4370 JANAK, WI LLIAMS, AND MOB U Z ZI

Phys. Rev. Letters 11, 541 (1963). Rev. B 1, 1318 (1970).

~
W. Budge, Phys. Rev. 181, 1024 (1969). 24F. Herman and K. Schwartz, in Ref. 19, p. 245.
D. Koelling, A. Freeman, and F. Mueller, Phys. A. M. Boring, Phys. Rev. B 3, 3093 (1971).

PHYSICA L RE VIEW B VOLUME 6, NUMBER 12 15 DE CEMBER 1972

Optical Constants of the Noble Metals

P. B. Johnson and R. W. Christy
Department of Physics and Astronomy, Dartmouth &ollege, Hanover, Ne'er Hampshire 03755
(Received 7 July 1972)
The optical constants n and k were obtained for the noble metals (copper, silver, and gold)
from reflection and transmission measurements on vacuum-evaporated thin films at room
temperature, in the spectral range 0. 5 — 6. 5 eV. The film-thickness range was 185-500 L.
Three optical measurements were 0 inverted to obtain the film thickness d as well as g and k.
The estimated error in d was +2 A, and that in yg, k was less than 0. 02 over most of the spec-
tral range. The results in the film-thickness range 250-500 A were independent of thickness,
and were unchanged after vacuum annealing or aging in air. The free-electron optical effec-
tive masses and relaxation times derived from the results in the near infrared agree satis-
factorily with previous values. The interband contribution to the imaginary part of the dielec-
tric constant was obtained by subtracting the free-electron contribution. Some recent theoret-
ical calculations are compared with the results for copper and gold. In addition, some other
recent experiments are critically compared with our results.

I. INTRODUCTION of the optical constants, however, differ signifi-

cantly as reported by different investigators, even
The optical constants of the noble metals have in the most recent and careful experiments. Since
been measured since the time of Drude, with con- the actual value of the optical constants is now
tinuing attempts to increase the experimental ac- yielding to theoretical calculation, it is important
curacy, for the purpose of comparison with the- to pin down the xperimental values with accurate
ory (and also for their practical application as measurements including estimates of error. Our
mirrors). It was seen from the earliest results purpose is to do so for the three noble metals,
that Drude's free-electron theory failed in the copper, silver, and gold, under similar conditions
visible and near-ultraviolet regions. After quan- for each metal, and to compare the results with
tum theory, it was recognized' that the absorption recent calculations and also with other experimen-
in the visible and ultraviolet probably was due to tal results.
transitions from the filled d bands into the sP con- The earliest measurements used Drude's polari-
duction bands. With more accurate measurements metric method. This has the advantage that ex-
of the optical constants and detailed band-struc- plicit expressions for the optical constants n, k
ture calculations, one could separate the free- can be derived in terms of measured quantities,
electron contribution from the effect of interband but the disadvantage that the results obtained de-
transitions, and try to identify the photon energies pend very sensitively on the surface preparation
at which structure occurs in the measured inter- of the sample. The advent of more powerful com-
band absorption with energy gaps in the theoretical puting techniques saw increased use of other ex-
band structure. This work confirms the im- perimental methods which depend less sensitively
portance of transitions from the d band, and is on surface conditions, but which require numerical
supportedby observations on the effects of stress' inversion of the equations for the optical constants.
and alloying. ~ Most recently, for copper and
goldnew measurements' ' havebeencompared ' " These include Kramers-Kronig analysis of normal-
incidence reflection measurements, reflection mea-
with quantitative calculations of the interband op- surements at different incident angles, and trans-
tical constants based on refined band-structure mission measurements through thin films. Each
calculations, taking into account the joint density of these has its own advantages and weaknesses.
of states and the transition-probability matrix Several investigations of copper, silver, and
elements throughout the Brillouin zone. gold have used the Kramers-Kronig technique
In the experimental measurements there has applied to normal-incidence reflection measure-
been reasonably good agreement on the wavelengths ments. A problem with this method arises from
at which structure occurs. The actual magnitudes the necessity of extrapolating the reflectance to
 OP TICAL CONSTANTS OF THE NOBLE ME TALS 4371

frequencies outside of the measured range. The from some other recent measurements. We will
choice of the extrapolation has a large effect on the compare our experimental data with these other
magnitude of the optical constants, although the recent experiments and suggest why we believe

tures occur are insensitive to it. '

photon energies at which various structural fea-
' Some other
investigations have used two reflection or trans-
our results are more accurate.
II. EVALUATION OF OPTICAL CONSTANTS

mission measurements. Two reflection measure- The optical constants n and k, as well as the
ments from opaque samples, at different angles of thickness d of thin metal films, were determined
incidence or with different polarizations, yield by inverting the normal-incidence reflection and
values only over a relatively small range of the transmission functions, Ro and To, simultaneously
spectrum. The reflection contours in the n-k with the transmission function for P-polarized
plane may be nearly parallel and thus give indef- light incident at 60 Tgp This method has re-
inite values of n and k. 0 Methods using normal- cently been described in detail and evaluated by
incidence transmission and reflection measure- Nestell and Christy. ~ '~4 One of the chief advan-
'
ments on thin films can also encounter this prob- tages of this method, a three para-meter fit, com-
lem, and may have additional trouble with accurate pared with other thin-film techniques which also
measurement of the thickness of the film. ~0 invert reflection or transmission functions, ' is
The method we adopted, of utilizing three mea- the good accuracy with which the film thickness
surements on thin films, can circumvent the above can be determined. The accuracy of interfero-
problems. Measurements of reflection and trans- metric thickness measurements is normally about
mission at normal incidence, together with p-pol- + 30 A, but if our three optical measurements are
arized transmission at an angle of 60', yielded inverted in an appropriate spectral range, then
accurate values for the two optical constants as the thickness can usually be obtained to within
well as accurately determining the film thickness. + 2 A. The appropriate range is one in which the
This technique is particularly convenient when a measured reflection and transmission values lie on
spectrophotometer is available and one is dealing contours which intersect steeply enough in the n-k
with such easily evaporated materials as the noble plane. Figure 1 shows how the three contours in-
metals. tersect for a suitable wavelength. The width of
In addition to the accuracy of the chosen method the contours represents the error in the measure-
of measurement, the preparation of the sample is ments. At wavelengths where n or k takes on val-
very important. Bulk samples are usually polished ues larger than 2, however, the two transmission
mechanically, chemically, or electrolytically, and contours often are nearly parallel, and the accura-
then they may be annealed. Mechanical polishing cy of the method is destroyed. Once the true film
may leave the surface structure distorted, and thickness has been determined from the average
annealing may reduce surface smoothness. ' Thin
films prepared by vacuum evaporation require no
polishing. Rapidly evaporated films will reproduce
the surface smoothness of the substrate and form
4.86 eV Ag
a smooth homogeneous surface. Films formed
at slower evaporation rates can have a discontinu- cj =304 A
ous agglomerated structure. 22 The optical prop-
erties of evaporated thin films have been found to
be the same as for bulk materials, provided the
thickness of the films is greater than about 200-
300 A.
We report values of the optical constants for
copper, silver, and gold in the spectral range
0. 5-6. 5 eV, measured with an oblique-incidence
thin-film technique. We make estimates of the
experimental error inherent in our measurements,
and also investigate the effect of variations in our 0
sample preparation. Our results will be compared
with the classical Drude theory in the free-electron
FIG. 1. Example of the intersection in the yg-0 plane
region, and more especially with some recent cal- of normal-incidence R and T contours with the p-polar-
culations of interband absorption for copper and ized T contour for 60' incident angle. The width of the
gold. Our results in the free-electron region contours, which represents the estimated experimental
agree satisfactorily with earlier experimental ~ .
"ror, is exaggerated by a factor of 5 for the normal-
work, but our interband results differ significantly incidence contours.
 P. B. JOHNSON AND R. W. C HRIS TY

of the values in the appropriate spectral range, since it is here that free-electron effects play a
it can be used along with any two of our three op- dominant role.
tical measurements in order to determine n and k Our three-parameter method gives us an ac-
over our entire spectral range. The normal-in- curate measurement of the film thickness, and
cidence reflection and transmission functions were immediately picks out the physical solution. Then,
usually used for this calculation; however, the using the known thickness along with the known
60' P-polarized transmission function was some- physical solution at certain wavelengths, we can
times combined with one of the other two measure- determine n and k throughout the entire spectrum
ments in order to improve or check on the e and using only two of our three measurements. Be-
k values. cause our theoretical reflection and transmission
Computer methods for inverting the complex re- functions2P refer to a semi-infinite substrate me-
flection or transmission functions have been sug- dium, the data must be corrected for reflection at
gested previously, with application to the inversion the back surface of the actual substrate. The por-
of two measurements to determine yg and k. tion of this reflection which entered the detector
The three-parameter method we used is based on in the R, measurement was small enough to be
an approximate Newton-Raphson technique of find- neglected. The correction in the T measurements
ing the intersections of the contours. It converges is approximately T= (T~ cos8, /Ta cos83)x (1 —Ra
rapidly and accurately. Other methods of invert- RaR+-T&), where subscript I refers to the film-
ing three measurements have been suggested, but coated substrate and B to the blank substrate.
they do not seem to have been widely used. For T«, R~ =0; for Tp the correction is still
Our inversion programs are actually quite general only a few percent. The refractive index of the
and can be used to determine n, k, and d from substrate was taken from the literature. '7 All of
any three transmission or reflection functions. the data reduction was done on the Dartmouth
We used Tp Rp and T6p since they are easiest time-sharing computer system.
for us to measure, and their intersections in the
III. EXPERIMENTAL CONSIDERATIONS
n-k plane are relatively steep.
Another important advantage of using three mea- All of our samples were deposited onto fused-
surements is that they automatically determine the quartz substrates at room temperature by vacuum
Physical solution. It is known that the reflection evaporation. The evaporations were perf orrned
and transmission contours intersect at several at pressures below 4& 10 6 Torr in an oil-pumped
points in the n-k plane, and an example of multiple bell-jar system. Before evaporation the substrates
intersections can be seen in Fig. 1. The problem were cleaned ultrasonically in Lakeseal Laboratory
of determining which is the physical solution may Glass Cleaner and then ultrasonically rinsed in
be difficult, and at times the wrong intersections distilled water. The substrates were dried with
have been reported. A rule of thumb that the a hot-air gun and placed immediately in the vacuum
solution ' with the lowest value of n should be system. The source materials deposited by
chosen seems to give the correct solution for the evaporation were 99. 99% pure metals. The gold
noble metals. Vrba ' points out that by changing was evaporated from a conical tungsten basket,
the film thickness the position of the physical-so- while the silver and copper were evaporated from
lution intersection will not change, while the other boats made of tantalum foil. In all cases the sub-
intersections will move to new locations, but this strate was masked until the source was at evapora-
approach requires different samples and several tion temperature, and the evaporation rate was
computations. Our three-parameter fit immediate- about 60 A/sec. This high rate of evaporation
ly picks out the correct solution since the three was necessary to ensure the surface smoothness
measurements Tp, Rp, and T6p always seem to of the films. The effects of evaporation rates
intersect at a unique point in z-k-d space which will be discussed in more detail later. Several
is the physical solution. of the gold films were annealed under a vacuum
Inversion techniques have also been applied to or a nitrogen atmosphere at 150'C for 9 —12 h.
two reflection measurements taken from bulk The effects of this annealing will also be discussed
samples or opaque evaporated films. Although in Sec. IV.
there is no thickness problem, the reflection con- Reflection and transmission measurements on
tours often do not intersect very steeply except in the films were made at room temperature with a
certain restricted areas of the n-k plane. Miller, Unicam SP700 double-beam recording spectropho-
Taylor, and Julien have determined the optimum tometer in the spectral range 0. 5-6. 5 eV. The
angle for reflection to be R and R' at 74 . This polarization for the transmission measurement at
method is accurate only when n& 3. 0 and k& 3. 2. 60' was provided by a pair of Gian-prism polariz-
Such a restriction rules out most of the infrared ers. In order to ensure identical beam geometry
region of the spectrum. This region is of interest and polarization, one polarizer was placed in the
 OP TICAL CONSTANTS OF THE NOBLE ME TALS 4373

values are shomn in Table I. We mill present

6— what f ollows in terms of & = &, + i &2, however,
where e = n .
The advantage of first considering
n and k is that the intersections of reflection and
transmission contours in the n-k plane are mag-
nified in the interesting region near the origin. 20
On the other hand, the dielectric constants E, and
E2 are closely related to the electronic structure
of the solids and are more directly comparable
0 with theory.
A. Copper

Our dielectric constants for copper are pre-

sented in Fig. 2. The curves represent the aver-
age of values from two films of thicknesses 29V
and 305 A, respectively. The agreement of the
two experiments is well within the estimated error,
shown by the width of the curves, mhich was cal-
I
culated by assuming that the accuracy of our trans-
2 3 4 5 6 mission and reflection measurements was + 0. 5%%uo

PHOTON ENERGY (eV) of full scale. The scales used were 0-10%, 0-20%%uo,
and 0-100%%ug.
FIG. 2. Dielectric constants for copper as a function
of photon energy. The width of the curves is represen- It has been shown that below a certain critical
tative of the instrumenta1 error. film thickness, which varies somewhat with the
material, the inferred dielectric constants depend
on thickness. In order to verify that our values
sample compartment and the other in the reference for e, and E~ are independent of thickness, me ran
compartment. Normal-incidence reflectance was another experiment on a 486-A-thick copper film.
obtained from tmo reflections off two identical These results agree within the error mith the val-
(simultaneously evaporated) samples. The light ues obtained from the thinner films. The thinner-
beam in the sample compartment mas reflected film values are more accurate, since the trans-
at near-normal incidence (about 12') off the first mission of the thick film was less than 1% at long
sample, and then again reflected at near-normal waveleng ths.
incidence off the second sample, displaced a fern
B. Silver
millimeters laterally, into the detector. Since
this arrangement lengthened the beam path by In Fig. 3 are shown plots of our complex di-
about a centimeter and tl)e beam is divergent, the electric constants for silver, determined from
loss of light into the detector was calibrated by films of 304- and 3V5-A thickness. The results
measuring freshly evaporated aluminum mirrors agreed within the experimental error, determined
whose reflectance is well. known. 3~ The correction as above, again shomi~ that &, and e2 are not
was about G%%uqin the reflectance, only slightly de- thickness dependent in this range.
pendent on mavelergth. For small angles, near-
C. Gold
normal-incidence unpolarized reflection measure-
rnents are effectively the same as true-normal-in- Our values for gold are presented in Fig. 4.
cidence measurements. The accuracy of the The curves represent the average of two experi-
spectrophotometer is about + 0. 005 for measure- ments on films of thickness 343 and 456 A. Again
ments above 0. 2 and about + 0. 0005 for measure- the agreement between the two experiments was
ments from 0 to 0. 1. The width of the contours within the experimental-error estimates, demon-
in Fig. 1 displays this error. strating that in this range there is no thickness
The optical measurements were begun within dependence of the inferred values of E~ and &2.
5 min after exposure to air. Any significant oxi- We also investigated the effects of annealing
dation of the films would ha, ve to occur in the first gold films. We picked gold since of the three no-
few minutes, since measurements on the films ble metals it is the least affected by oxidation. A
were reproducible for several days thereafter. complete set of reflection and transmission mea-
surements was made on the 343-A gold film before
IV. RESULTS
annealing. The sample was then annealed in a
All of our data analysis was done in terms of the nitrogen atmosphere at 150 C for 9 h. When the
complex index of refraction n =n+ik. Our average optical measurements were repeated after the
4374 P. B. JOHNSON AND R. W. CHRISTY

anneal, the preannealing results were almost ex- certain critical thickness, which for gold is about
actly reproduced. (The only measurable diHer- 250 A. In order to check this limit, we made a
ence occurred in the photon energy range 4. 5-5. 5 thinner gold film of I86-A thickness. The initial
eV, where the normal-incidence transmission mea- optical measurements on this film as evaporated
sur ement had decreased by about twice the esti- failed to converge to any values of n and k in the
mated error. ) visible or ultraviolet. The observed reflection
As mentioned, the inferred dielectric constants- and transmission contours did not intersect in the
are independent of film thickness only above a n-k plane, presumably because the film was not

TABLE I. Optical constants for copper, silver, and gold as well as the approximate errors in n and k.
Copper Silver Gold Error
eV
0. 64 l. 09 13.43 0. 24 14. 08 0. 92 13.78 +0. 18 +0. 65
0. 77 0. 76 11.12 0. 15 11.85 0, 56 11.21 +0. 08 +0. 30
0, 89 0, 60 9. 439 0, 13 10.10 0. 43 9. 519 +0, 06 +0. 17
1. 02 0. 48 8. 245 0. 09 8. 828 0. 35 8. 145 +0. 04 +0. 10
1. 14 0. 36 V. 217 0. 04 7. 795 0. 27 7. 150 +0. 03 +0. 07
1. 26 0. 32 6. 421 0. 04 6. 992 0. 22 6. 350 +0. 02 +0. 05
1. 39 0. 30 5. 768 0. 04 6. 312 0. 17 5. 663 +0. 02 +0. 03
1, 51 0. 26 5. 180 0. 04 5. 727 0. 16 5. 083 +0. 02 + 0. 025
1.64 0. 24 4. 665 0. 03 5. 242 0. 14 4. 542 +0. 02 +0, 015
1. 76 0. 21 4, 205 0. 04 4. 838 Q. 13 4. 103 +0. 02 +0, 010
1. 88 0. 22 3. 747 0. 05 4. 483 0. 14 3. 697 +0, 02 + 0. 007
2. 01 0. 30 3. 205 0. 06 4, 152 0. 21 3. 272 +0. 02 +0. 007
2. 13 0. 70 2. 704 0. 05 3. 858 0. 29 2. 863 +0, 02 +0. 007
2. 26 1. 02 2. 577 0. 06 3. 586 0. 43 2. 455 +0. 02 +0, 007
2. 38 l. 18 2. 608 0. 05 3. 324 0. 62 2. 081 +0. 02 + 0. 007
2. 50 1.22 2. 564 0. 05 3. 093 1.04 1. 833 +0. 02 +0. 007
2. 63 l. 25 2. 483 0. 05 2. 869 1.31 . 849
1. +0, 02 +0. 007
2. 75 1.24 2. 897 0. 04 2. 657 1.38 1. 914 +0. 02 + 0. 007
2. 88 1.25 2. 305 0. 04 2. 462 1.45 l. 948 +0. 02 *0.007
3. 00 1.28 2. 207 0. 05 2. 275 1.46 1. 958 +0. 02 +0. 007
3. 12 1.32 2. 116 0. 05 2. 070 1.47 1.952 +0, 02 + 0. 007
3. 25 l. 33 2. 045 0. 05 1, 864 1.46 1.933 +0. 02 +0. 007
3. 37 l. 36 1.975 0. 07. 1.657 1.48 l. 895 +0, 02 +0, 007
3. 50 1~ 37 l. 916 0. 10 1.419 1.50 l. 866 +0, 02 + 0. 007
3. 62 1.36 1. 864 0. 14 1.142 1.48 l. 871 +0. 02 + 0. 007
3. 74 1.34 1. 821 Q. 17 0. 829 l. 48 1. 883 +0, 02 +0, 007
3. 87 1.38 1.783 0. 81 0. 392 1. 54 1. 898 +0. 02 + G. 007
3. 99 1.38 1.729 1. 13 0. 616 1.53 . l. 893 +0. 02 +0. 007
4. 12 1, 40 1.679 1.34 0. 964 1.53 1.889 +0. 02 + 0. 007
4. 24 1.42 l. 633 l. 89 1.161 1.49 l. 878 +0, 02 + 0. 007
4. 36 1.45 1, 633 l. 41 1.264 1.47 1. 869 +0. 02 +0, 007
4. 49 1.46 l. 646 1.41 1.331 1.43 1. 847 +0. 02 +0. 007
4. 61 1.45 1.668 1.38 1.372 1.38 l. 803 +0, 02 +0, 007
4. 74 1.41 1.691 1.35 1.387 l. 35 l. 749 +0. 02 +0. 007
4. 86 1.41 1.741 ly 38 l. 393 1.33 1.688 +0. 02 +0. 007
4. 98 1.37 1.783 1.31 1.389 1.33 1.631 +0, 02 +0, 007
5. 11 1.34 l. 799 l. 30 1.378 1.32 l. 577 +0. 02 +G. 007
5. 23 1.28 l. 802 l. 28 l. 367 1.32 1.536 +0. 02 +0. 007
5. 36 1~ 23 1, 792 1.28 1, 357 1.30 1, 497 +0. 02 + G. 007
5. 48 l. 18 l. 768 1.26 l. 344 1.31 l. 460 +0. 02 +0. 007
5. 60 l. 13 1.737 1.25 1.342 1.30 1.427 +0. 02 +0. 007
5. 73 1. 08 1, 699 l. 22 1, 336 1.30 1, 387 +0. 02 +0. 007
5. 85 l. 04 1, 651 1. 20 l. 325 1.30 1, 350 +0. 02 +0. 007
5. 98 1.01 l. 599 1.18 1.312 1.30 1.304 +0. 02 +G. 007
6. 10 Q. 99 l. 550 l. 15 1.296 1~ 33 1.277 +0. 02 +0. QGV
6. 22 0. 98 1.493 1, 14 1.277 1.33 1. 251 +0. 02 +0. 007
6. 35 Q. 97 l. 440 1. 12 l. 255 1.34 1.226 +0. 02 +0, 007
6. 47 0. 95 1.388 1. 10 1.232 1.32 l. 203 +0. 02 +0. 007
6. 60 Q. 94 1.337 1, 07 l. 212 1.28 1.188 +0, 02 +0. 007
 OP THECAL CONSTANTS OF THE NOBLE ME TALS

Here N is the density of the conduction electrons,

mo is their effective optical mass, and r is the
relaxation time. Separating e into its real and
imaginary parts, we get
2 2
y Gog7
i 1 1+~2~2

2 7'
COq

(d(1+& & )
For metals at near-infrared frequencies one finds
&u» 1/v. Therefore
2 2
Cq=1-
y (dg,
2 = 1- Xp
and

~f
I I
P
1 2 3 5 6
where X~~=Ne~/vmga'and 7''=2vcr We. observe
PHOTON ENERGY (eV)
that the optical mass can be determined from the
FIG. 3. Dielectric constants for silver as a function experimental results for E~&, and v can then be
of photon energy. The width of the curves is represen- found from the results for &2 The free-electron ~

tative of the instrumental error. expression of & is useful only for photon energies
below the threshold energy for the onset of inter-
band transitions. Above this threshold energy the
homogeneous and continuous. After annealing this form of the E2 curve, in particular, depends on
film for 12 h at 150'C, the structure changed the specific band structure of the material.
enough to give convergent solutions for n and k. Using e~ = 1 —& /&~, we can determine mo for
The values for &z (and n) were well outside of the each of the noble metals from the slope of a plot
error estimates for thicker films. Although the
annealing apparently improved the uniformity of
the 186-A film, inferred values of the dielectric
constants are not representative of bulk gold. 6—
Thus bulk values for &, and E2 can only be obtained
from films whose thickness is about 300 g or
more.
V. DISCUSSION
A. Comparison with Theory

1. Free-Electron Region

The electronic transitions in a solid are more

directly related to the complex dielectric constant
a=&, +i&2, instead of the complex index of refrac-
tion n=n+ik. These are connected by E =n2, so
that &~ =n2- k2 and &2= 2nk. The dielectric con-
stant is determined by the relaxation time (or dc
conductivity) and the optical mass of the electrons
according to the Drude free-electron theory,
2
(d =1- (8 ((d + 3
I I

4
I

6
f/T)
PHOTON ENERGY (eV)
where
FIG. 4. Dielectric constants for gold as a function of
4mNe
(dp = photon energy. The width of the curves is representative
SZ 0 of the instrumental error.
4376 P. B. JOHNSON AND R. W. CHRIS TY

200

150

100

Op 0 2 3 4
(pm2)
0 FIG. 6. Imaginary part of the dielectric constants for
copper, silver, and gold, divided by wavelength vs the
(pm ) square of the wavelength.
FIG. 5. Negative of the real part of the dielectric con-
stants for copper, silver, and gold vs the square of the
The zeros in —&~ for silver and gold are
wavelength.
offset by 25 and 50„respectively. copper" and gold. "
structure calculations have recently been made for
Copper is of special in-
terest since band-structure calculations are easier
for it and since the transition matrix elements
of —&~ vs ~ in the infrared. Such plots for cop- have been calculated also. To compare our experi-
per, silver, and gold are seen in Fig. 5. It is mental value of E~ for copper with these recently
assumed that all the noble metals have one con- calculated values, we must subtract the free-elec-
duction electron per atom, corresponding to a full tron value Ez from the experimental result to get
d band with one free s electron. The results for the interband contribution &~. To calculate E~ we
copper, silver, and gold are shown in Table II. use the optical mass, mo= 1.49m, and the relaxa-
These results are relatively accurate since in the tion time, 7 =6. 87&10 "sec, for copper. Figure
infrared k» n and so —e, = k . The percentage 7 is a plot of our values for &~. Also included are
error in k is small compared with that in n.
Next, using the expression e~~= &~jX~ y' and plot-
ting eaj& vs &, we can determine r from the slope
of the graph in the infrared. We have plotted the
results for copper, silver, and gold in Fig. 6.
The results for v are also shown in Table II. The
error in these measurements is larger since the
percentage error in our value of n is large in the
infrared. Results for silver are the least accurate,
with an uncertainty in n of about + 40% of its value.
2. Interbund Absorption

The quantity e~ is most directly comparable

with calculations based on band theory. Band-

TABLE II. Optical masses and the relaxation times for

copper, silver, and gold.

mo (electron masses) ~ (sec) p I I I f

'] 2 3 6
Copper 1.49~0. 06 (6. 9+0. 7) x10" 5
P HOTON ENERGY (eV)
Silver 0. 96 +0, 04 (31+].2) x1O ~5 FIG. V. Our experimental values of &2~ for copper vs
photon energy (JC). Also included are plots of the theo-
'Gold O. 99+0. 04 (9. 3 ~0. 9) x 1O-" retical curves of Williams et al. (W) and Fong et al. (F).
 OPTICAL CONS TANTS OF T HE NOBLE ME TALS
I f i
finer structure may be washed out.
Au A relativistic band-structure calculation for
PS
gold has recently been made by Christensen and
Seraphin. Figure 8 is a plot of our experimental
values of ez and their theoretical curve. Their
theoretical curve is adjusted to agree at about
4. 5 eV, since they did not calculate the transition
matrix elements, but assumed them to be constant.
b The theoretical curve agrees well with our experi-
62
mental curve above 4. 5 eV and below 3 eV. The
4-eV peak in our data is also seen in the theory,
while the 3-eV peak in our data is seen as a
shoulder in the theoretical curve. As was the
case in copper, our data exhibited some fine struc-
ture which, although smaQer than our estimated
error, was seen in our experiments on both of the
thicker gold films. This fine structure consists
of a slight rise in &~ near 3. 5 eV. This rise is
also seen in Christensen and Seraphin's theoretical
calculation. The theoretical curve shown includes
PHOTON ENERGY (eV) transitions only from the two highest d bands. In-
cluding contributions from the lower d bands with
FIG. 8. Our experimental curve of &2 for gold vs
equal matrix elements would spoi1 the agreement
photon energy (JC). Also included are plots of Christen-
sen and Seraphin's theoretical curve (CS) and the experi- with experiment above 4. 5 eV, as was already
mental curves of Theye (Y) and of Pells and Shiga, (PS). pointed out by Christensen and Seraphin.
8. Comparison with Other Experiments

the theoretical curves of Fong et a/. '

and of Wil- Our values for the free-electron optical masses
liams, Janak, and Moruzzi. Our experimental shown in Table II agree with earlier values within
values lie between the two theoretical curves, our experimental error. Values recently sum-
somewhat closer to the results of Fong eI; a/. marized for copper have ranged from (l. 42+ 0. 05)
near about 5 eV. Both theoretical curves exhibit to (],. 45q 0. 06)~, for silver3 from 0. 96 to (1.03
the same general structure as our experimental a 0. 06)m, and for gold" from (0. 94 + 0. 01) to l. 04m.
curve. There are two dominant peaks in the &~
curve, one near 2. 5 eV and the other at about
5 eV. The magnitudes of the lower-energy peak
others ' '"
The values of the relaxation times reported by
show much more scatter. The dis-
crepancies with our values are in some cases be-
agree well; however, the sizes of the peak at the yond our estimated limits of error, but we do not
higher energy differ greatly. Qur results suggest wish to stress the reliability of our values for the
that the calculation of Fong et a/. may underesti- relaxation times.
mate the size of the matrix elements for the higher- A plasma resonance, which occurs if && = 0 and
energy transitions. In addition to these general &z«1 (or n= k and k& 1), is found in silver but not
features, our data also exhibit some finer struc- in copper or gold. ' ' ' In silver, however, it is
ture. Although this structure is comparable with not at the frequency predicted by the free-electron
the estimated experimental error, it appeared in theory (co~), but is shifted to lower frequency be-
the results from all three copper films, and there- cause of the positive contribution to &, of Eg.
fore we believe that it is real. Near 4. 5 eV there McAlister and Stern pointed out that the plasma
is a slight shoulder in our &~ curve. This shoulder resonance frequency is accurately observed as a
is seen in the theory of Williams et a/. ; Fong minimum in the transmission curve for P polari-
et al. seem to overestimate its size. Our data zation. Our minimum, at 3. 78+0. 02 eV, is in
also exhibit a small peak near 4 eV, which can be satisfactory agreement with McAlister and Stern's.
seen in the theory of Williams et a/. but not in the Turning now to the interband region, our re-
Fong et a/. results. This peak was previously sults for copper are compared in Fig. 9 with the
identified by Gerhardt. The size of this peak as experimental results for &2 obtained by Nilsson
well as the peak at 5 eV in the theory of Williams and by Pells and Shiga. ' We agree more closely
et a/. may be temperature dependent. The the- with the results obtained by Nilsson, who used a
oretical calculations do not contain thermal effects. Kramers-Kronig analysis on transmission mea-
On the other hand, our experimental results are surements taken from thin films of thickness
obtained at room temperature and some of this 400-600 A. The differences may be attributable
 P. B. JOHNSON AND R. W. C HRIS TY

8 at about 4 eV. The discrepancy is well outside

our experimental error. Taft and Philipp used a
Cu
Kramers-Kronig analysis, as did Decker and
Stanford. The differences in the results could be
due either to differences in the extrapolation of
the Kramers-Kronig integrals to higher and lower
b frequencies, or to a difference in measured re-
flectance values. Sample preparation could affect
the reflectance measurements since Taft and
Philipp used electrolytically polished samples,
while Decker and Stanford used evaporated films.
The data of Huebner et al. show the same sharper
peak of E~. They made two reflection measurements
on opaque evaporated films, but their inversion
technique was not as accurate as ours since the re-
sults were interpolated graphically. In addition,
the contours for the two measured reflections (at
o I
20' and 70') do not intersect very steeply in this
1 6
range.
P HOTON E NERGY (eV) Finally, our results for gold are compared with
FIG. 9. Our experimental curve of &2 for copper vs those of Theye ~ and Pells and Shiga. ' In addition
photon energy (JC). Also included are the experimental to our experimental Eaplotted in Fig. 8, we have
curves of Nilsson (N) and of Pells and Shiga (PS). also plotted Theye's and Pells and Shiga's values
for &2. Theye's values of &~ were measured by
inverting normal-incidence transmission and re-
either to the form of his Kramers-Kronig extrap- flection measurements from a 150-A-thick film.
olation or to the greater inaccuracy, + 20 A, in Her thin film agrees closely with our anomalous
Nilsson's determination of his sample thickness. 186-A film. Although her thin film is well crys-
The differences with Pells and Shiga are greater. tallized since it has been annealed, like ours, we
Their values of &2 are significantly larger, especial- do not think, for reasons stated in Sec. IV, that
ly in the spectral range from 4. 5 to 6 eV. This her value for ez is representative of bulk gold.
large discrepancy may be due to the surface prep-— As was the case in copper, Pells and Shiga's val-
aration of their samples, which were mechanically ues of &z for gold are significantly greater than
polished and then annealed under vacuum. Their ours. This difference might again be attributed
polarimetric method of determining &, and &~ is to their method of sample preparation. We may
'6 also note that we are unable to fit their results in
highly dependent on good surface preparation.
Comparison of our E, and &3 values for silver the near infrared to the free-electron theory.
is made with the results obtained by Decker and
Stanford, Taft and Philipp, and Huebner, Ara- VI. SUMMARY AND CONCLUSIONS
kawa, MacRae, and Hamm. ' The values of Deck-
er and Stanford agree well with ours. Their work Our thin-film method utilizing three optical
is the most recent extensive investigation of silver measurements was able to determine the average
using a Kramers-Kronig analysis applied to nor- film thickness to within a few angstroms, judged
mal-incidence ref lee tion measurements. Their by the scatter in the results. Then from an ap-
measurements were made on opaque evaporated propriate pair of optical measurements n and k
films about 1500 A thick, prepared in an ultrahigh could be determined to within less than about
vacuum 10 Torr, and all their measurements +0. 02 over most of the spectral range, with the
were made under a nitrogen atmosphere to reduce error estimate based on the instrumental accuracy
oxidation. Their fast evaporation rate, 50 A/sec, of the reflection and transmission measurements.
like ours, ensured uniformity of the film surface. The results for the optical constants were inde-
The close agreement of our optical constants and pendent of film thickness in the range 250-500 A.
theirs indicates, if the extrapolation of our results (The small transmission of still thicker films
to thicker samples is accepted, the validity of the introduced excessive error with our apparatus;
ref lee tance-extrapolation procedure used by Deck- films thinner than about 200 A were thought to be
er and Stanford beyond the measured reflectance inhomogeneous. ) Annealing of the reasonably
range. thick homogeneous films at 150 C had no observ-
The most striking difference in the data of Taft able effect on the optical measurements, nor
and Philipp is the sharper and higher peak of &z did aging these films in air for days. Thus we
 Op TICAL CONSTANTS OF THE NOBLE ME TALS 4379

conclude that our &, and &2 values are accurately ably well with our experimental results for cop-
representative of the bulk properties of the noble per. Fong's theory appears to be somewhat closer
metals. in magnitude around 5 eV, as is perhaps to be ex-
The free-electron behavior dominates in the pected since their theory had more empirical in-
infrared where n is small and k is large. Thus put than that of Williams et al. On the other hand,
our value of e~z= 2nk contains a large percentage the &~ curve of Williams et al. displays the ap-
error, but &~= -k is relatively accurate. The parent experimental fine structure more closely.
effective free-electron mass determines e „and Christensen and Seraphin did not calculate the
so our values of the effective mass are accurate transition matrix elements but assumed them to
to within a few percent and they agree with the be constant. The fit of their calculation to our
best previous values. The relaxation time ob- results for gold is poor at higher energies; but if
tained from &~, although not very accurate, also only the transitions from the two highest d bands
agrees reasonably well with previous values. are included, the fit is good. Thus it appears that
The interband absorption dominates in the visi- the probability of higher-energy transitions may
ble and ultraviolet where n and k are both of the be smaller than expected. It may be noted that the
order of 1. Thus we conclude that our values of experimental results are obtained at room tem-
&z are accurate to a few percent, and are to be perature, whereas the theoretical calculations do
preferred over previous values which disagree not contain any thermal effects; but it seems
with them in certain regions. The calculations doubtful that thermal effects could entirely ac-
of Fong et al. and Williams et al. agree reason- count for the discrepancies.

~N. F. Mott and H. Jones, The ProPerties of Metals 11, 643 (19V2).
2
and A/loys (Clarendon, Oxford, England, 1936; reprinted P. O. Nilsson, Appl. Opt. 7, 435 (1968).
by Dover, New York, 1958), p. 119. L. Holland, Vacuum Deposition of Thin Ei7ms (Chap-
2E. A. Taft and H. R. Philipp, Phys. Rev. 121, 1100 man and, Hall, london, 1966), pp. 208 and 244.
(1961). ~3C. Reale, Infrared Phys. 10, 173 (1970).
3H. Ehrenreich and H. R. Philipp, Phys. Rev. 128, 4J. E. Nestell, Jr. and B. W. Christy, Am. J. Phys.
1622 (1962). 39, 313 (19V1).
B. R. Cooper, H. Ehrenreich, and H. R. Philipp, 25V. Vrba, Czech. J.
Phys. B 19, 1429 (1971).
Phys. Rev. 138, A494 (1965). E. Kretschmann, Z. Physik 221, 346 (1966).
5U. Gerhardt, Phys. Rev. 172, 651 (1968). 2'D. W. Juenker, J. Opt. Soc. Am. 55, 295 (1965).
6B. F. Schmidt and D. %. Lynch, Phys. Rev. B 3, L. %'ard, A. Nag, and L. C. W. Dixon, Brit. J.
4o15 (1evi). Appl. Phys. 2, 301 (1969).
E. L. Green and L. Muldawer, Phys. Rev. B 2, 330 L. %ard and A. Nag, J. Phys. D 3, 462 (1970).
(ievo). D. Male, Compt. Rend. 230, 1349 (1950).
P. O. Nilsson, Physica Scripta 1, 189 (1970). M. G. Bouchard, in Basic Problems in Thin Film
D. H. Seib and W. E. Spicer, Phys. Rev. B 2, 1676 Physics, edited by R. ¹iedermayer and H. Mayer
(1evo). (Vandenhoeck and Buprecht, G'ottingen, 1966), p. 301.
G. P. Pells and M. Shiga, , J. Phys. C 2, 1835 (1969). 32Bichard Tousey, J. Opt. Soc. Am. 29, 235 (1939).
M. L. Theye, Phys. Rev. B 2, 3060 (1970). 33R. F. Miller, A. J. Taylor, and L. S. Julien, J.
~2P. O. Nilsson, Physik Kondensierten Materie 11, 1 Phys. D 3, 1957 (1970).
(19vo). R. F. Miller, L. S. Julien, and A. J. Taylor, J.
3C. Y. Fong, M. L. Cohen, R. R. L. Zucca, J. Phys. D 4, 1100 (1971).
Stokes, and Y. R. Shen, Phys. Bev. Letters 25, 1486 R. H. Huebner, E. T. Arakawa, B. A. MacRae, and
(19vo). R. N. Hamm, J. Opt. Soc. Am. 54, 1434 (1964).
A. R. %'illiams, J. F. Janak, and V. L. Moruzzi, 38D. Beaglehole, Proc. Phys. Soc. (London) 85, 1007
Phys. Bev. Letters 28, 671 (1972). (1965).
N. E. Christensen and B. O. Seraphin, Phys. Rev. ~American Institute Of Physics Handbook (McGraw-
B 4, 3321 (19V1). Hill, New York, 1957), Sec. 6, p. 31.
~~S. Roberts, Phys. Rev. 118, 1509 (1960). 38H. E. Bennett, M. Silver, and E. J. Ashley, J. Opt.
F. Stern, in Solid State Physics, edited by F. Seitz Soc. Am. 53, 1089 (1963).
and D. Turnbull (Academic, New York, 1963), Vol. 15, 3
H. E. Bennett and W. F. Koehler, J. Opt. Soc. Am.
p. 299. 5O, 1 (196O).
G. W. Bubloff, Phys. Bev. B 3, 285 (1971). 4
A. J. McAlister and E. A. Stern, Phys. Rev. 132, 1599
D. L. Decker and J. L. Stanford, J. Opt. Soc. Am. (ie63).
61, 679A (1971). 4~P. O. ¹ilsson, I. Landau, and S. B. M. Hagstrom,
20J. E. Nestell, Jr. and B. W. Christy, Appl. Opt. Phys. Rev. B 1, 498 (1970).