return the combined effects of the temperature dependence and changes in the
material. The results show p-type conduction (S > 0), linear S(T), and a positive Thomson coefficient
(dS/dT) up to melting temperature. The results also reveal an interesting linearity between dS/dT and
the conduction activation energy for mixed amorphous-fcc GST, which can be used to estimate one
parameter from the other. A percolation model, together with effective medium theory, is adopted to
correlate the conductivity of the material with average grain sizes obtained from XRD measurements.
XRD diffraction measurements show plane-dependent thermal expansion for the cubic and hexagonal
phases. Published by AIP Publishing. [http://dx.doi.org/10.1063/1.4996218]

I. INTRODUCTION layer is grown on a low doped p-type Si substrate. 300 nm

Ge2Sb2Te5 (GST) has been the most studied phase-
change material for phase change memory (PCM) devices.
The atomic structure of this chalcogenide material is very sta-
ble in crystalline phases, and it presents an electrical resistiv-
ity that is 5–6 orders of magnitude lower compared to the
resistivity of the material in the amorphous phase. The face
centered cubic (fcc) phase of the material and the hexagonal
(hcp) phase are characterized by very high conductivities due
to high carrier density.1,2 Thermoelectric transport—the cou-
pling of electronic and thermal transport—plays an important
role in the operation of PCM devices due to high current den-
sities and temperature gradients3–8 and has been observed as
asymmetric amorphization of GST line cells and polarity
dependent operation of PCM cells in general.3 However, data
on thermal conductivity and Seebeck coefficient of phase-
change materials are still limited. In this work, we report the
evolution of the Seebeck coefficient S(T) and the electrical
resistance R(T) characteristics of GST as the material is pro-
gressively changed from amorphous to crystalline fcc and
then to crystalline hcp. The Seebeck coefficient of a material
represents the open circuit voltage generated across a temper-
ature gradient along a sample, which can be approximated as
S ¼ dðV0 Þ=dðDTÞ for small temperature gradients.9,10
II. EXPERIMENTAL SETUP AND MEASUREMENT
PROCEDURE
The electrical resistivity and Seebeck coefficient of GST
are measured simultaneously using GST thin films with tung-
sten bottom contacts. A 1 lm thick silicon dioxide (SiO2)
deep trenches of the 2 mm  2 mm area are opened into SiO2
by optical lithography and reactive ion etching (RIE).
Tungsten is then deposited by physical vapor deposition
(PVD), and the contacts are formed using second optical
lithography and RIE steps. Bottom tungsten contacts were
found to provide better adhesion and contact to the GST
films, and the misalignment arising from the two lithography
steps is not important for these large-scale structures. Thin
GST films are deposited by elemental co-sputtering of Ge,
Sb, and Te and capped by a 10 nm layer of SiO2 to prevent
evaporation and oxidation (Fig. 1). GST has been found to
form good Ohmic contacts with most metals.11
The measurement setup used for simultaneous S(T) and
R(T) measurements up to high temperatures was described
in detail in Ref. 12. The bottom metal contacts are probed
with tungsten probes through the soft GST film and are con-
nected to an HP 4145B semiconductor parameter analyzer.
A distance of 20 mm between the two contacts is sufficient
to achieve a temperature difference of DT  10  C. The
measurements were performed in an enclosed chamber under
low vacuum (8  104 kPa) with nitrogen flow to minimize
the oxidation of the chuck and probe tips. Two K-type ther-
mocouples clamped very close to the contacts for each GST
sample were used to monitor the temperature on both sides
of the sample. The brass-alloy chuck is 10 mm thick, pro-
viding a good thermal inertia and linear temperature profile
on the surface between the probe locations. Two resistive
heaters inserted in the chuck are used to set the temperature
gradient (Fig. 2). In order to reach very high temperatures
(800  C), an inductive heater coil positioned below the

0021-8979/2017/122(12)/125104/9/$30.00 122, 125104-1 Published by AIP Publishing.

125104-2 Adnane et al.
setup is used to generate a high frequency alternating mag-
netic field that results in heating of the chuck through contact
with steel surrounding plates. The temperature gradient can
also be controlled with the inductive heater by moving the
coil from one side to another. The surrounding steel plates
form an oven-like enclosure for heat confinement and heat-
ing of the probes and probe arms to reduce the probe cooling
effect.
When a temperature gradient is set across the sample at a
given stabilized temperature, the I-V characteristics obtained
with the semiconductor parameter analyzer simultaneously
provide the Seebeck voltage, which is the x-axis intercept of
the line, and the resistance between the two contacts, which
is the slope of the line (Fig. 3).
In the experiments, the maximum temperature differ-
ence recorded between the probe arm and the surface of the
chuck does not exceed 150  C at a chuck temperature of
600  C. The probe needles (Cascade Microtech PTT-24/
4–25 tungsten needles) are 20 mm long, and its tip diame-
ter is 2.5 lm. An approximated geometry of a GST sample
with electrical contact (2D cylindrically symmetric tungsten
probe tip on sample tungsten bottom contact) is simulated in
COMSOL to estimate the effect of the cooling by the probe
arms via a 10 lm diameter tungsten contact on the surface of
the GST film (Fig. 1). A worst-case scenario is simulated by
setting the temperature of the top surface of a 1 lm thick
tungsten contact on the sample surface at 200  C, with the
bottom surface temperature of the sample (surface of the
chuck) at 400  C. The simulations show that outside the
small area where the probe tips come into contact (10 lm
diameter), the temperature of the sample surface is the same
as the chuck temperature (Fig. 4).
FIG. 2. Measurement setup schematic. (1) K type thermocouple, (2) probe
arm with a tungsten probe, (3) GST sample, (4) brass chuck, (5) cartridge
heater, (6) inductive heater coil, (7) ceramic glass base, and (8) surrounding
steel plates. The steel plate on the right is drawn partially transparent to
show the components inside.
J. Appl. Phys. 122, 125104 (2017)
FIG. 1. Top-view optical image and
schematic of the 100 nm thin film GST
sample. The film is deposited on top of
2 mm  2 mm W bottom contacts.
GST behaves like a p-type semiconductor,13 and the
conductivity ðrÞ can be expressed as
rðT Þ ¼ elp ðTÞpðTÞ; (1)
where e is the elementary charge, p is the carrier density, and
lp is the low field hole mobility. Both the carrier density and
the mobility are temperature dependent.
The measured resistance of the sample is scaled to resis-
tivity using the resistivity value of the sample measured at
room temperature using the van der Pauw (vdP) method14
and assuming that the geometry factor does not change with
temperature. The resistivity of a square shaped GST thin film
sample with known film thickness, from the same wafer, is
measured at room temperature using four probes to contact
the film surface [Fig. 5(a)]. According to the vdP technique,
the resistivity of the sample is given by
R14; 23 þ R12;43
q ¼ 4:532 tf ; (2)
2
where t is the thickness of the film, R14,23 ¼ V14/I23, and
R12,43 ¼ V12/I43, and the values of the coefficient f are tabu-
lated for arbitrary values of the resistance ratio R14,23/
R12,43.14
The measured resistivity of the 100 nm thick fcc-GST
film in a square shape [configuration in Fig. 5(a)], at room
temperature, previously heated up to 200  C, is 504 6 3
mXcm (assuming a 5% error on the film thickness).
A second sample from the same wafer, with 4 inline
metal contacts, as shown in Fig. 5(b), is also heated up to
FIG. 3. Typical I-V characteristic obtained at T ¼ 400  C and DT ¼ 4  C.
The x-axis intercept (V0) is the Seebeck voltage, and the inverse of the slope
is the resistance between the two contacts.
125104-3 Adnane et al.
FIG. 4. Numerical modeling of the cooling of the sample by the probes. (a)
Temperature of the sample simulated after 60 s sample heating to 400  C.
(b) Temperature of the simulated probe tip region while the tip surface is
maintained at 200  C. (c) Temperature on the surface of the sample across
the metal contact.
200  C, with the same heating rate (5  C/min), to measure
the R(T) characteristic during heating and cooling back to
room temperature. The resistance between the two inner con-
tacts on the surface of the sample is given by
R ¼ V23 =I14 : (3)
The results obtained from the 2 previous measurements are
used to determine the resistivity value of the film at 200  C
qroomT
q200 ¼ R200 : (4)
RroomT
This procedure is repeated for a second anneal temperature
(300  C) and the R(T) characteristic obtained while heating
matches the R(T) obtained during the previous cooling from
200  C (Fig. 6). The obtained resistivities at 200  C and
300  C are 75 mXcm and 11 mXcm, respectively.
The measured R(T) characteristics for different samples
are then scaled to resistivity (Fig. 7) using the resistivity
value at 200  C, which assumes the same geometry factor
measured at room temperature.
The sign of the Seebeck coefficient indicates the
majority carrier type (S > 0 for p-type conduction). The
FIG. 5. Schematics of measurement settings showing the position and the
numbering of electrical contacts for (a) the configuration for the room tem-
perature measurement and (b) 4-point configuration for the R(T) measure-
ment. (c) 2-point configuration for R(T) and S(T) measurements on 2
different samples.
J. Appl. Phys. 122, 125104 (2017)
FIG. 6. GST resistivity versus temperature measurements on the 200 nm
GST film using the 4-point in-line measurement, scaled from R(T) using the
vdP measurement at room temperature after heating and cooling down the
sample from 200  C and from 300  C at the same rates.
configuration shown in Fig. 5(c) is used for simultaneous
R(T) and S(T) measurements of two different samples during
the same run. The temperature difference DT ¼ Thot  Tcold
is the temperature difference between the two probe loca-
tions, which is varied between 10  C and þ10  C at given
stable average temperatures TAVG¼(Thot þ Tcold)/2.
III. RESULTS AND DISCUSSION
A. Continuous R(T) measurements
Continuously increasing temperature measurements of
R(T) were performed on 50 nm, 100 nm, and 200 nm (target
thicknesses) GST thin films up to the melting temperature
and scaled to resistivity q(T) using the room-temperature
value after anneal at 200  C as described above (Fig. 7).
During these measurements, the temperature was increased at
a constant rate of 5  C/min. The two transitions, from amor-
phous to fcc (155  C) and from fcc to hcp (365  C), corre-
spond to the turning points in the curves. The as-deposited
FIG. 7. GST resistivity versus temperature. All the curves represent the 4-
point resistance measurements obtained with the same heating rate of
5  C/min, scaled to resistivity using the resistivity value of the 200 nm film
at 200  C (dashed green line, data from Fig. 6).
125104-4 Adnane et al. J. Appl. Phys. 122, 125104 (2017)

amorphous films have a room-temperature resistivity of

8.7  102 Xcm, in agreement with reported values of
8  102 to 9  102 Xcm.4 The films typically start breaking
up and become discontinuous after melting (600  C). The
resistivity of the 100 nm GST film in the hcp phase is calcu-
lated to be slightly higher than that of the 50 nm and 200 nm
films after the second transition phase which can be attributed
to variations in film thicknesses. The drop in resistivity at
585  C as shown in Fig. 7 (inset) indicates the melting of
the film. The liquid resistivity of 1.14 mXcm observed after
this point is also in agreement with previously reported values
for GST.6,15 A film thickness reduction of 5% due to mate-
rial density changes after the first transition has been
reported16 but is not taken into account in our calculations of
resistivity.
The R(T) curves of the as-deposited amorphous GST
films show a drop in resistivity of more than five orders of
magnitude as the material transitions from amorphous to fcc
and hcp crystalline phases. In GST devices, this ratio is only
approximately four orders of magnitude.15 This difference is
attributed to the melt-quench amorphization process in
small-scale devices through electrical pulses, which leads to
lower atomic disorder and hence lower amorphous resistivity
to start with, compared to as-deposited amorphous films.

B. Simultaneous S(T) and R(T) measurements

In order to characterize the temperature-dependent resis-
tance R(T) and Seebeck coefficient S(T) of the material at
each crystalline state as the material progressively changes
from amorphous to crystalline, measurements during multi-
ple heating and cooling cycles with increasingly maximum
temperature were performed (Figs. 8 and 9). The R(T) and
S(T) characteristics were obtained simultaneously from 2-
point I-V measurements performed on 2 samples [Fig. 5(c)]
using a parameter analyzer. Due to the required thermal gra-
dient, and for practical experiment times, the measurements
are performed in 10  C increments from 40  C up to 300  C,
in 20  C increments above 300  C, and in 50  C increments
for the last measurement (up to 540  C). The final tempera-
ture was set to 180  C for the first cycle and was increased
by 20  C for each subsequent cycle.
The R(T) and S(T) results obtained simultaneously for
50 nm and 200 nm GST films are shown in Figs. 8 and 9. The
R(T) characteristics show the expected exponential decrease
in the resistance with increasing temperature, within each
crystalline state, as the material progressively crystallizes
from amorphous to fcc and from fcc to hcp.17,18 The conduc-
tivity of the film below the glass transition (100  C)19 fol-
lows an Arrhenius dependence (Fig. 8 insets)
r ¼ r0 eE=kT ; (5)
where E is the conduction activation energy, k is the
Boltzmann constant, T is the absolute temperature, and r0 is
defined as the minimum metallic conductivity.20 In the amor-
phous phase, the activation energy E corresponds to the
energy for sub-band hopping mechanism which does not
occur in fcc and hcp-GST based on UV/visible/NIR band
FIG. 8. Resistance measured in consecutive cycles to increasingly maximum
temperature for (a) 50 nm and (b) 200 nm GST films. The average tempera-
ture and temperature gradients are regulated at each step for 20 min while
measuring the resistance and the Seebeck coefficient. The insets in the
graphs show the resistance vs. 1/kT, from which the activation energies are
obtained. The dashed lines in the insets represent the resistance values of the
samples at the second transition temperature (Rc0). Each point is the average
of 15 measurement points, and for all tests but the first, the errors are
smaller than the data markers.
gap measurements.21 The activation energy for conduction
obtained for the amorphous phase from the first cycle,
E ¼ 0.417 eV, is in agreement with reported values,22 and it
decreases as the material crystallizes in the following cycles
of increasingly maximum temperatures. Given the large car-
rier concentration in GST,23,24 the overall decrease in the
resistance from one annealing cycle to another is expected to
be mostly due to the increase in mobility with crystallization.
After the fcc to hcp transition (last four R(T) curves), the
material shows a metallic behavior with the resistance
increasing with temperature due to mobility degradation, in
agreement with previous reports.5,16,25
Once the material starts crystallizing, and for each state
(corresponding to the material annealed at the previous cycle
maximum temperature), the Seebeck coefficient increases
linearly with temperature until further crystallization occurs
beyond the previous anneal temperature. Linear fits of these
regions of S(T) curves, with fixed zero intercept (0 mV/K at
125104-5 Adnane et al.
FIG. 9. Seebeck coefficient versus temperature measured simultaneously
with the R(T)s in Fig. 8 for (a) 50 nm and (b) 200 nm GST films. The insets
show the slopes dS/dT versus the conductivity of the sample at room tem-
perature. Standard deviations for each point are shown as error bars, but
these are not visible in this scale.
0 K, Fig. 10), result in very small relative standard deviation
errors on the slopes (0.6–3.7%).
The observed linear dependence of the Seebeck coeffi-
cient with temperature is consistent with the Boltzmann
approximation transport model for degenerate semiconduc-
tors, which results in26–28
 2=3
8p2 kB2  p
S¼ 2
m p T ; (6)
3eh 3p
where kB is the Boltzmann constant, h is the Plank constant,
e is the elementary charge, p is the carrier density, mp* is the
carrier effective mass, and T is the absolute temperature.
Figure 11(a) shows the conductivity activation energy
and dS(T)/dT as a function of anneal temperature for the
50 nm and 200 nm films. The derivative dS/dT, / mp*/p2/3,
shows a linear dependence on the conductivity activation
energy for the amorphous-fcc mixed phase, where this
energy is positive [Fig. 11(b)]. dS/dT is closely related to the
electrical conductivity, e2psp/mp*, where sp is the carrier
relaxation time. Decoupling the relative changes in the car-
rier concentration, effective mass, and relaxation time as the
material crystallizes would require Hall measurements. The
J. Appl. Phys. 122, 125104 (2017)
FIG. 10. Seebeck coefficient versus temperature measurement data on
50 nm and 200 nm thin films. The dashed lines represent the linear fit of the
data with the 0 lV/K intercept. The numbers in the legend represent the tem-
perature in Celsius to which the sample was previously annealed.
linear dependence of dS/dT with E, together with the linear
S(T) with S ¼ 0 V/K at 0 K, can still be used to empirically
estimate one transport parameter from the other. A few low-
temperature measurements of conductivity (to obtain the
activation energy) can be used to determine dS/dT, and
hence, S(T) or a single measurement of Seebeck coefficient
can be used to determine the conductivity activation energy
E. The different slopes of dS/dT vs. E for the 50 nm and
200 nm films may be due to a mismatch of the activation
energies for the two thicknesses since the anneal temperature
may not result in the same crystallinity state [Fig. 11(a)].
The positive Seebeck coefficient obtained for the GST
films confirms the p-type conduction of the material up to
the maximum measured temperature of 800 K. The S(T)
decrease with temperature for the amorphous phase, as well
as the increase with temperature for the stable hcp phase, is
in agreement with previous reports.5,6 For the fcc phase,
however, these multiple stepped measurements show that the
Seebeck coefficient also increases with temperature for each
crystalline state, up to the maximum temperature reached in
the previous cycle. The decrease in S(T) with temperature
observed for amorphous GST and fcc-GST after certain tem-
perature is therefore due to crystallization rather than bipolar
conduction which explains the S(T) turn-around in stable
semiconductors as they approach the intrinsic regime. A pos-
itive or negative slope in S(T) determines the direction of
asymmetry of the thermal profile (Thomson effect) that has
been observed in phase-change memory devices and has
important implications related to power and reliability of the
devices. The positive S(T) slope we observe for the crystal-
line material (both fcc and hcp) is consistent with observa-
tions of asymmetric amorphization of PCM line cells toward
the higher potential terminal.3,29,30
C. Average grain size from XRD measurements
Room temperature XRD measurements on pre-annealed
samples and in-situ XRD measurements (at stabilized tem-
peratures during anneal)31 were performed to correlate R(T)
125104-6 Adnane et al.
FIG. 11. (a) Activation energy and dS/dT as a function of anneal tempera-
ture. The activation energies for the 50 nm film are slightly higher than those
for the 200 nm film, probably due to the difference in the crystallinity of the
two films annealed at the same temperature. (b) Derivative of Seebeck ver-
sus temperature dS/dT versus the conductivity activation energy E for 50 nm
and 200 nm GST thin films, showing a linear dependence for the amor-
phous-fcc mixed phase region. Standard deviations for dS/dT are shown as
error bars. The dashed lines show the linear fits for the two samples, using
all points of positive conduction activation energy.
and S(T) with grain sizes in mixed-phase GST. Figure 12
shows the evolution of XRD patterns with temperature for
an as-fabricated amorphous 100 nm GST film as the chuck
temperature was increased from room temperature to
585  C in 100  C steps and decreased back to room tempera-
ture. The peaks in the pattern of the amorphous sample are
repeated in all patterns, suggesting that these originate from
the silicon substrate and sample holder. The fcc phase of the
sample at 200  C and 300  C is identified by the peaks at
25.5 , 29.4 , 42.4 , and at 52.6 , while the hcp phase exhib-
its different peaks at the angles 21.4 , 25.7 , 28.9 , 40.1 ,
42.8 , 48.5 , and 52.9 corresponding to different planes,
similar to what has been previously observed.32–36 After
cooling down the sample to room temperature, a new pat-
tern is acquired, and it shows a small shift to the right in all
hcp peaks, which is due to the thermal expansion of the
material at high temperatures.37 From these results, the
highest increase in the hcp planes spacing at 400  C is
1.45% for the (004) direction.
The average grain size at different temperatures was
obtained from the XRD patterns using Scherrer’s equa-
tion38,39 which relates the full-width-half-max (FWHM) b to
the grain size (gs)40
J. Appl. Phys. 122, 125104 (2017)
FIG. 12. XRD patterns of the 100 nm GST film annealed at different temper-
atures and then cooled down from 400  C to obtain the pattern for hcp-GST
at 30  C. The peaks around 47.5 and 55 , present in all patterns, likely orig-
inate from the sample holder and are not related to the GST sample.
0:9k
gs ¼ ; (7)
b cos h
where k is the X-ray wavelength and h is the diffraction
angle. XRD measurements have been previously reported for
hcp-GST annealed up to 330  C, and the average grain size
was found to be 40.3 nm.41
The data for average grain size versus temperature were
interpolated with a third order polynomial fit to correlate
XRD grain size results with R(T) and S(T) measurements
obtained after different anneal temperatures (Fig. 13). Grain
sizes calculated from impedance spectroscopy (IS) measure-
ments by Huang et al., assuming a brick model,42 fall in the
same range but show a significantly different trend with the
anneal temperature (downward curvature, square symbols in
Fig. 13).
D. Percolation model
Since the increase in average grain size in fcc-GST is
only from 18 nm to 30 nm, the drastic change in conduc-
tivity observed at the amorphous to fcc transition is likely
due to percolation paths that form in the material as both the
number of grains and the grain size increase. The electrical
conductivity of an inhomogeneous material formed by two
materials with different conductivities was described in the
late 1800s by Rayleigh43 and Wiener44 and then formulated
into the effective-medium theory by Bruggeman45 for vari-
ous shapes and by Landauer46 for spheres of conductivity rc
embedded in a material of conductivity ra. The model was
adopted recently for mixed phase amorphous-crystalline
chalcogenides such as GST.47 In this work, we model the
mixed phase amorphous-fcc GST as spheres of fully crystal-
line fcc-GST embedded in amorphous GST and use the aver-
age grain sizes obtained from XRD measurements as the
diameter of the spheres. The conductivity of the mixed mate-
rial in this case is given by45–48
125104-7 Adnane et al.
FIG. 13. Average grain size calculated from XRD measurements on the
200 nm thin GST film. In-situ measurements were done in 25  C steps from
150  C to 585  C (green circles). Room temperature measurements were
also done on different samples (from the same wafer) pre-annealed at differ-
ent temperatures for 10 minutes, with a heating rate of 2  C/min (red stars).
Interpolated data were calculated using the third order polynomial fit (black
dashed line) to obtain grain sizes for the anneal temperature values used in
the S(T) and R(T) measurements. Grain sizes calculated from room-
temperature impedance spectroscopy measurements by Huang et al. (blue
squares) are in the same range but show a significantly different trend with
anneal temperature.42
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 The resistance of fully crystalline fcc-GST is assumed to be
r ¼ 2rp  rq þ ð2rp  rq Þ2 þ 8ra rc (8)
4 constant with temperature (E ¼ 0 eV) since the activation
with
rp ¼ ð1  f Þra þ f rc ;
rq ¼ ð1  f Þrc þ f ra ;
where rc is the conductivity of the crystalline spheres, ra is
the conductivity of the amorphous matrix, and f is the crys-
tallinity fraction of the material (fraction of the crystalline
volume to the total volume of the sample). Using rc and ra
obtained from our R(T) measurements (r ¼ 1/q, qa ¼ 494
Xcm, and qc ¼ 0.01 Xcm, explained below), the calculated
mixed phase resistivity of GST (Fig. 14) shows a sharp drop
of 3 orders of magnitude at the percolation threshold of
33% crystalline fraction.
Solving Eq. (8) for f yields
ðr  ra Þð2r þ rc Þ
f ¼ ; (9)
3rðrc  ra Þ
which can be written in terms of the resistances assuming a
constant geometry factor C, r¼C/R as
ðRa  RÞðR þ 2Rc Þ
f ¼ ; (10)
3RðRa  Rc Þ
where R, Ra, and Rc are the resistances of the mixed phase,
the amorphous phase, and the fully crystalline fcc GST,
respectively.
The resistance of the mixed phase material at each crys-
tallinity fraction f follows an Arrhenius dependence:
J. Appl. Phys. 122, 125104 (2017)
FIG. 14. Calculated GST mixed-phase resistivity as a function of the crystal-
line fraction using Eq. (8) with amorphous and crystalline GST resistivities
qa ¼ 494 Xcm and qc ¼ 0.01 Xcm, respectively.
R ¼ R0 expðE=kB T Þ; (11)
where R0 and E are the pre-factor and conduction activation
energy of the mixed material. The parameters for amorphous
GST Ra0 and Ea are obtained from the first R(T) measure-
ment on the amorphous film (Fig. 8)
Ra ¼ Ra0 expðEa =kB T Þ: (12)
energy is observed to change from positive values for the
mixed amorphous-fcc phase to negative values for the mixed
fcc-hcp phase (see dashed lines in Fig. 8, insets). This con-
stant resistance value for fully crystalline fcc-GST is taken
as the resistance at the second transition temperature (the
zero of the second derivative of the R(T) ’envelope’ formed
by the last data point from each cycle)
Rc ¼ Rc0 : (13)
The different parameters, Ra0, Ea, and Rc0, are given in Table I
for the 50 nm and 200 nm thick films and are used to calculate
the crystallinity fraction f of the sample as a function of tem-
perature from Eq. (10).
Figure 15(a) shows the calculated crystallinity fraction f
along with the corresponding experimental S(T) slopes of
the mixed-phase material as a function of anneal temperature
for the 50 nm and 200 nm thick films. The crystallinity frac-
tion of the 200 nm thin film is slightly higher than that of the
50 nm film. Figure 15(b) also shows the experimental dS/dT
(/ m/p2/3) and room-temperature conductivity rRT (/ ms/p)
TABLE I. Amorphous and crystalline resistance pre-factors and activation
energies extracted from the R(T) measurements.
50 nm thin film 200 nm thin film
Ra0 (X) 77.05 56.99
Ea (meV) 419.8 376.7
Rc0 (X) 3558 1104
125104-8 Adnane et al.
FIG. 15. (a) Calculated crystalline fraction f and experimental dS/dT as a
function of anneal temperature for the 50 nm and 200 nm thin GST films. (b)
Experimental dS/dT (/ m/p2/3) and room-temperature conductivity rRT (/
ms/p) versus calculated crystalline fraction f for 50 nm and 200 nm thin film
samples.
versus calculated crystalline fraction f for the 50 nm and
200 nm thin film samples. These relations can now be used to
estimate S(T) and r(T) for a given amorphous-fcc mixed phase
material from a single, room-temperature value of the conduc-
tivity. rRT can be used to determine f and the corresponding
dS/dT and hence S(T), which can then be used to determine the
conduction activation energy E (Fig. 11) to obtain the full r(T).
Since the model for conductivity we have used assumes
spherical crystals in an amorphous matrix and f is proportional
to (gs)3 times the number of the grains in the sample, it is
interesting to look at the relationship between f and (gs)3
obtained from XRD (Fig. 16), which appears to show three
distinct regions for crystallization:
(1) 180  C < T < 220  C: the crystallinity of the sample
increases rapidly with slow grain growth (Df/Dgs3
¼ 1.1  104 nm3), suggesting that nucleation in the
material is dominant.
(2) 220  C < T < 300  C: f changes linearly with (gs)3 at a
lower rate compared to the previous region (Df/
Dgs3¼5.1  105 nm3), suggesting that the number of
crystals is approximately constant and the increase in
crystallinity is mainly related to the growth of the
crystals.
(3) 300  C < T < 320  C: the crystallinity again increases
rapidly with a small increase in grain size (Df/
Dgs3¼1.1  104 nm3), implying that the critical size
J. Appl. Phys. 122, 125104 (2017)
FIG. 16. Crystallinity fraction f for the 200 nm GST thin film sample as a
function of the cubic grain size (gs)3 obtained from XRD patterns for the
mixed phase amorphous-fcc region, pre-annealed at increasing anneal tem-
peratures (180  C to 320  C). The intermediate linear region suggests growth
dominated crystallization for this anneal temperature range.
of the crystals is reached and the increase in crystallinity
is mainly due to crystallization of the remaining gaps
between the crystals.
IV. SUMMARY
Simultaneous measurements of the temperature depen-
dent electrical resistance and Seebeck coefficient of GST
(Ge2Sb2Te5) thin films were performed up to 540  C.
Repeated measurements to increasingly maximum tempera-
tures allow characterization of the properties of each state (in
contrast to continuous measurements which show the convo-
luted effects of the temperature dependence of the transport
parameters and material crystallization during heating, as
also shown here up to melting temperature). The measured
resistance was scaled to resistivity using the 4 point room-
temperature resistivity measurement performed on a sample
of the same film. The R(T) characteristics measured at differ-
ent crystalline states follow an Arrhenius dependence with a
decreasing activation energy as the material crystallizes. The
S(T) results show p-type conduction until melting tempera-
ture and linear S(T) characteristics for each state above glass
transition temperature, in agreement with the degenerate
semiconductor transport model. The measurements also
show that both the mixed amorphous-fcc and the mixed
fcc–hcp phases exhibit a constant positive Thomson coeffi-
cient (dS/dT). A linear relationship between the slope of
S(T) characteristics and the activation energy of the
Arrhenius conduction for the GST material is also observed
for the mixed amorphous-fcc phase. This observation can be
useful to estimate the temperature-dependent conductivity or
Seebeck coefficient of GST from a single measurement of
the Seebeck coefficient or a few conductivity measurements
at low temperature points (to obtain the conduction activa-
tion energy). In-situ XRD measurements on a 100 nm
thin GST film sample showed the plane-dependent thermal
expansion of the hcp phase, with a maximum thermal expan-
sion observed for the (004) plane.
125104-9 Adnane et al.
11
A percolation model of conducting spheres in a lower
conductivity matrix (which predicts the sharp drop observed 12
in the resistance with a relatively small change in the grain
13
size, at a critical crystalline density threshold) was applied to
the measured data to relate the electrical transport parame-
14
ters and the crystalline grain sizes obtained from XRD meas- 15
urements. The R(T) results of mixed phase amorphous-fcc
are then expressed in terms of the amorphous and fully crys-
16
talline fcc temperature-dependent resistance values using the
effective-medium theory to obtain the temperature depen- 17
dent crystallinity fraction of the material.
This study focused on the properties of GST up to the
18
second phase transition (fcc–hcp), which is the more techno-
logically relevant range as phase-change memory devices do 19
20
not experience the slow fcc–hcp transition during switching.
A similar percolation and effective medium theory model
21
can however be applied to the mixed fcc–hcp phase after the
second transition for a fuller understanding of the crystalliza- 22
tion dynamics and transport properties in chalcogenide
23
glasses.
ACKNOWLEDGMENTS 24
The GST films were deposited at IBM T.J. Watson
Research Center and characterized at UConn. This work was
25
partially supported by the U.S. National Science Foundation
through Award Nos. ECCS 0925973 and ECCS 1150960. 26
The characterization and analysis efforts of L.A. were 27
28
supported by the U.S. Department of Energy, Office of
Science, Basic Energy Sciences under Award No. DE-
SC005038. L.A. also acknowledges a graduate fellowship 29
from the Graduate Assistance in Areas of National Need
30
(GAANN). F.D. and K.C. acknowledge graduate fellowships
from the Republic of Turkey Ministry of National Education. 31
A.C. was supported through an NSF GRFP fellowship. The 32
33
authors would also like to thank Simone Raoux, Norma Sosa,
and Matthew BrightSky for their contributions to sample
34
fabrication at the IBM T.J. Watson Research Center.
35
1
Z. Sun, S. Kyrsta, D. Music, R. Ahuja, and J. M. Schneider, Solid State
Commun. 143, 240 (2007).
2 36
D. Ielmini, in Proceedings of the International Conference on
Simulations of Semiconductor Processes and Devices (SISPAD) (IEEE, 37
2009), pp. 1–8.
3 38
D. T. Castro, L. Goux, G. A. M. Hurkx, K. Attenborough, R. Delhougne,
J. Lisoni, F. J. Jedema, M. A. A. In’t Zandt, R. A. M. Wolters, D. J. 39
Gravesteijn, M. A. Verheijen, M. Kaiser, R. G. R. Weemaes, and D. J.
40
Wouters, IEEE Int. Electron Devices Meet. 315 (2007).
4
H.-S. P. Wong, S. Raoux, S. Kim, J. Liang, J. P. Reifenberg, B. Rajendran,
41
M. Asheghi, and K. E. Goodson, Proc. IEEE 98, 2201 (2010).
5
J. Lee, T. Kodama, Y. Won, M. Asheghi, and K. E. Goodson, J. Appl.
42
Phys. 112, 14902 (2012).
6
T. Kato and K. Tanaka, Jpn. J. Appl. Phys., Part 1 44, 7340 (2005).
7 43
N. Ciocchini, M. Laudato, and A. Leone, IEEE Trans. Electron Devices
62, 3264 (2015).
8 44
J. Lee, M. Asheghi, and K. E. Goodson, Nanotechnology 23, 205201
45
(2012).
9 46
R. R. Heikes and R. W. Ure, Thermoelectricity: Science and Engineering
47
(Interscience Publishers, 1961).
10
G. S. Nolas, J. Sharp, J. Goldsmid, and H. J. Goldsmid, Thermoelectrics:
48
Basic Principles and New Materials Developments (Springer Science and
Business Media, 2013).
J. Appl. Phys. 122, 125104 (2017)
L. W.-W. Fang, Z. Zhang, R. Zhao, J. Pan, M. Li, L. Shi, T.-C. Chong,
and Y.-C. Yeo, J. Appl. Phys. 108, 53708 (2010).
L. Adnane, N. Williams, H. Silva, and A. Gokirmak, Rev. Sci. Instrum.
86, 105119 (2015).
A. Pirovano, A. A. L. Lacaita, A. Benvenuti, F. Pellizzer, and R. Bez,
IEEE Trans. Electron Devices 51, 452 (2004).
L. J. Van der Pauw, Philips Tech. Rev. 20, 220–224 (1958).
K. Cil, F. Dirisaglik, L. Adnane, M. Wennberg, A. King, A. Faraclas, M.
B. Akbulut, Y. Zhu, C. Lam, A. Gokirmak, and H. Silva, IEEE Trans.
Electron Devices 60, 433 (2013).
I. Friedrich, V. Weidenhof, W. Njoroge, P. Franz, and M. Wuttig, J. Appl.
Phys. 87, 4130 (2000).
D. Fugazza, D. Ielmini, G. Montemurro, and A. L. Lacaita, in Proceedings
of the International Electron Devices Meeting (IEEE, 2010), pp.
29.3.1–29.3.4.
C. David Wright, M. Armand, and M. M. Aziz, IEEE Trans. Nanotechnol.
5, 50 (2006).
J. A. Kalb, M. Wuttig, and F. Spaepen, J. Mater. Res. 22, 748 (2007).
N. F. Mott and E. A. Davis, Electronic Processes in Non-Crystalline
Materials (Clarednon Press, Oxford, 1971), p. 591.
F. Rao, Z. Song, Y. Cheng, M. Xia, K. Ren, L. Wu, B. Liu, and S. Feng,
Acta Mater. 60, 323 (2012).
A. Redaelli, A. Pirovano, I. Tortorelli, D. Ielmini, and A. L. Lacaita, IEEE
Electron Device Lett. 29, 41 (2008).
H.-K. K. Lyeo, D. G. Cahill, B.-S. S. Lee, J. R. Abelson, M.-H. H. Kwon,
K.-B. B. Kim, S. G. Bishop, and B. Cheong, Appl. Phys. Lett. 89, 151904
(2006).
A. Voraud, M. Horprathum, P. Eiamchai, P. Muthitamongkol, B.
Chayasombat, C. Thanachayanont, A. Pankiew, A. Klamchuen, D.
Naenkieng, T. Plirdpring, A. Harnwunggmoung, A. Charoenphakdee, W.
Somkhunthot, and T. Seetawan, J. Alloys Compd. 649, 380 (2015).
T. Siegrist, P. Jost, H. Volker, M. Woda, P. Merkelbach, C.
Schlockermann, and M. Wuttig, Nat. Mater. 10, 202 (2011).
M. Cutler and J. F. Leavy, Phys. Rev. 133, A1153 (1964).
G. J. Snyder and E. S. Toberer, Nat. Mater. 7, 105 (2008).
J. P. Heremans, V. Jovovic, E. S. Toberer, A. Saramat, K. Kurosaki, A.
Charoenphakdee, S. Yamanaka, and G. J. Snyder, Science (80-.) 321,
1457 (2008).
F. Dirisaglik, G. Bakan, A. Faraclas, A. Gokirmak, and H. Silva, Int. J.
High Speed Electron. Syst. 23, 1450004 (2014).
A. Faraclas, G. Bakan, L. Adnane, F. Dirisaglik, N. E. Williams, A.
Gokirmak, and H. Silva, IEEE Trans. Electron Devices 61, 372 (2014).
K. Cil, Dr. Diss. (2015).
D. Dimitrov and H.-P. Shieh, Mater. Sci. Eng. B 107, 107 (2004).
N. Kato, I. Konomi, Y. Seno, and T. Motohiro, Appl. Surf. Sci. 244, 281
(2005).
Y. Yin, H. Sone, and S. Hosaka, Jpn. J. Appl. Phys., Part 1 44, 6208
(2005).
A. Voraud, M. Horprathum, P. Eiamchai, P. Muthitamongkol, C.
Thanachayanont, W. Somkhunthot, and T. Seetawan, Surf. Coat. Technol.
291, 15 (2016).
K. S. Bang, Y. J. Oh, and S.-Y. Lee, J. Electron. Mater. 44, 2712 (2015).
C.-A. Jong, W. Fang, C.-M. Lee, and T.-S. Chin, Jpn. J. Appl. Phys., Part 1
40, 3320 (2001).
A. L. Patterson, Phys. Rev. 56, 978 (1939).
J. I. Langford and A. J. C. Wilson, IUCr J. Appl. Crystallogr. 11, 102
(1978).
B. Cullity, Elements of X-ray Diffraction (Addison–Wesley, Reading, MA,
1967), p. 262.
Y. Hu, H. Zou, J. Zhang, J. Xue, Y. Sui, W. Wu, L. Yuan, X. Zhu, S.
Song, and Z. Song, Appl. Phys. Lett. 107, 263105 (2015).
Y.-H. Huang, Y.-J. Huang, and T.-E. Hsieh, J. Appl. Phys. 111, 123706
(2012).
L. Rayleigh, London Edinburgh Dublin Philos. Mag. J. Sci. 34, 481
(1892).
O. Wiener, Phys. Z. 5, 332 (1904).
D. A. G. Bruggeman, Ann. Phys. 416, 636 (1935).
R. Landauer, J. Appl. Phys. 23, 779 (1952).
D.-H. Kim, F. Merget, M. Laurenzis, P. H. Bolivar, and H. Kurz, J. Appl.
Phys. 97, 83538 (2005).
R. C. McPhedran and D. R. McKenzie, Proc. R. Soc. London, Ser. A 359,
45 (1978).