Baca A.G., Ashby C.I.H. Fabrication of GaAs Devices

Подождите немного. Документ загружается.

Dry etching of GaAs and related alloys

Cl-containing gas for very long after etching, when the wafer has

already been heated by ion bombardment, should generally be

avoided. However, a slight amount of chemical etching may serve

to remove some of the ion-induced near-surface damage so a brief

opportunity for thermal reaction may improve surface electronic

properties.

5.6 PLASMA ETCHING AT VERY LOW ION ENERGIES

Although most dry etch processes employ a plasma to generate

highly reactive atoms and radicals, the term plasma etching (PE)

is generally applied to processes where there is negligible bom-

bardment with high-energy ions. In PE, the wafer sits within the

plasma but it is not on a powered electrode and few if any ions

with energies over 10 eV strike the GaAs surface (FIGURE 5.4).

The atomic-displacement threshold energy for GaAs is of the order

of 40 V, so processes at energies below that value do not produce

lattice damage that can degrade device operation.

FIGURE 5.4 Barrel etcher for

low-energy plasma etching.

Rates can be fast when samples are heated but proﬁles are

isotropic. Rates up to 5 μm/min can result at 5 mtorr and 60

◦

However, surfaces are virtually damage free. PE is not generally

used for regular GaAs etching because it lacks the primary advant-

ages of proﬁle control provided by energetic ion bombardment in

other dry etching techniques while involving much more expensive

and complicated hardware than wet etching.

Because PE is a low-damage process, it can sometimes be

used to remove surface damage produced by more commonly

employed dry processes. A PE process can be implemented in

a high-density plasma etcher while the reactant-generating ICP

or ECR power is on by reducing the power of the RF-powered

electrode to produce very low bias voltages (<10 V) at the end

of the etch. This can remove the thin surface layer that has been

damaged by ion bombardment during the etch. Etch rates are low

due to the low pressures (<10 mTorr) used in these processes, so

proﬁle degradation is avoided.

While PE is not very useful for GaAs patterning, it is very

useful for removal of residual organics from PR processes when

an oxygen plasma is used. The hardware for this process is some-

times referred to as an asher. The low ion energy avoidsdegradation

of the surface electronic properties of the GaAs while the oxygen

atoms remove trace organic contamination from the surface where

photoresist has been removed during PR pattern development

before etching. This permits better pattern deﬁnition during the

etch process. It is equally valuable after solvent removal of most

of the photoresist following an etch. Without this PE cleaning

152

Dry etching of GaAs and related alloys

of the surface, it would be difﬁcult to obtain highly reproducible

metal-semiconductor contacts.

5.7 CONVENTIONAL REACTIVE ION ETCHING (RIE)

Reactive ion etching (RIE) is the workhorse of dry etching tech-

niques for GaAs. The equipment involved is widely available

in most clean rooms and uses the most reasonably priced basic

hardware of any of the dry processes. Two geometries have been

commonly employed: planar and hexode (FIGURE 5.5). In the

planar geometry, which is by far the most common, the radio fre-

quency (RF) plasma is formed between two parallel plates with

one electrode powered and the other grounded. In the hexode geo-

metry, wafers are mounted on a hexagonal column located within

a grounded chamber. Most reactors today are of the planar variety.

In both cases, RF power is applied to the electrode upon which the

wafers sit.

(a)

(b)

FIGURE 5.5 Typical RIE

etchers. Wafer sits on powered

electrode: (a) Planar (b) Hexode.

Because of the RF character of the applied power, the actual

voltages present at the powered electrode vary with time during

the RF cycle, making the environment experienced by the wafer

more complex than is implied when describing the plasma system

in terms of input power or even applied bias voltage (FIGURE 5.6).

Typical commercial RF plasma hardware operates at a frequency

of 13.56 MHz. The rapidly varying potential allows the elec-

trons, which have very high mobilities, to respond to the varying

RF potential. However, the ions, being much heavier and pos-

sessing much lower mobilities, respond only to the time-averaged

potential, V

The time-averaged potentials at different locations between the

powered and grounded electrodes are shown in FIGURE 5.7. Ions

are accelerated in the ﬁeld and strike the wafer at energies between

e(|V

|+V

) and zero. In a typical reactor, where the area of the

grounded surfaces is large compared to the area of the powered

electrode, V

will be relatively low (20–50 V) and the maximum

energy of ions incident on the wafer is reasonably well described,

within about 10%, by V

. Because it is the energy of the incident

ions that is so important in determining the etching characteristics,

it is important to measure the DC potential of the cathode (powered

electrode) whenever possible rather than using the RF power alone

to describe an etching process.

The same RF powersupply both dissociates the source gasesinto

more reactive radicals and ions and provides the ion acceleration

voltage that determines the ion bombardment energy during the

etch cycle. This means that there is no independent or even semi-

independent control of the degree of dissociation in the plasma and

153

Dry etching of GaAs and related alloys

the energies of the ions striking the wafer. The wafer is etched by

neutrals and by ions whose energies are greater than the positive

space-charge potential of the plasma.

time

V(t)

FIGURE 5.6 Time-dependent

variation of voltages during several

RF cycles in an RIE reactor. V

plasma potential: time-average of

potential through most of plasma

and at grounded electrode; V

self-bias potential: time-average of

potential at powered electrode.

ave

cathode dark

space (sheath)

negative glow

FIGURE 5.7 Time-averaged

potential in a planar RIE etcher.

It is important to remember that a plasma is actually a com-

plex mixture of ions, electrons and neutral species. In etching

plasmas, the neutrals greatly outnumber the ions, and a complete

description of the plasma would require knowledge of the number

and identity of the neutral species in the plasma as well as the

range of ion energies and ion densities. In addition, there are ener-

getic neutrals which form when electrons are transferred between

energetic ions and neutrals. Depending on where in the sheath

the neutralisation occurs, the incident neutral energy will vary

widely. The degree of ionisation (ion/neutral ratio) is usually in

the range of 10

−4

–10

−6

, so it is clear why etch contributions

due to neutrals are so important and why it is so difﬁcult to

fully characterise what species are actually reacting at the wafer

surface.

The energies of individual ions striking the wafer will depend on

their history. As an ion passes through the sheath as it accelerates

towards the cathode, it may undergo collisions that alter its energy

and charge state. Some ions may be neutralised by collisions with

neutral species. This can produce an energetic neutral that moves

towards the wafer with an energy determined by how much it had

been accelerated prior to the neutralisation. The new ion resulting

from the collision will accelerate towards the surface and strike

with a fraction of the maximum possible energy. The result is

an energy spread of up to several hundred eV for the chemical

species striking the surface. Consequently, reaction at the surface

will involve a complex mixture of low-energy neutrals and ions

and higher energy neutrals and ions.

Since charge neutrality in the plasma must be maintained, most

ions are positively charged. In a typical RF plasma there may be

–10

ions/cm

. In contrast, the neutral density at 10 mTorr

(a typical RIE pressure) is 10

–10

molecules/cm

. Clearly,

neutral species are predominant and will play a vital part in any

RIE process while ions largely affect relative product desorption

rates. Ions accelerate product desorption where they strike, so

rates on bombarded surfaces become faster than rates on side-

walls, leading to anisotropic proﬁles. In general, chemical etching

by plasma-generated radicals with its associated high rates, low

anisotropy and low damage is dominant when pressures are high

and sheath voltages are low. As pressure increases, reactant con-

centrations increase, but at the same time ion energy decreases due

to the decreasing self-bias voltage. Since rates tend to decrease

with decreasing ion energy, increasing pressure ﬁrst increases

etch rates by increasing reactant supply, then decreases them

154

Dry etching of GaAs and related alloys

as the rates shift from reactant-supply dominated to ion-energy

dominated.

In contrast, sputtering is dominant at low pressures and high

sheath voltages. This increases both the anisotropy and the damage

while reducing the material selectivity of an etch process. Higher

ion ﬂuxes at a ﬁxed energy and higher ion energies at a ﬁxed

ﬂux increase the sputtering rates. Higher ﬂuxes and energies also

deposit more energy into the wafer and can produce substantial

heating if the wafer is not actively cooled. This will accelerate the

chemical reaction rates, including those with neutral molecular

species. The proﬁle evolution from the mostly chemical (low ion

energy) to mostly sputtering (high ion energy) is illustrated in

FIGURE 5.8.

chemical etching

both mechanisms

physical sputtering

FIGURE 5.8 Proﬁle evolution

depending on ion and neutral

contributions.

It is customary to describe process conditions in terms of pres-

sure, gas ﬂow rate, RF power and applied bias, but this is grossly

inadequate to deﬁne the chemical characteristics of a plasma. The

great importance of the neutral species, which are not directly

measured by any of the diagnostics on standard reactors, and the

variation in ion densities at different energies are the probable

origin of the markedly different etching rates obtained with nom-

inally identical RF power even in nominally identical reactors from

the same manufacturer. Rates reported in the literaturefora speciﬁc

set of conditions should always be taken only as guidelines and

starting points when transferring a process into a different reactor.

Each reactor must be individually characterised with a process for

reliable etching results to be obtained. Some RIE rates and their

associated reaction conditions are listed in TABLE 5.1.

5.7.1 Halogen-based plasmas for RIE

A variety of chemistries have been employed in RIE of GaAs, but

the most commonly used ones are halogen-based (F, Cl, Br, I).

The most common ones employ Cl

and BCl

, separately or in

combination. SiCl

is another good Cl source gas. Rates can vary

by two orders of magnitude. Processes with intermediate rates tend

to produce good surfaces. Higher etch rates, such as those produced

when Cl

is the sole source of Cl, tend to produce rougher surfaces

and decreased anisotropy.

Typical chemistries employ Cl-based source chemicals, such

as Cl

and BCl

, that are dissociatively excited in the plasma to

produce Cl atoms that react readily with the GaAs surface. The

removal of the products GaCl

and AsCl

is facilitated by the

simultaneous ion bombardment that sputters away the products

from the surface, producing much faster etch rates where the ions

strike than in regions without ion bombardment. Sputtering can

remove even partially chlorinated GaCl

and AsCl

(x = 1–3),

155

Dry etching of GaAs and related alloys

TABLE 5.1 Reactive ion etching rates for GaAs.

Etchant Rate Pressure Flow RF power Self Ref. Comments

(μm/min) (mTorr) (sccm) density bias

(W/cm

)(−V)

4.5 150 3–23 0.44 [a] via drilling

3.0 85 0.8 400 [b]

1.4 5 270 [c] vertical walls

0.15 2 40 0.1 150–160 [d] low damage, semi-isotropic

5Cl

: 3Ar 0.09 5 20 0.14 300 [e] equirate versus AlGaAs

and vertical proﬁle

4Cl

: 1Ar 0.80 5 200 [c] severe undercut

3Cl

: 1Ar 3.4 15 12 0.75 330 [f] isotropic

1Cl

: 5BCl

0.6 60 0.42 270 [g] high reproducibility

1Cl

: 1BCl

: 3Ar 0.9 15 100 – 300 [h] very smooth vias

BCl

0.017 11 20 0.20 [i] very smooth

1BCl

: 9Ar 0.28 15 0.27 300 [j] equirate versus AlGaAs proﬁles,

resolution, rates f(W,P gases)

SiCl

0.129 10 10 0.81 600 [k] no polymer ﬁlm

0.067 10 10 0.27 275 [k] no polymer ﬁlm, max. at 10 mtorr

1SiCl

/4SiF

0.31 60 10 – 60 [l] 490 : 1 versus Al

0.35

0.65

1CH

:5H

0.016 10 24 0.75 500 [m] donor passivation

:3H

0.036 10 43 1.1 500 [n,o]

:5H

0.038 10 41 1.1 500 [n,o] 2 : 1GaAs/AlGaAs

[a] K.P. Hilton, J. Woodward [Electron. Lett. (UK) vol.21 (1985) p.962]

[b] C. Gagne [Solid State Technol. (USA) (May 1988) p.187]

[c] E.L. Hu, R.E. Howard [J. Vac. Sci. Technol. B (USA) vol.2 (1984) p.85]

[d] B.S. Lee, H. Baratte [J. Electrochem. Soc. (USA) vol.137 (1990) p.980]

[e] S.S. Cooperman, H.K. Choi, H.H. Sawin, D.F. Kolesar [J. Vac. Sci. Technol. B (USA) vol.7 (1989) p.41]

[f] H. Yamada, H. Ito, H. Inaba [J. Vac. Sci. Technol. B (USA) vol.3 (1985) p.884]

[g] H. Tamura, H. Kurihara [Jpn. J. Appl. Phys. (Japan) vol.23 (1984) p.L731]

[h] K.J. Nordheden, D.W. Ferguson, P.M. Smith [J. Vac. Sci. Technol. B (USA) vol.11 (1993) p.1879]

[i] G.J. Sonek, J.M. Ballantyne [J. Vac. Sci. Technol. B (USA) vol.2 (1984) p.653]

[j] A. Scherer, H.G. Craighead, E.D. Beebe [J. Vac. Sci. Technol. B (USA) vol.5 (1987) p.1599]

[k] M.B. Stern, P.F. Liao [J. Vac. Sci. Technol. B (USA) vol.l (1983) p.1053]

[l] W.H. Guggina, A.A. Ketterson, A. Andideh, J. Hughes, I. Adesida, S. Caracci, J. Kolodzey [J. Vac. Sci. Technol. B (USA) vol.8 (1990)

p.1956]

[m] J. Werking, J. Schramm, C. Nguyen, E.L. Hu, H. Kroemer [Appl. Phys. Lett. (USA) vol.58, no.18 (1991) p.2003–5]

[n] V.J. Law, G.A.C. Jones, M. Tewordt [Semicond. Sci. Technol. (UK) vol.5 (1990) p.1001–3]

[o] V.J. Law, M. Tewordt, S.G. Ingram, G.A.C. Jones [J.Vac. Sci. Technol. B. (USA) vol.9, no.3 (1991) p.1449–55]

accelerating the total GaAs etch rate. This difference in etch rate

between surfaces with and without ion bombardment is the origin

of the proﬁle control available with dry etching. The undercutting

of the mask that results in wet etching from nearly equal vertical

and horizontal etch rates is avoided by increasing the vertical etch

rate through sputtering away the intermediate and ﬁnal reaction

products rather than relying solely on the natural volatility of the

ﬁnal products. Depending on the degree of sputtering in a given

process, proﬁles can be varied from isotropic undercut, as in purely

chemical processes, through vertical to angled, as in IBE without

any chemical contributions.

156

Dry etching of GaAs and related alloys

The highest rates are achieved with Cl

as the Cl atom source, but

can produce surface roughness and may exhibit low reprodu-

cibility in etch rates from run to run. The very high rates are due to

a strong chemical component, and reduced anisotropy and surface

roughness can be a problem. The performance of Cl

is improved

by the addition of other gases. Addition of Ar or BCl

increases

anisotropy by increasing the sputtering contribution while addition

of some H

toaCl

/Ar mixture increases both surface smoothness

and anisotropy while decreasing the etch rate somewhat. Addi-

tion of CH

to Cl

/Ar yields smoothness, anisotropy and a rapid

etch rate. However, addition of H-containing sources to the gas

mix produces H atoms that can passivate dopants and may require

post-etch annealing to reactivate them.

BCl

or BCl

/Cl

mixtures tend to produce smoother surfaces

than Cl

-only plasmas. The smooth morphology, good anisotropy

and high degree of reproducibility make the BCl

-based etches

some of the best. BCl

can also act as a getter for oxygen atoms

that are formed in the plasma from residual O

or H

O. The product

of this reaction is the volatile trimer B

. SiCl

may also serve

this gettering purpose. Therefore, if one must work with a non-

loadlocked reactor, use of a BCl

-based chemistry is especially

advantageous since even trace oxygen can retard the etching of

AlGaAs.

When etching heterostructures of GaAs and AlGaAs, either

highly selective etching of one versus the other or equirate etch-

ing of the two materials may be desired. Either may be achieved

with the selection of proper reactants and RF plasma conditions.

Selective removal of GaAs is needed for the gate recess of HEMTs

where the ohmic contact layer must be removed to expose the

donor layer for gate oxide deposition. Similarly in the emitter mesa

process for HBTs, the GaAs contact layer is removed from the

wide-bandgap emitter ﬁrst before etching the emitter and exposing

the base for contact formation.

Ga Ga

FIGURE 5.9 GaF

dimer and

AlF

octahedral matrix.

One of the best ways to achieve selective etching of GaAs versus

AlGaAs is through the addition of an F-containing reactant to the

gas mixture. The formation of relatively non-volatile AlF

on the

surface of the AlGaAs greatly reduces the etch rate. In contrast,

GaF

is readily removed from the surface so etching of GaAs

proceeds relatively unimpeded. The origin of this pronounced dif-

ference is found in the structures of GaF

and AlF

. While GaF

forms the dimer, (GaF

)

, which is quite volatile, AlF

forms an

extended octahedral network that must be broken up and mechan-

ically ejected by physical sputtering (FIGURE 5.9). Consequently,

at the lower bias voltages that one generally employs for selective

etching, the octahedral AlF

network can serve as a barrier to

further etching. The formation of a similar octahedral network

157

Dry etching of GaAs and related alloys

by InCl

accounts for the slower etch rates of InAs, InGaAs and

InGaP versus GaAs. Selectivity will be degraded at high ion ener-

gies where physical sputtering can dominate over the formation of

non-volatile surface chemical products.

A good source of F atoms is SF

. With SiCl

/SF

mixtures of

varying mole ratios, GaAs : AlGaAs etch ratios from 1 : 1 to 500 : 1

have been demonstrated. This large difference in etch rate permits

the use of AlGaAs as an RIE etch-stop layer, with the required

thickness of the AlGaAs layer being lower for higher Al mole

fractions. However, the ion-induced damage below the etch stop

may be problematic for some applications.

Etches employing SiCl

generally produce better surface mor-

phologies, greater anisotropy and less change in surface stoi-

chiometry than the analogous Cl

-based etches. SiCl

produces

little As loss, while Cl

-etched surfaces are quite As deﬁcient.

Since the surface stoichiometry can strongly affect the surface

electronic properties through the presence of a different distribu-

tion of surface states, the choice of etchant can be important to ﬁnal

device performance. There also appears to be less Cl-containing

residue on an SiCl

-etched surface. However, Si can be incor-

porated into the near-surface region by ion implantation at biases

above 250 V and can be present at the 10

/cm

level at the sur-

face, although the level drops rapidly to <10

/cm

at 30 nm with

a 250-V etch bias. While this may be advantageous for forming

good ohmic contacts on n-type material, it could be a problem

for p-type if the top surface were not etched away before contact

formation.

A frequently unintentional way to obtain selective GaAs etch-

ing is to have appreciable amounts of water or oxygen adsorb on

the reactor walls during sample loading without the aid of a load-

lock. A loadlock is an extremely valuable equipment feature when

any readily oxidised material, such as AlGaAs, is to be etched. It

consists of an antechamber to the actual etching chamber where

the sample can be inserted and the moist, oxygen-containing atmo-

sphere can be pumped away before transferring the wafer into the

plasma chamber. This avoids adsorption of signiﬁcant quantities

of water or oxygen on the plasma chamber walls, which sub-

sequently serve as sources of oxygen when the plasma discharge

is on. The great ease of oxidation of Al leads to the formation

of a thin aluminium oxide surface layer that retards etching of

the underlying AlGaAs even at the low oxygen concentrations

available from wall desorption. While GaAs will etch well in

non-loadlocked systems, more reproducible results are favoured

by using a loadlock system for these materials as well. With a

loadlock, equirate etching of AlGaAs and GaAs is possible when

desired.

158

Dry etching of GaAs and related alloys

When equirate etching is desired, the most important require-

ment is a load-locked reactor to avoid formation of aluminium

oxide on the surface. It is also important not to have residual

F present in the reactor. If F-based chemistries have been used

in a reactor prior to an equirate etching process, it will be import-

ant to run a reactor-chamber cleaning process before subjecting

valuable substrates to RIE processing in the reactor.

Other halogen-based plasmas that generate Br or I atoms will

etch GaAs, but these do not, in general, have major advantages

relative to Cl-based chemistries. A major practical problem is the

greater rate of corrosion of the process tool with Br or I.

There may be an advantage to performing a brief water rinse

after Cl-based etching to remove any residual Cl compounds from

the surface that might later contribute to corrosion problems.

5.7.2 Alkane-based plasmas for RIE

A useful alternative to Cl-based plasmas are plasmas employing

alkanes (C

) as the source of reactive radicals. The simplest

alkyl radical formed in these plasmas is the methyl radical, CH

Organometallic compounds, such as Ga(CH

)

are formed by

reaction with Al, Ga or In. These are the same compounds used as

sources of the Column III atoms in metalorganic chemical vapour

deposition (MOCVD). While In does not etch well with Cl-based

chemistries due to the formation of the non-volatile octahedral

(InCl

)

structure (similar in structure to AlF

in FIGURE 5.9),

it does etch well with the alkyl radicals formed in alkane plas-

mas. Consequently, these chemistries are important despite some

practical disadvantages.

The primary problem with alkane plasmas is the ready forma-

tion of polymers with the general composition (CH

)

. Care must

be taken to minimise polymer deposition on the GaAs wafer or

etching will stop as a thick polymer layer blocks access of the

GaAs to reactive alkyl radicals. This problem can be avoided or

minimised by the addition of H

to the gas mixture. The hydrogen

helps remove polymer by reacting with the polymer to form volatile

alkanes that readily desorb from the surface. The addition of heavy

ions, such as Ar, delays the onset of polymerisation and improves

surface morphology, but the chemical effects of hydrogen pre-

dominate in controlling polymer buildup. Rates initially increase

with increasing hydrogen as polymerisation is reduced, but further

increases in hydrogen lead to lower rates due to dilution effects. In

general, alkane/H

ratios between 0.2 and 0.4 work well, with the

optimum ratio following the order propane < ethane < methane.

While hydrogen reduces the polymer buildup, hydrogen passiv-

ation of dopants requires post-etch thermal anneals at temperatures

159

Dry etching of GaAs and related alloys

near 400

◦

C to restore electrical activity. The activation energy for

recovery of donor activity is about 1.75 eV so a 360

◦

C anneal for

5 min gives almost complete reactivation of Si dopants. The depth

of Si dopant passivation by H increases with increasing substrate

temperature during etching and with increasing etch time. Higher

ion energies make reactivation more difﬁcult. With C

etch-

ing at a 410 V self-bias, almost 500

◦

C is required for restoration

of initial carrier concentrations. This is the combined effect of

H dopant passivation and the damage caused by deep penetration

of the light H ion into the substrate by ion implantation.

While thermal annealing above 400

◦

C can reactivate dopants

by dissociating the bound H, the released H can diffuse deeper

into the device so trying to minimise H diffusion during etching

is advisable. In addition, annealing can cause vacancies to diffuse

deeper in the wafer, resulting in electrically active defects. Defect

reactions are discussed in some detail in Section 10.2.2. Annealing

techniques are discussed in Section 7.2.1.

An important practical problem with alkane etches is the build-

up of polymer on chamber walls where ion bombardment does

not occur. Even under optimised etching conditions, frequent and

regular oxygen plasma cleaning is essential.

5.8 HIGH-DENSITY PLASMA ETCHING (HDPE)

A single power supply is used in conventional RIE both to generate

the plasma and to providethe bias potential for accelerating the ions

into the wafer surface. Consequently, one does not have independ-

ent control of the energy of the incident ions and of the degree of

dissociation of the source gases into highly reactive neutrals and

ions. To achieve a high degree of dissociation, high RF powers

are needed but high powers also produce high bias voltages and

more damage. Since it is highly desirable in most cases to minim-

ise damage by minimising ion energy, one would like to be able

to produce a high degree of dissociation to get rapid etching while

keeping bias voltages below 200 V.

This problem has been solved in high-density-plasma etching

where the reactors employ separate power supplies to generate

the reactive plasma species and the ion acceleration voltage. The

use of two separate supplies permits semi-independent control of

the degree of dissociation and the incident ion energy so high etch

rates can be achieved with minimal damage. The ion acceleration

voltage is still provided by a capacitively coupled RF power supply

operating at 13.56 Mz and the wafer sits on the powered electrode.

However, the reactant source gases are dissociated into reactive

neutrals and ions using either electron cyclotron resonance (ECR)

160

Dry etching of GaAs and related alloys

excitationwithmicrowaves, typically at 2.45 GHz (FIGURE 5.10),

or using an inductively coupled plasma (ICP) typically operating

at 2 MHz (FIGURE 5.11).

Plasma

m wave

2.45 GHz

13.56 MHz

Magnets

FIGURE 5.10 ECR-HDPE reactor.

13.56 MHz

2 MHz

Plasma

FIGURE 5.11 ICP-HDPE reactor.

Although many of the same source gases are used in RIE and

HDPE, the actual plasma composition is radically different. HDPE

plasmas are much more highly dissociated into ions, free atoms

and other radicals, producing much faster etching at lower pres-

sures than with RIE. Ion densities in an HDPE reactor can exceed

those in conventional RIE by two to four orders of magnitude

(up to 10

ions/cm

). Typical pressures in HDPE reactors are

between 1 and 10 mTorr while RIE employs pressures of the order

of 5–100 mTorr. Some typical rates and their associated reaction

conditions are listed in TABLES 5.2 and 5.3.

The greatest advantage of HDPE versus RIE is the much lower

ion bombardment energies that are used. Whereas is it difﬁcult

to achieve good etch rates and feature proﬁles at 100-V bias with

RIE, a 100-V bias produces both good rates and vertical proﬁles

with HDPE.

In general, a reactive gas mixture and pressure that performs well

in ECR will also perform well in ICP. Comparable etch rates can

be obtained by proper selection of microwave and ICP powers to

permit translation from one platform to the other. A signiﬁcant

practical difference between ICP and ECR etchers is the use of

magnets to conﬁne the plasma and collimate the ions in the ECR.

This gives some extra degrees of control that can be effectively

used to tune the composition of an ECR-plasma deposition, such

as changing the O : N ratio in SiO

ﬁlms. For etching, however,

the need to tune the magnets in ECR is more bothersome than

helpful, and ICP hardware is easier to use.

From a practical perspective, the tuning of the magnets that

may be required to minimise reﬂected microwave power when

the plasma ﬁrst ignites can lead to some variation in performance

for very shallow etches where the plasma on-time must be short.

For very shallow etches, an appreciable fraction of total plasma

time may be involved in the tuning, leading to reproducibility

problems. For deeper etches requiring longer etch times, this is

not a signiﬁcant factor. ICP systems, in contrast, do not require

manual tuning and tend to stabilise faster, but in any plasma system

there are voltage excursions when the plasma ﬁrst strikes. Since

these are not identical from run to run, some variation will occur

for very shallow etching in ICP and RIE etches as well.

Selective etching of GaAs versus AlGaAs or InGaP can be

achieved in HDPE with the same chemistries that work in RIE,

but some selectivity is lost with the higher density plasmas. For

example, using 20% SF

in BCl

at 5 mTorr and only 50 W RF

power, selectivities of GaAs versus AlGaAs or InGaP that exceed

161