Ghodssi R., Lin P., MEMS Materials and Processes Handbook

Подождите немного. Документ загружается.

784 A.D. Raisanen

Fig. 10.23 Gas phase diffusion schematic. Process equipment is essentially a modiﬁed diffusion

furnace with provision for introducing dopant gases and dealing with toxic efﬂuents

10.4.2 Solid State Diffusion

Sources of dopant atoms can be also conveniently introduced by applying a thin

ﬁlm of material to the substrate to be doped. These thin ﬁlms are often glass mate-

rials containing boron or phosphorus. Films of phosphorus- or boron-doped glass

can be applied with LPCVD deposition equipment using a low-pressure silane and

oxygen chemistry (at about 400

◦

C and 300 mtorr) with a small amount of added

or B

. The dopant gas is often diluted in N

,orevenSiH

, to avoid hav-

ing to work with a 100% toxic gas cylinder under high pressure. Liquid dopants

such as Trimethylphosphite (P(OCH

)

and Trimethylborate (B(OCH

)

), available

from Air Products corporation of Allentown, PA, have become popular low-toxicity

substitutes for PH

or B

, respectively, and may be added to the SiO

deposition

process by entraining dopant vapor with N

or He in a bubbler. Typical doping levels

for boron- and phosphorus-doped glasses are 4–7 wt%, as higher levels of doping

are hygroscopic and can be difﬁcult to work with.

Spin-on liquid materials are a very convenient doping method requiring little

process equipment and no toxic material storage [57]. These materials are available

for Sb, As, P, and B doping from Honeywell Electronic Materials in Tempe, AZ,

Emulsitone Corporation in Whippany, NJ, or Filmtronics Corporation in Butler, PA.

A typical process used in the author’s laboratory is illustrated i n Fig. 10.24, where

the liquid dopant is ﬁrst spin-cast on a silicon wafer. Film thickness is generally not

critical. Excess solvent is driven off by a 15 min bake on a hotplate at 200

◦

C. The

coated wafers are loaded into a diffusion tube in a N

ambient and a diffusion pro-

cess is performed, with time and temperature selected to produce the desired doping

proﬁle. Finally, the dopant glass is removed with a 10:1 solution of hydroﬂuoric

acid.

10 Doping Processes for MEMS 785

Fig. 10.24 Use of liquid

spin-on dopant materials

Solid-source dopant materials are another convenient source of dopant that does

not require toxic materials. Boron and phosphorus solid diffusion dopant sources

are available from Saint-Gobain Ceramics in Amherst, NY. Donor wafers made

of glass or silicon carbide saturated with boron or phosphorus-rich materials are

added to a furnace wafer carrier with the silicon wafers to be doped as shown in

Fig. 10.25. Silicon wafers are sandwiched between dopant source wafers and are

used in a slightly oxidizing ambient at atmospheric pressure. At the high temper-

atures utilized for diffusion, dopants diffuse out of the solid sources and immerse

the substrates in an ambient rich in dopant atoms, allowing indiffusion of dopant

into the target substrate. A thin coating of borosilicate or phosphosilicate glass is

left on the silicon surface after annealing which must be removed with hydroﬂuoric

acid. Dopant wafers are usable f or dozens of runs and are inert, nonﬂammable, and

nontoxic.

An old but extremely effective method of producing metallurgical junctions for

contacts and other heavily doped applications is the so-called alloy contact illus-

trated in Fig. 10.26, where an aluminum metal thin-ﬁlm or small bead of material is

placed in contact with the semiconductor to be doped and contacted. Annealing the

contact above the Al–Si eutectic temperature at 577

◦

C allows some metal to diffuse

into the semiconductor, doping it p-type as indicated in Fig. 10.3. This forms a heav-

ily doped p-type region immediately below the metal, producing a diode in n-type

material (a), or an ohmic contact in p-type material (b). This technique also works

with indium metal, although it can be trickier due to the low melting point of indium.

Aluminum–antimony alloys can be used to form ohmic contacts to n-type material.

This technique was used early in the development of microelectronics to produce

diffused junctions for diodes and semiconductors, but was eventually replaced with

more controllable processes such as gas and solid-state diffusion [7].

786 A.D. Raisanen

Fig. 10.25 Use of solid-source dopant wafers in diffusion tube

Fig. 10.26 Formation of

metallurgical junctions for

self-doped contacts. Al or In

metal balls sintered on the

surface of n-type silicon

produce diodes, and on p-type

silicon produce ohmic

contacts

10.4.3 Masking Materials

Doping blanket ﬁlms or the entire surface of silicon wafers is straightforward, but

many devices require dopants to be placed in a speciﬁc area. This process requires

a mask material that can withstand the high temperatures encountered during the

diffusion process, which can run as high as 1250

◦

C for a deep diffusion in a SiC

furnace tube. The mask must also have a low diffusion coefﬁcient for the dopant

that is being introduced. In practice SiO

and Si

ﬁlms meet these requirements

10 Doping Processes for MEMS 787

[58], and are generally convenient to deposit in any laboratory performing silicon

processing. Silicon nitride can be deposited with dichlorosilane and ammonia in a

LPCVD furnace [51]. To make certain it adheres to the silicon wafer throughout

the diffusion temperature cycle, it is advisable to put a thin (250–400 Å) thermally

grown SiO

layer underneath. Openings in the nitride can be formed by standard

lithographic t echniques and plasma etching in CF

or SF

.SiO

masks can be grown

by thermal oxidation and subsequently patterned with HF or plasma etching, or they

can be grown in speciﬁc areas by the LOCOS (local oxidation of silicon) process

(which uses a nitride mask) [7, 59]. Nitride ﬁlms of about 1500 Å thickness are

typical, and oxide ﬁlms from 2500 Å– 1.25 μm are common. See Fig. 10.27.

Fig. 10.27 Masking during diffusion. The mask material must have a very low diffusivity for

the dopant species while being able to resist temperatures of 1100

◦

Cormore.SiO

and Si

are common mask materials. Openings in the mask in (a) allow dopant atoms to diffuse into the

silicon. Lateral diffusion under the mask results in diffused areas slightly larger than the openings

in the mask as shown in (b)

10.4.4 Modeling

A variety of software packages is available to accurately model diffusion of dopant

atoms in the solid state, many based upon the SUPREM series of simulation models

developed at Stanford University. Code for one- and two-dimensional simulation

is available from Stanford University directly [60] or incorporated into a variety of

commercial software packages such as ATHENA by Silvaco Corporation. Modeling

ultrashallow junctions for advanced CMOS devices requires the latest code to

account for dopant clustering, transient diffusion enhancement, and other effects

not incorporated in the simple Fick’s law (Equations (10.3)–( 10.5)). However, for

MEMS applications, detailed modeling of dopant tails, transient diffusion and other

second-order effects are often more than necessary, and simple one-dimensional

analytical expressions are usually adequate to model the process required for doping

a boron etch stop to produce a given silicon diaphragm thickness.

788 A.D. Raisanen

Fig. 10.28 Simulation of boron diffusion from a constant surface concentration of 1 ×10

−3

and 1 ×10

−3

at 1100

◦

C for 90 min. Solid lines are simulations using ATHENA, and discrete

markers are computed via Equation (10.4)

Figure 10.28 illustrates a comparison between an ATHENA simulation (solid

lines) and calculations from Equation (10.4) for a 90 min, 1100

◦

C boron diffusion

with constant surface dopant concentrations of 10

and 10

−3

. Agreement is

quite good and only departs at the highest doping densities, where concentration

effects and electric ﬁelds induced by the high dopant concentration lead to higher

effective diffusivities. Even these factors can be substantially accounted for by using

diffusivities adjusted for concentration effects or ambient environments other than

inert gases (e.g., oxidizing environments during diffusion). See data tabulated in

Properties of Silicon (INSPEC, London, 1988). Brigham Young University at the

time of this writing has an excellent set of simple online diffusion and implant proﬁle

calculation tools at their Electrical and Computer Engineering, which at the time of

this writing is hosted at http://cleanroom.byu.edu.

10.5 Ion Implantation

Ion implantation is a process for doping semiconductor substrates using ballis-

tic implantation of high-energy ions. It is more convenient than conventional

high-temperature diffusion processes as it allows simple control of dopant den-

sity by altering implant current and time, and good control of the dopant proﬁle

by altering ion energy and thus implantation depth. A broader range of masking

materials is possible as well, inasmuch as the mask does not have to withstand a

high-temperature process, increasing convenience in patterning and mask removal.

10 Doping Processes for MEMS 789

Signiﬁcant crystal damage occurs during the implantation process, which gener-

ally must be repaired by annealing processes to restore useful crystal structure and

activate the dopant atoms.

A high-energy ion affecting a solid will undergo a number of electronic and phys-

ical collisions until it comes to rest. In general, heavy ions, such as arsenic, do not

penetrate as far as light ions, such as boron. The interaction between incoming ions

and the silicon lattice is composed of two components, nuclear and electronic. The

nuclear interaction is essentially a physical collision between the incoming ion and

a l attice atom, and scales with the atomic weight of the ion. The nuclear interac-

tion also depends on energy, with a lower collision cross-section at high energies.

The electronic interaction is an interaction between the incoming charged ion and

the electronic structure of the solid. Motion of the ion through the solid produces

a realignment of electrons in the solid, gradually draining kinetic energy from the

ion as it passes. As the ion enters the solid, it loses energy from electronic interac-

tions, increasing the effective nuclear cross-section of the ion. A nuclear collision

will eventually occur, with the probability of that collision being maximized at some

depth below the surface at which the nuclear collision cross-section is maximized by

the electronic energy loss. Plots of the ion density deposited in a solid as a function

of penetration depth will have a characteristic peak depending on ion energy and

mass, with an approximately Gaussian proﬁle around the peak position, as illus-

trated in Fig. 10.29 for 100 keV implants in silicon calculated using ATHENA and

a simple Gaussian model.

Fig. 10.29 Simulated dopant proﬁle for ion-implanted boron, phosphorus, and arsenic at 100 keV

and 1 × 10

ions/cm

−3

. Lighter ions penetrate more deeply into the silicon. Calculated using

ATHENA

790 A.D. Raisanen

The depth of the Gaussian implant concentration peak and the width of the peak

(standard deviation), called the longitudinal straggle, is determined by the relative

electronic and nuclear scattering cross-sections for the ion and substrate atoms.

These cross-sections are tabulated for many i ons, and can be calculated to a high

degree of accuracy using software tools such as SRIM (stopping range of ions in

solids) [61]. SRIM calculations for ion range and longitudinal straggle are shown

in Fig. 10.30 for boron, phosphorus, and arsenic implants in silicon up to 300 keV.

These calculations may be used to estimate the position of the highest dopant con-

centration as well as the width of the implanted dopant distribution. The maximum

dopant contribution of the implanted distribution is related to the total implanted

doseby[59].

Q =

√

2πR

, (10.6)

where Q is total dose implanted (atoms/cm

), R

is the longitudinal straggle from

SRIM or Fig. 10.30, and C

is the peak concentration in atoms/cm

These simple calculations allow estimates to be made for the proper energy and

dose to provide a given implant proﬁle, but more sophisticated calculations are

best conducted using a simulation tool such as SUPREM IV [60]orATHENA

(Silvaco Corp, Santa Clara, CA). In particular, activation of dopants will broaden

the implanted proﬁle through diffusion, and ions implanted in a crystalline solid

undergo processes such as channeling, which are not accounted for by SRIM and

simple Gaussian proﬁle calculations.

Unlike diffusion methods, which result in the highest dopant concentration at the

surface in contact with the dopant source, ion implant processes can easily place the

ion dose well below the substrate surface, enabling structures such as deep buried

channels. Ions can also be implanted through overlying structures at sufﬁciently

high energies. Ion implant offers unique process opportunities by decoupling dopant

concentration (ion current) and dopant proﬁle (ion energy), which is not possible

with basic diffusion processes.

10.5.1 Equipment

Ion implantation is performed by a small particle accelerator that ionizes atoms

from a solid or gaseous dopant source. To produce a beam of dopants, a suitable

source material must ﬁrst be vaporized and ionized. A simple ion source can be

constructed by using a hot ﬁlament to ionize a gaseous feedstock containing dopant

atoms. Historically, BF

gas is used to produce boron dopants, PH

is used for

phosphorus implants, and AsH

is used for arsenic implants. Because many of these

gas dopant materials are highly toxic, there is a trend toward safer delivery systems

than high-pressure gas bottles, such as the subatmospheric Safe Delivery Source

systems pioneered by Advanced Technology Materials Corp. Solid materials can be

utilized as source materials by direct thermal sublimation or vaporization in a more

complex electric arc or inductively coupled plasma source.

10 Doping Processes for MEMS 791

Fig. 10.30 Calculated ion range and longitudinal straggle, in angstroms, for boron, phosphorus,

and arsenic ions in silicon. Calculated using SRIM [61]

Once a stream of ions is produced from a source, a mass selector extracts the

desired dopant from the ionizer and accelerates the ions to the desired energy.

Mass selectors may be conventional magnetic sectors as illustrated in Fig. 10.31

or quadrupole mass ﬁlters. Focusing optics produce a collimated beam that can be

directed onto the substrates, and rastering electrostatic or magnetic optics scan the

beam across the wafer surface. In high-current machines, the substrates themselves

are mechanically scanned across the beam. At high beam currents an electron ﬂood

source is utilized to neutralize the surface charge produced by the ion bombardment,

and a wafer cooling system may be used to prevent excessive temperature increase

of the wafers and subsequent damage to masking materials.

792 A.D. Raisanen

Fig. 10.31 Schematic of a medium-current ion implant system

Specialized implanters are available for speciﬁc tasks. Medium-current

implanters operate well from about 5–200 keV ion energy, with currents up to a

few milliamps. Individual substrates are typically held stationary, and the beam

is electronically rastered over the surface by an X−Y scanning system. Medium

current machines are good for producing moderate implant doses (up to about

(>1 × 10

− 5 × 10

ions/cm

) with high throughputs, suitable for such appli-

cations as channel stop and threshold adjust implants. Obviously, higher doses

are obtainable by simply using longer exposures and accepting lower process

throughput. High-current implanters are optimized for producing very large doses

(>1 ×10

−5 ×10

ions/cm

) for source/drain implants, polysilicon doping, and

SOI wafer production. Typically high-current implanters mount a batch of wafers

on a cooled rotating stage assembly. The ion beam is rastered in a linear or fan pat-

tern rather than in two dimensions, and the stage is rotated and translated through

the linear beam pattern to uniformly irradiate each wafer.

A third variation of the basic ion implanter technology is the low-energy

implanter, optimized for producing ion beams at low energies, <10 keV.

Conventional implanters have difﬁculty in obtaining high-current beams at the

low energies useful for producing very shallow dopant proﬁles, so low-energy

implanters have been developed with short beam paths and optics optimized for

low-energy, high-current ion beams.

10.5.2 Masking Materials

A wide variety of masking materials are usable with ion-implant processes. Silicon

dioxide, silicon nitride, polysilicon, and metals are sometimes used for implant

masks, especially in “self-aligned” processes that place implanted species in silicon

laterally adjacent to the edge of an overlying structure. Conventional photoresist

is probably the most commonly used material as it is easy to apply, pattern, and

10 Doping Processes for MEMS 793

remove after the implant is complete, in contrast to the hard masking materials

required for high-temperature diffusion processes. Resist damage or burning can

be an issue with high-current implants, and aggressive removal methods with oxy-

gen plasmas or sulfuric acid/peroxide stripping solutions are often required to strip

the damaged polymer. Candidate materials for masks, or even complete stacks of

dissimilar materials, can be easily simulated for ion stopping effectiveness using the

SRIM [61] simulation software.

10.5.3 Modeling

Simple calculations of implanted proﬁles may be performed using software tools

such as SRIM [61] or by estimating implant depths from tabulated cross-section

values [61, 62]. For relatively coarse applications, such as generation of a heav-

ily doped boron stop layer or production of a heavily doped polysilicon layer,

these estimates will often sufﬁce. However, thermal cycles must be included after

implant as dopants will diffuse through the crystal during activation with conven-

tional thermal methods. This can be approximated by Gaussian broadening the

implanted proﬁle via Equation (10.5), or by simply using a modeling tool such as

SUPREM.

For more precise work including perturbations such as ion channeling, amor-

phous layers, or implant through oxides or other surface ﬁlms, a 1-D or 2-D process

simulator is indispensable. For example, Fig. 10.32 illustrates a high-dose boron

implant of 1 ×10

B11+ at 100 keV, before and after an anneal in an inert environ-

ment of 90 min at 1100

◦

C. As-implanted, the boron proﬁle has a signiﬁcant shoulder

at higher depths due to ions “channeling” through preferred directions in the crystal

structure [63]. One such channel is shown in Fig. 10.33 through a silicon–diamond

structure. At the correct angle, long “tunnels” extend through the crystal struc-

ture allowing ions to penetrate anomalously deep with little chance of scattering.

The presence of this shoulder is not predicted by range calculations in Fig. 10.30,

which would lead to predictions of the implanted proﬁle being shallower and of

higher concentration than it actually is. The channeling effect can be suppressed

by implanting through a thin screening oxide and adding a small tilt to the sub-

strate to prevent ions from entering at the channeling angle. Simulation codes based

on SUPREM can simulate the channeling effect by using dual-Pearson implant

proﬁles [64].

10.5.4 Crystal Damage

Ion scattering from a solid through electronic interactions produce relatively lit-

tle damage to the crystal structure, but actual collisions between ion cores during

nuclear interactions can cause severe crystal damage. These collisions are essen-

tially elastic in nature, and sufﬁciently high-energy events can cause a cascade of