Fahlman B.D. Materials Chemistry

Подождите немного. Документ загружается.

best candidates for CVD, as they are often less susceptible to premature hydrolysis

during transfer to the CVD system. Co-reactant gases such as NH

S, H

O, etc.

are often added in order to assist the removal of organic moieties from the organo-

metallic precursor molecule, preventing their incorporation in the growing ﬁlm.

The use of mixed precursors adds complexity to the control over ﬁlm stoichiometry.

The different volatilities of the precursors often lead to irreproducible vapor-phase

concentrations. As a result, ﬂow rates have to be carefully controlled to give ﬁlms of a

desired composition. Another problem that is posed by the use of mixed precursors is

the variable rate of hydrolysis/oxidation, resulting in morphologically varied ﬁlms.

Hence, it is most preferable to use a precursor that contains as many desired building

blocks of the thin ﬁlm as possible. For the ﬁlms deposited during IC fabrication,

common precursors include silane (SiH

) with co-reactant gases of H

(for polysilicon

ﬁlms), NH

(for Si

), and O

(for SiO

). For tungsten ﬁlms, a 1:3 WF

precursor

mixture is used; TiN is deposited using a TiCl

/NH

combination (Eq. 18):

Figure 4.62. CVD/ALD delivery methods for liquid precursors. Shown are (a) carrier gas bubbling of a

volatile precursor, (b) vaporization of a volatile precursor, and c) direct liquid injection (DLI) of a volatile/

non-volatile precursor.

308 4 Semiconductors

6 TiCl

þ 8NH

!

500



6 TiN þ 24 HCl þ N

ð18Þ

Figure 4.63 illustrates the chemical structures of a variety of precursors that

may be used for CVD/ALD applications.

[57]

Metal alkoxides (M(OR)

) and

Figure 4.63. Molecular structures of commonly used CVD precursor classes. Shown are: (a) metal

b-diketonate (acetylacetonate, acac) complex to grow a metal oxide ﬁlm (H

as the coreactant gas

yields a metal ﬁlm); (b) a heteroleptic (more than one type of ligand bound to the metal) b-diketonate

complex to yield a Cu ﬁlm; the ancillary ligand helps prevent oligomerization, enhancing volatility;

coreactant gas(es) used – respective ligands are b-ketoiminato, b-diketiminato, amidinato, and

guanidinato; (d) a metal azolato complex commonly used to deposit lanthanide metal ﬁlms; metal

cyclopentadienyl (cp) complexes, (e), and dialkylamido complexes, (f), are often used to deposit a

variety of metallic or compound thin ﬁlms.

4.2. Silicon-Based Applications 309

halides represent the earliest known examples; mixed metal alkoxide precursors

(e.g., for bimetallic thin ﬁlm growth) may be synthesized by the following route

(A ¼ Group 1 metal):

M(ORÞ

þ AOR ! AM(ORÞ

xþ1

ð19Þ

y AM(ORÞ

xþ1

þ M

! M

½MðORÞ

xþ1



þ y AClð20Þ

Metal b-diketonate complexes have been used extensively to deposit thin ﬁlms

of most metals and their oxides. Their low cost and high volatility make them

wise candidates for precursors; however, they generally result in large concentra-

tions of carbon incorporated into the ﬁlm. Replacing one oxygen with -NR sub-

stituents results in the b-ketoiminato ligand system. The greater basicity of the

nitrogen-donor ligands relative to oxygen analogues has been reported to increase

the volatility through enhanced metal-ligand bonding.

[58]

Replacing both oxygen

groups with -NR moieties results in the b-diketiminato ligand, which provides

further thermal stability of the resultant metal comple x, while offering a tunable

volatility by varying the imine substituents.

The volatility of a precursor is related to its molecular weight (i.e., London

Dispersion intermolecular forces), as well as its degree of oligomerization. As

such, it is also common to have a metal complex that features bulky ligands

[59]

the use of an ancillary Lewis base (e.g., alkene, alkyne, triorganophosphine, vinyl-

silane, etc.) to prevent oligomerization. Such a complex with more than a single

type of ligand surrounding the central metal is known as a heteroleptic complex; in

contrast, homoleptic complexes have only a single ligand type, such as Al(C

)

Another effective method to enhance precursor volatility is to incorporate partially

and fully ﬂuorinated ligands. The enhancement may be rationalized either by an

increased amount of intermolecular repulsion due to the additional lone pairs or that

the reduced polarizability of ﬂuorine (relative to hydrogen) causes ﬂuorinated

ligands to have less intermolecular attractive interactions.

Prior to CVD studies, it is essential that the potential precursor undergo thermo

gravimetric analy sis (TGA) and differential scanning calorimetry (DSC) to deter-

mine the most suitable temperature regime for thin-ﬁlm growth (Figure 4.64).

A sharp TGA curve with no remaining residue indicates that the precursor vaporizes

without signiﬁcant ligand decomposition – most desirable for CVD applications.

Further, if the mass loss via TGA occurs concomitantly with a single endot herm

via DSC, then sublimation of the precursor is occurring without decomposition

(Figure 4.64, bottom) – ideal behavior for CVD applications.

In addition to simple mass-loss vs. T investigations, TGA may als o be applied to

accurately determine the enthalpy of sublimation (DH

sub

), as well as the sublimation

temperature (T

sub

) of a solid.

[60]

In these investigations, the mass loss through

sublimation (m

sub

) will be constant at a given temperature, as long as the phase

change occurs without appreciable decomposition. Hence, by measuring the mass

loss over a variety of isothermal regions, a plot of log(m

sub

1/2

) vs.1/T may be

310 4 Semiconductors

generated, which readily yields values of DH

sub

and T

sub

from the slope and

y-intercept, respective ly (Eq. 21):

log m

sub

ﬃﬃﬃ



0:0522 DH

sub

ðÞ

0:0522 DH

sub

ðÞ

sub



log

1306



ð21Þ

It is often desirable to deposit ﬁlms that possess more complex stoichiometries.

These ﬁlms may include mixed-metal, and binary/ternary metal oxides and sulﬁdes

that are of importance for emerging applications for catalysis and microelectronics.

Figure 4.64. Top: Thermogravimetric analysis of organomagnesium CVD precursors, indicating the

dependence of molecular structure on its decomposition temperature. Bottom: Tandem TGA/DSC

analyses showing a single endothermic event (sublimation) that coincides with the mass loss. The

ligand abbreviations are: dpm ¼ 2,2,6,6-tetramethyl-3,5-heptanedionate, TMEDA ¼ N, N, N

, N

tetramethylethylenediamine, hfa ¼ 1,1,1,5,5,5-hexaﬂuoro-2,4-pentanedione, HTEEDA ¼ H(N, N, N

-tetraethylethylene diamine). Reproduced with permission from Chem. Mater. 2005, 17(23), 5697.

4.2. Silicon-Based Applications 311

Although these materia ls may be generated through use of individual species, such

as Y, Ba, and Cu b-diketonates for YBa

7d

superconductor ﬁlms,

[61]

it is most

advantageous to use a single-source precursor for thes e applications. The ﬁxed ratio

of component metals in the individual precursor molecules offers a unique route

toward stoichiometric control over thin-ﬁlm and nanoparticulate growth. Further,

lower deposition or post-annealing temperatures are often required for single-source

precursors relative to their co-reactant analogs.

Perhaps the greatest “Achilles heel” of using single-source precursors is the

complex and expensive synthetic procedures that are often required. Although this

is usually overcome in laboratory-scale processes, this represents a major limitation

for industrial scale-up consi derations. Recently, a one-step reaction scheme was

designed to yield a host of single-source precursor molecules that were suitable

for the CVD growth of I–III–VI semiconductors (Figure 4.65).

[62]

Subsequent CVD

studies using the ternary sulﬁde precursors were performed using an aerosol delivery

methodology, with ﬁlms of comparable purity to those generated by highly

expensive procedures such as PVD from co-evaporation of source metals.

Next-generation patterning: “soft” lithography

The design of transistors, with feature sizes currently much less than 100 nm, will

need to move beyond traditional photolithography in order to continue Moore’s

Law and the miniaturization of electronic devices. In addition to the exorbitant cost

of photolithography, this technique is not amenable for the patterning of large and

[{PPh

}

Ag(SMe)

In(SMe)

]

[{PPh

}

Ag(SMe)

In(SMe)

]

E = S, R = Et,

L = PPh

, M=Cu

E = Se, R = Ph,

L = PBu

, M=Cu

E = Se, R = Ph,

L = PPh

, M=Cu

E = S, R = Et,

L = PBu

, M=Cu

4NaER + InCl

+ MX + 2L

E = S, R = Et, L = PBu

, M=Ag

E=S, R=Me

L = PPh

, M=Ag

E = S, R = Me, L = PPh

, M = Cu

[{PPh

}

Ag(SMe)

In(SMe)

]

[{PPh

}

Ag(SMe)

In(SMe)

]

[{PPh

}

Ag(SMe)

In(SMe)

]

[{PPh

}

Ag(SMe)

In(SMe)

]

[{PPh

}

Ag(SMe)

In(SMe)

]

Figure 4.65. Synthetic pathway for single-source precursor design. Reproduced with permission from

312 4 Semiconductors

nonplanar substrates – of importance for future 3D devices. Though advanced

techniques such as deep UV (DUV), extreme UV (EUV), and focused-ion beam

(FIB) lithographies (see Chapter 7) are able to push the resolution limit s of pattern-

ing to well below 100 nm, they are much too expensive for the low-cost, high-

volume processing that is required for commercial applications.

[63]

Within the last decade, comparatively inexpensive and scalable techniques known

as “soft lithography” have been the focus of much development. Patterning of a

substrate is afforded by using a master elastomeric (see Chapter 5) stamp that contains

a nanostructured pattern, known as a relief, on its surface. Contrary to photolithogra-

phy, the resolution of the ﬁnal pattern is not limited by light diffraction, but only

depends on the dimensions of the relief structures – typically fabricated in the master

by electron-beam lithography (Figure 4.66). Typically, the mold (or stamp) is com-

prised of poly(dimethylsiloxane) (PDMS), which allows for intimate contact between

the mold/substrate surfaces, even if nonplanar substrates are used. More recently,

other polymers have been developed for this application such as polyimides, poly-

urethanes, and a variety of substituted siloxanes – especially ﬂuorinated analogues

due to easy release after molding, and lack of swelling by organic solvents.

The technique of replicating a master pattern is aptly termed replica molding.

In theory, the resolution of the replica will be identical to the master. However, due

to the “soft” natu re of the mold, the nanoscale features of the relief may become

distorted due to polymer shrinkage (e.g., solvent evaporation, in situ cross-linking,

mechanical deformation), or interfacial phenomena between the mold and master

surfaces (e.g., differing thermal expansions, adhesive forces

[64]

). In contrast, a hard

mold of Si or quartz exhibits signiﬁcantly less distortion due to their solvent/

chemical resistance, and thermal stabilities at temperatures sufﬁcient to cause

polymer cross-linking. Hard molds, used for step-and-ﬂash imprint lithography

(SFIL, Figure 4.67) and nanoimprint lithography (NIL, Figure 4.68),

[65]

are com-

monly used to pattern mater ials such as CDs, DVDs, and holographic images on the

front of most credit cards.

[66]

A common application for elastomeric molds is for micro- (m-CP) or nanocontact

printing, where a self-assembled monolayer (SAM) is placed on both planar

[67]

and

curved

[68]

surfaces via contact with the reliefs on the mold (Figure 4.69). SAM s will

be an important architecture for the next generation of nanostructured materials.

The archetypical example of a SAM is the chemisorption of alkylthiols on a gold or

silver surface, which results in self-assembly/alignment into a 3D forest array

(Figure 4.70). Applications for SAMs span a number of ﬁelds from sensors to

high-density storage; a recent precedent illustrates the selective adsorption and

spontaneous alignment of carbon nanotubes (CNTs)

[69]

and directed growth of

nanowires

[70]

from SAMs.

In order to improve the stamping resolution of the elastome ric stamp, there have

been recent improvements in both the stamp and “molecular inks” (e.g., alkylthiols,

silanes). In particular, traditional PDMS exhibits a relatively high elasticity that

limits possible relief linewidths; on the other hand, small molecular weight inks

4.2. Silicon-Based Applications 313

Figure 4.66. Comparison of (a) conventional photolithography/electroplating with (b) soft lithography.

Shown in (b) is replica molding which consists of the formation of a PDMS stamp, and subsequent

replication of a master in a photo- or thermally curable prepolymer. Reproduced with permission from

Gates, B. D.; Xu, Q.; Stewart, M.; Ryan, D.; Willson, C. G.; Whitesides, G. M. Chem. Rev. 2005, 105,

314 4 Semiconductors

exhibit diffusion during patterning. Hence, the following complementary strategies

have been employed:

(i) Using a composite two-layer stamp comprised of a 30 mm hardened PDMS

coating on a 2–3 mm thick PDMS support

[71]

; especially in tandem with sharp,

V-shaped grooves

(ii) Using high molecular weight inks such as dendrimers

[72]

and biological mole-

cules (e.g., proteins)

Using a combination of the above modiﬁcations has now extended nanocontact

printing to well below the 30 nm regime – even as low as 2 nm!

[73]

Though PDM S has dominated the ﬁeld of soft lithography, DeSimone and cow-

orkers have introduced ﬂuorinated perﬂuoropolyether (PFPE) elastomers, which

offer better solvent resistanc e and cleaner mold-release characteristics.

[74]

This

Transparent template

(mold)

Planarization layer

Resist dispenser

Imprint resist

Imprint fluid fills

template pattern

UV blanket

exposure

Exact replica of

template pattern

Substrate

Step 1: Orient template and substrate

Step 2: Dispense drops of liquid imprint resist

Step 3: Lower template and fill pattern

Step 4: Polymerize imprint fluid with UV exposure

Step 5: Separate template from substrate

Figure 4.67. Schematic of step-and-ﬂash imprint lithography, SFIL. The SEM images shown in (b) and

resultant imprint in a resist material. Reproduced with permission from J. Microlithogr., Microfabr.,

Figure 4.68. Schematic of nanoimprint lithography (NIL), and SEM images of a nanopatterned mold and

arrays imprinted in poly(methyl methacrylate). Reproduced with permission from J. Vac. Sci. Technol. B

4.2. Silicon-Based Applications 315

Figure 4.69. Top: Schematic of the general process of nanotransfer printing. SEM image B illustrates a

20-nm grooved gold layer transferred onto a GaAs surface. Image C shows a multilayered stack of 20-nm

thick layers of parallel grooves; the channels in adjacent layers are aligned perpendicular to one another.

Society. Bottom: illustration of the diversity of microcontact printing. Shown is the printing of

hexadecanethiol (HDT) via: (a) a planar surface with a planar stamp (I: SAM printing, II: etching, III:

deposition), (b) large-area printing on a planar surface with a rolling stamp, and (c) printing on a

nonplanar surface with a planar stamp. Reproduced with permission from Angew. Chem. Int. Ed. 1998,

316 4 Semiconductors

version of soft lithography is coined PRINT – Particle (or Pattern) Replication in

Nonwetting Templates. Figure 4.71 illustrates the processes involved in the PRINT

technique. A master template with the desired features is ﬁrst prepared via e-beam

lithography or photolithography/etching of a Si substrate. A homogeneo us coating of

a photocurable liquid PFPE is then applied and photochemically cross-linked to yield

a mold that may be peeled away from the master. To generate a 2-D array of particles

or free particulates, the mold is ﬁlled with an appropriate liquid via capillary ﬁlling

without wetting the land area around the cavities. The liquid in the mold is then

converted to a solid using evaporation, curing, lyophilization or other mode of phase

transition, and the array may then be transferred to another surface. It should be noted

that this process is also amenable for the replication of naturally-occurring objects

such as virus particles,

[75]

block copolymer micelles, and protein particles.

[76]

The PRINT technique is analogous to micromolding in capillaries (MMIC) or

solvent-assisted micromolding (SAMIM), with the use of perﬂuorinated polymers

rather than silicones (Figure 4.72). The use of PFPE instead of PDMS molds offers

the following beneﬁts (Figure 4.73)

[77]

(i) Lower surface energy and solvent resistance, which allows one to selectively ﬁll

the nanosized cavities of the mold without wetting the surrounding area – giving

rise to a diverse range of particulate morphologies and surface functionalization

(Figure 4.74), of extreme importance for n anomedicine applications

[78]

;

(ii) Non-stick properties allow the organic particles to be easily removed from the

mold.

Figure 4.70. Schematic of a self-assembled monolayer (SAM), illustrating the organization of the alkyl

chains via van der Waal interactions, and the close-packed array of sulfur atoms on the gold surface.

Reproduced with permission from Xia, Y.; Whitesides, G. M. Angew. Chem. Int. Ed. 1996, 37, 550.

4.2. Silicon-Based Applications 317