Defect formation and evolution in high dose oxygen implanted silicon


Material Information

Defect formation and evolution in high dose oxygen implanted silicon
Physical Description:
vii, 244 leaves : ill. ; 29 cm.
Venables, David, 1959-
Publication Date:


bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1992.
Includes bibliographical references (leaves 237-243).
Statement of Responsibility:
by David Venables.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001806291
notis - AJN0124
oclc - 27804173
System ID:

Full Text








Dedicated to my parents,

Dr. John D. Venables


Kathryn E. Venables


I would like to extend my sincere thanks to my advisor,

Dr. Kevin S. Jones for his advice and encouragement and for

allowing me the freedom to find my own way. Thanks are also

due to my doctoral committee members, Dr. Robert DeHoff, Dr.

Paul Holloway, Dr. Robert Park and Dr. Robert Fox for their

support and encouragement.

Dr. Fereydon Namavar of SPIRE Corporation generously

supplied samples in support of this research and shared his

knowledge of the subject. Dr. Stan Bates enthusiastically

provided assistance in cracking the mysteries of "high

resolution" x-ray diffraction, and Dr. Augusto Morrone was

always available to help with the electron microscopes,

whether high resolution or otherwise. I extend my gratitude

to each of them.

To Peter Lowen, Steve Rubart, David Hoelzer and my

other friends and coworkers I am also indebted for making my

stay in Gainesville more pleasant. My thanks go especially

to Wishy Krishnamoorthy for his friendship, help and

encouragement and for the many lively discussions and good

times. No one could ask for a better friend.

Last, but by no means least, I wish to thank my

parents, John and Kathryn, and my brothers, Lee and Jess,


for their loving support, encouragement and care. Their

continued faith in me has been a source of strength and

comfort in sometimes difficult times. Finally, I wish to

acknowledge my father for setting standards of scientific

excellence and ethics that will always be my goal to







Motivation and Objectives .
Background . .
Scope of the Present Work .


Implantation and Annealing .
Analytical Methods . .


Strain Measurements . .
Microstructural Observations .
Discussion . .










. 1

. 22

. 26
. 26

S. 3026
. 26

. 69

. 70
. 95
. 151

. 189

. 206




Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



David Venables

August 1992

Chairman: Kevin S. Jones
Major Department: Materials Science and Engineering

The effect of dose and annealing temperature on the

strain state and defect microstructure of oxygen implanted

silicon were investigated by high resolution x-ray

diffraction (HRXRD) and cross-sectional (XTEM) and plan view

(PTEM) transmission electron microscopy. The dose was

varied from 1x1016 to 9x1017 cm-2 and the annealing

temperature ranged from 700 to 13000C.

HRXRD results indicated that two strain layers existed

in the as-implanted samples. At the lowest dose a

unidirectional lattice expansion perpendicular to the

surface occurred as a result of excess implantation-induced

self-interstitials and interstitial clusters in the vicinity

of the projected range. The unidirectional nature of this

expansion was destroyed with increasing dose and with

annealing as dislocations, stacking faults and SiO,

precipitates formed in the buried layer.


For doses of lxl0"7 and 3x10i7 cm-2, an association of

lattice vacancies and oxygen atoms in the near-surface

region gave rise to a unidirectional lattice contraction at

the surface. Annealing at 700 to 9000C reduced the

magnitude of the lattice contraction as oxygen containing

cavities developed in the surface layer. Large prismatic

1/2<110> dislocation half-loops that formed as a result of

primary ion damage and the stress state of the surface layer

annealed out by 10500C as the cavities filled in with SiO2.

At higher doses the oxygen precipitated as SiO, and was

no longer associated with vacancies in the surface layer of

as-implanted samples. The precipitates provided an

interstitial source for extrinsic Frank loops, which

apparently unfaulted upon annealing at 210500C. The

remaining dislocations were pinned by the precipitates

resulting in a threading dislocation (TD) density of

-108 cm-2 after 13000C annealing. It was concluded that

controlling near-surface precipitation is the most important

factor in reducing TD densities.

These results led to the development of a multiple

3x1017 cm2 implant, low temperature (9000C, 0.5 hr)

annealing procedure designed to prevent near-surface

precipitation by ensuring that oxygen is accommodated by

cavities. After a final high temperature anneal the TD

density was reduced from 108 cm-2 for a comparable single

implant to less than the PTEM detection limit of 105 cm-2.



Motivation and Objectives

Ion implantation has played a significant role in the

development of silicon based electronics technology,

primarily as a means of selectively doping areas of the

substrate. This important application of ion implantation

is largely limited to low ion doses (1011 cm-2 to 1016 cm-2)

and room temperature implants. In the past decade, however,

ion implantation has been applied in quite a different

context--that of compound synthesis by very high dose ion


Although the field of ion beam synthesis is relatively

new, a variety of systems have been studied in an effort to

produce electrically conducting [1,2], semiconducting [3] or

insulating [4-6] buried layers. The overwhelming research

emphasis has been on high dose implantation of oxygen into

silicon to form a buried layer of electrically insulating

SiO2. This silicon-on-insulator (SOI) technology is known

by the acronym SIMOX (Separation by Implanted Oxygen)[6].

The basic SOI structure consists of a thin (-1500 to

2500 A) film of single crystal silicon on an electrically

insulating substrate. The technological motivation for this

kind of structure stems from the expectation of enhanced

performance of devices fabricated in the thin surface

silicon layer for radiation environments [7,81 and for

ultra-large scale integrated circuit (ULSI) applications

[8,9]. These advantages include radiation hardness for

aerospace applications and immunity from latch-up and

elimination of the short channel effect which allows for

greater packing densities and faster devices for ULSI

applications [7-9]. Although there exist a number of

alternative SOI technologies (e.g., silicon-on-sapphire

(SOS), zone-melt recrystallization (ZMR) and wafer bonding),

SIMOX has rapidly emerged as the most promising means of

realizing a viable SOI technology [9,10].

The SIMOX process typically consists of implanting

oxygen at about 150 keV to a dose of up to 2x101 cm-2 at

elevated temperature. Amorphous, stoichiometric SiO2

precipitates around the ion projected range beneath a

damaged, but still monocrystalline surface layer of silicon.

After high temperature annealing (-13000C), the SiO2 forms a

continuous, planar buried layer overlayed by a thin (-1500

to 2500 A) layer of single crystal silicon. The primary

disadvantage of SIMOX, until recently, has been the high

density (-109 cm-2) of threading dislocations that remain as

residual defects in the surface layer where the devices are

to be fabricated.

Dislocations have a deleterious effect on the

performance of minority carrier devices such as bipolar

junction transistors by reducing the minority carrier

lifetime and acting as a leakage path [11-13]. It is

therefore of technological interest to determine how to

avoid their formation. Recent innovative processing schemes

have been found to produce low threading dislocation

densities [14,15]; however, there is no clear understanding

why these methods are successful. It is therefore difficult

to make improvements on them or to develop new methods that

are more commercially viable.

The enormous amount of research devoted to the

traditional low dose, dopant implants in silicon has led to

a good basic understanding of implantation-induced defects

for low doses. Efforts to understand the mechanisms by

which threading dislocations form in SIMOX have been

hampered by a lack of such fundamental knowledge regarding

elevated temperature, high dose implants. Much of the

research that has been done in this area has focused on

very high doses where the microstructures are decidedly

complex and correspondingly difficult to interpret. In

addition, the evolution of the defect microstructure as a

function of annealing temperature is poorly understood

because of a heavy emphasis on very high temperature


In view of these considerations, the overall objective

of this investigation was to develop a more fundamental

understanding of defect formation and evolution in high dose

oxygen implanted silicon with the specific objective of

developing an improved understanding of the mechanisms by

which threading dislocations form in SIMOX. These goals

required that the gap between our considerable knowledge of

dopant implants (low dose, room temperature) and our

relative lack of knowledge about ion beam synthesis (high

dose, high temperature) be bridged. Accordingly, the effect

of increasing dose from the upper end of the dopant regime

(1016 cm2) to well into the ion beam synthesis regime (9x1017

cm-2) was investigated. In addition, the evolution of the

defect microstructure at each dose was investigated for low,

intermediate and high temperature anneals. This led to an

improved understanding of the mechanisms by which threading

dislocations form. This in turn led to the development of a

new process for producing low dislocation density SIMOX, and

to proposed improvements in both the new method and in

existing methods.


In this section the relevant literature on

implantation-induced defects will be reviewed. After a

basic introduction to the ion-implantation process, the

secondary defects produced in low dose implants will be

reviewed. For high dose implants the solid solubility of

the implanted species is exceeded, and therefore the

precipitation of oxygen in bulk silicon is considered next.

Finally, the relevant literature on defect formation in high

dose oxygen implants is discussed.

Ion Implantation Fundamentals

When energetic ions strike an elemental solid, two

basic processes occur within the solid [16]: implantation

and damage production. The impinging ion penetrates some

distance below the surface before coming to rest because it

loses energy in a series of collisions with lattice atoms

(nuclear stopping) and electrons (electronic stopping). The

implanted ions are usually distributed about a mean range

(depth) in a Gaussian like form [17]. The mean range

depends chiefly on the ion mass and kinetic energy and the

mass of the target species. Lighter and/or more energetic

ions penetrate further than heavier or less energetic ions.

Damage is produced in the target as a result of the

energy transferred to target atoms involved in elastic

collisions with the moving ion [16]. The displacement

energy of a silicon atom is about 15 eV, so, in a simple

collisional model, it is evident that a single ion with keV

energy can displace many lattice atoms before coming to

rest. Moreover, many of the displaced target atoms have

sufficient kinetic energy to displace additional atoms, thus

creating multiple-collision cascades. Sufficiently intense

damage of this kind can lead to amorphization of an

originally crystalline target. In this respect, however,

light and heavy ions produce different effects because light

ions suffer predominantly electronic losses and heavy ions

lose more energy through nuclear collisions [18].

Amorphization of silicon by heavy ions occurs by the

process of direct-impact amorphization. Individual heavy

ion tracks create discrete amorphous zones which begin to

overlap as the dose increases. Eventually, when a critical

dose is reached, this process results in a continuous

amorphous layer. This layer may be either a surface or

buried layer, depending on the total dose. Since simple

collisional theories cannot account for the large size of

the discrete zones, displacement and thermal-spike models

for the damage process have been developed. In contrast,

light ions in silicon only produce small, discrete defect

clusters of a size consistent with simple linear cascade

models. Amorphization by the accumulation and overlap of

regions of discrete defects is a much less efficient

process. Thus, light ions require a larger dose to induce

amorphization than heavy ions [18].

The degree to which amorphization occurs is highly

dependent on the temperature of the target. Competition

between the damage production rate and thermal self-

annealing effects determine the final defect type and

density. In this regime, the ion dose rate becomes an

important parameter because it essentially determines the

damage production rate. At sufficiently high implant

temperatures amorphization can be completely suppressed,

although primary ion damage still exists. The high-dose

oxygen implants used in SIMOX processing are usually

performed at 5000C to avoid amorphization.

The details of nuclear and electronic stopping have

been sufficiently well modeled that it is possible to

calculate ion range profiles and deposited energy profiles

for given ion/target combinations with reasonable accuracy.

Brice [19] employed a statistical model to tabulate

normalized range and damage density distributions for a

variety of ion/target combinations. Several Monte Carlo

simulation computer programs (TRIM [17] and MARLOWE [20,21])

are available that do not employ Brice's statistical

approximations and are therefore more realistic. These

techniques build up profiles by following one ion at a time

and summing over the results of many such one ion histories.

They are therefore quite time consuming. The TRIM code

assumes the target atoms are distributed randomly and

therefore cannot predict crystal orientation effects such as

ion channeling, which can significantly broaden range and

damage profiles [22]. However, implants of crystalline

samples that are tilted off of major channeling orientations

show good agreement with TRIM calculations [17].

None of these methods account for chemical effects

associated with precipitation of the implanted species so

caution must be exercised in their application to medium and

high dose implants [23]. They are not at all applicable to

very high dose implants such as those used in stoichiometric

dose SIMOX where a continuous buried layer is synthesized.

Low Dose Implantation

The primary damage produced by ion implantation can

evolve into a wide variety of extended defect configurations

upon post-implantation annealing. For low dose implants

into silicon, the many aspects of secondary defect evolution

have been the subject of extensive research. Recently,

Jones et al. [24] suggested a unified classification system

that places secondary defects into five basic categories

that distinguish between the origins of the implant damage.

Type I damage arises when an amorphous layer does not form

and so is typical of light ion implants at room temperature.

Extended defects that evolve upon annealing are usually

extrinsic dislocation loops that are thermally stable. When

an amorphous layer is produced by the implantation, Type II

damage forms adjacent to the amorphous/crystalline

interface. Annealing typically produces a discrete layer of

extrinsic dislocation loops that are thermally stable under

most conditions. Imperfect regrowth of the amorphous layer

by solid phase epitaxy during post-implantation annealing

gives rise to Type III defects. These may take the form of

"hairpin" dislocations or microtwins. If the amorphous

layer was originally buried, then regrowth occurs at both

the top and bottom interfaces. Type IV defects form at the

depth where the two advancing interfaces meet. They take

the form of large dislocation loops. Type V defects form

when the solid solubility of the implanted ion in silicon is

exceeded. They usually take the form of precipitates, but

dislocation loops that evolve into half-loop dislocations

have also been observed.

Type I and Type V defects are of particular interest

for SIMOX implants since amorphization, and thus

amorphization-related defects (Type II, III and IV) are

often suppressed by the use of high implant temperatures.

For low-dose implants, Type I defects can take a variety of

forms. For room temperature ion implants the as-implanted

morphology typically consists of point defect clusters.

High resolution x-ray diffraction [25,26] and wafer

curvature [27] measurements show that these defects cause a

net lattice expansion, indicating that the region is

interstitial dominated. Upon annealing, these clusters

evolve through a series of intermediate defect

configurations, resulting eventually in the formation of

dislocation loops, dipoles and, if the damage level is

sufficiently high, into a dislocation network. Jones et al.

[24] have suggested that dislocations will form only when a

critical dose has been exceeded. However, recent Rutherford

backscattering spectroscopy measurements [28] have shown

that the appropriate criterion for dislocation formation is

attainment of a critical number of displaced silicon atoms

integrated over the depth of the implant. This number does

not vary with ion species, at least for heavy ion implants

into silicon. Once the critical integrated number of

displaced atoms has been exceeded, the dislocation density

increases with increasing dose.

The form of the primary ion damage changes for elevated

temperature (5000C) irradiation with protons and electrons

[29-31]. The defects consist of faults on {113} planes;

however, the nature of these defects has been subject to

dispute [32-38]. They are thought to consist either of

silicon interstitials [34-36] or of coesite [32,331, a high-

pressure phase of silica. They become unstable upon

annealing above 8500C.

The nature of Type V damage clearly depends on the

implanted species. The defects formed as a consequence of

exceeding solid solubility include precipitates involving

the implanted species and often dislocations and stacking

faults associated with the precipitation reaction. Since

high-dose SIMOX implants are specifically designed to ensure

that oxygen exceeds solid solubility, the next section will

review the oxygen precipitation process in bulk silicon.

Oxygen in Silicon

Since oxygen is the most common contaminant in

Czochralski-grown silicon, its solubility, diffusivity and

precipitation behavior in bulk silicon have been extensively

studied. Oxygen in solid solution normally occupies an

interstitial site [39] and the maximum solubility [40] is

about 2xl018 cm-3 at 14100C, the melting point of silicon.

Single stage annealing treatments of crystals pulled from

the melt have shown that the primary precipitation path

depends on the temperature range used [41,42]. In the

temperature range 4850C to 7500C, rod-like defects occur

which are similar to the {113) defects in electron-

irradiated samples. They are thought to be coesite, a high-

pressure phase of silica. However, this interpretation has

been disputed and an alternative explanation on the basis of

silicon interstitial condensation has been proposed, as

discussed above. At 6500C to 10500C platelike precipitates

occur on (100} planes, and this is the dominant

precipitation path. They are square-shaped and apparently

consist of amorphous SiOx, with x close to 2. At higher

temperatures precipitation does not occur because of the low

oxygen supersaturation unless nuclei have already been

formed with a first-stage anneal at lower temperature.

These two-stage anneals produce larger octahedral shaped

precipitates of amorphous SiOx. The facets occur along

{111} and {100} planes and there is no strain field

associated with the precipitates.

Accompanying the precipitation of SiOx in either form

are a variety of extrinsic defects [41,42]. At lower

temperatures (56000C), distorted extrinsic Frank loops,

known as "loopites", appear on {111} planes in the vicinity

of the precipitates. At higher temperatures (up to 9500C),

faulted Frank loops with b=1/3 nucleate on the

precipitates. Prismatic punching of small perfect loops

also occurs for temperatures above 9000C.

The presence of these extrinsic defects has been

explained on the basis of a strain accommodation mechanism

for the growing SiO, precipitates. The molecular volume of

SiO2 is 2.25 times the atomic volume of silicon so that the

precipitation of SiO2 involves a large volume change and a

correspondingly large strain energy. Hu [43,44] proposed

that this strain could be relieved by the ejection of excess

silicon atoms as interstitials. This implies that for every

oxygen atom that joins an SiO2 precipitate, 0.625 silicon

interstitials must be emitted to totally relieve the strain.

A quantitative comparison of the number of interstitials

bound by extrinsic "loopites" and the number of molecules of

SiO2 formed provided experimental verification of this

mechanism [32,41]. Thus, high oxygen concentrations lead

not only to the formation of amorphous SiO2 precipitates,

but also to extrinsic extended defects in the silicon


High Dose Oxygen Implantation

High dose oxygen implantation as a means of producing

SOI structures has been the subject of much research since

Izumi et al. [6] demonstrated that a buried oxide layer

could be produced and that the surface silicon layer could

support MOS devices. The explosion in SIMOX research that

occurred in the mid-1980's has identified the

microstructural effects of many of the key variables, such

as implant temperature, dose rate and anneal temperature.

Much of the research has concentrated on very high,

stoichiometric doses. A stoichiometric dose corresponds to

the peak of the implanted oxygen profile reaching 66 at.%,

i.e., the composition of SiO2 For a 200 keV implant this

requires a dose of 1.2x1018 cm-2 [10]. In general,

stoichiometric doses at energies in excess of 100 keV result

in the formation of a buried layer of stoichiometric SiO,

overlayed by a defective, but still single-crystal, surface

silicon layer. Both the surface silicon layer (SS) and the

buried oxide (BO) undergo drastic changes upon post-

implantation annealing. The nature of the final

microstructure, however, is highly dependent on the as-

implanted morphology, which in turn is determined by the

implantation conditions.

The implant temperature was found to have the most

pronounced effect on the microstructure of the implants. In

general, the as-implanted morphology falls into three basic

categories which depend on the implant temperature: 1) a

more or less random distribution of amorphous oxide

precipitates extending through a defective, but single-

crystal surface layer for implant temperatures between 450

and about 5500C [45-49]; 2) a multilayer structure

consisting of amorphous silicon sandwiched between defective

single-crystal silicon and the buried oxide for temperatures

below 4500C [50-53]; and 3) for temperatures above 5500C a

distribution of cavities in the near-surface region [15,


The most frequently studied implantation conditions

have been stoichiometric dose implants with an implant

temperature between about 450 and 5500C. The as-implanted

morphology consists of a continuous buried oxide at the

projected range while in the surface silicon layer, numerous

amorphous oxide precipitates form, with precipitate size and

number density increasing from top to bottom in the layer

[49]. This graded size distribution parallels the

increasing oxygen concentration with depth in the layer. As

Stoemenos and Margail [45] point out, both the nucleation

and growth rates of homogeneously precipitated oxide should

depend strongly on the oxygen concentration and thus, this

gradation of size and number density with depth is scarcely


Post-implantation annealing of this structure leads to

significant rearrangement of the oxygen profile [46,48].

The small oxide precipitates in the SS layer dissolve while

the buried oxide thickens. At intermediate annealing

temperatures (e.g. 11500C), large oxide precipitates are

observed adjacent to the SS/BO interface with dislocations

running between them [45,46]. Additional dislocations

extend to the surface of the sample. At higher annealing

temperatures, all of the precipitates dissolve. The SS/BO

interface sharpens; however, the buried oxide is often

interspersed with "islands" of unoxidized single-crystal

silicon. The threading dislocation density in the surface

layer is typically about 109 cm-2 [55].

Stoemenos and Margail [45] have explained the

redistribution of oxygen in terms of classical nucleation

theory in which a critical precipitate size, re, exists as

a result of a balance between the chemical driving force for

precipitation and the opposing strain energy and surface

energy terms. At 11500C, the smaller precipitates near the

surface are below the critical size and dissolve, with the

oxygen being incorporated into the larger precipitates near

the buried oxide. At higher temperatures, these

precipitates also become subcritical in size and dissolve,

with the buried oxide acting as a precipitate of effectively

infinite radius to which all of the oxygen atoms migrate.

A more detailed model [60] accounts for the effect of

vacancy and interstitial concentrations on the critical

radius. The point defect concentrations influence the

precipitation process, at least at lower anneal

temperatures, by affecting the strain accommodation

mechanism. Interstitial supersaturations suppress

precipitation (large re) because it is more difficult to

overcome the nucleation barrier by emitting interstitials.

Likewise, a vacancy supersaturation will promote

precipitation (smaller r,) by making the strain

accommodation process easier. However, for a very high

temperature anneal (-1300 to 14000C), the point defect

supersaturations rapidly decrease to near-equilibrium values

and all of the precipitates must initially grow during the

annealing stage due to the high interstitial oxygen

concentration in the matrix. Once this level is reduced,

the critical radius becomes large and the precipitates can

dissolve. The initial growth of the oxide precipitates is

accompanied by a high Si interstitial emission rate, which

could act as a source for nucleation and/or growth of

extended defects.

The as-implanted SS layer is very complex and it is

therefore difficult to study the defect microstructure with

the TEM. Krause et al. [61] originally reported the

presence of small dislocation loops in the structure. Later

high resolution TEM studies [49,571 indicated that short

stacking faults, in apparent association with SiO2

precipitates were present and that the defects grew longer

with higher implant temperature. Upon annealing at 10500C

for 2 hours, stacking faults have been observed to extend to

the surface from oxide precipitates in the SS layer [57].

In a different sample, it was observed that only

dislocations were present after an 11500C, 2 hour anneal

[49]. Krause et al. [49] suggest that the presence of a

free surface near the oxide precipitates helps stabilize

these defects and thus is an important factor in defect


The second regime is for implant temperatures below

4500C where amorphization of the silicon occurs. White et

al. [50] studied the effect of implant temperature for a

3x1017 cm'2 substoichiometric dose at 200 keV. Implant

temperatures 5 4000C resulted in the formation of an

amorphous silicon layer which consumed all of the surface

silicon at 250C, and formed an increasingly thinner discrete

buried layer as the temperature increased. At higher

implant temperatures dynamic annealing of the implantation

damage completely prevented amorphization. Annealing at

6000C allowed limited regrowth of the amorphous layer except

where the oxygen concentration was highest, in which case a

layer of microtwins was formed as the solid phase epitaxy

process broke down. Subsequent annealing of the samples

with a microtwin layer at 14000C produced thin, continuous

buried oxide layers with sharp interfaces and no islands of

unoxidized silicon within the buried oxide. Microtwins

remained at the edges of the buried oxide. For samples with

higher implant temperatures, where no microtwin layer was

formed, the buried oxide layers had rough interfaces and

contained numerous islands of silicon. The islands were

attributed to the premature segregation of oxygen as a

result of fewer precipitate nucleation sites in the absence

of an amorphous layer.

If the initial anneal temperature is too high (>

10500C) the amorphous layer recrystallizes as

polycrystalline silicon, rather than as microtwins [50-53].

Tuppen et al. [53] report that for stoichiometric doses,

amorphization is suppressed when the implant temperature is

above 450 to 4750C. The redistribution of oxygen in the

surface layer is similar to the behavior of the 450 to 5500C

implants, and the threading dislocation density after high

temperature annealing is typically about 109 cm2.

The third microstructure type consists of oxygen

containing cavities in the SS layer [15, 54-59]. El-Ghor et

al. [54] have explained this microstructure in terms of

void-formation theories for irradiated metals. Evidently,

the voids can only form over a very narrow range of

implantation temperatures (600-6750C) due to dynamic

annealing effects. The most striking effect of annealing

this structure is that the cavities dissolve and a low

threading dislocation density is obtained. Etch-pit

measurements on stoichiometric dose implants after 13000C, 6

hr annealing have shown that the threading dislocation

density is <104 cm-2. Visitserngtrakul et al. [57] suggest

the dislocation reduction mechanism is related to the

absence of precipitates near the free surface while El-Ghor

et al. [15] suggest it is due to a dislocation blocking

effect of the cavities and their strain-free nature.

These high temperature implants also have several other

unique microstructural features. At the buried oxide

interface with the surface layer, 400 to 1400 A long planar

defects on {111} planes form [58,62]. These have been shown

to consist of a series of discontinuous stacking faults,

both extrinsic and intrinsic in nature. They are postulated

to form by a shear process in response to stresses generated

by the growing buried oxide [62].

The effect of dose on the formation of the near surface

cavities has also been investigated [57-59]. They form at

doses as early as 3x107 cm-2 in the top 1500 A or so of the

surface layer, and increase in size as the dose increases.

Their presence causes substantial surface roughness in the

as-implanted condition; however, the surface becomes planar

again upon annealing at 13000C [59].

In general, the effect of dose rate on SIMOX

microstructures has been somewhat difficult to separate from

the effects of implant temperature because ion beam heating

has often been the sole means of controlling the wafer

temperature. No differences were found for implants at 10

or 1 mA/cm2 [58] where the wafer temperature was partially

controlled by external heating. However, for high energy

implants (300 keV) at very low beam current densities (1.5

PA/cm2) with the wafer in a channeling orientation a very

unusual microstructure was observed [63,64]. The surface

silicon layer contained an ordered array of oxide

precipitates adjacent to the buried oxide. The ordering

consisted of 2 nm precipitates spaced 5 nm apart in a simple

cubic arrangement. Upon annealing, the oxide precipitates

dissolved, and the threading dislocation density was <105

cm-2. This drastic reduction in dislocation density was

explained as arising from the ordered structure confining

any dislocations to the vicinity of the buried oxide, thus

preventing them from threading to the surface. This method

has not been exploited for the production of dislocation-

free SIMOX because commercial implanters must use high dose

rates and are not designed to enable channeling implants.

The effect of dose on the threading dislocation density

has also been investigated. The first systematic study of

sub-stoichiometric doses was by Homma et al. [65]. They

implanted oxygen at 150 keV over a range of doses and then

grew homoepitaxial layers on the surface. Etch-pit

measurements were used to measure the dislocation density as

a function of dose. It was discovered that above a critical

dose of between 4 and 6x1017 cm-2 dislocations formed, while

for lower doses no dislocations intersected the surface.

This result has been confirmed by Reeson et al. [66].

Hill et al. [14] applied this result to a novel method

of making SIMOX. They sequentially implanted and annealed

the wafers with oxygen doses below the critical dose between

13000C, 6 hr anneals. The result was a dislocation-free

(<103 cm2 ) surface silicon layer overlaying a continuous

buried oxide.

Hill et al. [14] tentatively suggested three possible

explanations for the existence of a threshold dose. These

included tensile strain build-up in the wafer with

increasing dose, the existence of a threshold oxygen

precipitate size for punching out dislocation loops and the

build-up of implantation damage in the wafer.

Numerous other researchers have confirmed that this

sequential implant and high temperature anneal method does

indeed produce low threading dislocation densities [67-69].

The disadvantage of this technique, however, is the many

long anneals at high temperature that are required to

produce a single SIMOX wafer. This not only increases the

contamination level of the wafers, but it also makes it

difficult to produce low-dislocation-density SIMOX


Thus, three methods for producing low dislocation

density SIMOX are known; however, at least two of them have

severe drawbacks from a commercial production point of view.

The low dose rate channeling implants require long implant

times, thus reducing wafer throughput, while the multiple

implant method requires many long-time, high-temperature

anneals which may increase the contamination level of the

wafers and also reduce wafer throughput. The high

temperature implants that produce cavities in the surface

layer appear to be the most promising from a production

point of view; however, it is not yet clear whether the

threading dislocations densities are consistently below 104

cm .

Although the existence of a threshold dose for TD

formation has been established, there has been little

systematic work on defect formation and evolution in this

dose range. In addition, the effect of annealing at lower

temperatures has not been extensively investigated, since

the emphasis in annealing studies has been on the

redistribution of oxygen at high temperatures. As a result

there is no clear understanding of how threading

dislocations form, in spite of the fact that several methods

exist to avoid their formation. It is therefore difficult

to make improvements on existing methods or to develop new

methods that may be more commercially viable.

Scope of the Present Work

A review of the relevant literature has shown that in

spite of the existence of several methods to avoid threading

dislocations in SIMOX, there is no clear understanding of

how TD's form, or how these.methods avoid their formation.

As a result, it is difficult to make improvements on the

existing methods or to develop new methods that may be more

commercially viable. This lack of knowledge concerning TD

formation mechanisms stems in part from the very complex

microstructures of high-dose implants and a general lack of

fundamental knowledge concerning defect formation and

evolution in high-dose implants.

An integral part of the approach used in this

investigation was an attempt to study a simplified system

before proceeding to the analysis of the very complex

microstructure of the high-dose implants. The

simplifications imposed were of two basic kinds. First, the

effect of increasing dose on the defect microstructure was

investigated, starting with much lower doses than are used

in ion-beam synthesis. Secondly, it was clear from a review

of the relevant literature that defect evolution as a

function of annealing was poorly understood because of a

heavy emphasis on very high temperature anneals (13000C) in

most previous SIMOX studies. Thus, the effect of annealing

at intermediate and high temperatures at each dose was

studied. This approach also served to help bridge the gap

between our considerable knowledge of secondary defects

associated with low-dose, dopant implants in silicon and our

relative lack of knowledge concerning the high-dose, high-

temperature implants used in ion beam synthesis.

The primary investigation tools were high resolution

x-ray diffraction (HRXRD) and transmission electron

microscopy (TEM). These techniques provided complementary

information about defects in the implanted samples. In so

much as point defects and their clusters produce strain in

the lattice, the HRXRD measurements were able to provide

information about the extent and distribution of these

defects. In addition, the TEM observations were able to

provide detailed information about the configuration and

location of extended defects in the ion implanted samples.

Detailed information about these experimental techniques,

along with the implantation and annealing conditions

employed are given in Chapter 2. Results from a general

investigation of how to correctly interpret the HRXRD

results are deferred to the Appendix.

Chapter 3 presents the results of HRXRD measurements

and TEM observations on the effect of dose and annealing

treatment on the strain state and microstructure of single

implants of oxygen in silicon. A full discussion of strain

relief and defect formation and evolution processes in high-

dose oxygen-implanted silicon is presented in the final

section of the chapter. This discussion includes a

consideration of those factors that are important in

determining the threading dislocation density of SIMOX


This more detailed understanding of defects in SIMOX

led to the development of a new multiple implant/low

temperature annealing process for producing low threading

dislocation density SIMOX. The results of this new process

are presented in Chapter 4, together with a discussion of

possible future work in this area. Finally, the major

conclusions of this investigation are summarized in

Chapter 5.


Implantation and Annealing

All samples used in this study were nominally <100>

oriented single crystal p-type silicon wafers. These wafers

were implanted with 16"0 at 160 keV at a beam current density

of 10 pA/cm2 on an Eaton-Nova NV-160 ion implanter by SPIRE

Corporation of Bedford, MA. The wafer fixture of the

implanter was designed so that the incident ion beam was

perpendicular to the wafer surface which would be,

nominally, a channeling direction for the <100> wafers. The

wafers were bare, i.e. no screen oxide was used.

Single implants were obtained at total doses of 1x1016,

3x1016, 1x101, 3x1017, 6x1017 and 9x1017 cm-2 to investigate

how defect formation and evolution changed as the dose was

increased from dopant-type implant doses (<1016 cm-2) to the

higher doses (21017 cm-2) used in ion-beam synthesis.

Samples to be annealed were capped with 2000 to 5000 A of

Si02 deposited by chemical vapor deposition at about 3500C.

Annealing was performed in a tube furnace under flowing N2

at temperatures of 9000C, 11500C and 13000C for 30 minutes

or 6 hours. Defect evolution was further investigated in

the 3x107 cm-2 samples with 30 minute anneals at 7000C,

8000C and 10500C. Additional multiple implant/anneal

samples were prepared by implanting wafers at 3x017 cm-2,

annealing at the desired temperature, re-implanting an

additional oxygen dose and annealing a second time. This

procedure allowed the manipulation of the surface silicon

defect microstructure and proved to be an important tool for

understanding defect evolution mechanisms in this study.

The ion beam current density was held constant at

10 pA/cm2 for each implant in order to minimize variations

in effects associated with the dose rate. These effects

include the rate of damage production, the rate of internal

oxidation and the equilibrium wafer temperature during the

implant. Since no external source of wafer heating was

used, the wafer temperature was determined entirely by the

latter effect. In view of the importance of implant

temperature in determining the microstructure of high dose

implants (as discussed in Chapter 1), it was necessary to

evaluate the heating rate and equilibrium temperature for

the implant conditions used in this work.

According to Parry [70] the wafer temperature should be

controlled by the input power of the ion beam, radiative

heat losses at the wafer surface and conductive heat losses

to the wafer holder. In the absence of special heat sinking

holders, the first two factors should dominate. Under these

circumstances the simple model of power balance developed by

Smith [71] should be applicable. In this model, the power

density input by the ion beam, Pb/A, is equal to the heat

absorbed by the wafer plus the heat lost by the wafer to

radiative cooling:

Pb/A = pCl dT/dt + 2oa(T4, Ts4) (2-1)

where p is the density of silicon (2.32 g/cm3), C is the

specific heat of silicon (702 J/kg-K), 1 is the wafer

thickness (0.05 cm), dT/dt is the instantaneous heating

rate, a is the Stefan-Boltzman constant (5.67x10-8 W/m2-K4),

E is the "effective" emissivity (taken to be 0.35 for

silicon in a typical implant station), T, is the

instantaneous wafer temperature and T, is the temperature of

the surroundings. Following the suggestion of Parry [70] T,

is taken to be about 1000C. For 160 keV singly charged ions

incident at 10 pA/cm2 the power density is 1.6 W/cm2.

Numerically integrating equation 2-1 yields the plot in

Figure 2-1 of wafer temperature as a function of implant

time and dose. The results show that the equilibrium wafer

temperature was approached to within 60C in less than 45

seconds. The lowest dose in this work was 1xl106 cm-2 which

corresponds to an implant time of 160 seconds. Thus, even

the lowest dose sample was held at the equilibrium

temperature for a significant majority of the implant time.

The calculation also shows that the theoretical equilibrium

temperature was 5350C, which is in reasonable agreement with

the experimentally measured value of about 5000C obtained at

Oxygen Dose (cm-2)
Oxygen Dose (cm )

1013 1014 1015

1 16

100 101 102

1017 1018


Implant Time (sec)

Figure 2-i.

Calcuated wafer temperature rise due to ion-beam
heating as a function of implant time and dose.






100 -



SPIRE Corporation [F. Namavar, private communication]. The

slightly higher theoretical temperature is likely due to

neglecting conductive losses and to the approximate values

used for and Ts in the above equation.

Analytical Methods

A thorough analysis of defect formation and evolution

in the SIMOX process requires knowledge about point defects

and their clusters as well as extended defects. No single

experimental technique can provide detailed information

about each of these defect categories. Therefore, it was

necessary to consider a variety of experimental techniques.

Transmission Electron Microscopy

Extended defects such as dislocations, precipitates and

stacking faults are most effectively studied by transmission

electron microscopy and this was used as a primary

experimental tool. Both cross-sectional and plan-view

specimens were examined.

Cross-sectional samples were prepared by the method of

Jones [24]. Cleaved pieces, 1 cm x 1 cm, were glued to a

sacrificial silicon wafer with Crystal Bond thermal adhesive

and diced with a diamond watering blade into 0.25 mm thick

slices. The slices were then epoxied together with Gatan

G-l epoxy and cured for 5 minutes at 1000C. Prior to this

step, the slices were arranged so that the original wafer

surfaces were facing each other with two additional slices

placed on either side to increase the total width of the

sample. The samples were mounted to a polishing jig with

Crystal Bond and ground on both sides with 5 pm alumina grit

until the edges of the sample began to disappear. A copper

grid, 3 mm in diameter with a 1 mm x 2 mm slot was then

epoxied to the ground samples to provide mechanical

stability. Final thinning was performed by bombarding the

samples with 5 key Ar+ ions in a Gatan ion mill until an

electron-transparent region at the site of the original

wafer surface was obtained. This ion-milling process

typically required 3 to 5 hours before the sample was

thinned sufficiently for TEM.

Plan-view specimens of silicon are usually obtained by

chemical jet polishing from the back side of the wafer.

However, it was found that the presence of a non-continuous

buried oxide layer in lower dose samples complicated this

procedure because of preferential etching. Therefore,

chemical jet thinning was only used until light was visible

through the sample, but no hole was formed. The samples

were then ion-milled from the back side until a hole was

obtained, which usually took 30 to 60 minutes. The jet

thinning was performed with a South Bay Technologies jet

thinner using a 3:1 HNO3/HF solution. The front side of the

samples were protected against chemical attack with a layer

of paraffin wax. The wax was removed in a heptane solution

after chemical thinning. During ion milling, the front

side was protected against contamination with a sheet of

plastic. After the ion milling was concluded, the samples

were ready for TEM examination.

Two electron microscopes were used. A JEOL 200CX with

operating in the TEM mode was employed for the diffraction

contrast studies which constituted the vast majority of the

work. A JEOL 4000FX with a point-to-point resolution of

1.95 A was used for phase-contrast experiments in a few

instances where lattice imaging was required to assist in

identifying defects in the implanted samples.

A variety of imaging techniques were used in the study.

Conventional two-beam bright field imaging allowed for an

overview of the defect microstructure and for Burgers vector

analysis of dislocations. With only two beams excited (the

transmitted beam and one diffracted beam) the contrast from

dislocations and stacking faults in elastically isotropic

materials is well understood from the standpoint of

dynamical theory [72,73]. If R is the displacement vector

of the defect and g is the reciprocal lattice vector of the

operative reflection then the defect is invisible for g.R=0

and visible for all other values of g-R. For dislocations,

R is just the Burgers vector, b, and the contrast consists

essentially of a thick dark line. Stacking faults appear as

alternating bright and dark fringes bounded by partial

dislocations, where R is the fault displacement vector.

This simple principle is the basis for experimentally

determining the displacement vector of defects. In

practice, the images are taken with a slight deviation (s <

1\tg, where s is the deviation parameter and tg is the

extinction distance) from the exact Bragg condition because

this results in a maximum in transmitted intensity.

Overviews of the defect microstructure were obtained

with either gn2 or g400 in the <110> oriented cross-sectional

samples and with gn2 in the <100> oriented plan views.

Burgers vector determinations were accomplished with

standard tilting experiments. At least two cases of

"effective invisibility" of the dislocations were obtained.

This redundancy was necessary because mixed dislocations in

isotropic materials and any dislocations in anisotropic

materials such as silicon (anisotropy factor=1.56) can show

residual contrast when g-b=0 and a judgement must be made of

when the dislocation is effectively invisible.

Threading dislocation densities were obtained from

plan-view specimens using a g220 reflection. Experimentally,

it was found that the dislocations had b such that all were

visible on a g220 reflection. Therefore, no correction for

invisibility was necessary. The lower limit on the

measurable dislocation density depends on the total area

viewed. As a practical matter, this limit is about 10s to

106 cm-2

Weak-beam dark field images of the defect

microstructure were also obtained. This technique provides

a high contrast image of dislocations with vastly improved

spatial resolution compared to conventional bright field

images. With the third-order diffracted beam excited, the

first-order beam is used for centered dark field imaging,

producing a g-3g weak beam image. Provided s-2x10-2 A-i, the

dislocation image is within 20 A of the core with a width of

about 15 A compared to an image width of ~-g/6 (-160 A for

g220 in silicon) for bright field images [72]. The narrower
image allowed closely spaced defects to be imaged.

Voids or cavities in the samples were imaged according

to the rules of Ruhle and Wilkens [74]. They have shown

that in a focused conventional bright field image, cavities

are essentially invisible, but that by under- or over-

focussing a kinematical (i.e., s>>0) image the cavities will

show either bright or dark contrast at the edges. An under-

focussed image taken with a g400 diffraction vector under

kinematical conditions (s=g) was found to produce the best

cavity contrast in the samples in this study. The volume

fraction of cavities in a sample was estimated by measuring

the diameter of each cavity, calculating the volume assuming

a spherical cavity, summing the volumes of each cavity and

dividing by the volume of material in which the cavities

were found.

High Resolution X-ray Diffraction

Radiation-induced point defects and clusters have been

studied by a wide variety of spectroscopic techniques

including infrared spectroscopy, Raman spectroscopy and

Rutherford backscattering spectroscopy. The present work

has pioneered the use of a relatively new technique, high

resolution x-ray diffraction, in the analysis of radiation-

induced point defects in high-dose implants. The technique

allows a three-dimensional analysis of the sign and

magnitude of lattice strain as a function of depth in the

surface layers of a material. In so much as point defects

produce lattice strain, HRXRD should be a sensitive and

powerful tool for analyzing irradiation-induced point

defects and point-defect clusters. In view of this

potential, a considerable effort was devoted to the

application of HRXRD to strain analysis of the ion-implanted


Principles of operation

A rocking curve (RC) is a plot of the diffracted x-ray

intensity around the Bragg angle of a particular

crystallographic plane of a single crystal material. If a

thin surface layer of slightly different lattice parameter

exists, diffraction maxima from both the layer and the

substrate occur. A measurement of the angular separation of

these peaks will provide the lattice parameter difference

through Bragg's law. However, the natural wavelength spread

(spectral divergence) and size of a normal x-ray source is

in general too large to allow these closely spaced peaks to

be resolved. As a result, it is necessary to use one or

more beam-conditioning crystals to collimate and/or

monochromate the x-ray beam.

Traditionally, a single beam-conditioning crystal has

been used to collimate the x-ray beam for the second, sample

crystal. The primary advantages of the high-resolution

five-crystal diffractometer over this traditional

double-crystal method are its improved resolution and its

versatility. Double-crystal systems require that the

reference crystal have the same d-spacing as the sample in

order to obtain good resolution (about 8 arc-sec). This

requires changing the reference crystal for every new

material or reflection of interest. In contrast, the

five-crystal instrument can obtain an RC for any reflection

at any angle of any single crystal material without

modification or realignment of the monochromator with a

resolution of about 5 arc-sec for 0, < 150 degrees. These

attributes arise from the properties of the four-crystal

monochromator, which were first enunciated by Dumond [75].

The most effective way of understanding the operation

of this, or any other monochromator is to use Dumond's

"transparency" diagrams, an example of which is shown in

Figure 2-2. The diagram is a two dimensional plot of

Bragg's Law, nX = 2dsinO,. With d, the plane spacing, and

n, the reflection order, taken as fixed by the reflection of

interest, the x-ray wavelength, X, is plotted as a function

of 8O. However, since every reflection has an intrinsic


low intensity high intensity




i 1

low intensity

0 L --i
0 10 20 30 40 50 60

Figure 2-2. Dumond diagram for a single x-ray reflection.
The area enclosed by the parallel curves
represents the range of 0 and X values were
appreciable x-ray intensity occurs.

angular spread of 08 values given by the full width at half

maximum (FWHM), the plot is of two parallel curves separated

by the FWHM. The area between these curves represents the

range of X and 8B values where appreciable x-ray intensity

is obtained, while outside of the curves the intensity falls

off rapidly. It should be noted that these diagrams are

only schematic in nature and the FWHM is greatly exaggerated

for illustrative purposes.

The action of two or more crystals can be discovered by

superimposing the appropriate Dumond diagrams for the

crystals. In this respect, the orientation of the crystal

faces with respect to each other is important (see Figure

2-3). If the faces are parallel (known as the (+,-)

setting) then an increase in the Bragg angle of crystal 1,

01, will produce an increase in angle, 62, at crystal 2, of

equal magnitude. The Dumond diagrams are then superimposed

with 80 increasing in the same direction, as shown in Figure

2-4a. For the anti-parallel (+,+) case an increase in 81

corresponds to a decrease in 82 so that one of the curves

must be plotted in the opposite way in the superposition, as

shown in Figure 2-4b.

A rocking curve is obtained by rotating one crystal

with respect to the other through a range of rocking angles.

This can be schematically illustrated by sliding one diagram

with respect to the other. The x-ray intensity exiting the

second crystal to a detector is simply the product of the



Figure 2-3. Schematic diagrams of the two possible crystal
arrangements in a double-crystal diffractometer.
a) Parallel (+,-) setting; b) Anti-parallel (+,+)

02 2


0 10 20

30 40 50 60




3: 1
3 1









0- 82
60 50 40

0 10 20 30

6, -

30 20 10 0

40 50 60

Figure 2-4. Dumond diagrams for the two possible crystal

arrangements of a double-crystal system.
a) Parallel (+,-) setting; b) Anti-parallel (+,+)

0 10 20 30 40 50 60


reflectivities of each crystal at each value of rocking

angle, i.e., the mathematical correlation of the two

reflectivity curves. This is represented on the Dumond

diagram by the "window" of overlapping area within the FWHM

of each curve at each rocking angle for the wavelength of

interest. By mentally performing this rocking action with

the diagrams in Figure 2-4 it is immediately apparent that

the parallel setting produces a sharp rocking curve since

the window of overlap occurs only over a narrow range of

rocking angles, independent of wavelength. In contrast, the

window area in the anti-parallel setting persists over the

entire range of rocking angles. The rocking curve would

thus explore the spectral linewidth of the x-ray source,

giving rise to a very broad and diffuse curve. It is for

this reason double crystal diffractometers can only operate

in the parallel, non-dispersive setting. A further

restriction is that the d-spacings of the two crystal

reflections must be similar, otherwise the window of overlap

is broadened and the resolution of the rocking curve is

severely degraded, even in the parallel setting.

The five-crystal diffractometer overcomes these

difficulties by setting the crystals in such a way that both

collimation and monochromation occur. The crystals are

fixed (i.e., not rocked with respect to each other) in the

(+,-,-,+) setting shown in Figure 2-5. Crystals 1 and 2 are

fixed in the parallel setting so that their reflectivity



Source Sample

Figure 2-5. Schematic diagram of the (+,-,-,+) setting of
the four crystal monochromator/collimator.



3 I l I I I

FWHM crystal 1 or 2
....... FWHM of exit beam

overlap window

< low intensity


low intensity

0 I x I I

0 10 20 30 40 50 60

1 -

Figure 2-6. Dumond diagram of two crystals in the parallel
(+,-) setting fixed at the Bragg angle. The
dashed line represents the narrower FWHM of the
exit beam resulting from the superposition of the
two reflectivity curves.

curves completely overlap as shown in the Dumond diagram in

Figure 2-6. As a consequence the output of the first two

crystals is simply the square of their individual

reflectivity curves. Since the reflectivity is always < 1,

the exit beam is reduced in angular width, thus providing

better resolution, but with reduced intensity. Crystals 3

and 4 are also in the parallel setting and behave in a

similar manner to crystals 1 and 2. Monochromation is

achieved by arranging crystals 2 and 3 in the anti-parallel

position. From the Dumond diagram in Figure 2-7 it is

apparent that with this arrangement only a narrow spectral

width, corresponding to the small region of overlap of the

four reflectivity curves, is allowed to pass. When the

fifth crystal, the sample to be analyzed, is rocked with

respect to this fixed arrangement of the beam-conditioning

crystals, the result is a rocking curve of very narrow FWHM

that is nearly independent of the natural linewidth of the

x-ray source and of the Bragg angle. Thus, the four-crystal

monochromator/collimator allows any reflection of any sample

crystal to be analyzed with very high resolution, but at the

cost of reduced intensity.

The Philips high-resolution x-ray diffractometer used

in this study employs two U-shaped <110> oriented Ge blocks

for the monochromator/collimator in an arrangement developed

and patented by Bartels [76,77]. The high intrinsic

reflectivity of Ge minimizes the intensity loss while the

8- 02 and 83

3 -

10 20 30 40 50

01 and 62 -

Figure 2-7. Dumond diagram for the four crystal monochromator
with the crystals in the (+,-,-,+) setting.

<110> orientation allows the diffractometer to operate in

two different modes. When the (440) reflecting planes are

used the resolution is about 5 arc-sec, but the intensity is

very low. A factor of twenty gain in integrated intensity

is possible by using the (220) planes of the Ge crystals

with, however, a loss of resolution to about 12 arc-sec.

For general work, the lower resolution mode was used in this

study. In certain circumstances it was found necessary to

operate in the high resolution mode to resolve closely

spaced diffraction peaks.

In order to obtain a meaningful rocking curve it is

necessary to accurately align the specimen and to optimize

the rotation, R, and tilt, 0, angles. Figure 2-8 shows a

schematic of the diffractometer arrangement. First, the

sample is adjusted along the z-axis by means of a micrometer

so that the surface lies at the center of the diffractometer

axis. The remaining alignment procedures require the

acquisition of a series of rocking curves at different

values of R and 4.

The necessity of determining the appropriate value of R

arises from the likelihood of a misorientation, or miscut,

between the sample surface and the (004) planes, as

illustrated in Figure 2-9. Rocking curves are measured with

respect to the angle of incidence to the specimen surface,

o, not with respect to the Bragg angle. Thus, as the

specimen is rotated, the measured values of peak position,


I--- X'

Detector stage




Figure 2-8. Schematic diagram of high resolution

[001] normal


sample surface

(004) planes

Figure 2-9. Schematic diagram of the effect of wafer
misorientation on the diffraction geometry.

0, will vary according to [78]:

0 = OB + asin(R + Ro) (2-2)

where a is the miscut angle and Ro is arbitrary. At

R+Ro=n7, where n=0,1,2 ... c=0, and the correct Bragg

angle is known.

The situation becomes more complicated if there is an

additional miscut between the strain layer and the

substrate. In this case both the substrate and strain layer

peaks vary sinusoidally with R but are out of phase. The

angular difference, Ac, between peaks is then a function of

R so that the calculated strain appears also to be a

function of R. Distinguishing which is the true value of

the peak separation can be done in two different ways.

First, a series of rocking curves can be used to

evaluate c as a function of R for both the strain layer peak

and the substrate, taking the true value of 0( at the

appropriate R value for each, according to equation 2-2. An

alternative method is to obtain rocking curves at an

arbitrary rotation angle, Ri, and at R+K7. The true values

of C are then just the average of the C values for the two

measurements since sin(R1+Ro)=-sin(R1+Ro0+x) and

Coavg= (0R1+0) /2=08+4a (sin (R+R) +sin (R+R0o+) )

avg=OB+0=8B (2-3)

After finding the appropriate rotation angle, the tilt

angle must be optimized so that the diffracted beam is

centered on the detector window. A series of rocking curves

at different # values are taken until the FWHM of the

diffracted peaks is minimized and the intensity is

maximized. With both ) and R determined, the final rocking

curve can be obtained.

When the angular difference between epilayer and

substrate peaks is small, the x-ray detector can remain

stationary at the theoretical value of 28, and still be able

to measure both peaks because of the detector's relatively

wide window. Rocking curves obtained in this manner are

known as o scans since only c is varied. If, however, the

angular difference is very large, the detector must move as

the scan progresses, producing an o/20 scan. The scans can

be made in either a continuous motion for quick scans or in

increments down to 0.9 arc-sec per step for precision


Experimental procedures

Specimens for HRXRD analysis were cleaved from the

implanted wafers to a size of about 1 cm by 1 cm and were

given a buffered oxide etch to remove any surface oxide

film. They were affixed to glass slides with paraffin wax

and the glass slides were then taped to the diffractometer

specimen holder. This mounting method was thought to reduce

the risk of spurious strain measurements arising from

accidental straining of the wafers during the mounting


The system used Cu K. radiation (X=1.541A) obtained

from a generator operating at 40 kV and 40 mA. Rocking

curves were routinely obtained with the diffractometer

operating at 12 arc-sec resolution to take advantage of the

increased intensity obtainable. The high-resolution mode

was used for some samples with very closely spaced peaks.

The specimens were physically aligned as described above.

Rotation optimizations of the symmetric (004) reflection of

each sample consistently indicated that there was no miscut

or misorientation between the substrate and strain layer,

although the wafers themselves did show a miscut from the

nominal <100> orientation. An experimental example of one

of these R optimizations is shown in Figure 2-10. Since

there was no phase difference between the peaks, rotation

optimizations were obtained by finding the minimum or

maximum value of w as a function of R. The correct R value

was then taken to be at 900 away from this value. The tilt

angle, #, was then varied until the substrate peak exhibited

a minimum FWHM and a maximum peak intensity. Final rocking

curves were then obtained from the symmetric (004)

reflection with the diffractometer operating in the 0 mode.

These rocking curves were used for comparison to the

simulated rocking curves and they allowed for the evaluation

of the perpendicular mismatch.

Asymmetric rocking curves (e.g., (044), (115) and

(444)) at both glancing incidence and glancing exit angles

17 -2
1x1 0 cm as-implanted

"0- 0.04-

-40 0 40 80 120 160


Strain layer peak
O Substrate peak
^ 34.8


a 34.6 ,(O
0 N 10

Figure 2-10. Experimental R optimization curve showing the

-40 0 40 80 120 160

Rotation Angle, R (deg.)

Figure 2-10. Experimental R optimization curve showing the
that the substrate and epilayer are not
misoriented with respect to each other but that
the wafer does have a miscut.

were obtained to evaluate the parallel mismatch component.

Although these also provide the perpendicular component, the

symmetric (004) rocking curves were still necessary for

comparison with the rocking curve simulations. In addition,

the independently obtained values for the perpendicular

mismatch served as a check that the peak position

assignments were correct. The curves were obtained in the

0/20 mode because of the increased peak separation in the

asymmetric geometry.

Experimentally, it was found that the R optimization

procedure was greatly complicated for the asymmetric curves

because the optimal # value showed considerable variation as

R was changed. This effect was so pronounced that

frequently it was not possible to see any diffraction peaks

for a given value of R. As a result, an alternative method

was adopted to account for the wafer miscut. This procedure

involved obtaining two rocking curves, one at an arbitrary

value of R where the diffraction peaks were visible, and a

second at 1800 away from the first value. As explained

earlier, the average of these two 0 values is then the

correct one. This procedure was used for both the glancing

incidence and glancing exit measurements so that evaluation

of the parallel mismatch required a total of four different

rocking curves.

Many of the measured diffraction effects were weak in

intensity. In order to reduce statistical fluctuations,

which vary as the square root of the total number of counts,

long counting times were required. For a diffraction peak

intensity of 10 counts per second (cps), the fluctuations

would be about 3% for a count time of 100 sec/step, 10% for

10 sec/step and 30% for 1 sec/step. Counting times were

typically 10 sec/step with a step size of 1.8 arc-sec for

the low resolution work and 30 sec/step with a step size of

0.9 arc-sec for the high resolution scans. These counting

times led to total scan times of 4 to 20 hours, depending on

the total angular range covered.

Rocking curve analysis

The Phillips high resolution diffractometer has been

used almost exclusively for the analysis of relatively

uniform, thin epitaxial layers grown on single crystal

substrates by a variety of techniques. These RC's typically

consist of a strong diffraction peak from the substrate, a

diffraction peak from the epitaxial layer (if it's not

lattice matched) and additional intensity maxima associated

with finite thickness effects. Under these circumstances,

the interpretation of the rocking curves is relatively

straightforward. One of the conclusions of this study was

that in many cases the rocking curves of ion-implanted

samples could be interpreted in a manner directly analogous

to the treatment of thin epitaxial layers. Therefore, the

analysis of rocking curves obtained from hetero-epitaxial

layers will be briefly reviewed before proceeding to a

discussion of analysis methods for rocking curves of ion-

implanted samples.

Experimentally, it is observed that the lateral

constraint of the substrate causes the epitaxial layer to be

tetragonally distorted to some degree, depending on the

extent of relaxation that has occurred in the layer. For an

(001) oriented material that is cubic in bulk form, this

implies that the lattice cell dimension in a direction

perpendicular to the surface is different from that of the

cell edges parallel to the surface and that all three cell

dimensions are different from the bulk lattice constant.

The fractional change in lattice parameter of the epitaxial

layer thus depends on the direction in the layer. Rocking

curves obtained from HRXRD allow these two components to be

directly evaluated.

For an (001) oriented, tetragonally distorted crystal,

one can obtain from Bragg's law and the geometry of

tetragonal crystals the following relations for the (hkl)


a1 = Xl (2-4a)
2sin0 cos(

al = al2 (h2 + k2) cot2 (2-4b)

where a1 is the perpendicular cell dimension, a, is the

parallel cell dimension, X is the x-ray wavelength, 0 is the

Bragg angle for the (hkl) planes and # is the angle between

the (hkl) and (004) planes. Differentiating these equations

and rearranging yields [76]:

(Aa/a)1 = tan#d) cotOdO (2-5a)

(Aa/a), = -cotod) cotOd8 (2-5b)

where # and 0 are now taken to be reference values for the

cubic substrate of lattice parameter a. It is customary to

refer to these as the "x-ray strain" or "strain" in the

layer. However, it must be noted that they are actually the

fractional change in lattice parameter of the epitaxial

layer with respect to the substrate and not with respect to

the bulk lattice parameter of the layer. Thus, they are the

perpendicular and parallel components of lattice mismatch,

not the actual elastic strain in the layer.

Evaluation of the two mismatch components requires

knowledge of the change in planar tilt angle, A), and the

change in Bragg angle, AO, for the (hkl) planes of the layer

with respect to the substrate. Both AO and AO are directly

obtainable from HRXRD provided rocking curves are obtained

at both glancing incidence and glancing exit angle for the

(hkl) reflection. Assuming that any miscut in the sample

has been taken into account by appropriately rotating the

sample then the angles are given by:

0 = o + O (2-6a)

C = O* 0- (2-6b)

AO = Am + Aw- (2-6c)

A = Ag~ Am- (2-6d)

where oV and o are the angular position of the substrate

peaks at glancing incidence and exit, respectively, and Aw

is the angular difference between the layer peak and the

substrate peak. The sign convention used here requires that

all differences be taken as epitaxial layer value minus

substrate value. With this convention, substitution into

equations 2-5a and 2-5b yields the perpendicular and

parallel mismatch.

It should be noted that any asymmetric reflection

(i.e., when Co* co) can be used to obtain the parallel and

perpendicular components of mismatch. If, however, the

symmetric (004) reflection is used, only the perpendicular

component can be evaluated. With 0=0 and O=0subst the result

(Aa/a)1 = -cotWsubst Ao (2-7)

Thickness fringes also typically appear in RC's of

epitaxial structures. These arise from constructive and

destructive interference of the x-rays within the epitaxial

layer. Bartels [76] gives the approximate thickness of a

thin layer, t, as:

t = ksin(20 C) (2-8)

where A0 is the fringe spacing.

In contrast to the above analysis for thin epitaxial

layers, interpretation of rocking curves from ion-implanted

single crystals has been assumed to be much less

straightforward. The primary differences are that the

strain is usually not distributed homogeneously throughout

the layer which distorts the peak shapes and that

implantation induced disorder depresses the intensity of any

diffraction maxima. As a result, it may not be obvious what

the origin of a series of maxima in the rocking curve might

be. Two general approaches to these difficulties have been


Afanasev et al. [79] have developed a Fourier analysis

method for extracting an effective layer thickness and an

average strain value from a rocking curve. The technique is

analogous to crystal structure determinations using

Patterson functions. In both instances, however, a complete

solution to the problem is not possible because the

experimentally accessible quantities are sufficient to

evaluate only the amplitude of the Fourier coefficients, not

their phase relationships. The assumptions in their

analysis also limit the general applicability of the

technique. Specifically, their analysis assumes a single

strained layer located at the surface that is thin enough

with respect to the extinction distance that kinematical

theory applies. In addition, it requires that the

diffraction effects be relatively strong so that the Fourier

transform will be sensitive enough to the perturbations

introduced by the strain layer. Neither of these

requirements was met by the implanted samples in this study.

By far the most widely used technique for analysis of

implanted layers, and the one used in this study, is that of

computer simulations of the rocking curves using a series of

assumed strain distributions. This rather tedious trial and

error technique is normally carried out until the calculated

rocking curve matches the experimental one, in which case

the trial strain distribution is assumed to be the correct

one. Such simulations have been performed using kinematical

[80], semi-kinematical [81,82] and dynamical theories of

x-ray diffraction [83-85].

The kinematical and semi-kinematical simulations treat

the strain layer as a small perturbation on a kinematically

or dynamically diffracting substrate. The purely

kinematical approach is entirely inadequate for Bragg

diffraction from the thick single crystal substrate under

any circumstances. The semi-kinematical treatment is valid

only if the strain layer thickness is much less than the

extinction distance of the operative reflection. For the

(004) reflection of silicon the extinction distance is about

3.7 pm while the strain layers in the implanted samples

extend to approximately 20% of this depth. It was therefore

judged necessary to use dynamical theory in this study.

The dynamical simulations were carried out using the

RADS computer program, available commercially from Bede

Scientific Instruments, Ltd. of Durham, UK. RADS uses the

Takagi-Taupin [86-88] formulation of dynamical theory to

calculate rocking curves based on user specified trial

composition distributions of hetero-epitaxial films. The

composition variation with depth is specified as discrete

layers of variable thickness but each with uniform lattice

parameter. The Takagi-Taupin equations are solved for the

ratio of diffracted to incident beam amplitudes for each

layer. The amplitude ratios are then matched at each layer

proceeding from the substrate upward to the surface, thus

producing the final amplitude ratio distribution. The

reflectivity (i.e., the rocking curve) is then taken as the

square modulus of the amplitude ratio exiting the top


The program was written for double-crystal x-ray

diffraction analysis of hetero-epitaxial layers grown on

single crystal substrates. In order to use the program for

the present study some adaptations were necessary, although

the software itself could not be modified. These

adaptations fell into two categories, those associated with

the analysis of ion implanted layers rather than hetero-

epitaxial layers and those associated with the use of HRXRD

instead of a double-crystal system.

In order to adapt the program to the analysis of ion

implanted specimens it was necessary to specify a depth

dependent strain distribution. Two mythical elements were

created for this purpose. Big-silicon (Big-Si) was

identical to silicon in all respects except that it had a

lattice parameter 1% larger than silicon. Similarly,

Little-silicon (Lit-Si) was created with a lattice parameter

1% smaller than silicon. Strain distributions were thus

specified in terms of the percentage of Big-Si in a Big-

Si/Si alloy or in terms of the percentage of Lit-Si in a

Lit-Si/Si alloy. In this manner it was possible to

construct both positively and negatively strained layers and

to have the strain vary with depth in the sample. Each

layer was made to be completely coherent with the underlying

layer in keeping with the experimental result that the

parallel mismatch was effectively zero in all cases. With

the parallel mismatch assumed to be zero, the perpendicular

mismatch is given by:

(Aa/a) = x (1 + V) = 1.77x (2-9)
100 (1 v) 100

where x is the atom fraction of Lit-Si or Big-Si and v=0.278

is Poisson's ratio for silicon.

A major shortcoming of the RADS program is that it does

not use Debye-Waller factors in the calculations. For ion

implanted materials, it has been shown that the large number

of displaced atoms depresses diffracted intensities [82,84].

This effect has been successfully modeled in rocking curve

simulations by the introduction of a depth dependent static

Debye-Waller factor. Since the lattice strain is normally a

result of the atomic displacements, the static Debye-Waller

factor is usually directly proportional to the strain [26].

In the absence of the ability to model this effect with

RADS, a perfect match between experimental and calculated

rocking curves was not possible.

An additional effect that could not readily be

incorporated in the simulations was Huang scattering from

extended defects (i.e., dislocations, stacking faults etc.).

Huang scattering appears as a broad elevation of intensity

in the tails of Bragg diffraction peaks [89]. When the

Huang intensity exceeds that of the subsidiary diffraction

peaks around the substrate Bragg peak it is no longer

possible to measure the subsidiary peaks. This effect

became very pronounced only at very high defect densities

where the microstructure was more amenable to analysis by

TEM than by HRXRD.

Several modifications in procedure were also needed to

enable the simulations to be compared to experimental

rocking curves obtained from HRXRD rather than from a

double-crystal system. The Takagi-Taupin equations must be

solved separately for x-rays polarized perpendicular to (o

polarized) and coplanar with (K polarized) the plane of

incidence. In double-crystal systems the incident radiation

is essentially unpolarized so RADS simulates the final

rocking curve by averaging the a and x polarized

reflectivity curves. In HRXRD, however, the multiple

reflections off of the beam-conditioning crystals can be

expected to lead to some polarization of the x-ray beam

incident on the sample surface.

This possibility was investigated by simulating rocking

curves for the Ge beam conditioning crystals. The Dumond

diagram for the four-crystal monochromator (Figure 2-7)

indicates that the effect of the multiple reflections can be

approximately treated by taking the final reflectivity as

the fourth power of the reflectivity of a single Ge crystal.

Figure 2-11 shows the calculated single crystal and four-

crystal reflectivities for both the a and x components for

the (440) Ge crystal setting. The integrated intensity of

the x component is only 1.9 % of the a component. An exact

analysis would require that the rocking curve of the fifth

crystal (the sample) be taken as the average of the a and n

components weighted according to the above results.

However, it was found that the relatively less intense x

component contributed so little to the final result that it


-50 0 50
Aw (arc-sec)

Figure 2-11.

Calculated reflectivity curves for a single
(440) oriented Ge crystal and for the four
crystal monochromator.
a) X-rays a polarized; b) X-rays K polarized.







(a) 0
single crystal

:four crystal --






10-6 L



could be ignored. Accordingly, rocking curve simulations

were only performed for the a polarized component.

An additional complication related to the use of HRXRD

entailed the manner in which peak broadening due to the

angular width of the incident x-rays was modeled. For

double-crystal systems this effect is accounted for by

calculating the reflectivity of both the reference crystal

and the sample crystal and mathematically correlating the

two curves. Since the FWHM of the reference crystal is

often similar to that of the sample, the effect can be quite

pronounced. In the HRXRD case, the effect was modeled by

correlating the calculated reflectivity curve of the sample

with the R4 curve for the Ge crystals in a separate computer

program outside of RADS. As a check on the validity of this

procedure, an experimental rocking curve of un-implanted

<100> silicon was obtained for comparison with the

calculated curve. The comparison of these curves in Figure

2-12 shows that the fit was very good. The FWHM of the

experimental curve was 5.9 arc-sec compared to 5.5 arc-sec

for the calculated one.

Another source of peak broadening common to both HRXRD

and double-crystal systems is that due to wafer curvature

arising from the presence of lattice strain at the surface

of the wafer. The finite width of the x-ray beam (1 mm for

the HRXRD) in combination with this curvature results in

peak broadening. RADS models this effect by performing a







0 -

Figure 2-12.

-20 0 20 40
Ao (orc-sec)

Comparison of experimental and simulated (004)
rocking curves for an <100> oriented bulk
silicon wafer.

user specified n-point smoothing operation on the calculated

rocking curve. This procedure was used for the HRXRD

results also.

A final very important consideration was the uniqueness

of the strain distribution obtained by this trial and error

fitting routine. Tsai et al. [90] have shown that under

certain circumstances the solution obtained in this manner

is not unique. Accordingly, a systematic investigation of

how the calculated rocking curves varied with the height,

width, depth and gradient of trial strain distributions was

performed. A detailed description of this investigation

appears in the Appendix. The investigation served to

identify the circumstances under which the computer matching

technique was unreliable. Moreover, it provided the basis

for correctly interpreting the rocking curves, even in those

cases where the technique was relatively insensitive to

variations in the strain distributions. In particular, it

was found that most of the rocking curves could be

interpreted in a straightforward manner in direct analogy to

the analysis of hetero-epitaxial layers presented above. As

discussed in the Appendix, it was found that the RC

simulations could be used to identify the Bragg diffraction

peak from the strain layer. Experimental measurements of

the position of these peaks then provided an estimate of the

parallel and perpendicular mismatch through equations 2-5a

and 2-5b. Finally, it should be noted that correlation of

the HRXRD results with cross-sectional TEM of the as-

implanted and annealed samples provided further assurance

that the experimental rocking curves were being interpreted


In view of these considerations, two general fitting

procedures were followed. When a detailed fit between

experimental and calculated rocking curves was either not

required or not credible, only the a polarized curve was

simulated and no further effort was made to account for the

instrumental effects discussed above. In several instances,

where a more precise strain distribution was needed, it was

possible to obtain a credible match between calculation and

experiment. In these cases both the instrumental broadening

and curvature effects were used in obtaining the final

simulated curves.


An integral part of the approach used in this

investigation was an attempt to study a simplified system

before proceeding to the analysis of the very complex

microstructure of the high-dose implants. The

simplifications imposed were of two basic kinds. First, the

effect of increasing dose on the defect microstructure was

investigated, starting with much lower doses than are used

in ion-beam synthesis. Secondly, it was clear from a review

of the relevant literature (Chapter 1) that defect evolution

as a function of annealing was poorly understood because of

a heavy emphasis on very high temperature anneals (213000C)

in most previous SIMOX studies. Thus, the effect of

annealing at intermediate and high temperatures at each dose

was studied. This approach also served to help bridge the

gap between our considerable knowledge of secondary defects

associated with low-dose, dopant implants in silicon and our

relative lack of knowledge concerning the high-dose, high-

temperature implants used in ion-beam synthesis.

The primary investigation tools were high resolution

x-ray diffraction (HRXRD) and transmission electron

microscopy (TEM). These techniques provided complementary

information about defects in the implanted samples. In so

much as point defects and their clusters produce strain in

the lattice, the HRXRD measurements were able to provide

information about the extent and distribution of these

defects. In addition, the TEM observations were able to

provide detailed information about the configuration and

location of extended defects in the ion-implanted samples.

Accordingly, this chapter presents the results of HRXRD

measurements and TEM observations on the effect of dose and

annealing treatment on the strain state and microstructure

of single implants of oxygen in silicon. A full discussion

of strain relief and defect formation and evolution

processes in high-dose oxygen implanted silicon is presented

in the second section of this chapter. This discussion

includes a consideration of those factors that are important

in determining the threading dislocation density of SIMOX

material. This more detailed understanding of defects in

SIMOX led to the development of a new multiple implant/low

temperature annealing process for producing low threading

dislocation density SIMOX. The results of this new process

are presented in Chapter 4. Finally, the major conclusions

of this investigation are summarized in the final chapter.

Strain Measurements

In order to study the strain state of the oxygen

implanted samples, high resolution x-ray diffraction

measurements were made. Rocking curves were obtained from

the (004) symmetric reflection and various asymmetric

reflections to analyze both the perpendicular and parallel

components of strain. Since the strain distributions in

ion-implanted samples were expected to be non-uniform, it

was necessary to use dynamical x-ray diffraction rocking

curve simulations from trial strain distributions to

correctly interpret the results. A systematic investigation

of the effects of such strain distributions on calculated

rocking curves was performed so as to produce general

guidelines for the correct interpretation of rocking curves

from ion-implanted samples. The details of this investi-

gation appear in the Appendix. However, a summary of the

resulting interpretation guidelines is given in Table 3-1.

Effect of Dose

A series of (004) rocking curves from the as-implanted

samples showing the effect of increasing dose appear in

Figure 3-1. The results indicate that two distinct strain

layers exist in the as-implanted samples. In the vicinity

of the projected range of the implanted oxygen ions the

lattice undergoes an expansion creating a buried, positively

strained layer while at higher doses, a second layer,

located at the surface undergoes a lattice contraction.

After comparing these results to the microstructural

observations and to TRIM calculations, it will be shown that

these strain layers may be interpreted in terms of an excess

Table 3-1. Information available directly from rocking curves.

Feature in Rocking Curve

Information Available

1. Diffraction peaks displaced
from substrate peak at
a. lower angle

b. higher angle

2. Strain layer peak
a. symmetric

b. asymmetric

3. Secondary thickness fringes
on strain layer peak
a. present

b. absent
FWHM>100 arc-sec

FWHM<100 arc-sec

4. Primary thickness fringe
spacings are
a. equal

b. unequal

positive strain (i.e,lattice

negative strain (i.e, lattice

homogeneously strained layer
strain estimated by
E=-cot Ao

inhomogeneously strained layer
(but not necessarily an asym-
metric strain distribution)
maximum strain estimated by

buried layer, depth given by
unique strain distribution
not, in general, possible

surface layer
unique strain distribution

indeterminate case
requires further investigation

homogeneously strained layer
thickness given by

inhomogeneously strained layer
total thickness crudely
estimated by ttot>sinO/Awsin2O
width of maximum strain region
crudely estimated by




1 0 0 4 0 0 W
101 0

-800 -600 -400 -200
Aw (orc-sec)





-600 -400 -200 0
Au (orc-sec)



Figure 3-1. The effect of dose on (004) rocking curves
of as-implanted samples.
a) IxlO01 cm-2; b) 3x1016 cm 2;






10 0 -200
-400 -200


- -1

10-1 I 2
-400 -200

0 200 400 600 800
Aw (arc-sec)

0 200 400 600 800
Ac (arc-sec)

Figure 3-1--continued
c) 1x017 cm-2; d) 3x1017 cm2;

17 -2
3x10 cm as-implanted

pr =p -2650ppm


-1000 0 1000 2000
Aw (orc-sec)

-1000 0 1000 2000
Ac (orc-sec)



Figure 3-1--continued
e) 6x10'7 cm2; f) 9x107 cm-2.










of implantation-induced self-interstitials beneath the

surface and to an association of vacancies and oxygen atoms

in the surface layer.

At the lowest dose, 1x1016 cm2, the rocking curve

consists of a series of intensity maxima at lower angles

from the strong substrate peak (Figure 3-la). According to

the rules outlined in the Appendix, the lowest angle peak

corresponds to Bragg diffraction from a strain layer and the

additional peaks between it and the substrate peak are

finite thickness fringes. The angular separation of the

substrate and strain layer peaks is 550.4 arc-sec from which

the maximum perpendicular strain in the distribution may be

estimated as E_=+3850 ppm. The positive sign of the strain

indicates that the strained layer has undergone a lattice

expansion with respect to the underlying silicon substrate

in a direction perpendicular to the surface. The irregular

spacing of the thickness fringes and the asymmetric nature

of the strain layer Bragg peak indicate that the strain is

distributed inhomogeneously in the material. From the

spacing of these thickness fringes the total width of the

strain layer is crudely estimated to be >3080 A while the

width of the maximum strain region is crudely estimated to

be <1610 A.

An important feature of this rocking curve is the small

"bumps" on the strain layer peak. These "bumps" are

secondary thickness fringes which indicate that the strain

layer is buried beneath the surface of the sample. To an

excellent approximation the depth of the bottom edge of the

strain maximum is given by the spacing of these secondary

fringes as 4490 A. Thus, the maximum strain region extends

from about 2880 A to 4490 A. This range of depths

approximately corresponds to the projected range of the ions

(3800 A) and to the damage distribution maximum (3100 A) as

predicted by TRIM '88. This region should be dominated by

interstitial atoms (both oxygen and self-interstitials).

The positive sign of the strain in this layer is consistent

with the presence of interstitials, which would force

adjacent atoms apart, thus increasing the measured lattice


In general, it is difficult or impossible to use the

trial and error matching technique to obtain a credible

strain distribution from buried layers such as this one.

Therefore, a detailed match between experimental and

calculated RC's was not attempted and a more detailed

comparison with TRIM '88 predictions was not possible.

The RC simulation study presented in the Appendix

showed that the angular position of the strain layer peak in

both asymmetric and symmetric reflections led to a self-

consistent set of values for E1 and E1. Therefore,

experimental (044) RC's at glancing incidence and exit

angles were obtained from this sample. These yielded values

of +3810 ppm for the perpendicular component and -14 ppm for

the parallel component of mismatch. The experimental error

is approximately 50-100 ppm. Thus, the parallel lattice

parameter of the strain layer is identical to that of the

silicon substrate, within experimental error. These results

indicate that the strain has been accommodated by a

tetragonal distortion of the lattice with no relaxation

parallel to the surface. This finding is in agreement with

other x-ray measurements on ion-implanted silicon [90,91].

Apparently, the lateral constraint of the substrate is

sufficient to prevent significant relaxation from occurring

for the strain level in this buried layer.

Increasing the dose to 3x106 cm-2 resulted in a rocking

curve without a readily identifiable strain layer

diffraction peak (Figure 3-1b). The only peaks visible in

the RC are thickness fringes which are mostly at lower angle

from the substrate peak. The progressive decrease in

intensity of these peaks as Ao becomes more negative is

likely the result of a large increase in the static Debye-

Waller factor in the strain layer. The magnitude of the

static Debye-Waller factor reflects the extent to which

atoms in the crystal are displaced from their equilibrium

position. Thus, it appears that with increasing dose the

implantation-induced atomic displacements have increased

significantly. The RC also shows an elevated background

intensity which is likely due to Huang scattering from

extended defects in the strain layer. This effect also

contributes to the inability to resolve the strain layer


A careful examination of the RC indicates that there

are nine peaks visible above the background level. In the

RC simulations, the Bragg diffraction peak from the strain

layer was always found to be the outermost peak. However,

it is quite possible that some of the peaks are lost in the

elevated background intensity of this experimental RC so

that the outermost visible peak may only be a thickness

fringe, rather than the strain layer peak. Thus, only a

lower-bound estimate of the maximum strain can be obtained

from the position of the outermost visible peak. The

angular separation is Aw=-647.6 arc-sec which corresponds to

an estimated maximum strain of E>+4535 ppm. Once again the

strain is positive indicating that a lattice expansion

perpendicular to the surface has occurred.

The suppression of the peak intensities from this

buried layer was further accentuated when the dose was

increased to 1x1017 cm-2 (Figure 3-lc). Only a broad

elevated intensity was observed from the experimental (004)

RC at angles below the substrate peak, which prevented a

quantitative evaluation of the strain distribution from the

buried layer. It should be emphasized that the layer is

still likely to be strained; however, it is no longer

possible to measure the strain with the HRXRD technique.

The loss of the ability to measure the strain is likely due

to the presence of many extended defects as well as a high

degree of atomic disorder in the buried layer.

A striking, and rather surprising difference with the

lower-dose samples was the presence of a pronounced Bragg

diffraction peak at higher angle from the substrate peak in

this x1017 cm-2 implant. The estimated maximum strain is

E=-995 ppm, i.e., this strain layer has undergone a lattice

contraction perpendicular to the surface. The origin of

such a lattice contraction is not immediately apparent.

The full width at half maximum (FWHM) of the peak is

135 arc-sec. Therefore, according to Table 3-1, the absence

of secondary thickness fringes on the peak can be taken as

an indication that the strain layer is located at or very

near the surface of the sample. The asymmetry of the Bragg

peak indicates that the strain distribution is non-uniform

and asymmetric. Rocking curves from the (115) reflection at

glancing incidence and exit angles showed that the parallel

mismatch with respect to the substrate was Ei=-19 ppm, i.e.,

the layer is tetragonally distorted with no relaxation,

within experimental error.

The results summarized in Table 3-1 suggest that for a

surface strain layer such as this one, the trial and error

matching of calculated and experimental rocking curves

should produce a credible strain distribution. The results

of a trial and error fitting procedure for the lxl07 cm-2

as-implanted sample are shown in Figure 3-2. Instrumental

broadening effects were included in the calculated RC, as

discussed in Chapter 2. The experimental and calculated

curves agree quite well, except at angles below the

substrate peak where scattering from extended defects in the

buried layer could not be modeled appropriately by the RADS

simulation program. The corresponding strain distribution

for the surface layer, where the simulations were

successful, shows a maximum strain of -1275 ppm which is 280

ppm higher than the value estimated from the diffraction

peak separation and Bragg's Law. This discrepancy is a

general phenomenon that occurs in all rocking curves from

thin layers of material [92,93]. The maximum strain region

extends from the surface to a depth of 1000 A while the

total width of the distribution is 2000 A. This is

considerably shallower than either the projected range

(3800 A) or the maximum of the damage profile (3100 A).

Increasing the dose further to 3x1017 cm-2 produced an

increase in the estimated maximum perpendicular strain to

E=-2650 ppm for the surface strain layer (Figure 3-1d).

Once again a trial and error fitting procedure was followed,

producing the results in Figure 3-3. The maximum strain is

-2830 ppm, which is 7% higher than the estimated value, and

extends from the surface to a depth of 1200 A. The

strainthen decreases monotonically to zero over a further

distance of 1000 A for a total thickness of 2200 A, about

200 A deeper than the 1x1017 cm-2 sample. Measurements of

....... experimental RC
S-- calculated RC
a. 2



10 -, I I I ,
10- 1
-200 0 200 400 600 800
Ad (arc-sec)

-- -1500


L -5000


0 I- i ,i ,I- ,- ,i-,,
0 500 1000 1500 2000 2500
Depth (A)

Figure 3-2. Rocking curve simulations for the 1x1017 cm-2
as-implanted sample.
a) Experimental and calculated rocking curves;
b) Strain distribution for calculated rocking
curve in a).

0 200 400 600
AC (arc-sec)


0 500 1000 1500 2000 2500 3000
Depth (A)

Figure 3-3.

Rocking curve simulations for the 3x1017 cm-2
as-implanted sample.
a) Experimental and calculated rocking curves;
b) Strain distribution for calculated rocking
curve in a).











the asymmetric (044) reflection at glancing incidence and

exit angles showed that the parallel strain component was

+8.0 ppm. Thus, once again, the layer is tetragonally

distorted with no parallel relaxation, within experimental


Further increases in dose to 6x1017 cm2 (Figure 3-le)

and 9x1017 cm-2 (Figure 3-1f) resulted in only a broad

elevated intensity region at higher angle from the substrate

peak. This suggests that the lattice contraction layer at

the surface could still exist. However, a high degree of

atomic disorder and/or a high density of extended defects in

the layer have reduced the intensity and increased the FWHM

of the Bragg peak while elevating the background level so

that the Bragg peak from the strain layer is no longer

measurable. This finding is of particular interest to the

problem of threading dislocation (TD) formation in the

surface silicon layer because both doses are above the

critical dose for TD formation. Thus, these changes in the

RC's of the surface layer may be related to the formation of

threading dislocations or at least to the precursors of

threading dislocations.

Some additional insight into the formation of extended

defects in the surface silicon layer was provided by an

additional wafer implanted at 3x1017 cm-2. This second

implanted wafer, which will be referred to as wafer B, was

obtained in order to perform multiple implant and anneal

experiments, the results of which are detailed in Chapter 4.

HRXRD measurements (Figure 3-4) on the as-implanted wafer

still show the surface lattice contraction layer. However,

the strain-layer diffraction peak is considerably broader

than that of the original wafer (FWHM=680 arc-sec vs. 57

arc-sec) and the magnitude of the estimated maximum

perpendicular strain decreases from -2650 ppm to -2220 ppm.

The peak broadening suggests that some sort of extended

defects may begin to form at the 3x107 cm-2 dose.

Comparison with the 6x107 cm-2 and 9x1017 cm-2 results further

suggests that this may represent the beginning of the

defect-introduction process that has such a pronounced

effect on the RC's of the higher dose samples.

This variability in strain state for wafers implanted

under ostensibly identical conditions may be attributed to

several factors. First, some variability in wafer

temperature can be expected from implant to implant when

beam-heating is the sole source of temperature control [70].

Moreover, this dose is very close to the critical dose for

threading dislocation formation (about 4x1017 to 5x1017 cm-2)

in the surface silicon layer [65]. As a result, slight

variations in the implant conditions might be expected to

lead to relatively large strain state and microstructural

changes in this dose range. In this respect, wafer B

appears to represent a transitional stage between the sub-

and super-critical dose implant regimes.




100 L

0 400 800 1200


Aw (arc-sec)

Figure 3-4. Experimental (004) rocking curve for 3x1017 cm-2
as-implanted wafer B.

Effect of Annealinq Temperature

The results so far indicate that two distinct strain

layers exist in the as-implanted samples. These layers are

distinguished from each other by their location (buried vs.

surface) and by the sign of the strain (positive vs.

negative). Upon analyzing the strain state of samples

annealed at 9000C, 0.5 hr and 13000C, 6 hr, it became

evident that the annealing behavior of the layers was

significantly different also.

The annealing behavior of the buried layer is

illustrated in Figure 3-5 for the 1x106 cm-2 sample which

exhibited only the buried strain layer in the as-implanted

condition. The (004) RC for the 9000C anneal is compared

with an RC for bulk, unimplanted silicon in Figure 3-5a. The

RC shows no evidence of Bragg diffraction peaks from a

strain layer but does show long tails in intensity on either

side of the substrate Bragg peak. This is a classic example

of Huang scattering [89] from extended defects such as

dislocations or stacking faults. Apparently, the

unidirectional strain state in the as-implanted sample has

been destroyed by the formation of extended defects with

long-range strain fields extending in all directions. After

annealing at 13000C, 6 hr these Huang tails are gone.

Indeed, the rocking curve is virtually identical to that of

bulk, un-implanted silicon (see Figure 3-5b). Apparently,

-500 0 500
Ao (arc-sec)

-500 0 500
AwC (arc-sec)

Figure 3-5.

Effect of annealing on the (004) rocking curves
of the ixl106 cm-2 samples.
a) Comparison of 9000C, 0.5 hr annealed sample
with bulk silicon; b) Comparison of 13000C, 6 hr
annealed sample with bulk silicon.






100 1







16 -2
(b) 1x10 cm
1300*C, 6 hr
....... bulk silicon



the extended defect density is very low after the high-

temperature anneal.

Rocking curves of all of the other 9000C annealed

samples also showed these Huang tails from extended defect

formation in the buried layer. Likewise, annealing at

13000C, 6 hr also resulted in RC's very similar to that of

bulk silicon. The only discernable difference between the

various doses after the 13000C anneal was that the FWHM of

the substrate peak increased slightly with increasing dose.

This data is summarized in Figure 3-6, from which it is seen

that for doses 21xl017 cm-2, the FWHM is slightly higher than

that for bulk silicon. This trend may be related to the

increased presence of amorphous SiO, precipitates with

increasing dose. This would cause some diffuse scattering,

leading to a small increase in the FWHM of the measured


In contrast to the annealing behavior of all other

doses, the two doses with a distinct diffraction peak from

the surface layer still showed a Bragg diffraction peak

after annealing at 9000C. Figure 3-7 shows (004) RC's for

the 1xl017 and 3x1017 cm2 samples after 9000C annealing. A

diffraction peak is visible only as a shoulder on the

substrate peak in the Ix1017 cm-2 rocking curve while the

surface strain-layer diffraction peak is more clearly

resolved in the 3x1017 cm- rocking curve. The estimated

maximum perpendicular strain is -385 ppm and -525 ppm for

oxygen implanted
annealed 1300C, 6 hr

bulk silicon

bulk silicon

S f I I l I I ..




Dose (cm-2)

Figure 3-6. Full width half maximum (FWHM) of (004) substrate
peak as a function of dose after 13000C, 6 hr




-00 -200
-400 -200






100 I
-400 -200

0 200 400 600 800
Au (arc-sec)

0 200 400 600 800
Aw (arc-sec)

Figure 3-7. Effect of 9000C, 0.5 hr annealing
a surface strain layer peak.
a) 1x10i7 cm-2; b) 3x017 cm-2.

on samples with

the Ixl017 and 3x1017 cm2 samples, respectively. This

represents a strain reduction of 61% for the ixl107 cm-2

sample and 80% for the 3x017 cm-2 sample in comparison to

the as-implanted values.

Rocking curves of the (444) asymmetric reflection at

glancing incidence and exit angles showed that the

perpendicular strain component was +10.8 ppm for the 3x1017

cm-2 sample after 9000C annealing. Thus, the layer remained

tetragonally distorted after annealing at 9000C, 0.5 hr, in

sharp contrast to the buried layer where the unidirectional

strain state was destroyed upon annealing and only Huang

scattering was evident. Apparently, the defects responsible

for the surface lattice contraction do not evolve into

extended defects such as those in the buried layer upon

9000C annealing. It was not possible to obtain the

parallel strain for the ixl017 cm-2 9000C annealed sample

because of difficulties in resolving the strain layer peak

when it is so close to the substrate peak. However, it

appears reasonable to suppose that it behaves in a similar

manner to the 3x1017 cm2 sample.

This strain-reduction process in the surface layer of

the 3x107 cm-2 sample was studied further as a function of

temperature. Figure 3-8 is a plot of the measured

perpendicular strain as a function of annealing temperature

for a series of 0.5 hr'anneals. The strain-reduction

process has already begun at 7000C and is exhausted by

17 -2
3x10 cm

-3000 7/




600 800 1000 1200 1400

Anneal Temperature (C)

Figure 3-8.

Estimated maximum perpendicular strain as a
function of annealing temperature for samples
implanted at 3x017 cm-2.

I .- as-implanted