EQUI PARTITION PRINCIPLE OF CANON I CAL ENSE M B LES 793

from zero, i.e. , To conclude, we have derived an equipartition

principle of generalized canonical ensembles and shown
BQ; 8 Bd,.
it to be useful in the calculation of certain interesting
~P j ~&n Av ~Pi ~&n Av averages of functions of the parameters of a system.

x.
BQ;

~&n
(a, +(a,&A) 0a&'

~&n
(a, +(a;)A,)
Av
AO,
Xeedless to say, the averages thus found often can be
obtained by other methods perhaps more directly.
This is, of course, true also of the equipartition of
xvith i, j=1, 2, I (11). energy in equilibrium statistical mechanics.

PH YS I CAL R EVI EW VOLUME 91, NUM HER 4 AUGUST 15, 1953

. Electrical Properties of Titanium Dioxide Semiconductors

ROBERT G. BRECKENRIDGE AND WILLIAM R. HOSLFR
National BNreue of Standards, 8'ashington, D. C.
(Received February 24, 1953)

%measurements have been made of the electrical resistivity and peratures combined with a scattering by ionized impurities pre-
Hall coefficient of semiconducting rutile both in the form of dominant at low temperatures. The low values of the mobility in
ceramics and single crystals from 190'C to +500'C. The all samples indicate an anomalously large effective electron mass.
samples were reduced in pure hydrogen for various times and In the crystalline samples the mobility in the c direction is approxi-
temperatures to provide a range of resistivities, all corresponding mately twice as great as in the a direction. The data on the
to rather large numbers of charge carriers. The single crystal temperature dependence of the electron concentration may be
samples were measured in different orientations to study the direc- represented by a sum of two Boltzmann terms, indicating two
tional dependence of resistivity and mobility. From these measure- types of donor centers. It is suggested that these centers may be
ments the variations of mobility and charge carrier concentration oxygen ion vacancies at which one or two electrons are trapped
with temperature have been determined. The mobility data can as Ti+' ions. An energy level diagram is proposed which explains
be satisfactorily described 'in terms of an electron scattering by many of the electrical and optical observations on rutile semi-
the optical modes of lattice vibration predominant at high tem- conductors.

I. INTRODUCTION of charge carriers in a semiconductor, two measure-

' ~T ments are needed. The measurements chosen here are
has been known for some time that titanium dioxide
~ ~ of nonstoichiometric composition containing an the electrical resistivity and Hall coeKcient over a wide
excess of titanium is an n type semiconductor with temperature range. Such measurements have been made
properties showing promise of practical utility. ' ' These on both ceramic and single crystal samples. Both for
convenience in making the Hall effect measurements
previous studies on titania semiconductors, however,
had not obtained information about the electron mo- and because of the practical interest in the properties
of relatively good conducting titania semiconductors,
bility and the number of charge carriers in this material
although this information is needed for a detailed our measurements have been concerned with samples
analysis of the conduction process. Recently synthetic that, have been reduced in hydrogen at elevated tem-
single crystals of titanium dioxide in the form of rutile
peratures so that their resistivities are in the range 0.1
have become available, ' and it seemed desirable to to 1000 cm. The study of these reduced materials is also
useful in that it is possible to observe the variation of
investigate the electrical properties of semiconducting
crystals and compare them with the properties of highly the properties with temperature independent of changes
in the oxide composition which is the predominant
purified titanium dioxide ceramic semiconductors. Simi-
factor in studies on the pure oxide. As a result of this
larly, since rutile is tetragonal, an investigation of the
electron mobilities in the two crystal directions would simplification it is possible to observe a number of
be expected to be instructive. details in properties that are normally masked by larger
For such a study of the electron mobility and number effects or unmeasurably small. It will be shown that
the observations on these highly conducting materials
' W. Physik. Z. 36, 749 (1935).
' W. Meyer,
Meyer and H. Neldel, Physik. Z. 38, 1014 (1937). can tentatively be correlated with other studies on
' M. Earle, Phys. Rev. 61, 56 (1942). rutile to produce a reasonably consistent picture of the
4Von Hippel, Breckenridge, de Bretteville, and Brownlow,
Massachusetts Institute of Technology Laboratory for Insulation
energy levels in rutile.
Research Report N. D.R.C. 540, October, 1945 (unpublished).
' H. K. Henisch, Elec. Commun. 25, 163 (1948). II. PROPERTIES OF TITANIUM DIOXIDE
'

' R. G. Breckenridge and W. R. Hosier, J. Research Natl. Bur.

Standards 49, 65 (1952). Titanium dioxide crystallizes in three modifications:
' National Lead Company, South Amboy, New Jersey; Linde
Air Products Company, North Tonowanda, New York. rutile, anatase, and brookite. Of these, rutile is the form
 R. G. 8 RECKENRI DGE AND W. R. HOSLER
with the observations of Cronemeyer and Gilleo" and "
also agrees with observations of the reflection from
natural rutile single crystals made by Liebisch and
Rubens" who extended their measurements to 300@.
It has been shown" that the measurements of Liebisch
and Rubens can be satisfactorily explained by assuming
two fundamental vibrations at 180 cm ' and 500 cm ',
~ Ts+ 1 the one at 180 cm ' corresponding to a deformation of
0o= the Ti02 groupings, and that at 500 cm ' to a valence
vibration of the titanium relative to the oxygen.

III. TITANIUM DIOXIDE SEMICONDUCTORS

When the titanium dioxide is treated so as to cause
Fto. 1. C'rystal structure of rutile (TiO&). a loss of oxygen in the lattice either by heating in a low
oxygen pressure or in hydrogen, the material becomes
I

an n type semiconductor whose properties depend on

stable at temperatures greater than 820'C and is the the extent of oxygen loss. ' Several previous observa-
one obtained on firing ceramic samples of TiO2. The tions on the conductivity of unreduced rutile show a
crystal structure of rutile is tetragonal with a= 4.4923A dependence on the equilibrium conditions of tempera-
and c= 2.8930A, ' as shown in Fig. 1. The structure may
be considered as built up from somewhat distorted Ti06
octahedra, the octahedra forming chains in the c direc-
tion, each octahedron sharing an edge with the adjacent
members of the chains. The crystal density is 4.26
0. 8-
g/cms from x-ray measurements.
07
The dielectric properties of rutile are rather unusual
and have been studied on both single crystals and s 0.8
so
a
ceramics. It was found by Schmidt' that in single ga5
crystals of rutile the dielectric constant in the c direction
was 173 and in the a direction 89. An average value of
114 for a randomly oriented ceramic is computed from
these results in rough agreement with the value cu 100 I
found experimentally for densely 6red ceramic samples. " O.

a8
I
.4
I
.5
I
.8
I

Z.e
III
.910
I I
2
I I I

8
I I
8 7 8
When care is taken to prevent the loss of oxygen during W4VKLEMGT8(MICRONS)

firing of the ceramics, the dielectric loss of rutile is very FIG. 2. Optical density of clear rutile.
low, values of tan5=0. 0003 at 10' cps" being typical.
This corresponds to a room temperature resistivity ture and oxygen pressure. Earle' has found that
cu 10"0 cm.
The optical properties of rutile are also of interest, o ~P02
particularly as they shed some light on the electrical with n=4. 3 for oxygen pressures greater than 30 mm
properties to be considered later. The refractive index Hg. At lower pressures the n was variable ranging from
at 5896A of rutile is 2.903 in the c direction and 2.616 2 at low temperatures to 3 at higher temperatures.
in the a direction. "The optical transmission in rutile Karle's measurements of conductivity as a function of
has been measured for single crystals by the Radi- temperature showed an exponential relation
ometry Section of the. National Bureau of Standards
(Fig. 2) and will be described in detail elsewhere. It ~ ~lLE/kT
was found that there is a strong absorption band at
with dE= 1.7 ev for fired ceramics. A similar tempera- -

about 4100A at room temperature with no other

ture dependence with values of DE=1.834 ev for
marked absorption in the visible region of the spectrum. T)900'C and DE=1.526 ev for T&900'C has been
Measurements at infrared wavelengths show a strong reported by Cronemeyer" for single crystals of unre-
absorption at about 7 microns. This is in agreement
'~ D. C. Cronemeyer and M. A. Gilleo, Phys. Rev. 82, 975
R. W. G. Wyckoff, Crystal Structures Handbook (Interscience (1951).
Publishers, New York, 1948). "D. C. Cronemeyer, Massachusetts Institute of Technology,
~ W.
Schmidt, Ann. Physik 9, 919 (1902); 11, 114 (1903). Laboratory for Insulation Research Technical Report 46, August,
' E.g. , von Hippel, Breckenridge, Chesley, and Tisza, Ind. Eng. 1951 (unpublished); Phys. Rev. 87, 876 (1952). .

Chem. 38, 1097 (1946). '4T. Liebisch and H. Rubens, Sitzber. preuss. Akad. Wiss. ,
"A. Schroder, Z. Krist. 67, 485 (1928). Physiic-math. Kl. 211 (1921).
 ELECTRI CAL PROPERTI ES OF Ti02 SE M I CON DU CTORS 795

duced rutile, the same values of AE being found for both Although no detailed. investigation of the reduction
crystal directions. process was attempted, an eGort was made to treat the
When the material is reduced an increased optical samples under controlled conditions. For reduction, the
absorption at longer wavelengths is noted, the crystals samples of both ceramic and single crystal were placed
become blue in color and on continued reduction opaque. on a small platinum cradle in a combustion boat to
This is shown in Fig. 3 for a slightly reduced crystal allow free access of the hydrogen to the entire sample
(2.5 min in H2 at 600'C) measured at the National and placed in a 1-inch diameter quartz tube, heated by
Bureau of Standards. It will be noted that the absorp- a tube furnace, the temperature of which was controlled
tion of 0.41p, and 7p, are apparently unchanged by the to &5 'C. A preliminary calibration showed that the
treatment, but a broad absorption band centered at indicated and actual temperature agreed within 5 'C ~
about 1.2p. is introduced. Prior to the present work no at the center of the furnace. In order to control the time
detailed studies have been reported on the conductivity of reduction the furnace was heated to the desired tem-
and Hall coeKcient of such reduced materials. However, perature with purified helium gas Qowin, g slowly
the measurements available have indicated a typical through the tube, then pure hydrogen obtained by
exponential variation of the conductivity with tem- passing tank hydrogen through a dry-ice trap and a
perature LEq. (2)$ with an activation energy AE 0.2 charcoal trap at liquid nitrogen temperature was
"
0.3 ev4' between room temperature and about admitted to the system. After the desired reduction, the
200'C. A similar activation energy was found from hydrogen was again swept out with helium, and after
studies of the thermoelectric power. ' These preliminary a further heating at the peak temperature for about
studies have now been extended sufficiently to allow a twenty minutes to insure sample homogeneity, they
rather detailed picture to be drawn of the electrical were cooled to room temperature under helium. The
properties of titania semiconductors. ceramic samples were reduced for ten minutes at tern-
peratures ranging from 600'C to 900'C and for three
IV. EXPERIMENTAL PROCEDURE different times at 800'C. Appropriate conductivities
The ceramic samples studied in this investigation were usually obtained in the single crystals by reduction
were fired from very high purity rutile by the Porcelain for five minutes for the same range of temperatures
and Pottery Section of the National Bureau of although in a few cases a longer time was required. The
Standards. A spectroscopic analysis of the samples precautions described were, in general, sufhcient to
showed that the chief impurity was alumminum present produce a reasonably regular dependence of the number
in amounts ~& 0.001 percent, all others being ~& 0.0005 of charge carriers, etc. on the times and temperatures of
percent. The firing temperature was 1450'C, and the reduction although in a few cases anomalous values
samples were held at this temperature two hours. The were found for some unknown reason.
density was measured" to be 4. 10 g/cm'. The color of The electrical conductivity and Hall eGect measure-
the fired ceramic was rather gray, indicating some loss ments were made on a dc instrument of conventional
of oxygen in the firing. Using a small diamond saw, the design shown schematically in Fig. 4. The magnet
fired ceramics were cut into rectangular parallelepipeds produced a homogeneous field over a volume much
cu 1 mm &( 3 mm &( 10 mm. After cutting, they were larger than the sample, the field strength being ca 4000
lightly ground with No. 600 Alundum and washed. gauss at the sample with a gap of 1 ', in. The field was

The rutile crystals studied included a crystal ob- measured with a Leeds and Northrup Auxmeter. The
tained from the National Lead Company and one from temperature was measured by a calibrated Chromel
-
the Linde Air Products Company. Spectroscopic anal- Alumel thermocouple in close proximity to the sample.
yses of the crystal showed the presence of Ba as a chief The measurement of resistivity was made by the
impurity cu 0.01 percent, minor impurities of Al and
Mg present in amounts of 0.001 to 0.01 percent, with
LO
traces of Ag, Cu, Ca, and Fe in both cases. The Linde
crystal also contained a minor amount of Pb. The de-
alkalies, if present, are in amounts less than 0.05 percent. 08
The crystallographic axes of these boules were deter- d7
mined by an x-ray study in the Mineral Products p 04-
K
Section of the Bureau, and samples of approximately g 05
the same size as the ceramics were cut from the boule
using the diamond saw. Cuts were made with the c axis
in the thickness direction and with the c axis parallel
with the length of the sample. These crystals were also CLI
lightly ground with Alundum and washed. III
"Determined by the Capacity, Density and Fluid Meters
Ck5
I
.4
I I I
4 .6 .7 8 SIO
I I

WAVELENGTH (MICRONS)
545878
I I I I I

Section of the Mechanics Division of the National Bureau of

Standards. FIG. 3. Optical density of slightly reduced rutile.
796 R. G. 8RECKENRI DGE AND Ml. R. HOSLER
I
by a resistance box and was measured from the voltage
TYPE K2 TC,
drop across a precision 8-ohm resistor. All emf's were
POT ENTIOME TER
RES.
savpd measured with a I eeds and Northrup Type K-2
potentiometer.
HALI EME The sample holder consisted of a thin quartz plate
ST'D. carrying small binding posts for the electrical leads.
PRI, ERR.. RESISTOR
These leads were 20-mil platinum-rhodium wire shar-
pened and sprung into contact with the sample. They
~ OV DC. IIOV AC. VARIAC
were then soldered to the sample with indium metal.
HIGHTEMP.
L==) Even though measurements were made at temperatures
ELECTROMAGNET CONTROL
( above the melting point of the indium, no difficulties
were experienced since the indium wets the Ti02 and is
LOW TEMP.
APPARATUS
held in place by its surface tension. The quartz plate
LIQUID
was bolted to a half-round brass backing plate to aid
in maintaining a uniform sample temperature. The
FIG. 4. Schematic diagram of equipment for Hall effect brass backing plate was supported on a rod and the
and resistivity measurements.
electrical leads were brought out from the sample holder
potential probe method. A three-electrode system was through TeAon spacers mounted on the rod. For high
used for the Hall effect measurements; in this way temperatures the rod was brass, but for low tempera-

I I I I I I I I I I I I I
I I I I [ I I I I I I I I

IO. CI O
IO.

O
I.O
O
I- I-
I-
CA I- 800'Go -o-o C DIR REDUCED IND. MIN.
CO 900'C o-o--o C Co 700 II 4
850 o
DIRECTION o
gol o--o Q O.I 650 I5 a

800 II REDUCED lL 800
= = A
II
6 II

700 o~
750 ++--+ 4
650 x --a
o II

DIRECTION
5 MINUTES 70P
650
x -x
-x
+ -+
+
II
o
6
N
II

QOI' I
O.OIp
0 I 2 3 4 5 6 7 8 I 2 3 4 5 6 7 8 9 !0 II
'
l2 I3
IO II l2 l3 I/T Kx IO
I/T'K X IO
FIG. 7. Resistivities of reduced Linde Air Products Company
FIG. 5. Resistivities of reduced National Lead Company
rutile single crystals. rutile single crystals.

difficulties with electrode asymmetries are eliminated. tures it was of Bakelite in order to provide thermal
The primary current through the sample was controHed insulation.
The low temperatures were obtained by cooling the
sample holder in a small covered Dewar Qask containing
liquid nitrogen. . This Qask in turn was surrounded by
O a larger Dewar Aask also containing liquid nitrogen.
IO.
High temperatures were obtained with a small tube
furnace. In both cases temperature fluctuations were
8 avoided by enclosing the sample.
O
O
yn&
In the measurement of conductivity, some potential
O 'o0 & probe studies were made to investigate the possibility
I-
z og 0'o of field distortion in the samples. No such eGects were
O O. l P
o:
g'o noted with ields of the magnitude used for measure-
0 o-oo C
O
o
ooo
800
DIRECTION, 900
850
C

REDUCED
ment ((0.
5 volt/cm). The variation of Hall effect with
OMM
+--+
700 5 MINUTES primary current was also investigated and found to be
O.OI
+ 4 DIRECTION, 750
x -x x 650 linear for currents used in these measurements.
l I l I 1 I I I I I I
0 I 2 3 4 5 6 ? 8 9 IO I I l2 I3 V. RESULTS
I/Tk X IO

Fzc. 6. Hall coe%cients of reduced National Lead Company The results of measurements of the resistivities of
rutile single crystals. National Lead Company crystal samples plotted as
 ELECTRICAL P ROPERTIES OF Ti02 SEMI CONDUCTORS 797

logp in 0 cm as a function of the reciprocal of the ab- I I I I I I I t I I I I

solute temperature are shown in Fig. 5. A linear de- +~+ + '++--+ +

+
V -+--+--+--- +-- +
pendence of logp as a function of 1/T is not observed;
Io
instead, the resistivity varies only slightly over the
entire temperature range in a manner resembling that X
O

""
found for degenerate or partially degenerate semicon-
ductors in other cases. With the apparatus available I

o--~
it was not possible to observe the intrinsic conductivity --o--
~e- -o-o
g~-o
at high temperatures since the material begins to re- VI
V)
600
roo
C REDUCED IQ MIN.
Io
oxidize at temperatures much above 500'C in a, ir. cr O.I 800 II
5
800 IQ
The Hall voltages for these samples were measured as 800 20
a function of temperature and the Hall coefhcient R 900 Io

in cm'/coulomb calculated from the relation 0

I

I 2
I

4
I I

5
I

6 7
I I

S 9
I I I I I
IO II I2 Q
I/TK x IQ
8=10 sVIrt/HI, (3)
F&G. 9. Resistivities of reduced rutile ceramics.
where VH is the Hall voltage in volts, t the sample
thickness in cm, II the magnetic 6eld in gauss, and the I Figs. 9 and 10. In all cases the same characteristic
current in amperes. For the sample sizes used the features are found,
shorting effect of the current electrodes is small. For all From the measured values of the Hall coeScient R,
the number of charge carriers e, present in the sample
may be calculated using the relation

IOO
where G(t) is a function of the Fermi level varying
between 1 for nondegenerate samples to 8/3Ir for com-
~~ IO,O pletely degenerate samples. From evidence to be de- "
c3
scribed later in this paper, it may be shown that, even
I- 0 A4
0
though the numbers of charge carriers are large, the
~ I.O samples are not degenerate; hence, G(I) was assumed
IJI
to be 1. Then the number of charge carriers per cm' is
C)
REDUCED IND. MIN. given by
+ Q. l o
15
I5 II
II, = / 40X 10rs/E
II II
II
I5 II w
where R is in cm'/coulomb.
I
Similarly, the Hall mobility, IIlr, in cm /volt sec is
0 I 2 5 4 5 6 7,
lo'
8 9 IO II 12 l5. given by the relation
I/T'K x

FIG. 8. Hall coeScients of reduced Linde Air Products = 1/IMp = (8/3Ir) (R/p),
IIJl
Company rutile single crystals.
for p in ohm
' cm '.
samples the sign of R was always negative, indicating
I I .
conduction by electrons. The measurements of Hall Ioo
I I I I I I I I I I

+
voltage were made in two of the three orientations
possible for rutile. Those designated as c direction are 03
g++
O
with the primary current parallel to the c axis, those r /+

designated as a direction are with perpendicular to c I O Io

O

and II parallel to c. Appropriate size samples with the I-

2
-&x-- J-

third possible orientation could not be cut from the O

-Lo
IJ
crystals on hand. . The values of logR plotted as a IJ
IJI

function of the reciproca, l temperature are given in 8 REDUCED 10 MIN.

II
Io II

Fig. 6. For all samples logR changes very slowly with //

800 Io
Q.I /x 8 QQ x--x--x 20 1I

temperature at low temperatures but shows a more o

/0
9QQ o--o-o Io
pronounced linear dependence on the reciprocal tem- /

perature at somewhat higher temperatures.

Similar measurements on the I inde crystal are shown I 2 5 4 5 6
I/T Kx lo
7 8 9 IO II l2 I5

in Figs. 7 and 8 and on the pure ceramics samples in

Fzo. 10. Hall coe%cients of reduced rutile ceramics.
"C.
'
L. Pearson and J. Bardeen,
Phys. Rev. 75, 865 (1949).
's
G. Busch and H. Labhart, Helv. Phys. Acta 19, 463 (1946). K. Shifrin, J. Phys. (U. S.S.R.) 8, 242 (1944).
 R. G. 8 RECKENRI DGE AND W. R. HOSLER
) OP1 this paper indicate that the eRective electron mass is
Samole Dir. Temo. Min.
o much larger than the free electron mass, and, if the
~ " A 700
Nat. Lead C
650
5
5 eRective masses found from the mobility data are used
& linde C 700 5
A 700 5 in Eq. (8), it is found that our samples are not degen-
'I
+ Ceramic
"
'-900 l0
X 600 l0 erate in the temperature range covered. The values of
t
ORO n&', n&', AE&, and dE2, obtained by 6tting the two
terms to the observed values of n, at high and low
temperatures, are given in Table I. The activation
energies show a dependence on nbut the dependence
+ is not as clearly indicated as in the silicon case, and "
actually is in the opposite'direction, i.e., AE increases
to" for increasing n, .
X
C3
~ y0 VII. ELECTRON NOBILITIES IN RUTILE
~ ~
OP

C3 The values of the Hall mobility pII found using Eq.

O
CP (6) are of the order of 0.1 to 1.0 cm'/volt sec at room
CP )pal
O temperatures for both the single crystal samples and
O the ceramics, although the ceramics have the smaller
values. An interpretation of the observations on electron
O mobilities in rutile may be given using the theoretical
'x treatment of the scattering of slow electrons by the
x
IO x
-"--x--x co-workers. ""
optical lattice vibrations developed by Frohlich and
This work showed that the microscopic
mobility p is given as
p= 2;(e"r- 1),
I I I l l
where A, is a constant and 8 is the Debye temperature.
6 8 IO I2 In reference 20 it was shown that
I/T Kx lp
1 h' v3 E,'E'
Fio. 11. Electron concentrations in reduced Ti02 samples. A, =-- (10)
(m*)& e 2~'" K,' E' (kO)'
VI. ELECTRON CONCENTRATIONS IN RUTILE
where E,' and E' arethe static and optical dielectric
The number of charge carriers n, at various tempera- constants. In the absence of any detailed information
tures was calculated from Eq. (5). Representative as to the relation between pII and p, in this material,
results are shown in Fig. 11 for the various samples. A
reasonably satisfactory description of the change of n, TABLE I. Properties of rutile semiconductors.
with temperature can be given for all samples with a
m*
simple expression of the form b, E& AE2 nP n2o
A
cm2/volt p =
~ // oe BE1(2TLT+$$0e AE 2I2kT)
Sample ev ev No. /cmo
National Lead Company crystal
No. /cm' sec m No. /cmo

700 c 0.011 0.14 2.73 X10to 1.29 X102o 0.063 37 7.60 X 10&o
with BE1(&AE~ and ni'(&n2'. This suggests two sources 800 c 0.0 0, 12 2.90 X10&7 1.44 X 10' 0.055 43 5.16 X10
850 c 0.018 0.20 1.82 X 10 5.91 X 10 0.011 125 2 60 X102o
of conduction electrons, one of which has a very small 900 c 0.018 0.19 7.95 X10 1.01 X 10&& 0.010
0.033
133 1.18 X10&o
650 a 0.0094 0.21 1.06 X10 4. 18 X10&o 60 1 20 X10uo
activation energy so that it was considerably ionized 750 a 0.0043 0.25 5.41 X10 8.41 X10 0.033 60 9.85 X10
even at liquid air temperature. For most of the samples Linde Air Products Company crystal
satisfying this relation, the values of n, were large 650 c 0.0091 0. 11 6.39 X10'o 3.43 X10 0.077 34 8.42 X10
700 c 0.0 0.12 4.50 X10 1.30 X10 0.094 29 5.00 X 10'o
enough to indicate a degenerate behavior as calculated 800 c 0.0044 0.16 4.42 X10 1.47 X10 0.016
0.050
46 1.39 X 10&o
2.02 X10'o
650 a 0.0065 0.18 8.62 X101o 2.83 X 10'o 46
from the well-known relation between n, and the 700 a 0.0062 0.17 8.38 X10''r 2.59 X20 0.044 49 5.80 X 10'o
800 a 0.010 0.15 6.87 X10 1.29 X10 0.0084 114 2.33 X102o
degeneracy temperature,
National Bureau of Standards ceramic
600-10 0.0056 0 14 1 13 X1017 8 37 X10 0.023 77 4.21 X102o
700-10 0.0078 0.14 7.66 X10'& 4.04 X10' 0.046 48 2.35 X102o
800-10 0.0084 0.12 2.12 X10 1.58 X102o 0.033 60 1 32 X102o
900-10 0.0069 0.16 5.07 X10'o 1.83 X10 0.021 81 2.94 X102o
8005 0.0037 0. 14 2. 19 X10'o 2.86 X10M 0.038 55 2. 15 X10oo
. 800-20 0.0053 0.16 2, 94 X10 4.64 X102o 0.030 64 3.13 X102o
if the eRective electron mass m* is taken to be the free
electron mass; hence the validity of the Boltzmann "H. Frohlich F. Mott, Proc. Roy. Soc. (I ondon)
and N. A171,
expression was surprising. However, observations on 496 (1939).
the electron mobility in rutile to be described later in Frohlich, Pelzer, and Zineau, Phil. Nag. 41, 221 (1950).
 ELECTRI CAL PROPERTIES OF TiOg SEMICONDUCTORS
they may be assumed equal without serious error. %e IOO,

thus may use the experimentally determined values of /

Piattice
A to 6nd the effective electron mass if a value of 8 is e-670
P Impurity
available. Ng R.lsxlp A =0.0%
The choice of 8 may be made in several ways. Studies
of the speci6c heat of rutile" have indicated that an
Einstein function with a characteristic 8 of 670'K will
describe the heat capacity measurements in the tem-
perature range of interest here.
An independent estimate of 0 may be obtained from,
the infrared studies previously mentioned. ' In ref- " CJ
erence 19 it was shown that 4j
V7

0 I,P

where ) ~ is the longest wavelength of the optical absorp-

tion in the infrared. Using the band at 180 cm ', 8 is
found to be 936'K and 8, is 1161'K. It was found that
somewhat better agreement at higher temperatures
could be obtained for almost all samples using 0= 670'K
O.I
instead of the larger 8's calculated from Kq. (11), so
this value was adopted.
A plot of logL(p/A;)+1j as a function of 1/T for a
6 8 IO 12
number of samples is shown in I'ig. 12. The values of A IiT'K x IO
for the various samples were chosen to give the best
FIG. 13. Impurity and lattice scattering of electrons in
agreement at temperatures slightly above room tem- reduced rutile ceramics.
. perature. As will be shown, the deviations observed at
low temperatures can be explained as an impurity differing from those calculated from Kq. (11), should be
scattering contribution. The most pronounced devia- used in the two directions for the single crystal results,
tions at high temperatures are observed for the crystal while in the ceramic an average value would be antici-
samples. This is probably an effect of the crystal pated. Such an average presumably is obtained from the
anisotropy. Diferent values of 8, although possibly heat capacity measurements.
Of particular interest are the e6ective electron masses
Sa mole Dii. TemrdNin. A calculated from the empirical values of A in Kq. (10).
o Mat. Lead C 700 5 0.070
II II
A 650 5 030 . Using 8=670'K and averaged values for E,' and ', E
C 700 5 094 . one Ands values of m* to be 30 to about 100 times the
j
A Linde
II
+ Ceramic
A 700 5
QOO 10
044
02 I free electron mass. These results are given in Table I.
!000 x st
600 .023
-t)
IO
Scattering of the electrons by acoustic vibrations is
P A(e~
8~ 670 apparently negligible in these materials, since the data
cannot be satisfactorily described by a-T ' term which
is predicted for this eGect. 22
At low temperatures the eGects of scattering by
IOO ionized impurities becomes important. Conwell and
%eisskopf" give for this scattering
1 m' e'Nlm*'* (3K,'kT~ '
ln 1+(
. (12)
Ijr 2' E "(kT)'* ( e'NI*
l

lo
Using the appropriate value of m* from the lattice scat-
tering term and a value of 83 for E,', it is possible to
fit observed mobility values at the lowest temperature,
where the contribution of the lattice scattering is
negligible, with reasonable values of Ez, the number of
scattering centers/cm' (Table I). It is usually assumed"
I l

0 2 4 6 8 to
I/Tt X iO
that the total mobility may be given as
1/p=1/pl. +1/pl.
I

FIG. 12. Electron mobilities in reduced Ti02 samples. (13)

~E.g. ,W. Shockley, Holes and Electrons in Semicondlctors
"H. J. McDonald and H. Seltz, J. Am. Chem. Soc. 61, 2405 (D. Van Nostrand Company, Inc. , New York, 1950).
(1939). ~ E. Conwell and V. F. Weisskopf, Phys. Rev. 77, 388 (1950).
800 R. G. BRECKENRIDGE AND %. R. . HOSLER
A typical result of such a calculation is shown in Fig. 13 oxygen vacancies. There may in addition be present in
for the 800' five-minute ceramic sample, with quite the oxide true foreign impurities that will behave
satisfactory agreement with observations over the similarly. " 0
entire range of temperatures studied. The values of I There are a number of reasons for believing that in
given in Table I vary roughly according to the degree Ti02 the donor centers are not interstitial titanium
of reduction, which suggests that ionized donor centers atoms. Ehrlich" has shown that the rutile lattice is
are the chief source of the low-temperature scattering preserved quite precisely for about 1 percent reduction,
although there may be a considerable contribution and no marked changes were noted until the material
from the foreign ions known to be present. was about 10 percent reduced. This-would seem unlikely
An interesting feature of the mobility observations for an excess of an interstitial atom of radius ca 1.4A
is the similarity in the mobility values found between but is quite reasonable for ions as similar in size as Ti+'
the ceramic samples and the single crystals. In previous and Ti+' located on a normal site. The radii quoted by
work on oxide semiconductors in the sintered form, it Pauling'~ are 0.69A and 0.68A, respectively.
has usually been assumed that the low values of p There is also a considerable amount of thermody-
resulted from trapping of the charge carriers at grain namic evidence that the product of the oxide decom-
boundaries. '4 From these measurements it would position is not the free metal. It has been shown by
appear, at least in a very densely 6red rutile sample, Richardson and Jeffes" that the free energy of disso-
that the grain boundaries are not the predominant ciation of most simple oxide systems can be represented
eAect, although this need not be true if the grains are with an accuracy of a few percent over wide temperature
in poor contact as in a pressed powder sample, or even ranges by simple expressions of the form
in a lightly sintered ceramic. Similarly, frequency TAS',
AG'= AH'
effects such as noted in reference 24 can still be attrib-
uted to the presence of grain boundaries and inhomo- with AH and AS independent of temperature. The
geneous grains. actual values of AH' and AS' are, however, less accurate
These results show that there is a diGerence in elec- than those obtained from more detailed formulas. These
tron mobilities in the two directions studied as would be authors give a value of 66' for the chemical reaction
predicted by Eq. (9) assuming equal effective masses
for the two directions and previously given values of E' Ti02 Ti+Og, (15)
and 8; the calculated value of the ratio p, /p, being 1.8
nE= 217 500+41.4T,
AG'= RT 1 (16)
at 500'K. This is illustrated for pairs of samples having
about the same numbers of charge carriers and pre- valid from 298'K to 2080', IC with an accuracy of &10
sumably nearly equal numbers of scattering centers, percent. Since for this system E=I'02, this relation
for example, the 800' Linde samples, where the mobility gives a calculated oxygen pressure even at 2000'K of
ratio y, /p has an average value of 1.8 and varies only only 1.9&(10 "atmos, indicating a very small degree of
slowly with temperature. Similar values are found for dissociation. We can obtain a similar expression for the
the National Lead Company crystals; the 700 c and free energy of the reaction
650 a samples give an average ratio of 2.0 but with a
greater temperature variation. The slightly greater 2 Tio~ Ti203+-', Ou, (17)
values of the ratio found experimentally presumably using the data of Nasu" for the equilibrium constant
arise from the larger effective mass found in the a of the reaction
direction.
VIII. DISCUSSION 2 Tiog+H, Tip03+Hgo. (18)
%hile it is possible to give a rather detailed picture of He found for this reaction
the electrical properties of the elementary semicon-
ductors in terms of a simple band picture involving logipE= 1.9305 2754/T, (19)
filled and conduction bands with one donor or acceptor
"
impurity level, or possibly both, such a simple model
valid for temperatures of 1022'K to 1282'K. Expressing
this equation in terms of free energy, we have
cannot explain the variety of observations on rutile.
The conductivity at high temperature of e type oxide AG'= 12 600
'8.$3T. (20)
semiconductors is believed to arise from the thermal
Combining Kq. (20) with an expression for the free
decomposition of the oxide resulting in the introduction
of donor "impurities" in the form of an excess of the 25 E.
g. , E. J. W. Verwey, in Semiconducting Materials (Butter-
metallic constituents of the oxide. This excess may be worths Scientific Publications, London, 1951).
2' P. Ehrlich, Z. Elektrochem. 45, 362 (1939).
present as interstitial metal atoms or as metal ions of a 27 L.
Pauling, Nature of the Chemical Bond (Cornell University
diferent valence in the normal lattice accompanied by Press, Ithaca, 1940).
28 F. D. Richardson and
J. H. E. Je6es, J. Iron and Steel Inst.
24
E.g. , P. H. Miller, in Semiconducting materials (Butterworths (London) 160, 261 (1948).
Scientific Publications, London, 1951).
- ~ N. Nasu, . Sci. Repts. Tohoku Imp. Univ. 25, 510 (1936).
 ELECTR I CAL P ROPE RTI ES OF Ti02 SE M I CON DU CTORS
energy of the dissociation of water, " using the 1s hydrogen-like wave functions including Z
AG'= 118 000+26.75K, and E', the exchange integral E is found to be
(21)
E= (Z'e'/aii*) e &(p+1), (24)
we obtain for the free energy of the reaction (17)
EGo= 71 600+22.20T where p=Zr~e/aii*. For the value of rye indicated, E
(22) is computed to be
1.6&(10 '~ erg, corresponding to
Since for the reaction (17) E~=I'o2r, this indicates a a resonant frequency of jump between nuclei of 4.8&(10'
partial pressure of oxygen at 2000'K of 6.9)&10 ' sec '. A similar calculation for the exchange of an
atmos, hence a degree of dissociation sufhcient to electron between two singly ionized centers with the
account for the observed conductivity of rutile at same separation leads to an even greater frequency of
elevated temperatures. Thus it seems reasonable to jump and larger exchange energy.
assume that the source of electrons in semiconducting It seems, then, that the wave functions overlap suf-
rutile is Ti+' ions in the lattice ficiently to produce a narrow impurity conduction
band, and under these conditions a large e6ective mass
To preserve electrical neutrality in this situation
there must be present oxygen defects; and, since these shoul. d be observed. "
An estimate of the e6ective mass
defects can act as electron traps, it would be anticipated
that the Ti+' ions would be on sites adjacent to the
oxygen vacancy. The oxygen vacancy has an effective
charge of +2, thus it is possible to have present in the
crystal neutral centers (0.2 Ti+')', singly ionized
centers (OTi+')+, or free vacancies (O)~. Since two ~ Ov++

types of donor centers were found in the conductivity

studies, it seems reasonable to assume that the centers ~ Low ri+']
are those suggested by the structural possibilities, the [ov 2Ti+a]
AE & 0.0l ev
neutral centers being those with the small activation

energy for ionization AE& 0.01 ev and the singly
ionized centers those with the larger energy AEG 0.2
ev. In addition to the large number of centers produced hE 5.67ev hE*5.05ev
by the reduction, there may be present a considerable
number of free oxygen vacancies, probably of the order
of 10'r/cm', produced along with a corresponding
number of vacant Ti+4 sites as normal Schottky lattice
defects either in the firing of the ceramic or in the growth
0
of the single crystal by the Verneuil furnace method.
There is an interesting consequence of this high
density of impurity centers in a medium of remarkably Fio. 14. Thermal equilibrium energy-level diagram of
semiconducting rutile.
high dielectric constant, EA= 114. Because of this high
dielectric constant an electron is very loosely bound to
may be made from the value of E since it was shown in
the center and has a considerable probability of jump
the reference cited for a cubic lattice
to a neighboring vacancy. . A rough estimate of the
exchange frequency can be made based on a simple P m*/m= 5'/2mEa', (25)
calculation analogous to that given by Pauling and
Wilson" for the hydrogen molecule ion H&+. Our model where a is the edge length of the crystal unit, here pre-
effective
consists effectively of two helium nuclei embedded in a sumably the impurity center separation, since we have
continuous, homogeneous dielectric with constant already assumed an average, cubic distribution of the
E'=114, and a single electron shared between the two vacancies. This estimate, based on obviously approxi-
nuclei. If we assume a density of O, ~ of the order of
mate values of u and E, gives a value of P of 8.2 which
is at least reasonable since the values of P found range
10"/cm', the mean separation r~~ between the vacant
from 27 to 122. It is qualitatively evident that higher
sites will be 215A. The radius of the 6rst Bohr orbit for
mobilities and smaller masses should be
the electron on one of the nuclei is given as
observed in the c compared to the a direction since the
Gii = E Ge/Z~ (23) dielectric constant, and hence the extent of exchange
interaction, is larger in the c direction.
where Z is the nuclear charge, (= 2) and aii is the normal
A proposed energy-level diagram which describes
Bohr orbit. For our situation ao~ is found to be 30.2A. many observations on rutile is shown schematically in
Following the variation treatment outlined by Pauling Fig. 14. The electronic states indicated correspond to a
~ L. Pauling and E. B. Wilson, Introduction to Quantum "F. Seitz, Modern Theory of Solids (McGraw-Hill Book Com-
Meehoriks (McGraw-Hill Book Company, Inc. , New York, 1935). pany, Inc. , New York, 1940), p. 303.
802 R. G. 8RECKENRI D GE AND W. R. HOSLER

state of thermochemical equilibrium of the system. The Observations by Dr. R. F. Blunt of this laboratory
diagram provides a ready explanation for the change in on the bleaching of the colored rutile on cooling to liquid
the activation energy for conduction observed by air temperatures suggest that the band at 1.2p (1.0 ev)
Cronemeyer. " The very high-temperature branch may arises from the process
now be assigned to a true intrinsic conduction process
hv
that can take place without decomposition of the oxide;
(P . T;+3)+~ P+++, (30)
the lower branch is that due to the oxide decomposition
producing oxygen vacancies and Ti+' ions. There is This work will be described in detail elsewhere. We
some thermodynamic evidence that justifies this assign- have observed the thermal activation energy assigned
ment. For our purpose Eq. (17) may be written, in-
to this process to be 0.2 ev, i.e., almost exactly in the
cluding the ionization of the Ti+"s, as predicted ratio. This would suggest that the 0.41'
0 '-', 02+0+++2c. (26) (3.02-ev) absorption corresponds to an excitation of an
Then electron from (0.2 Ti+')' or (OTi+')+ centers to the
E=Po2'Xo++X ' (27) conduction band since these are presumably about 0.62
ev apart. This assignment provides an explanation for
At moderate temperatures (&~900'C) the number of observations of Cronemeyer on the temperature de-
0,++ formed by the reaction is small compared to the pendence of the photo response. It was found that the
constant number presumed present, i.e., Xo,++ 10 '; photoconductivity showed a shift of the maximum
hence, response to higher energies as the temperature was
X, = (E/Xo.++) '*Po& lowered; and, , at 103'K, a splitting into two maxima.
We may use the free energy expression (22) to obtain The suggested model would predict a similar eRect
the value of Eand find on converting to e the number since at room temperatures the majority of the centers
per cubic centimeter, present would be singly ionized, and thus the photo-
conductivity maximum would be centered at the energy
n, /cm'=8. 68&(10"(Xo++)V'o2 'e '"~'"r (29) corresponding to the separation between the singly
predicting a thermal activation energy in excellent ionized centers and the conduction band. As the tem-
agreement with that observed by Cronemeyer" as well perature is lowered, the number of neutral centers
as a pressure dependence in good agreement with that would be increased at the -expense of the singly ionized
found by Earle. ' centers, shifting the maximum photoresponse to greater
The correlation of the various optical observations energies corresponding to the energy diRerence between
with the thermal effects is dificult in rutile because of the neutral centers and the conduction band.
the great diRerence between the optical and static die- If the energy for the optical transition corresponding
lectric constants. Mott and Gurney" have indicated to the 3.67-ev thermal process is computed using the
that the optical activation energies are proportional to same ratio of 5:1, the absorption for the intrinsic
1/E', while the thermal activation energies should be process should occur at about 0.067' (18.7 ev), in the
proportional to 1/E, '. This would predict for rutile normal region for absorption in the transparent oxides. "
that hE, ~q;~/AEu, ~, ~= 5. 1 as an average value. However, no crystals have shown transmission at wave-
While Cronemeyer observed photoconductivity related
lengths shorter than 0.41p, , indicating that the "im-
to the absorption at 0.41+ (3.02 ev), it seems unlikely
purity" absorption may overlap the fundamental
that it can be due to the excitation of an electron from
the filled to the normal conduction band since this absorption. This is very likely in the present situation

would require that AE, ~&,& AE&h, and would leave
&
since the absorption band presumably arises from transi-
no process assignable to the 3.67-ev thermal activation tions directly to the conduction band. It would seem
energy. that further exploration of the absorption' in the
32 N. F. Mott and R. W.
Gurney, Electronic Processes in Ionic
ultraviolet, particularly at very low temperatures,
Crystals (Clarendon Press, Oxford, 1940). might be of considerable interest.