Growth of Silicon Crystals Free from Dislocations

William C. Dash

Citation: J. Appl. Phys. 30, 459 (1959); doi: 10.1063/1.1702390

View online: http://dx.doi.org/10.1063/1.1702390
View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v30/i4
Published by the AIP Publishing LLC.

Additional information on J. Appl. Phys.

Journal Homepage: http://jap.aip.org/
Journal Information: http://jap.aip.org/about/about_the_journal
Top downloads: http://jap.aip.org/features/most_downloaded
Information for Authors: http://jap.aip.org/authors

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 Journal
of
Applied Physics
Volume 30, Number 4 April,1959

Growth of Silicon Crystals Free from Dislocations*

WILLIAM C. DASH
General Electric Research Laboratory, Schenectady, New York
(Received August 29, 1958)

The dislocations which occur in silicon crystals grown by the terms of climb. In a super-saturation of vacancies, an edge dis-
Czochralski method with and without a crucible are studied by location can be forced completely out of a crystal; a screw will
etching and copper decoration techniques. A representative remain straight or may be distorted into a coaxial helix. By growth
crystal with about 500 dislocations per cm2 is examined in detail along {100) or (111) axes under conditions which favor a vacancy
to provide general information about their distribution. Sources supersaturation, screw as well as edge and mixed dislocations:can
of dislocations which are important at particular stages of growth propagate to the surface and be eliminated. The observations are
are isolated and means found to minimize or eliminate them. The consistent with this interpretation. When the dislocations are
sources include generation by thermal shock from surface damage eliminated, the growing crystal can withstand large temperature
or residual dislocations in the seed; poor epitaxy at the seed; fluctuations without further generation of dislocations. A pro-
plastic generation in the bulk of the crystal; effects of segregation cedure is outlined for the growth of dislocation-free silicon crystals.
of impurities. Plastic generation is the only source found in the Germanium crystals free of dislocations have been grown by a
bulk of a crystal. "Spontaneous" formation of dislocations from direct application of the method. It is believed that the conclusions
excess vacancies is not found. Impurity segregation causes some reached during the course of this investigation may be applied to
generation at the tip end of crystals. For certain crystallographic the development of methods for growing dislocation-free crystals
directions it is possible to eliminate residual dislocations and of other substances.
thenceforth grow a dislocation-free crystal. This is interpreted in

I. INTRODUCTION In a recent note," the writer reported that silicon

ILICON has a number of properties which permit crystals free from dislocations have been grown from
S simple and reliable investigations of the origin of the melt. It is the purpose of the present article to
discuss the origin of dislocations in silicon and to
dislocations. Single crystals of high purity can be grown.
The points of emergence of dislocations through the describe a procedure by which dislocation-free crystals
surface can be detected by etch pits formed on any can be grown.
crystallographic plane.! Specimens can be prepared and II. EXPERIMENTAL PROCEDURE
handled conveniently at room temperature without
disturbing the internal structure. 2 Decoration of screw, The observations to be presented were made using
edge, or mixed dislocations by precipitated copper zone-refined silicon 6 with resistivity of SOQ-cm or greater
makes possible their direct observation in the bulk by as starting material. The crystals were grown by the
use of an infrared image tube.!·3.4 Czochralski technique from a quartz crucible or from a
silicon "pedestal"." These methods are described
* An abbreviated version of this article has appeared in the briefly in the following.
report of a conference on the growth of crystals at Cooperstown, A schematic diagram of the quartz crucible and
New York, in August, 1958: Growth and Perfection of Crystals,
edited by Doremus et al. (John Wiley and Sons, Inc., New York, graphite resistance heater is shown in Fig. 1. Each
1958), p. 361. chamber contains argon at atmospheric pressure. A
1 W. C. Dash, J. Appl. Phys. 27, 1193 (1956).
2 W. C. Dash, J. Appl. Phys. 29, 228, (1958).
seed is lowered into the melt to initiate growth, then
3 W. C. Dash, Dislocations and Mechanical Properties of Crystals
(John Wiley and Sons, Inc., New York, 1957), p. 57. (1958), corroborate the decoration and etch pit techniques for
• X-ray methods recently reported by A. R. Lang, J. App!. identifying dislocations in silicon.
Phys. 29, 597 (1958); J. B. Newkirk, Phys. Rev. 110, 1465 (1958); Ii W. C. Dash, J. Appl. Phys. 29, 736 (1958).
and Bormann, Hartmann, and Irmler, Z. Naturforsch. 13a, 423 6 E. A. Taft and F. H. Horn, J. Electrochem. Soc. 105,81 (1958).

459
Copyright © 1959 by the American Institute of Physics.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 460 WILLIAM C. DASH

its growth. Studies of the 9 !l absorption at room tem

perature in pedestal-grown crystals have shown that
1__ 1',1 _ _- MOLYBDENUM SHAFT oxygen could not be detected. Although the oxygen
content of floating-zone crystals is also low, they gen-
erally have too many dislocations to permit detailed
_-SEED studies. For the purposes of the present investigation,
the pedestal crystals are of interest because the growth
conditions differ considerably from those maintained
in withdrawal from a crucible. Despite this, the con-
siderations which apply to the growth of dislocation-
\:b:S::;::'::::;i;l:7 FU SED free crystals by either method are shown to be closely
QUARTZ
related.
"Meltbacks," illustrated in the inset in Fig. 1, have
also been utilized in some experiments. To make a
meltback the lower end of a bar is melted by radiation
from the walls of the furnace. Surface tension holds the
tl ~_ _ rt--''''L.IT -CUP
drop of molten silicon, generally about! gram, which
RESISTANCE is then recrystallized by lowering the temperature. The
HEATER
oxygen content in the meltback is found to be below

FIG. 1. Schematic diagram of quartz crucible furnace The lower

chamber is insulated with quartz wool. Both upper and lower
chambers contain an argon atmosphere. The arrow in the inset
indicates a "meltback" made from the lo'ver end of a bar of silicon
suspended in the furnace.

withdrawn at a rate of about 2 millimeters per minute

with rotation at about 10 rpm. The approximate tem-
perature gradient under typical growth conditions was
determined by inserting a thermocouple into a hole in
a partially grown specimen, lowering this assembly into
the melt as a seed, and measuring the temperature as
it was withdrawn. The gradient thus determined was
about 12SoC per cm. It was found to be linear for about
4 centimeters over which the diameter was approxi- FIG. 2. Typical single crystal
imately constant. Figure 2 shows a typical crystal grown from quartz crucible.
weighing about 20 grams.
The "pedestal" method was devised to avoid oxygen
contamination7 •8 and at the same time achieve the high
perfection attainable with the Czochralski technique.
In this method the melt is an inductively heated mound
held on top of a solid silicon support by surface tension
and electromagnetic levitation. The support is sectored
to inhibit electromagnetic coupling to the pedestal. A
seed is inserted and the growing crystal is withdrawn
at a rate which may vary from 3 cm per minute at the
start to 3 or 4 mm per minute during the major part of
7 Kaiser, Keck, and Lange, Phys. Rev. 101, 1264 (1956) have
reported that the oxygen content of silicon withdrawn from quartz
can be determined by infrared absorption of 9 J.L. The concentration
is generally around 10'8 cm-'.
8 R. A. Logan and A. J. Peters [J. App!. Phys. 28, 1419 (1957)J
have reported that in some cases the presence of oxygen may lead
to errors in the identification of dislocations by etch pits. They
found, however, that in these specimens identification by copper
decoration was not affected.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF Si CRYSTALS FREE FROM DISLOCATIONS 461

the limit detectable by 9 J.L absorption. An additional

feature is that some residual impurities are concentrated

~ ~
in the bottom tip, permitting examination under more
favorable conditions. (110)
In order to obtain an estimate of the number and t
distribution of dislocations in a crystal, it is etched in a
mixture consisting of 50% HF solution, 70% HNO a 5
solution and glacial acetic acid in the ratio of 1: 3: 10 by
Fm. 3. Schematic 2
volume. [This mixture is hereafter referred to as diagram of (110)
(1 :3: 10); two other mixtures, with volume concen- crystal with a {111}
trations (1:3:0) and (1:3:6) are used for special pur- twin plane which
changed the axial Twin
poses to be described.J The time required to form orientation to ap-
identifiable pits is one to two hours; pits several hun- proximately (113).
dred microns deep are formed by etching overnight. The specimen was 7
cut up into the indi-
The crystal is then cut into parallel-sided slabs, cated sections for

8~
lapped smooth and chemically polished in (1: 3: 0) decoration and ex-
amination.
solution. A water solution of copper nitrate with copper
equivalent to about 10-6 to 10-5 atom percent of the

'f~
silicon is spread in droplets over one side of each slab
and evaporated to dryness. They are then put into a
hydrogen atmosphere in a quartz tube and heated at
about 950°C for 30 minutes and quenched by dropping
onto a quartz plate. Rapid cooling is necessary to pre-
vent formation of large precipitate particles which 9 10 II
result in poorly defined dislocation lines. The contact (a) (b)
with the quartz plate makes possible cooling below the
temperature of luminosity of slabs weighing as much as
ten grams in a few seconds with no apparent change in Figure 3(b) illustrates how the crystal was cut for
the number or distribution of dislocations. Faster examination of the decorated dislocations. In what is
to follow only slices 3, 6, 7, and 8 and a portion of slice
quenchi~g, for example by contact with a copper plate,
results m too light a decoration for convenient 10 are illustrated, since these show all the important
examination. features. Saw cuts and polishing account for a removal
Following decoration, the slabs are optically polished of about one mm of material from between the slices'
on the flat sides and viewed with a microscope and this must be taken into account in tracing the lines fro~
one slice to another.
infrared image converter as described in reference 1.
Figur~ 4 shows dislocations in slices 3 and 10, viewed
Over a hundred silicon crystals have been examined
either by etching or by both etching and decoration. in a [110J direction. (The following comments apply
The discussion will be limited to specimens which con- specifically to slice 3; the lower part of the figure cor-
tain a maximum of 103 dislocations per cm2 , because the responding to slice 10 is discussed later.) Although the
arrays in less perfect crystals are complicated and diffi- dislocations present here are too numerous for extensive
analysis, some interesting observations have been made.
cult to analyze. Two of the specimens contained one
and two twins, respectively. These crystals were In the center the density of dislocations is relatively
selected for study because twins form crystallographic low: they are present mainly in the form of long lines.
barriers to glide, thereby making it possible to separate Many are over a centimeter in length and run parallel
motion and multiplication from other origins of to each other and to the [110J axis. This is a close-
dislocations. packed crystallographic direction in the diamond struc-
ture. It is also the preferred orientation for both screw
III. DETAILED EXAMINATION OF A CRYSTAL and "60°" dislocations during plastic deformation at
WITH ABOUT 103 DISLOCATIONS PER Cm" temperatures around 900°C. 1 ,3
A specimen which was decorated and examined In addition to the long [110J dislocations, many
throughout most of its volume contained a single (111) loops are observed near the seed end and near the outer
twin boundary about half-way along its length, as shown e~ges of the slab. Isolatej. loops can be seen on edge
schematically in Fig. 3(a). The twin was nucleated at lymgJn the (11!land (111) planes. Others, presumably
the point marked A, possibly by a speck floating on in (111) and (111) planes, appear to be parallel to the
the surface of the melt, after which it grew completely plane of the sample because of the large depth of focus.
~cross ~he crystal. Except for the twin, this specimen H is very likely that these loops result from a light
IS typIcal of crucible-grown crystals of moderate plastic deformation produced by thermal stresses,
perfection. because they resemble the noncrystallographic loops

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 462 WILLIAM C. DASH

twin. Dislocations which are essentially parallel to the

crystal axis appear as lines of equal length parallel to
the long axis of the elliptical section. A number of loops
on (111) glide planes can be seen; their "pinning" dis-
locations extend out of the plane of the loops. From this
figure the dislocation density is estimated to be of the
order of 500 cm-2•
Figure 6 is a photograph of slice 7 containing the twin.
The most striking observation is the large number of
long straight lines, mostly [Ol1J and [101J directions.
As nearly as can be determined, these lie at the position
of the twin. (The twin is not independently identifiable
by decoration except where dislocations lie at or very
near to it.) In most cases it is found that the lines ter-
minate at the periphery at one end, leave the twin in
the interior and pass through either the upper or lower
surface. A few have been found to lie parallel to the
twinning plane over their whole lengths, with both ends
emerging at the periphery.
Figure 7 is a photograph at a higher magnification of
the region at (D4) , illustrating the intersection of
several lines from above the twin with the twinning
plane. The arrows indicate two points of intersection.
It is probable that many of the (110) lines in the twin-
ning plane are partial dislocations. This would account
for the apparent immobility of these dislocations.
A possible mechanism for the formation of som~ of
these lines is suggested by the loop lying on a (111)

FIG. 4. Upper part shows decorated dislocations in slice 3 of the

specimen in Fig. 3. Many long lines lie accurately parallel to the
(110) axis. Lower part shows dislocations in slice 10, with axial
orientation approximately (113). The sample was broken in
handling after decoration, then pieced together, resulting in the
jagged dark areas. Most of the dislocations in the lower specimen
diverge from the axis and are soon eliminated. Some of the dis-
locations appear to terminate; the ends of these were curved and
consequently were cut off in preparing the specimen. Maximum
width is 1.3 cm. The dark specks in this photograph and others
are caused by lint and dust which were difficult to eliminate
entirely.

found in crystals slowly deformed at temperatures

FIG. 5. Dislocations in slice 6 parallel to {111}. There are
considerably higher than 900°C. 3 approximately 500 dislocations per cm2 • Minimum diameter is
Figure 5 is a photograph of slice 6 cut parallel to the 1.3 cm.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF Si CRYSTALS FREE FROM DISLOCATIONS 463

plane at the center of the section in Fig. 6. The loop

appears to be expanding on the glide plane as a result
of local stresses. Part of it, indicated by the arrow, lies
in the twinning plane as nearly as can be determined
and extends in a [llOJ direction. Further motion of this
part of the loop is not possible because the glide plane
ends abruptly at the intersection with the twin. If the
stress were applied for a sufficiently long time, the line
at the twin would extend to the edge of the section at
the right.
Several "pileups" of dislocations, which are a series
of lines of similar orientation on the same glide plane
held apart by mutual repulsion, are clearly visible at
(A3) in Fig. 6, and in the stereoscopic view in Fig. 8.
Many other such groups are not quite so easily distin-
guishable but can be detected from etch pits at the
surface as shown in Fig. 9. Each group of these disloca-
tions probably originated at a single source some
distance above the twin.
At (C5) in Fig. 6 there are hexagonal networks in the
twinning plane. These are seen in higher magnification FIG. 7. Enlarged view of region near D4 in Fig. 8. The arrows
indicate the points where dislocationsfro m above enter the twin
in Fig. 10. They appear to have formed from the inter- plane and abruptly change direction to [OIl]. Arrow length cor-
action of moving dislocations from two different sources responds to about 0.2 mm.
in planes parallel to the twin. Hexagonal networks of
this nature are conspicuous by their rarity in the .l\Iany of the dislocations which intersect the twinning
silicon crystals studied. plane from above pass on through. Some have rather
abrupt changes in direction as at (C3) in Fig. 6; others
have little deviation. Abrupt changes probably occur
because the intersection with the twin is a pinning
point-the parts above and below move independently
under local stresses. Those which get through to slice 8
are seen in Fig. 11. The total number is about half as
great as that in slice 6. It is probable that this large de-
crease is due to the fact that the twin serves as a barrier
to propagation of the dislocations by deformation.
An observation of considerable importance can be
made from slices 3 and 10 in Fig. 4 representing the
regions above and below the twin, respectively. (Slice
10 was broken in handling and had to be pieced to-
gether. Cracks are responsible for the large dark areas.)
The marked tendency of the dislocations to follow the
growth axis in the upper part of the specimen is not
evident in the lower part. Examination of slices 9, 10,
and 11 indicated that most of the dislocations grew
toward the surface of the crystal, resulting in a con-
tinuously decreasing concentration except for a small
amount of plastic generation. This is not a result of a
change of growth conditions, but rather of crystal-
lography, since the lower part of slice 3 and upper part
of slice 11 grew at the same time. Moreover, the shape
of the solid-liquid interface, identified by diffuse groups
FIG. 6. Dislocations in slice 7 containing twin. Long rows of of copper specks faintly visible in Fig. 4, is everywhere
parallel lines lie in (110) directions at or very near twin plane. convex toward' the melt and does not change noticeably
Arrow points to a loop forming a [110J line at twin plane. Mini-
mum diameter is 1.25 cm. in going through the twin plane. This observation gives

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 464 WILLIAM C. DASH

FIG. 8. Stereoscopic
view of a portion of slice
7 containing twin. Sev-
eral groups of "pileups"
are clearly visible.

a clue to a mechanism for the elimination of dislocations. IV. SOURCES OF DISLOCATIONS DURING
This is discussed more fully in a later section. GROWTH OF A CRYSTAL
The specks of copper precipitate mentioned above A. Generation of Dislocations in the
are commonly observed in regions remote from dis- Seeding Process
locations. They are more evident in the specimen in
There appear to be four sources which introduce dis-
Fig. 16. They are usually grouped with a periodicity
locations into the crystal at the inception of growth on
corresponding to the amount of growth in a single
a seed. These are listed and described in the following.
revolution of the crystal. The nuclei at which they form
may be aggregates of oxygen or other impurities or 1. Propagation by growth of the dislocations which
possibly vacancy clusters. The size of the specks is very emerge from the seed at the solid-liquid interface.
sensitive to the rate of cooling after decoration-rela- 2. Multiplication by thermal shock of the dislocations
tively slow cooling gives rise to large specks, and vice in the seed and propagation by growth of those which
versa. These specks are discussed further in the follow- emerge at the solid-liquid interface.
ing in relation to pedestal-grown crystals. 3. Generation from surface damage.-Deformation

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF Si CRYSTALS FREE FROM DISLOCATIONS 465

studies reported in references 1 and 3 indicated that

groups of dislocation loops frequently entered from the
surface where there was no evidence of a grown-in dis-
location to serve as a source. A characteristic of arrays
of this type is irregularity of spacing and overlapping
of consecutive loops, as seen in Figs. 3 and 4 of reference
3. This differs markedly from the coherence and regu-
larity of spacing of loops observed to originate from a
Frank-Read source (Figs. 5 and 6 of reference 3).
Recently we discovered that the surface arrays originate
from vestiges of surface damage introduced in preparing
the specimens for the deformation experiments.
The effect can be demonstrated by using the meltback
method to simulate the thermal conditions which occur
in a seed. A dislocation-free crystal was lightly scratched
with a tungsten carbide scriber at a point several mm
from one end. The region below the scribe mark was
then melted as shown in the inset in Fig. 1, then allowed FIG. 10. He'aglnal network of dislocaticns at Bs in Fig. 8.
to freeze. Upon decoration, it was found that incoherent Diameter of field is about 2.5 mm.
dislocation loops had expanded from the region of the
scratch into the bulk of the crystal (Fig. 12). There is
little doubt that these loops were formed when the
crystal was heated in the meltback furnace rather than
when it was scratched or decorated with copper. If the
specimen is etched in (1: 3: 6) for about 15 minutes after
scribing, the damaged material is preferentially re-
moved; the loops do not form when it is heated. The
lines in Fig. 12 have the shape characteristic of deforma-
tion well above the decoration temperature. s It is
believed that dislocations are generated in the region
very near the scratch, and that they expand into the
bulk at high temperature under the influence of thermal
stresses. This experiment demonstrates the necessity of
careful etching and handling of the seed.
4. Imperfections formed by poor epitaxy.-Disloca-
tions can form at the seed or in other parts of the
crystal during growth without multiplication of those
already present. This occurs when liquid comes in con-

FIG. 11. Dislocations in slice 8 below twin. Only about half of

the dislocations in slice 6 have been propagated into this region.
Most of the others were removed at the twin. Minimum diameter
is about 1.2 cm.

tact with a solid surface which is contaminated or at too

Iowa temperature to permit perfect epitaxy. The effect
is easily demonstrated with a meltback. Surface tension
draws the liquid up the bar as the size of the drop
FIG. 9. View of etch pits on external surface of slice 7. Vertical increases. When the drop is allowed to freeze before the
groups represent pileups. High concentration of pits in diagonal entire tip is melted through, the result is as shown in
line indicates position of twin plane. Periodic horizontal shadings Fig. 13. The position of the original surface of the bar is
represent the amount of crystal withdrawn in one rotation. Spacing
between outer rows of pits is about 0.5 mm. indicated by the arrows.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 466 WILLIAM C. DASH

B. Dislocations Generated in the Main

Volume of the Crystal
Another twinned crystal gives further information
about the origin of dislocations when growth is well
under way. In this specimen two closely-spaced twins
lie parallel to the [l1OJ growth axis as sketched in
Fig. 1S(a), beginning at the surface of the crystal
several millimeters below the splice. Since there is an
even number of twins, the orientation of the crystal on
either side of the twins is the same. The scalloped
external shape of the upper part resulted from abrupt
increases in the temperature of the melt which were
required in order to prevent solidifi.cation from a cold
spot at the wall of the crucible. Upon etching this
crystal in (1: 3: 10) it was discovered that all of the dis-
location etch pits were confined to one side of the twins.
Figure lS(b) shows decorated dislocations in a longi-
tudinal slab cut perpendicular to the twins. Only two
small loops (see arrows) are to be found in the lower
part of the slab. It is likely that these came from screws
which "leaked" through the twin barrier, since they
can move on any {111} plane. In contrast, there are in
the upper part a number of long [110J dislocations

FIG. 12. Generation of dislocation loops from surface damage.

A dislocation-free crystal was scribed (dark region in center) and
heated in the arrangement used for making meltbacks. Loops
expanded into the surrounding region. No loops form if the scribed
region is etched away before heating. Width of field is about
0.25 mm.

The region near the arrow at left in Fig. 13 is shown

at higher magnification in Fig. 14. Many twins, detect-
able on the surface, appear in Fig. 14 as a series of
crystallographically oriented lines. Traces of the twin
planes can be identified internally after decoration
because of dislocations which lie parallel to them as in
Fig. 6. In addition, a number of dislocations can be seen
extending down into the region where the epitaxy was
good, whence they are propagated by growth and
multiplied by thermal stresses. Although the original
crystal was not completely free from dislocations, most
of those in Figs. 13 and 14 can be traced back to the
region of poor epitaxy.
The incoherent structure is probably caused by con-
taminants at the surface such as an oxide layer or con-
ceivably by a polycrystalline silicon layer formed by
condensation from the vapor. Generation of disloca-
tions by this process is important only at the seed under FIG. 13. A decorated meltback which demonstrates the effect
of poor epitaxy. The arrows indicate the surface of the original
normal growth conditions. However, specks of solid bar. Surface tension caused liquid silicon to climb up the sides of
matter either floating on the surface or suspended in the bar, where it solidified before proper wetting could take place.
Most of the dislocations in the lower region came directly or by
the melt might cause a similar effect at any time during plastic generation from the region of poor epitaxy. Maximum
growth. diameter is about 5 mm.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF Si CRYSTALS FREE FROM DISLOCATIONS 467

propagated by growth, as well as large concentrations

~
EED =RAPHED ~.
DOUBLE
of tangled lines. These appear to have been generated ENO _ ,~A~~~ ~ _ TWINS

by mUltiplication processes under thermal stresses fol- [110) --- - -- - - r--.-..... -

lowing the abrupt changes in melt temperature. t
"Spontaneous" generation from the surface is notably I em
absent in the lower part of the specimen in Fig. 15. (a)
Since surface damage is unlikely to occur in growth by
the Czochralski method, no deformation is expected
from this source.
The absence of spontaneous internal generation of
dislocation loops is also noteworthy. Seitz9 has pointed
out that with a sufficiently high vacancy supersatura-
tion in a growing crystal, aggregates might form and
eventually collapse into disks bounded by dislocations (b)
of Taylor-Orowan character. Continued condensation
on the edge of such a disk nearest the interface would FIG. 15. Dislocations in a [11OJ crystal with two closely spaced
twins. Deformation from thermal stresses has occurred repeatedly
result in the formation of a pair of parallel edge dislo- on only that side of the twins which contains dislocations. The
cations running perpendicular to the interface. Pris- other side has only two small loops (marked by arrows). These
were probably screws, which can move on any glide plane,that
matic dislocations can be formed under special condi- came through the twins from the other side. Other lines on lower
tions described in Sec. vc. If prismitic dislocations do part of specimen are scratches on the surface. No "spontaneous"
form by vacancy aggregation in as-grown silicon, their generation of dislocations occurred either internally or at the
surface of this half of the crystal.

diameter can be no greater than about 2 micron,;, the

limit of resolution of the decoration technique.
From the ,;pecimen in Fig. 15 it can be seen that
thermal stresses which are sufficient to cause appreci-
able plastic strain when dislocations are present cause
no plastic strain under identical conditions but in the
absence of dislocations. A similar observation was made
on a crystal which was completely withdrawn from the
melt during growth and re-inserted a few seconds later.
A decorated slice from this specimen is shown in Fig. 16.
The line indicated by the arrows identifies the interface
between the material grown prior to withdrawal and
that grown afterward. The crystal was free from dis-
locations above this line. None formed from the thermal
shock upon re-insertion. In a crystal containing a few
dislocations, treatment such as this specimen received
results in the appearance of a large density of disloca-
tions. Freedom from dislocations of a part of the crystal
in Fig. 16 including the region of thermal shock was
confirmed by A. R. Lang using his x-ray technique. 4
From the specimens in Figs. 15 and 16 one observa-
tion deserves further emphasis: dislocations were gen-
erated only from other dislocations. This greatly simpli-
fies the problem of growing crystals with high perfection.

C. Concerning Dislocations Generated by

Segregation of Impurities
The bottom tip of a crucible grown crystal or of a
meltback frequently shows evidence of poor structure.
FIG. 14. Higher magnification view of region near arrow at left A photograph of a typical "sprout" which has been
in Fig. 16. The region with poor epitaxy is highly twinned. The
twins are made visible by dislocations parallel to them as in Fig. 6. etched lightly is shown in Fig.17. Examination by infra-
Many dislocations extend from this region into the main part of red transmission of an undecorated sprout reveals a
the meltback. Width of field is about 0.8 mm.
filamentary segregated phase shown in the upper part
9 F. Seitz, Phys. Rev. 79, 890 (1950). of Fig. 18. The lower part of this figure shows bire-

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 468 WILLIAM C. DASH

The volume of the "sprout" in undoped crystals

varies from 1 to 50 mmS from specimen to specimen. In
general, it is of order of one percent of that found in the
SiC-doped crystals, which would suggest a lower limit
of concentration of the segregating impurity of 10-6 to
10--7 atom fraction in the average crystalY
Segregated impurities in the "sprout" can affect the
dislocations further up the crystal because of differen-
tial thermal contraction. The high strain seen by bire-
fringence in Fig. 18 is capable of causing plastic defor-
mation in the region above the sprout if there are dis-
locations present. This probably accounts for the
increased concentration of dislocations at the lower end
of many crystals. In specimens in which all of the dis-
locations have been eliminated, no generation of dis-
locations occurs in the bulk of the crystal as a result of
segregation at the tip. However, there is some evidence
from etch pit studies that generation does occur in the
immediate vicinity of the filaments.
Tiller12 has postulated that segregation of solutes on
a microscopic scale throughout the growth of a crystal
causes the introduction of dislocation lines at the bound-
ing surfaces of the segregate. He calculated that 103 to
108 dislocations per cm2 might be expected to form by
this process, depending upon impurity content and
growth conditions. Goss et al.J3 have observed a macro-
scopic effect analogous to this caused by fluctuations
in the freezing rate in Si-doped Ge.

FIG. 16. A decorated slice from a dislocation-free crystal that

received a severe thermal shock-extraction from the crucible for
several seconds and re-insertion. The rows of copper specks,
possibly nucleated by oxygen or vacancy clusters, have a spacing
which corresponds to the amount of material withdrawn in one
revolution, and delineate the successive interface positions. Maxi-
mum diameter is 1.7 em. The position of the interface on re-
insertion is indicated by the arrows. No dislocations were gen-
erated by the shock.

fringence in the region of the filaments. Figure 19 is a

transmission electron micrograph of a filament, made
on a polished specimen about 2 microns thick. Mor-
phologically similar filaments of segregated material are
found throughout the volume of the last several grams
of a crystal withdrawn from a melt into which a small
single crystal of silicon carbide had been placed. Meas-
urements of weight change of the SiC indicated a solu-
bility of 2 to SX 10-5 atom fraction in the liquid.I o
Filaments with the same general appearance have FIG. 17. Lightly etched "sprout" end of silicon crystal. The
been found in Ge-doped crystals. Other impurities may perfection changes abruptly here because impurities are concen-
trated and form a second phase. Maximum diameter of sprout is
also produce a similar effect, and indeed impurities in about 2 mm.
the silicon carbide crystals used may have been respon-
II H. A. Papazian and S. P. Wolsky [J. App1. Phys. 27,1561
sible for the observations. This is considered to be (1956)J reported mass spectrometric evidence of carbon in silicon
unlikely, however, because the total impurity content with a concentration around 1019 em-a. This figure appears to be
in the SiC dissolved was probably no more than 10-8 much higher than expected from Hall's measurements, and from
the observation that segregation occurs. The discrepancy is not
atom fraction of the total silicon. understood.
12 W. A. Tiller, J. Appl. Phys. 29, 611 (1958).
10 R. N. Hall, J. Appl. Phys. 29, 914 (1958). 13 Goss, Benson, and Pfann, Acta Met. 4, 332 (1956).

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF Si CRYSTALS FREE FROM DISLOCATIONS 469

No evidence of Tiller's mechanism has thus far been

found in the main bulk of silicon crystals which were
either undoped or which contained boron at concen~
trations up to about 10-4 atom fraction. The effect of
more strongly segregated impurities has not been
studied.

V. CONSIDERATIONS INVOLVED IN
IMPROVING PERFECTION
A. Preparation of the Seed
The initial dislocation concentration immediately
after the commencement of withdrawal of the crystal
is determined by the following effects, which have been
described in Sec. IV.
(1) Propagation by growth of dislocations initially
FIG. 18. (a) Infrared transmission of an undecorated sprout,
present in the seed. sowing filaments of a segregated impurity. Separation of tips of
(2) Generation of dislocations by thermal stresses in clusters is about 3 mm. (b) Photograph of the specimen in (a)
the seed from dislocations already present and between crossed polarizers.
propagation of these by growth.
(3) From residual surface damage in the seed. The number of dislocations generated by plastic flow
(4) From poor epitaxy of the newly crystallized will depend upon the nature of thermal stresses while
material. the temperature of the crystal is in the plastic range.
Billig14 has analyzed this problem for the growth of
These sources can be minimized by care in selection and
germanium and silicon crystals. However, Figs. 15 and
preparation of the seed, use of a small cross section at
16 indicate that thermal stresses cause plastic flow only
the tip, and by allowing sufficient time for complete
when dislocations capable of motion are present. A
melting of the original surface of the seed where it con-
moderate number of dislocations can be tolerated in
tacts the liquid.
the early stages of growth provided that there is a
B. Effect of Crystallographic Orientation
on Plastic Generation
The preferred crystallographic orientations minimize
the number of plastically generated dislocations which
expand from a source in the solid down to the growing
interface. For a crystal of given diameter, the larger
the minimum angle between the growth axis and {111}
planes, the smaller will be the probability that disloca~
tions formed by thermal stresses will expand, intersect
the interface and be propagated. Thus (100) and (111)
orientations are more favorable than (110) or (112).
Table I lists four low index directions in order of FIG. 19. Transmission
electron micrograph of a
preference. branched filament of segre-
It should not be concluded from Table I that growth gated material in sprout.
along a (100) or (Ill) axis will in itself assure high per- Other lines are surface
scratches, produced in pre-
fection. The dislocations which actually emerge at the paring specimen. Length of
solid-liquid interface may be propagated indefinitely filament is about 30 p..
under suitable thermal conditions regardless of axis.
Figure 20(b) is an example of this in a crystal with (100)
growth axis.
TABLE 1.

Smallest angle between axis and a

Crystal axis {Ill) glide plane
------
(100) 35.26°
(111) 19.48°
(112) 0° (one set of {111) planes)
(110) 0° (two sets of {U1} planes)
14 E. Billig, Proc. Roy. Soc. (London) A235, 37 (1956).

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 470 WILLIAM C. DASH

in" of vacancies in the solid when a temperature gra-

dient exists near the solid-liquid interface. This occurs
if the vacancies which are present in equilibrium con-
centration at the solid-liquid interface are not given
sufficient time to diffuse to the surface. 9 ,16 The resulting
supersaturation can be relieved by the climb of disloca-
tions or by the condensation of the vacancies into
aggregates, as discussed by FrankP Higher rates of
growth and larger temperature gradients lead to greater
supersa tura tions.
At the present time there is insufficient quantitative
information about vacancy concentrations and diffusion
coefficients at fixed temperatures or in thermal gradients
to allow the number of excess vacancies to be calcu-
lated. However, studies of the perfection of pedestal-
grown crystals are qualitatively consistent with the
statements made about rate of growth and thermal
gradient. The perfection depends critically upon the
way in which the initial stages of growth occur. This is
illustrated in Fig. 20, which shows decorated disloca-
tions in longitudinal slabs of two different specimens
grown under nearly identical conditions. In Fig. 20(a)
the diameter of the first part of the crystal is kept small,
which is accomplished in this method by a rapid initial
withdrawal rate (about 3 cm per minute); the disloca-
tions are completely removed after about 5 mm of
growth. The small diameter results in a larger tempera-
ture gradient in the first case because of lower efficiency
of heating the withdrawn crystal in the electromagnetic
field. In Fig. 20(b) the diameter increases rapidly below
the splice; the dislocations persist and multiply as
(a) (b)
growth proceeds. The withdrawal rate where the diam-
FIG. 20. Dislocations near the seed end of two pedestal crystals eter of the specimens is comparable is approximately
illustrating the effect of different growing conditions. (a) Initial
withdrawal rate about 3 cm per minute to maintain small diameter 4 mm per minute in both cases. The first specimen, for
and high temperature gradient. All dislocations have been re- which the initial growth rate and thermal gradient
moved in about 5 mm. Note the "island" of copper specks at were both large relative to the second, is quickly freed
lower end. (b) Initial withdrawal rate about 4 mm per minute.
Diameter increases rapidly, dislocations propagate and multiply of dislocations, whereas the second has a large number
by thermal stresses. of dislocations. This behavior is always observed in
crystals grown in this manner.
mechanism for removing them. When none remain, The comparison of the specimens in Fig. 20 is con-
radial thermal gradients and the shape of the solid- sistent with the idea that a high thermal gradient and
liquid interface no longer influence the dislocation high rate of growth produces a vacancy supersaturation.
density. This in turn favors climb of dislocations with an edge
component; a pure edge dislocation can be driven com-
c. Effect of Temperature Gradient and pletely out of the crystal. It is important to note, how-
Rate of Growth ever, that under the same conditions a screw dislocation
will not be translated in a direction perpendicular to its
Climb of dislocations, caused by a supersaturation of
axis. When the excess of vacancies becomes as large as
vacancies, provides a mechanism for removing the dis-
1% of the equilibrium concentration, a screw will climb
locations which are present in the early stages of growth.
into a helix. 18 ,19 Growth along a (110) direction has the
The excess vacancy concentration is related to the tem-
perature gradient and to the rate of growth. It is not disadvantage that a screw dislocation which happens
possible to say a priori how the gradient alone will 16 A. G. Tweet [J. App!. Phys. 29, 1520 (1958)] has reported
affect the diffusion of vacancies from the liquid into the evidence of a supersaturation of vacancies in germanium crystals
solid at very small growth rates. ln The continuous grown in a high temperature gradient.
17 F. C. Frank, Deformation and Flow of Solids (Springer-Verlag,
withdrawal of the crystal, however, causes a "quenching- Berlin, 1956), p. 73.
18 Amelinckx, Bontinck, DeKeyser, and Seitz, Phil. Mag. 2, 355
16 A. D. LeClaire, Phys. Rev. 93, 344 (1954). The author is (1957).
indebted to F. S. Ham for discussions of this problem. 19 J. Weertman, Phys. Rev. 107, 1259 (1957).

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF Si CRYSTALS FREE FROM DISLOCATIONS 471

FIG. 21. Climb of dislocations following simultaneous diffusion of copper and gold at 1200°C. Dislocations in screw orientations form
tightly wound spirals parallel to (110) orientations. A series of prismatic loops probably formed from a glide dislocation can be seen near
bottom. Large loops parallel to (Ill) plane probably had a different origin: they may be stacking faults bounded by partial dislocations.
Diameter of large loops is about 50 1'.

to be oriented parallel to this axis cannot be cast off by possible that gold ions change the normal vacancy
climb. concentration. 2! Further experimental results are
A number of observations illustrate these arguments. necessary to clarify the detailed mechanism.
a. Long undeviated dislocations parallel to the For the purposes of the present investigation, several
growth axis are very common in [110J crystals, as observations can be made from the specimen shown in
can be seen in Figs. 4 and 15(b). Copper decoration Fig. 21: Helices with strictly (110) orientations are
of deformed crystals demonstrated that many such formed. From the analyses by Amelinckx et at.ls and by
dislocations in a typical [llOJ crystal were in the Weertman!9 they were formed from nearly pure screw
screw orientation. dislocations. The sequence of prismatic loops seen on
b. Tweetl 6 has found helical dislocations with axes edge near the bottom of the photograph was apparently
aligned parallel to the growth axis in [110J ger- formed from a glide dislocation, since their Burgers
manium crystals grown in a high temperature vector is (110). The large loops lie on {1ll} planes;
gradient and then brought to room temperature these presumably have a different origin. They may be
over a period of about 10 to 15 minutes. regions of stacking fault surrounded by partial
c. A qualitative idea of the effect of an excess of point dislocations.
defects on the configuration of dislocations in sili- Prismatic loops can be seen in the process of forming
con has been obtained by simultaneous diffusion from a glide dislocation in Fig. 22. Successive loops are
of gold and copper into a specimen at about being cut off from a helix as a result of inter-turn repul-
1200°C. 20 The mechanism is not understood. It is sion and expansion of the helix with respect to the
remainder of the dislocation. As the loops are cut off,
20 A preliminary account of these studies was presented at the
Chicago Meeting of the American Physical Society, March 27, their spacing increases because of mutual repulsion. A
1958. See Bull. Am. Phys. Soc. Ser. II, 3, 106 (1958). A discussion
of the mechanism of generation of prismatic loops from a glide 21 F. A. Kroger and H. J. Vink D. Phys. Chem. Solids 5, 208
dislocation was published by William C. Dash, Phys. Rev. Letters (1958)J discuss the effect of impurities upon vacancy concentration
1,400 (1958). in solids.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 472 WILLIAM C. DASH

locations during growth is shown in the specimen in

Fig. 23. The long, smoothly curved lines in the center
and on the right-hand side are confined largely to {111}
planes. These were probably generated by glide several
hundred degrees below the melting point. In contrast,
those on the left-hand side show considerable evidence
of climb. The zig-zag lines near the arrow are shown in
the inset at higher magnification. They are irregular
spirals with sections which are primarily edge alter-
nating with sections which are primarily screw. The
circumstances that led to the formation of these con-
figurations are not known in detail. The curvature of
the edge portions is indicative of a vacancy supersatu-
ration of the order of 10-4, as computed from the
Bardeen-Herring climb formula22 for an edge dislocation
with a radius of 1 mm. The interpretation of this
curvature quantitatively in terms of supersaturation is
questionable, however, since steady-state conditions
were not maintained.
VI. A PROCEDURE FOR GROWING DISLOCATION-
FREE SILICON CRYSTALS
The results discussed in the preceding section suggest
FIG. 22. Generation of prismatic loops from a glide dislocation that by proper selection and preparation of the seed
following simultaneous copper and gold diffusion at 1200°C.
Distance between outermost loops is 150 p.. and by establishing growth conditions which favor the
elimination of residual dislocations by climb, crystals
largely free from dislocations may be grown. A proce-
process such as this could account for the very regularly dure for achieving this in crucible- or pedestal-grown
spaced prismatic loops in Fig. 21. crystals is outlined below:
In summary, an excess of point defects tends to lock A seed crystal as free as possible from dislocations is
a screw dislocation parallel to the Burgers vector, desirable; a perfect crystal is not essential. Mechanical
whereupon it can propagate undeviated a long distance. operations at room temperature required to prepare
With even a slight helicoidal shape it becomes difficult the seed do not affect the bulk perfection. 2 Surface
to move by glide. If the growth axis is not parallel to damage, including relatively severe cracks from saw
the Burgers vector, the screw will eventually grow out cutting, is readily removed by careful etching with
to the surface. (1:3:6) followed by a chemical polish with (1:3:0). A
In one specimen with a [100J growth axis a single small diameter over all or a long taper to a small
straight [100J dislocation over 10 cm long was found. diameter at the tip of the seed is desirable in order to
In several other [100J crystals a few dislocations have minimize thermal stresses upon splicing and to mini-
persisted under conditions which favor climb. In order mize the total number of dislocations at the interface.
to be consistent with the ideas expressed here, the A (111) orientation is preferred, although in most cases
Burgers vector of these dislocations would have to be a (100) orientation is also satisfactory. .
[100]. However, further information is required in When the seed is inserted into the melt, suffiCient
order to be certain that the Burgers vector of the dis- time and a high enough melt temperature should be
locations is [100]. No preferential alignment parallel provided to assure complete melting of the wetted
to the growth axis has been found in [111J or [112J surface and therefore removal of possible contaminants
crystals grown under conditions which favor a vacancy which might generate dislocations.
excess. The diameter of the crystal during its initial growth
The usual behavior encountered in crystals grown should be kept as small as possible to provide a maxi-
with an axis other than (110) is a continuous decrease mum opportunity for dislocations to grow out. This
in dislocation concentration with growth if excessive requires that the temperature of the melt be kept
plastic generation is avoided. This was mentioned relatively high at first. The growth of the initial small
earlier in connection with the twinned specimen de- diameter portion is maintained for a distance equal to
scribed in Sec. III. Many of the dislocations below the several times its diameter.
twin plane in that specimen were grown out to the The melt temperature can then be lowered to increase
surface in a short distance, as seen in the lower part of
Fig. 4. 2'.l J. Bardeen and C. Herring, Imperfection~ in Nearly Perfect
Crystals, edited by W. Shockley et al. (John Wiley and Sons, Inc.,
Another illustration of the gradual removal of dis- New York, 1952), p. 261.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 GROWTH OF 51 CRYSTALS FREE FROM DISLOCATIONS 473

the diameter of the crystal. No fundamental limitation ~OTO:RAPHEDAREA

~;
on maximum diameter is expected; none has been found
with the largest crystals thus far grown, namely SO [IOOJ-- :>
grams in weight with a maximum diameter of 2.2 cm. lem
When the initial stages have been carried out properly, (a)
the perfection is insensitive to fluctuations in diameter
or to other details of further growth.
The specimen shown in Fig. 2 is a typical example of
a dislocation-free crystal grown from a crucible. Numer- FTG. 23. IDisloca-
ous [100J and [111J crystals, as well as one with [110J tions in a [100J speci-
men viewed in the
orientation completely free of dislocations over at least [Ol1J direction. A
90% of their length have been grown from a crucible in gradual improvement
this way. in perfection occurs
with continued
Many [100J and [111J crystals grown by the pedestal growth. The long,
method have likewise been found to be free from dis- smoothly curved lines
locations over most of their length. The orientation is were probably formed
by plastic flow sev-
especially important in this case because twins caused ral hundred degrees
by poor epitaxy around the periphery of the seed com- below the melt tem-
perature. The zig-zag
monly occur, but are easily cast off when the axis is not lines near the arrow,
parallel to a {111} plane. shown at higher mag-
Several Ge crystals free of dislocations have been nification in the inset,
are composed of sec-
grown by a direct application of the above procedure. tions which are alter-
A CP-4 etch on {111} planes was used to determine nately primarily edge
the presence of dislocations. 23 This is not quite as satis- and primarily screw.
This is believed to
factory as a decoration process, since only the surface indicate climb and
can be examined. However, Tweet's16 results on Ge glide in the region of
the growing inter-
crystals grown under similar conditions give confirma- face. Width of region
tion that an accurate determination of dislocation shown in inset is
density can be made. There can be little doubt that the about 0.4 mm.
qualitative behavior of dislocations during growth of
both germanium and silicon is the same.
VII. DISCUSSION
At the time of writing, approximately 20 silicon
crystals free from dislocations of detectable size have
been grown from a quartz crucible and a comparable
(b)
number by the pedestal method. As mentioned earlier,
dislocation loops less than 2 microns in diameter would
escape detection by the methods used. Although the Aggregation of vacancies might occur at places suffi-
growth conditions (growth rate, temperature gradient, ciently removed from dislocations and from surfaces.
rotation, etc.) differ greatly between the two methods, The randomly distributed specks of copper found in
these findings fit a comparatively simple pattern. With the center of the pedestal-grown specimen in Fig. 20(a)
the exception of the external effects discussed above, might have been nucleated by clusters of vacancies
plastic deformation accounts for the generation of all of which were too far from the surface to enable them to
the dislocations in a crystal grown as described. Climb
be removed. Some etching properties of the Ge crystals
caused by vacancy supersaturation modifies the dis-
tribution and provides a mechanism for eliminating grown by Tweetl 6 have a similar geometrical distribu-
dislocations. The seed preparation and initial growth tion to the copper specks in Fig. 20(a).
must be carried out in a way which minimizes plastic When all of the dislocations have been eliminated,
flow and removes the residual dislocations. When this there is no longer any advantage in maintaining growth
is done, the remainder of the crystal is free from dis- conditions which favor climb. There may in fact be
locations despite considerable changes in growth vari- disadvantages if aggregation of vacancies is undesirably
ables. The withdrawn and re-inserted specimen in large. For this reason it may be necessary at this stage
Fig. 16 demonstrates that a crystal entirely free of dis- to use an after-heater and to cool slowly to room tem-
locations can be grown from a clean, dislocation-free perature. This would tend to remove a large fraction of
seed.
the excess vacancies which otherwise might aggregate
23 Vogel, Pfann, Corey, and Thomas, Phys. Rev. 90, 489 (1953). internally.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 474 WILLIAM C. DASH

In view of the success of Frank's24 screw dislocation VIII. SUMMARY

mechanism for the growth of crystals, we may consider The sources of dislocations in all phases of the growth
briefly its applicability to the present case. Frank's of silicon crystals by the Czochralski method have been
mechanism satisfactorily explains growth from vapor investigated by etching and copper decoration tech-
or solution under quasi-equlibrium conditions, where a niques. Poor epitaxy and surface damage in the seed
supersaturation of 1% or less is sufficient to make are important sources at the commencement of growth.
ordinary crystals grow. The step formed on the surface In the bulk of the crystal only plastic generation forms
where a screw dislocation emerges provides a self- dislocations which propagate by growth. No prismatic
perpetuating edge where atoms can be accommodated loops of detectable size formed by vacancy aggregation
without the necessity of two-dimensional nucleation are found. Segregation of some impurity, possibly
each time a layer is completed. In growth from the melt silicon carbide, in the form of dendrites has been found
by the methods used in this investigation, near-equilib- at the tip of most crystals. When no grown-in disloca-
rium conditions are not maintained; a substantial tem- tions are present, the dendrites have no effect on the
perature gradient exists in the solid. The interface can bulk perfection.
become supercooled by whatever amount is required to If sufficient care is taken in the selection and prepa-
produce two-dimensional nucleation. It would be of ration of the seed and in the initial stages of growth,
interest to compare the temperature of the solid-liquid the growing crystal will contain only those dislocations
interface during growth of a dislocation-free crystal and which have propagated by growth from the seed.
of crystals with different numbers of dislocations. The Removal of these can be achieved by growing a small
effect of orientation on interface temperature would diameter portion wherein conditions which favor climb
also be worthy of study. are maintained. Under these conditions, an edge or
The result of the present experiments can be helpful mixed dislocation is readily eliminated. A screw dis-
location will become locked parallel to its axis and may
at least as a guide for improving the perfection of other
form a helix in a sufficient vacancy supersaturation. To
crystals. Surface damage has been shown to be an
cast off screws it is important that all screw orientations
important source of dislocations. This can be much make a finite angle with the growth axis. (100) and (Ill)
more troublesome in crystals which deform readily at axes are preferred for this purpose. When all dislocations
room temperature. Great care in handling the seed and have been removed, further generation does not occur
the grown crystal is essential. The growth conditions despite considerable variation of growth conditions.
required for the success of the climb mechanism of Germanium crystals have been grown free from dis-
eliminating dislocations will vary from one material to locations by a direct application of the procedure out-
another. For some substances, it may be necessary to lined for silicon. It is believed that the procedure will
grow at a slow rate with a low thermal gradient and prove to be of value for improving the perfection of
other crystals as well.
small diameter; this would prevent too large a super-
saturation of vacancies and subsequent generation of ACKNOWLEDGMENTS
prismatic dislocations if such a process were found to The author is greatly indebted to F. Hubbard Horn,
be important. In crystals in which partial dislocations who provided the author with a large number of silicon
are found, the climb mechanism may be ineffective be- crystals over a period of several years. He is indebted
cause a large surface energy of misfit would be created if also to L. J. Keifer, R. F. Reihl, and R. Zmijewski, who
if they were to move out of their glide planes. 22 However, grew and prepared many other specimens for this
for many substances it might be possible to use the pro- investigation.
cedure outlined for silicon to grow dislocation-free The x-ray investigation of dislocations in silicon by
crystals with only such modifications as are required by A. R. Lang of Harvard University and the electron
dislocation geometry. Germanium is an example of a microscopy by A. S. Holik are gratefully acknowledged.
The author also wishes to thank J. W. Mitchell for
crystal for which no modification is necessary.
valuable discussions and criticism of the manuscript.
U For a discussion of this mechanism see Burton, Cabrera, and
W. W. Tyler and A. G. Tweet also contributed very
Frank, Phil. Trans. Roy. Soc. (London) A243, 299 (1951). helpful criticisms of the manuscript.

Downloaded 03 Aug 2013 to 160.94.45.157. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions