Koster G., Rijnders G. (Eds.) In situ Characterization of Thin Film Growth

Подождите немного. Документ загружается.

164 In situ characterization of thin ﬁlm growth

drift region, and isolated from the main scattering chamber by an aperture.

By controlling the time interval between the applications of voltage steps

to the two sets of deection plates, a pulsed ion beam with pulsed duration

adjustable in the 10–1000 ns range is produced. A nal set of direct current

deection plates is then utilized to steer the beam to the desired location

on the sample. All beam lines are differentially pumped by turbomolecular

pumps. Two channel electron multiplier detectors are positioned at angles of

165° and 25° relatively to the ion beam direction, and the reectron for the

MSRI analyzer is placed at a scattering angle of 60°

(Krauss et al., 1998).

Each detector is differentially pumped and sees the sample through a 1 mm

diameter aperture. All detectors and associated optics are located sufciently

far away from the sample so that they do not block the deposition ux. By

Neutral detector

DRS detector

Differential pumping

apertures

Kaufman ion

source

Substrate

Telefocus

ion source

MSRI reﬂectron

analyzer

ISS detector

Target carousel

Detector

6.5 Schematic of the TOF–ion scattering and recoil spectroscopy

system constructed by Krauss et al. (1998), indicating the location of

the ion beam line, and ISS, DRS, and MSRI detectors.



165In situ ion beam surface characterization

differentially pumping the space between the apertures, it is possible to

maintain a detector vacuum approximately three orders of magnitude lower

than that of the sample environment.

6.4 Studies of ﬁ lm growth processes relevant to

multicomponent oxides

Ferroelectric perovskite oxide thin  lms such as Pb(Zr

1–x

(PZT)

and SrBi

(SBT) have been studied as potential candidates for non-

volatile ferroelectric random access memories (NVFRAM). Because of

the high crystallization temperature of ferroelectric SBT  lms (>700 °C),

SBT integration with Si-based technology requires a stable metal barrier

heterostructure. One possible stack structure for a bottom electrode is Pt/

Ti/SiO

/Si thin  lm. Pt is often used as the capacitor electrode material

because of its high electrical conductivity and because Pt lattice parameters

give a good epitaxial relationship for the ferroelectric layer growth. The

main functions of Ti are to improve adhesion of Pt to SiO

, and prevent Pt

direct contact with silicon to form Pt silicides. Yet the previous studies of

Pt/Ti layer stacks in an oxygen atmosphere at high temperatures showed

their instability. During heating Ti diffused into the Pt layer along the grain

boundaries and became incorporated in the SBT layer (Sreenivas et al., 1994).

Ti interactions with SBT layer in turn affect composition, microstructure and

electrical properties of the SBT-based capacitors.

In case of Pt/Ti/SiO

/Si stack, a relatively poor depth resolution of RBS

limits information on whether the diffusion process results in the appearance

of Ti atoms on the topmost Pt surface. The latter is not a desirable scenario

for the initial stages of SBT  lm growth. Mass spectroscopy of recoiled

ions (Krauss et al., 1998) has higher depth resolution than RBS, and it

was therefore used to obtain a detailed picture of the surface Pt/Ti/SiO

/Si

composition (Fig. 6.6). One can see that the surface composition in vacuum

is dominated by H, C, O, Pt, and Ar, and small amount of Ti. The relative

chemical inertness and high mass of Pt results in lower sensitivity of Pt

detection than that for Ti, Si, and H. As the temperature increases, the Pt

signal decreases and nearly disappears, while at the same time the Ti signal

increases signi cantly and a large Si peak appears. It is estimated that after

700 °C anneal, at least 80% of the surface consists of Ti and Si. It is interesting

that if a similar annealing procedure (700 °C ) is conducted in an oxygen

(

= 5 ¥ 10

–4

torr) environment, only a small Si signal is detected on the

surface. It has been proposed by Bruchhaus et al. (1992) that this may be

due to the Tio

diffusion barrier formation at the Ti/Si interface as a result

of oxygen diffusion via Pt grain boundaries.

Diffusion processes are also extremely important in novel high-

k dielectric

material structures. When different metals are deposited on a HfO

(or Hf



166 In situ characterization of thin ﬁlm growth

silicate)/SiO

/Si stack, additional interactions may occur both at the metal/

high-k and high-k/Si interfaces, which may either improve or degrade

electrical device performance. Several metals have been investigated with

the goal of improving electrical behavior of the overall gate stack. It was

shown by Preisler et al. (2004) that in W/HfO

gate stacks the tungsten can

under certain conditions introduce excess oxygen, resulting in interfacial SiO

growth at the hfo

/Si interface. It was suggested that Ti acts as an oxygen

getter when used as a gate metal on HfO

/SiO

and ZrO

/SiO

dielectric stacks

(Kim et al., 2004). Using in situ Ti deposition and MEIS measurements, it

was shown that Ti metallization of HfO

/SiO

/Si stacks reduces the SiO

interlayer and (to a more limited extent) the HfO

layer. MEIS backscattering

spectra for an as-deposited HfO

/SiO

/Si stack after 4.2 nm Ti deposition in

situ and after a subsequent anneal to 300 °C in UHV are presented in Fig.

6.7 (Goncharova et al., 2007). After Ti deposition, the Hf, Si and interface O

peaks are shifted from their surface energy peak positions to lower energies,

due to energy loss in the Ti layer. Simulations for an HfO

/SiO

/Si stack after

Ti deposition (Fig. 6.7b) conrm that the Ti layer thickness is uniform. A

distinct surface O peak (~106 keV) and the low O yield between the surface

and interface peaks (106–107 keV) indicate that the oxygen concentration in

the Ti layer is small (Ti-O

, x < 0.10), with enhanced oxidation occurring at

the surface (2–3 Å of Ti

) and at the Ti/HfO

interface (5 Å TiO

, 0.5 < x

< 1). From MEIS spectra simulations, it was also concluded that Si atoms

0 5 10 15 20 25 30

Time of ﬂight (µs)

Counts

1600

1400

1200

1000

800

600

400

200

MSRI 10 keV Ar

Pt (150 nm)/Ti (50 nm)/

SiO

(200 nm)/Si

(b) 450 °C

(a) 25 °C

6.6 MSRI spectra obtained by Krauss et al. (1998) for a 10 keV Ar

primary ion incident on multilayered structure with Pt (150 nm)/Ti

(50 nm)/SiO

(200 nm)/Si. Note that the signal intensities in the MSRI

spectra do not directly correspond to the surface coverage.



167In situ ion beam surface characterization

initially present in the interfacial SiO

layer incorporate into the bottom of

the high-k layer. Some evidence for Ti–Si interdiffusion through the high-k

lm in the presence of a Ti gate in the crystalline HfO

lms has also been

reported, as illustrated in Fig. 6.8. This diffusion is likely to be related to

defects in crystalline HfO

lms, such as grain boundaries.

As deposited

15 min, 300 °C

Interface

Surface

¥10

Yield

104 108 112 116 120 124 128

Energy (keV)

(a)

As deposited

(b)

After UHV annealing

0 25 50 75 100

Depth (Å)

(c)

CompositionComposition

1.0

0.5

0.0

1.0

0.5

0.0

6.7 (a) Medium energy ion spectra for a Ti/HfO

/SiO

/Si(001) stack

directly after metal deposition (solid line) and subsequent to a 300 °C

UHV anneal (open symbols). (b) and (c) Best ﬁt depth proﬁles for Ti,

Hf, Si and O (adapted from Goncharova et al., 2007).



168 In situ characterization of thin ﬁlm growth

6.4.1 In situ analysis of Hf oxide growth

The initial stages of HfO

growth are critical for the development of alternative

to SiO

high-k materials because dielectric layer properties depend on the

control of initial layer growth and the interface composition and defects.

The nal thickness of Hf oxide or silicate layer can be as small as 20–40 Å;

therefore conventional RBS analysis with a resolution of ~100 Å, although still

useful to detect total number of Hf atoms deposited at each ALD deposition

cycle (Delabie et al., 2005; Wang et al., 2007), is little use in obtaining a

quantitative picture of the initial deposition steps, and interface composition.

To assist the understanding of each individual parameter (temperature,

substrate termination, cycle duration) during ALD growth, MEIS was utilized

to measure elemental depth proles of deposited species after each cycle in

situ (Chang et al., 2005; chung et al., 2007). As can be seen from a proton

backscattering energy spectra, in Figs 6.9 and 6.10, a gradual increase of Hf

peak intensity is well correlated with the Si peak shift to the lower energy.

This behavior is consistent with systematic increase of the HfO

layer

thickness. However, ALD growth deviates from ideal linear growth behavior

at the initial stage due to the island growth. This can be easily concluded

from the comparison of calculated ion yield for the HfO

ideal monolayer

and experimental curve for the same Hf peak areal density (Fig. 6.9). For a

more detailed interpretation of the HfO

growth mechanism, the Hf peak in

MEIS energy spectrum can be calculated using simulation software. It was

shown by chang et al. that initial interaction between incoming precursors

of Hf and oxidizing molecules (H

o, or o

) are strongly dependent on the

surface preparation prior to ALD growth: the growth on the H-terminated

surface is nonlinear and the growth on the chemical oxide surface is linear

(Fig. 6.11). Based on the MEIS results mentioned above, an initial growth

model has been proposed (Fig. 6.12).

6.8 Schematic showing the proposed O, Si and Ti diffusion processes

across (a) as-deposited, (b) crystallized HfO

, following the interfacial

SiO

reduction after 600 °C annealing, (c) after 1000 °C annealing.

Ti–O

TiSi

Ti–Si–O

Ti–Hf–Si

SiO

TiO

HfO

2–x

HfO

2–x

2–

Si (001) Si (001) Si (001)

(a)

(b) (c)

HfSi

TiSi



169In situ ion beam surface characterization

According to this model island-like growth is predominant up to 15 cycles,

after which, the rst 1 monolayer (ML) forms completely. As a result, island-

like growth occurs during the initial HfO

growth and islands are merged

Ideal 1 ML

Simulation

1 cycle

2 cycles

3 cycles

4 cycles

5 cycles

7 cycles

10 cycles

15 cycles

20 cycles

30 cycles

97 98 99

Energy (keV)

Hf peak

Ideal 1 ML

Avg. 1 ML

Ion yield (counts)

900

600

300

6.9 Medium energy ion scattering spectra of Hf atoms for layers

grown on a thermal oxide obtained by the group of Moon (Chang et

al., 2005). Atomic layer deposition growth deviates from ideal linear

growth behavior at the initial stage because of island-like growth

contribution.

6.10 Medium energy ion scattering spectra of Si atoms for the HfO

layers grown on a thermal oxide. After 15 cycles, the leading edge

of the Si peak is moved and the ﬁrst 1 ML is completely formed.

A small shoulder indicated by the arrows is observed, which is an

indication of slight silicate formation (Chang et al., 2005).

0 cycles

1 cycle

2 cycles

3 cycles

5 cycles

7 cycles

10 cycles

15 cycles

20 cycles

HfO

2 ML

Si peak

88 89 90 91 92

Energy (keV)

Ion yield (counts)



170 In situ characterization of thin ﬁlm growth

into a continuous atomic layer with increasing ALD cycles, summarized

in Fig. 6.12. This study (Chang et al., 2005) was helpful in understanding

interfacial reactions that occur during the initial growth stage and to control

the reactions for the application of high-

k dielectrics.

6.4.2 Studies of the initial stages of Al oxide ﬁlm growth

on GaAs

Silicon has been the material of choice in the development of CMOS devices

for many years, and if it is replaced by other high electron- and hole-mobility

semiconductors, such as GaAs or InGaAs, new high-

k dielectric layer and

interface passivation composition will have to be established. The quality

of Ga and As native oxides is poor for these applications. As a result, it is

well known that difculty in obtaining high performance is related to the

challenge of depositing stable passivating layers on GaAs and InGaAs, which

leads to high density of defects and Fermi-level pining. All these factors have

slowed their integration into mainstream applications. Chemical cleaning and

subsequent passivation were used in the past (Kim et al., 2006) to remove

the poor-quality GaAs oxide layer. One potentially promising method – the

so-called ‘self-cleaning’ approach – is to choose deposition precursors and

conditions so that native Ga and As oxide layer can be removed during

initial cycles of dielectric growth. The feasibility of such an approach was

illustrated in the case of Al oxide deposition on GaAs surface using ALD

Hf coverage (10

Hf/cm

)

Hf coverage (¥10

Hf/cm

)

1.0

0.8

0.6

0.4

0.2

0.02

0.00

0 1 2 3 4 5

Number of ALD cycles

H-terminated surface

Chemical oxide surface

1 nm full density

0 20 40 60 80 100

Number of ALD cycles

6.11 Hf coverage, measured by in situ medium energy ion scattering

(Chung et al., 2007), as a function of the number of HfO

deposition

cycles, for the thermal oxide. The inset shows the calculated

maximum island height obtained from simulation.



171In situ ion beam surface characterization

with the trimethylaluminum (TMA) precursor and in situ characterization

of the growing layer by MEIS and X-ray photoemission spectroscopy (Lee

et al., 2009, 2010).

Owing to the similarity in masses, the Ga and As signals overlap in MEIS

(Fig. 6.13), yet after TMA exposure, a strong Al signal is observed and the

cumulative Ga + As peak is shifted towards lower energy, which will be

consistent with the formation of AlO

lm. At the same time there is little

change in the oxygen areal density. By analyzing the line shape of all O,

Al, Ga and As peaks in MEIS spectrum, Lee et al. (2009, 2010) were able

to show that initial samples can be modeled as 10 Å thick Ga-rich native

oxide layer on top of GaAs substrate, while after 5 Å of Al oxide deposition

this interfacial oxide layer is reduced to 7 Å.

Furthermore, stoichiometry of the growing AlO

layer and interfacial

GaAs oxide layer changes as the deposition sequence progresses. Figure

6.14 shows the areal density of Al and O of samples after preheating

(320 °C) and after various full cycles of TMA and water. Although the

O density changes little initially, the Al density increases a lot (0 to

1.5 ¥ 10

atom/cm

), indicating that the oxygen in the Al oxide layer

is not from oxygen (water) exposure but from the GaAs native oxide.

1 cycle

3 cycles

4 cycles

~10 cycles

~15 cycles

>30 cycles

2.8 ¥ 10

Hf atoms/cm

30% coverage of 1 ML

Average

1 ML

Completed 1 ML

Similar morphology

6.12 A model suggested by Chang et al. (2005) for the growth

mechanism of an ALD HfO

layer grown on a thermal oxide at the

initial growth stage. Island-like growth occurs at the initial stage

and the growth then follows a slightly deviated ideal linear growth

behavior.



172 In situ characterization of thin ﬁlm growth

After two cycles a different pattern is observed: both Al and O signals

increase linearly with a ratio of 2:3, as expected for a stoichiometric Al

lm.

6.13 (a) Medium energy ion scattering spectra for a GaAs sample

taken after preheating at 320 °C for ~30 min., and after exposure

to 1 trimethylaluminum pulse, and (b) layer models depicting the

approximate composition of the ﬁlms (Lee et al., 2010).

0 10 20 30 40 50

ALD cycles

Areal density (10

at/cm

)

6.14 The areal density of Al and O of Al

/GaAs samples: after

preheating (320 °C), and after 1, 2, 5, 10, and 50 cycles of TMA and

water (Lee et al., 2010).

320 °C preheated

1 TMA pulse exposed

Ga + As

100 105 110 115 125

Energy (keV)

(a)

Yield (a.u.)

0.3

7 Å

3.4

5 Å

2.3

10 Å

GaAs

(b)



173In situ ion beam surface characterization

6.4.3 Initial stages of ferroelectric perovskite oxide thin

ﬁ lm growth

Since it has been reported that Ti degrades the electrical properties of SBT-

based capacitors (Schindler et al., 1997), the experiments were focused

on studying the initial growth of SBT  lms on modi ed Pt surfaces with

diffused Ti and Si species, and on alternative stable electrode structures such

as Pt/TiO

/SiO

/Si, Pt/Ta/TiO

/SiO

/Si, and RuO

/Si

/Si. These experiments

were carried out at 700 °C under

= 5 ¥ 10

–4

torr, since these are suitable

parameters to deposit SBT  lms using sputter deposition.

Figure 6.15 shows that when SBT  lms are deposited at different

temperatures in an oxygen atmosphere at

= 5 ¥ 10

–4

torr on Pt/Ti/

SiO

/Si multilayered structure, neither Ti nor Si segregation is observed, and

there is close correlation between deposition conditions, Bi concentration

in the  lm and electrical properties. If one considers BST crystallographic

structure (Krauss et al., 1998), it is possible to see that Bi is either missing

from all lattice sites or only selected lattice sites. The MSRI results indicate

almost complete loss of Bi during SBT deposition on Ti-containing surface

at elevated temperature. Independent ISS study with incident 10 keV Ar

primary beam was performed during heating in oxygen up to 570 °C of

44 Å thick SBT  lm deposited directly on a Ti layer on a MgO substrate. It

showed that some of the Bi atoms may be still retained in the  lm. Ta and

6.15 MSRI spectra during SBT deposition onto a Pt (150 nm)/Ti

(50 nm)/SiO

(200 nm)/Si substrate at various temperatures in an

oxygen background at 5 ¥ 10

–4

torr (Krauss et al., 1998).

0 5 10 15 20 25 30

Time of ﬂ ight (µs)

Counts

500

400

300

200

100

(b) T = 600 °C

(a) T = 700 °C

MSRI during

SrBi

deposition

on Pt/Ti/SiO

/Si

