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

Подождите немного. Документ загружается.

Cleaning and passivation of GaAs and related alloys

given by

carriers = αI

exp[−αx] dx (3.1)

where I

is the incident light intensity, x is the depth and α is

the absorption coefﬁcient. The value of α is calculated from the

complex index of refraction (n + ik) as α =4πk/λ. The optical

constants of many semiconductors for visible through UV light at

room temperature are found in Aspnes and Studna [9]. When the

excitinglightpossesses energygreatly in excessofthe bandgap, the

resulting carriers are initially created at sufﬁciently high energy

that there can be carrier multiplication during their relaxation to the

ground state. This may require consideration when short UV light

is employed.

Bandgaps at 300 K

eV Type dE

/dT × 10

AlP 2.43 i −3.5

AlAs 2.16 i —

AlSb 1.6 d −4

GaP 2.25 i −5.5

GaAs 1.43 d −4.5

GaSb 0.69 d −4.1

InP 1.28 d −4.6

InAs 0.36 d −3.3

InSb 0.17 d −2.9

ZnS 3.8 d −4to−5

ZnSe 2.58 d −7.2

ZnTe 2.28 d −5

CdS 2.53 d −5

CdSe 1.74 d −4.6

CdTe 1.50 d −4.1

Si 1.11 i −2.3

Ge 0.67 i −3.7

4.3

8–10

SiO

For a direct-gap semiconductor like GaAs, with an absorp-

tion coefﬁcient in excess of 10

/cm, a pronounced gradient in

photogenerated carrier concentrations exists. For indirect semi-

conductors like Si and GaP, the absorption coefﬁcients of <100/cm

at energies near the bandgap lead to a relatively uniform generation

of carriers throughout the material. Where light is absorbed, there

is an increase in free carriers, leading to an increase in conductivity.

While the light continues to be absorbed, the carrier concentra-

tions remain at new, higher equilibrium values. When the light is

removed as a carrier-generation source, the system decays back

to its thermal equilibrium populations. The time dependence of

the photoconductivity can be used to measure recombination life-

times that include both radiative and non-radiative contributions.

Photoluminescence measurements can also be used to study the

recombination processes non-destructively in wafers that have not

been fabricated into device or test structures. This will be discussed

further in Section 3.3.1.3.

In general, there are three very important mechanisms whereby

carriers can recombine to restore and maintain an equilibrium con-

centration. The ﬁrst is direct recombination in which the energy

released by the recombination is emitted as light. This may or may

not also involve the emission or absorption of lattice phonons. The

radiation may be from direct band-to-band recombination emitted

at the bandgap energy or, at lower temperatures, the free carriers

may form an exciton (an electron-hole pair) that emits at an energy

slightly less than the bandgap energy. Radiation can be emitted

from “free excitons” and from excitons bound to shallow donors

or acceptors; the characteristic energy of emission can identify the

type of exciton that is the source of the light. The typical radiative

recombination lifetime for a direct-gap semiconductor is of the

order of 10

−9

The second is trap-mediated recombination. Either holes orelec-

trons can be “trapped”, i.e. captured into a deep-level trap whose

Cleaning and passivation of GaAs and related alloys

energy often lies near the middle of the forbidden energy gap. This

occupied trap subsequently captures a carrier of the opposite type.

When the trap energy is near the middle of the gap, recombination

is non-radiative. Since crystal defects, such as As antisites, As or

Ga vacancies, lattice dislocations and impurity atoms are distrib-

uted throughout the crystal, this is a characteristic mode of bulk

non-radiative recombination.

The third is surface recombination. Similar to bulk non-radiative

recombination, this is mediated by capture of carriers into deep

levels. The primary distinction is that the traps are associated with

the surface. While the radiative recombination and the bulk trap-

mediated recombination are inherent properties of the wafer as

grown or as modiﬁed by dopant in-diffusion or ion implantation,

the surface recombination is potentially variable by the manner in

which the surface is treated during processing. As will be discussed

later in Chapter 5 on Dry Etching, near-surface lattice damage by

energetic ions during dry etching increases surface-recombination

rates and can thereby degrade device performance.

The situation is generally described for an n-type semicon-

ductor as follows. For a rigorous mathematical treatment, we can

refer the reader to most textbooks on semiconductor materials.

Auger recombination processes are ignored in this discussion. The

simpliﬁed presentation below is merely for illustrative purposes.

Recombination rates are proportional to the excess minority-

carrier concentration since both carrier types are required for

recombination to occur and there is, by deﬁnition, an ample sup-

ply of the majority carriers. For an n-type semiconductor, the net

recombination rate per unit volume is given by

tot

= (p − p

)/τ

(3.2)

where τ

is the lifetime of the minority carriers. Direct recombina-

tion is proportional to the number of free electrons and the number

of holes in the valence band

dir

= k

dir

np (3.3)

In direct recombination, both the energy and momentum are

conserved without the generation of any phonons to enable the

conservation. While direct recombination is very important for

light-emitting devices, it is not a surface-mediated process and we

will not discuss it further here.

Trap-mediated recombination depends on the trap concentra-

tion and trap energy. Traps can be in the bulk or at the surface.

Some general considerations apply to all trap types. To promote

recombination, both types of carrier must be trapped in very close

Cleaning and passivation of GaAs and related alloys

proximity at the same time. A trap can capture a carrier that passes

through a certain area near it; this is the capture cross-section (σ

and σ

) and may differ for electrons and holes but is generally

of the order of atomic dimensions, i.e. 10

−15

. For capture to

occur, the trap must be unoccupied by the particular type of carrier.

Therefore, the rate of carrier capture is proportional to the concen-

tration of traps that are unoccupied. For a trap concentration, N

the concentration of unoccupied traps is determined by the Fermi

function, F

, for the probability that the trap will be occupied. For

electrons, the concentration of available traps is given by

avail

= N

(1 − F

) (3.4)

where

= 1/(1 + exp[(E

− E

)/kT]) (3.5)

with E

being the trap energy and E

being the Fermi energy. The

capture rate for electrons is given by

capture

= v

(1 − F

) (3.6)

where v

is the thermal velocity of the carrier. Carriers can escape

or be emitted from the traps before the opposite carrier type is

captured. Although it is common to use the term “recombina-

tion centre”, the more accurate term is “generation-recombination

centre” since both processes can occur. At equilibrium, the rates

of escape (emission) and capture are equal and the proportion of

occupied traps is

Electron-occupied traps = N

(3.7)

When the trap energy is closer to the conduction band, the electron

has a great probability of escaping the trap before a hole is trapped

and recombination can occur. This increases the efﬁciency of

midgap states for promoting non-radiative recombination relative

to states with energies nearer the conduction band.

Similar equations can be written for holes. The rate of hole

capture by an electron-occupied trap is given by

hole capture

= v

(3.8)

while the rate of hole escape is given by

hole escape

= v

(1 − F

) exp[(E

− E

)/kT] (3.9)

Cleaning and passivation of GaAs and related alloys

As in the case of electrons, hole escape is more likely to occur

before recombination can annihilate it if the trap energy is nearer

to the valence band. For both holes and electrons, trap energies in

the middle of the forbidden gap are especially effective for pro-

moting recombination. The net recombination rate at traps under

low-injection conditions, assuming that σ

= σ

, can be

approximated as

trap

(np − n

)

p + n + 2n

cosh[(E

− E

)/kT]

p − p

(3.10)

In the dark, n and p are the thermal equilibrium values, n

and p

Under illumination, these both increase to n = (n

+ Δn) and p =

+ Δp) where Δn = Δp are the photogenerated carriers. When

the light is removed, recombination proceeds to restore the thermal

equilibrium concentrations.

Key points to note are the necessity for trapping both types of

carriers at the same time and the dependence on the position of the

Fermi level relative to the trap energy. Midgap traps are especially

effective at promoting recombination.

Surface recombination is a special case of trap-mediated recom-

bination. The loss of the third dimension leads one to discuss

the effectiveness of a particular surface condition in promoting

recombination in terms of a surface-recombination velocity, S (to

accommodate the cm/s units), rather than a lifetime, τ, with 1/s

units.

The net recombination rate at the surface at low injection,

assuming equal cross-sections for holes and electrons, can be

approximated by

surf recomb

− p

)

1 +[2n

/(n

+ p

)] cosh[(E

− E

)/kT]

= S(p

− p

) (3.11)

where p

is the hole concentration at the surface and E

is the

energy of the surface state. As with bulk traps, those states loc-

ated in the middle of the gap are most effective in promoting

recombination.

The equilibrium free carrier-concentration in the near-surface

region can be markedly altered from that in the bulk through

recombination mediated by surface states. This can be an espe-

cially serious problem for optoelectronic devices, since surface

recombination is non-radiative in character and reduces the car-

rier supply for light emission through radiative recombination.

Cleaning and passivation of GaAs and related alloys

HBTs (Chapter 9) can also suffer when non-radiative recombina-

tion increases since the gain of the transistor is reduced because

extra base current is required to provide the holes for recombina-

tion. While deep-level defects, such as EL2, within the bulk of the

material can also promote non-radiative recombination, the mature

state of GaAs technology has reduced the severity of this problem.

While deep-level defects such as EL2 can promote non-radiative

recombination, very little EL2 is produced during high-quality

epitaxial growth. While EL2 is still present in bulk crystals, it

generally does not migrate into the epitaxial layers. In present-day

bulk growth, its concentration is near 10

/cm

and the uniformity

is controlled to provideauniformresponse to ion implantation. The

solution to the surface state problem is not dependent on excel-

lent growth methods because the number and character of surface

states can be, and generally are, altered by each step in the fab-

rication sequence for a GaAs device. Midgap surface states can

be especially deleterious due to their high efﬁciency at promoting

recombination and the presence of the surface ﬁeld that sweeps

carriers towards the surface as long as they lie within the depletion

depth (d in FIGURE 3.3). The continuing effort to ﬁnd effective

passivations to reduce the number of midgap states is directed at

this serious problem.

The importance of a lowdensity of midgap states is evidencedby

the extreme differences in surface-recombination velocitybetween

the SiO

/Si and native-oxide/GaAs interfaces. Typical surface-

recombination velocities for Si are less than 100 cm/s, with values

even as low as 0.1 cm/s having been obtained. In contrast, typ-

ical surface-recombination velocities for GaAs range from 10

cm/s; this minimum 1000-fold difference makes the proxim-

ity of a surface a major factor in the operation of GaAs devices.

GaP typically has surface-recombination velocities that are 3–10

times lower than GaAs while InP has values near 10

cm/s. Since

surface-recombination velocity can be thought of as essentially an

inverse lifetime of the carrier at the surface, the deleterious effect

of such large values is obvious. A fruitful approach to reducing sur-

face recombination velocity effects is seen in the ledge passivation

of HBTs using GaAs/InGaP instead of bare GaAs and achieving

a higher current gain (β = I

3.3.1.3 Photoluminescence as a diagnostic of surface quality

Because of the ability of midgap surface states to shift the balance

between radiative and non-radiative recombination, photolumin-

escence (PL) is a widely used technique to measure changes in

surface-state densities that accompany effective passivation. The

non-destructive nature of PL and its ease of use have made it a

Cleaning and passivation of GaAs and related alloys

widely and sometimes incorrectly used tool for preliminary assess-

ment of the quality of a particular passivation approach. The object

of this section is to discuss how one can avoid an excessively

simplistic use of PL that can lead to wrong conclusions about

surface quality.

PL spectroscopy generally employs a relatively monochromatic

light source, such as a laser, to photoexcite an excess of free

carriers (equal numbers of electrons and holes); it then monitors

the intensity and wavelength of light that results from the radiative

recombination of the free carriers. A variation is photolumines-

cence excitation (PLE) spectroscopy, where the wavelength of the

excitation source is swept while monitoring the luminescence out-

put; this allows identiﬁcation of speciﬁc absorption processes that

lead to free carriers.

Important factors for interpreting

PL studies

1) Exciting wavelength.

2) AR coating effects.

3) Loss of passivant by photolysis.

4) Possible H effects on defects and

dopants.

As mentioned in Section 2.5.1, PL spectroscopy can be used

to identify shallow-level impurities and to qualitatively gauge the

number of deep-level defects in the bulk of a wafer. Here we will

focus on its use to measure surface quality and, more speciﬁcally,

changes in surface quality due to passivation treatments.

Photoluminescence intensity measurements are best used to

provide a semi-quantitative rather than an absolute comparison

between different treatments of samples. In general, the intens-

ity of band-edge emission is monitored at a single wavelength

corresponding to the bandgap energy with the slits on the spectro-

meter set with sufﬁcient width that slight shifts in the band-edge

will not affect the total intensity measurement. Intensity meas-

urements are most informative when all the samples have been

derived from a single wafer. Quantitative comparisons between

different wafers can be seriously misleading due to differences

in the balance between radiative and non-radiative recombination

in the bulk resulting from differences in the nature and number

of bulk defects in different wafers and also due to differences in

dopant concentrations. This may be true even for wafers cut from

the same crystal boule.

There are several very important factors to consider in analys-

ing the meaning of PL measurements of surface quality. The most

important experimentalfacttoconsider in understanding the mean-

ing of a particular set of PL experiments is the doping level of the

material being studied. It is generally assumed that the detected

light is being emitted from sample depths that are deeper than the

depletion region, d, and that changes in intensity largely reﬂect

changes in the depletion depth. Simplistically, d = N

d,a

where N

and N

are the concentrations of donors and accept-

ors, respectively, and N

is the concentration of surface states.

This assumption is made because carriers that are photogenerated

in the depletion region are swept to the surface by the surface ﬁeld

Cleaning and passivation of GaAs and related alloys

and effectively consumed there by non-radiative recombination.

The PL intensity, I

, is proportional to the excitation intensity at

a particular depth and is given by

∝ I

exp[−αd] (3.12)

where α is the absorption coefﬁcient at the excitation wavelength

and I

is the incident light intensity. The normalised PL intensity

ratio is given by

PL ratio = exp[−α(d

passivated

− d

native oxide

)] (3.13)

This value will be larger for a given surface-state density when

the doping level of the material is lower. There can be orders of

magnitude variation in the normalised PL for slightly doped versus

highly doped materials that have the same surface state density. A

lightly doped or semi-insulating sample may display a 100-fold

change in PL intensity while a heavily doped material with the

same change in surface state density may only increase in PL

intensity by a factor of 2.

The same wavelength must also be used to directly compare the

results of two studies without calculating the effect of different

penetration depths of the excitation light. For example, excita-

tion at 458 nm (Ar laser) will not excite the sample to as great a

depth as would excitation at 633 nm (HeNe laser) due to the lar-

ger absorption coefﬁcient at 458 nm. Consequently, the light with

the larger absorption coefﬁcient will be more sensitive to changes

in the depletion depth and will make a passivation appear more

effective than it might actually be if this is not taken into account.

The intensity of the excitation light should be maintained

relatively low since many species, including sulphur, can be photo-

lytically released from the surface and higher intensities accelerate

this problem. Thus, the optical measurement itself can degrade the

surface, especially if the illumination is performed in the presence

of oxygen. It is a wise policy to ﬂow a curtain of nitrogen gas

across the surface of the material while it is being illuminated to

prevent photolysis and loss of the passivant.

An additional factor that must be considered is whether there is

an alteration of the apparent PL intensity due to the presence of

thin ﬁlms of the appropriate thickness to serve as anti-reﬂection

(AR) coatings. When light strikes the surface of a semiconductor,

a fraction enters the material to be absorbed and another fraction

is reﬂected from the typically specular surface of the wafer. For

an uncoated GaAs wafer, typically 35% of the incident visible

light is reﬂected. However, when a ﬁlm thickness closely cor-

responds to a multiple of (incident wavelength/4), the reﬂection

Cleaning and passivation of GaAs and related alloys

from the surface is reduced to nearly zero and the effective excit-

ation intensity is increased. When passivants of some appreciable

thickness are being applied or when a monolayer-type passivant

is being encapsulated with, for example, a dielectric to increase

its longevity, the proper interpretation of any PL intensity changes

requires a knowledge of the excitation and emission wavelengths

and the ﬁlm thickness in case there are AR coating effects

involved.

An additional effect that may lead to misleading PL intensities

is the presence of atomic H during the treatment of the wafer.

If the H diffuses into the material and binds to defects and/or

dopants, thereby reducing bulk non-radiative recombination, the

characteristic PL intensity of the bulk material can change in an

uncontrolled and generally unquantiﬁable manner. This will make

PL intensity ratios essentially uninterpretable. While this H-effect

is “passivation” in the third sense as described in the introduction

to this chapter, it is not the goal of the surface treatment being

applied and may be a hindrance rather than a help in obtaining

good device operation.

For heavily doped materials, an alternative spectroscopic

approach can be used to measure changes in the depletion depth.

When doping levels exceed about 3 × 10

/cm

, Raman scattering

results from both the optical phonons (LO and TOmodes) and from

plasmon modes, commonly designated L

and L

−

. These phonon

modes are shifted to higher and lower frequencies relativeto the LO

mode depending on the doping level of the material (FIGURE 3.4).

The relative intensities of the coupled plasmon modes, L

and

Plasma

–

1000

800

600

400

200

16 17

log (electron concentration, cm

–3

)

Raman shift (cm

–1

)

18 19

FIGURE 3.4 Doping level dependence of plasmon modes of GaAs.

Cleaning and passivation of GaAs and related alloys

−

, change as the depletion depth changes. It is assumed that

the reduced carrier concentration in the depletion region results in

only the uncoupled LO phonon at 293 cm

−1

, typical of GaAs with

donor levels below about 2 × 10

/cm

being present in the deple-

tion region, and that the coupled L

mode is present in the bulk.

Modelling the relative intensities of the two provides a measure

of the depletion depth. In addition, shifts in the L

frequency can

reveal when the passivant is also serving as a donor (FIGURE 3.5).

For example, we have observed that photodeposited sulphur acts

as a donor after the deposition of an ECR-plasma silicon nitride

on top of the passivant and increases the near-surface doping level

from 1.9 × 10

to 2.2 × 10

/cm

, as measured by the L

line in

Raman spectra excited at 514.5 nm; a more electropositive form

of sulphur also appears in XPS spectra. Depletion depths were

observed using the L

−

/LO intensity changes at 458 and 514.5 nm

to decrease from 63 nm, corresponding to a surface-state density

of 1.1 × 10

/cm

, to 18 nm, corresponding to 3.3 × 10

/cm

surface states. An increase in depletion depth to 85 nm, indicating

a surface-state density increase to 1.5 × 10

, occurred with the

plasma deposition of the silicon nitride without any sulphur. This

demonstrates the ability of plasma exposure to degrade the surface.

PL intensities and Raman ratios are actually indirect meas-

urements of the band bending, so other things besides surface

states that affect band bending can also alter the PL or the Raman

spectrum. It is important to remember that band bending is caused

by the total surface charge. This charge is the combination of

Raman shift (cm

–1

)

0 100 200 300 400 500 600 700

–0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

intensity

FIGURE 3.5 Normalised Raman spectra from n-GaAs doped to

(a) 1.82 × 10

and (b) 2.0 × 10

/cm

. Shift in L

shows doping level;

relative intensities of LO and L

show depletion depth.

Cleaning and passivation of GaAs and related alloys

surface-trappedcharge due to surfacestateslyingwithin the forbid-

den gap and surface-ﬁxed charge, i.e. ionic charge on the surface

that may not be related to the surface traps. With only a surface-

ﬁxed charge, the Fermi level can still move freely. In contrast, with

sufﬁciently high densities of surface-trap states, the Fermi level

is pinned and neither bulk doping nor externally applied electric

ﬁelds can move it.

Band bending due to surface ﬁxed charge can repel minority

carriers sufﬁciently to reduce the surface-recombination velocity.

The change in the surface-recombination velocity as a function

of forward-injection parameters can provide a measurement of

surface-ﬁxed charge and help distinguish contributions from ﬁxed

chargeand surface states. Better results will be obtained if injection

of carriers by absorption of light is avoided when making such

electrical measurements.

3.3.2 Chalcogenide passivation: S and Se

The chalcogenides, S and Se, produce advantageous changes in

the surface electronic properties of GaAs. While Te is also a chal-

cogenide, it has not been a popular choice for improving surface

performance of GaAs. Two advantages attend the sulphidation of

GaAs: there is a change in the number and type of midgap sur-

face states, which improves surface electronic properties, and the

surface sulphur retards the reoxidation of a GaAs surface, which

facilitates device fabricationwhen an oxide-free surfaceis required

for the nextprocessing step. Calculations for the (100) surface have

predicted that the Ga-S interaction produces states outside the for-

bidden gap while some As-S states lie within the gap but near the

conduction band where their electronic effects are less severe [10].

Consequently, evena fully sulphided ideal surface will still possess

some undesirable in-gap states, but the situation should be greatly

improved relative to the native oxide. Both solution-phase (wet)

and gas-phase processes have been used to sulphide the surface.

3.3.2.1 Solution-phase processes

The most common solution-phase sulphur-based passivation

employs ammonium sulphide. Aqueous solutions are quite basic

(pH > 8) so they both dissolve the native oxide, exposing the

bare GaAs surface, and provide a source of sulphur. Immersion

for 15 min in aqueous (NH

)

S (3 M or 20 wt%) at room tem-

perature is sufﬁcient to passivate the surface. Often some excess

elemental sulphur is added (typically 4%), with the reactant then

designated as (NH

)

. Some heat the solution to as much as

◦

C; however, this leads to etching of the surface [11], with