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

Подождите немного. Документ загружается.

Cleaning and passivation of GaAs and related alloys

Transient conditions while the plasmas are ﬁrst striking can pro-

duce a different ﬁlm composition and/or density near the GaAs

interface. These are manifest in a higher refractive index for thin

ﬁlms. For PECVD ﬁlms, the index decreases with increasing ﬁlm

thickness up to a thickness near 150 nm, and then remains rel-

atively constant. Similar behaviour is seen in ECR ﬁlms. It is

especially important to keep this change in refractive index in

mind when growing AR coatings for optical applications. Since

the transients may be highly tool-dependent and can even vary

signiﬁcantly on the same tool, it can be helpful to grow test ﬁlms

to the calculated thickness, measure effective refractive index at

that thickness and then ﬁne-tune the thickness right before device

fabrication to get optimum AR performance.

The ion bombardment energy can have a signiﬁcant effect on

the leakage current for HBTs. Surface damage can increase current

conduction between the emitter-base and base-collector regions.

These damage-related current channels are more important at high

base-emitter voltages and high-injection levels. Ion damage will

also degrade FET performance.

The ﬁlms that are produced from the same nominal source gases

and plasma conditions may actually be quite different when depos-

ited in different ECR tools. Plasma tools can be constructed with

different materials and with different designs. Since the plasma

interacts with the chamber walls in a way that changes the actual

plasma composition, it is possible to deposit a ﬁlm that is NOT

what you might expect based on the composition of the source

gases. A good example is the deposition of SiN

and SiO

using ECR plasmas with SiH

and N

as the source gases. In one

ECR tool with minimal wall contributions, this forms partially

hydrogenated SiN

. In another tool, oxygen might be extracted

from the quartz window anodized Al parts, through which the

microwave radiation is launched into the plasma chamber, or from

the alumina ceramic liner of the chamber. Even a little oxygen

in the plasma will be incorporated at a signiﬁcant level to give a

partially hydrogenated oxynitride. The source gases for SiO

are

SiH

and N

O. No appreciable nitrogen incorporation is observed

because of the strongly favoured formation of Si-O versus Si-N

bonds (Si-O = 124.7 kcal/mol versus Si-N = 105 kcal/mol).

In virtually all cases, some amount of hydrogen is incorporated

in the dielectric. We have deposited ECR ﬁlms of nominally SiN

and SiO

with as little as 7% hydrogen and PECVD “SiN

”

ﬁlms that were as much as 25% hydrogen. There tends to be more

H retained in PECVD ﬁlms grown at lower wafer temperatures.

The presence of excessive H can be a problem for post-deposition

processing. For example, ﬁlms with a high H content can

“bubble” due to hydrogen gas release and fail when the wafer

112

Cleaning and passivation of GaAs and related alloys

is subjected to rapid thermal annealing (RTA) to form good metal

contacts.

It is possible to grow ﬁlms that are Si-rich. These display higher

refractive indices (n > 2.0) and lower resistivity. In PECVD,

increasing the silane fraction increases the Si content of the ﬁlm. In

ECR, deposition at lower ECR powers will increase Si content for

a ﬁxed gas composition; for example, decreasing the microwave

power from 400 to 160 W increases the index from 1.8 to 2.2.

Dielectric ﬁlms are often characterised by measuring their

refractive index and assuming that the desired composition has

been attained when the “proper” refractive index is achieved.

Unfortunately, this is often an erroneous assumption. Refractive

index is a measurement of an optical property, not a composition.

It is affected by elemental composition, lattice structure, porosity,

etc. Addition of some Ar to the plasma will increase the refractive

index for a ﬁxed set of reactants and plasma conditions because it

servestodensify the ﬁlm. Addition of Ar also causes plasmas to run

in a more stable fashionandthereforeincreasesrun-to-run reprodu-

cibility. Films with the same composition but greater porosity will

have lower refractive indices. Films that contain Si-rich regions

will have higher refractive indices. For example, while the refract-

ive index of polycrystalline Si

is 2.0, it is possible to grow

ECR-deposited ﬁlms with approximately the same index that are

actually Si

2.8

ﬁlms with 7% H and SiO

with 60% Si,

20% N, 10% H, 7% O and 3% C. One PECVD “silicon nitride”

with index 2.0 was actually 35% Si, 40% N and 25% H. Although

it is a poor choice as the main diagnostic for composition, it can

be used very effectively as a measurement of the reproducibility

of a process since it combines the effects of composition and ﬁlm

density. Hopefully, a speciﬁc process will produce uniform values

of these parameters when it is operating reproducibly.

Lower index ﬁlms tend to etch faster than higher index ones,

due either to differences in composition or to the lower index ﬁlms

being less dense. The effect is more pronounced in wet etchants

than dry, with a 5- to 20-fold increase over that for true Si

. Dry

etch rates are generally less than two-fold faster for the PECVD

ﬁlms. Higher H-content ﬁlms tend to etch faster. Having the same

refractive index does not mean there will be the same wet etch

rate. A factor of two spread in rate with 48% HF was obtained for

several PECVD ﬁlms all with a refractive index of 2.0. This all

shows how important it is to test the actual etch rate of the ﬁlms

deposited in your own hardware before attempting to etch them

during the fabrication of a valuable device.

Film stress is another important variable. ECR ﬁlms tend to

be lower stress than PECVD ﬁlms, but variations in this can

occur. Since high stress can lead to shifts in electronic properties

113

Cleaning and passivation of GaAs and related alloys

within a chip, e.g. offset voltages, resistance of thin-ﬁlm transist-

ors, pinch-off voltages, hot electron degradation and mechanical

failure, it is generally desirable to use low-stress ﬁlms.

Conformality of the dielectric can be an issue especially when

the dielectric is deposited on a mesa, which can be either photores-

ist or semiconductor. With PECVD depositions on a mesa, a proﬁle

with indentations near the mesa base (FIGURE 3.20(a)) often

forms. This non-uniform ﬁlm thickness can be used to advantage

in some contact formation schemes. On the other hand, uniform

ﬁlm thickness may be desired on all surfaces. While conformal

deposition is possible with PECVD, the room-temperature ECR

processes tend to automatically deposit in a more uniform fashion

(FIGURE 3.20(b)).

(a)

(b)

FIGURE 3.20 (a) Deposit with

non-uniform thickness with

PECVD. (b) Uniform thickness,

conformal deposit with ECR.

Pinholes can be a serious problem in dielectric deposition. These

can be caused by small particles on the wafer surface. These

particles may have been on the surface of the wafer as loaded in

the tool, or they may result from the deposition of particles within

the reactor. It is especially important to keep the deposition sys-

tem walls free of deposits that can ﬂake and shed particles onto the

wafer surface. Excessively rapid deposition of the ﬁlm increases

the likelihood of pinhole formation.

The use of silane as the Si source raises both safety and repeat-

ability concerns if there is even a tiny leak in the silane gas line.

Due to the pyrophoric nature of SiH

, ﬁres and explosions are a

possibility with major leaks. Even a very small leak can lead to the

deposition of “sand”, i.e. tiny SiO

particles, in the line that can

gradually clog mass ﬂow controllers and make the plasma com-

position vary slowly over time, leading to irreproducibility in ﬁlm

composition.

3.4 CONCLUSION

The most important characteristic to remember when dealing with

cleaning and passivation issues is that materials properties at the

atomic level are what really matter. Composition and bonding

conﬁgurations at the monolayer level can totally dominate the

properties that affect device operation. Thinking “molecularly”

and careful attention to process details at that level will contribute

to achieving reproducible and reliable device results.

REFERENCES

[1] M.-G. Kang, H.-H. Park [Vacuum (UK) vol.67 (2002) p.91–100]

[2] K.R. Zavadil, C.I.H. Ashby, A.J. Howard, B.E. Hammons [J. Vac. Sci.

Technol. A (USA) vol.12 (1994) p.1045–9]

114

Cleaning and passivation of GaAs and related alloys

[3] P.R. Lefebvre, L. Lai, E.A. Irene [J. Vac. Sci. Technol. B (USA) vol.16

(1998) p.996–1001]

[4] C.J. Sandroff, F.S. Turco-Sandroff, L.T. Florez, J.P. Harbison [J. Appl.

Phys. (USA) vol.70 (1991) p.3632–5]

[5] S.I. Yi, P. Kruse, H.M. Hale, A.C. Kummel [J. Chem. Phys. (USA) vol.114

(2001) p.3215–23]

[6] M. Passlack, M. Hong, J.P. Mannaerts [Appl. Phys. Lett. (USA) vol.68

(1996) p.1099–101]

[7] H. Hasegawa [Jpn. J. Appl. Phys. (Japan) vol.38 (1999) p.1098–102]

[8] S.M. Sze [Semiconductor Devices: Physics and Technology (John Wiley

and Sons, New York, 1985)]

[9] D.E. Aspnes, A.A. Studna [Phys. Rev. B (USA) vol.27 (1983) p.985–1009]

[10] T. Ohno, K. Shiraishi [Phys. Rev. B (USA) vol.42 (1990) p.11194–7]

[11] J.-W. Seo, T. Koker, S. Agarwala, I. Adesida [Appl. Phys. Lett. (USA)

vol.60 (1992) p.1114–6]

[12] D. Wu, L. Liu, J. Marcano, Y. Darici, N. Paul, S. Mergui [Mater. Sci.

Eng. B (Switzerland) vol.46 (1997) p.61–4]

[13] C.I.H. Ashby, K.R. Zavadil, A.J. Howard, B.E. Hammons [Appl. Phys.

Lett. (USA) vol.64 (1994) p.2388–90]

[14] C.I.H. Ashby, K.R. Zavadil, A.G. Baca, P.-C. Chang, B.E. Hammons,

M.J. Haﬁch [Appl. Phys. Lett. (USA) vol.76 (2000) p.327–9]

[15] M.R. Ravi, A. DasGupta, N. DasGupta [IEEE Trans. Electron Devices

(USA) vol.50 (2003) p.532–4]

[16] S.M. Sze [Physics of Semiconductor Devices, 2nd Ed. (John Wiley and

Sons, New York, 1981)]

[17] M. Terman [Solid-State Electron. (UK) vol.5 (1962) p.285–99]

[18] K. Remashan, K.N. Bhat [Semicond. Sci. Technol. (UK) vol.17 (2002)

p.243–8]

[19] D. Shoji, T. Miura, M. Niwano, N. Miyamoto [Appl. Surf. Sci.

(Netherlands) vol.130–132 (1998) p.441–6]

[20] S. Hohenecker, T.U. Kampen, T. Werninghaur, D.R.T. Zahn, W. Braun

[Appl. Surf. Sci. (Netherlands) vol.142 (1999) p.28–32]

[21] S. Anantathanasarn, H. Hasegawa [Appl. Surf. Sci. (Netherlands) vol.216

(2003) p.275–82]

115

Chapter 4

Wet etching and photolithography of GaAs

and related alloys

Chapter scope p.117

Mechanism of wet etch

processes p.118

Rates and proﬁles p.118

Diffusion control p.121

Reaction-rate control p.123

Aging of etching solution p.124

Practical wet etching p.125

Photoresist issues p.127

Basic wet etches p.132

Acidic wet etches p.133

Metal etches p.135

Compositional selectivity p.136

GaAs versus AlGaAs selectivity p.137

Selectivity versus other materials p.139

Effects of doping type p.141

Electrolytic effects in wet

etching p.142

Effects of defects and damage p.144

Conclusion p.144

References p.144

4.1 CHAPTER SCOPE

Since most III–V devices require the exposure of buried layers

for metal contact formation and/or the formation of mesa struc-

tures for electrical isolation from adjacent devices on the same

wafer, etching will always be a critical part of their fabrication.

“Wet” (liquid-phase) etches are the simplest to implement since

they require only simple benchtop glassware (or plasticware if

HF is involved). There are, however, both advantages and dis-

advantages to using wet etching processes. The key issues to be

considered in selecting a particular etch process for any application

are rate, uniformity, selectivity, critical dimension control, feature

proﬁle and surface damage that can degrade electronic properties

and, therefore, device performance.

The most favourable characteristic of wet etches is their ability

to cause virtually no surface electronic damage. When superior

surface electronic properties are most important, a wet process

will clearly be preferable to a dry process, which often degrades

surface electronic properties by ion bombardment.

The least desirable wet etch characteristic for many applications

is the unavoidable tendency to undercut etch masks, making pre-

cise dimensional control more difﬁcult. This is observed whether

there are nearly equal vertical and lateral etch rates (isotropic

etching) or etch rates that depend on the speciﬁc exposed crys-

tal planes of the semiconductor (crystallographic etching). Precise

control of critical dimensions is made more difﬁcult by the fre-

quently strong dependence of wet etch rates on temperature,

solution agitation and crystallographic alignment of the mask.

These aspects of wet etching, which will be discussed in some

detail in this chapter, present both difﬁculties and opportunities

for extra control.

A more detailed comparison between wet and dry (gas-phase)

etch processes is found in Chapter 5. In general, if feature proﬁle

and tight control of dimensions are less importantthan surfaceelec-

tronic quality, a wet etch may be the process of choice. If vertical or

117

Wet etching and photolithography of GaAs and related alloys

controlled-angle proﬁles and small dimensions are needed, a dry

etch that achieves the best compromise between feature proﬁle and

damage may be preferred. In this chapter, the issues of importance

in selecting and using a wet etch process will be discussed.

Advantages of wet etching

1) No surface electronic damage

2) Many chemistry choices

3) Chemical selectivity

4) Crystallographic etching

5) Inexpensive, simple equipment

Disadvantages of wet etching

1) Limited dimensional control

2) Vertical proﬁles hard to get

3) Solution waste disposal

When a pattern is to be etched, some method of protecting

or masking portions of the surface from the etchants must be

provided. Photolithography is the standard way of providing etch

masks for pattern delineation. Many issues related to photolitho-

graphy are independent of whether a wet or dry etch process is to be

used. Therefore, we will address many common photoresist issues

in this chapter. Some mask issues that are of importance only for

dry etching will be addressed at appropriate points in Chapter 5.

4.2 MECHANISM OF WET ETCH PROCESSES

The underlying chemical mechanism of most wet etching of GaAs

is the oxidation of the surface to form Ga and As oxides, fol-

lowed by the dissolution of these oxides by chemical attack with

acids or bases. Many GaAs etches call for the addition of hydro-

gen peroxide (H

) to serve as an oxidising agent to promote

the formation of the surface oxides. This is especially needed

for n-type material. Removal of electrons from the surface by an

oxidising agent like H

leaves an electron-deﬁcient, “hole-rich”

surface that reacts readily to form the oxide. Some other chemicals

that have been used as the oxidising agent for etching GaAs are

HNO

, CrO

and Br

. An alternative way to create

a hole-rich surface in n-type material is to shine light on the sur-

face. This can produce photochemical etching of n-type material.

When ample quantities of chemical oxidising agent are present in

solution, this is usually not a major effect. However, with dilute

solutions, especially in bright light, photochemical etching may

be an unintentional consequence if samples are left in solution for

long periods of time. Under special conditions, local galvanic cells

may arise that can etch GaAs without light or a speciﬁc oxidising

agent.

Steps in wet etching

1) Diffusion of reactants to surface.

2) Formation of oxide on surface.

3) Acid or base attack on oxide and

oxide dissolution.

4) Diffusion of products from surface.

Overall rate controlled by step 2)

Proﬁles controlled by step 3).

4.3 RATES AND PROFILES

The most important consideration in selecting a wet etch is the

proﬁle of the etched features. Since this is controlled primarily by

whether the etching chemistry is reaction-rate limited or diffusion

controlled, a brief discussion of reaction mechanisms is warranted.

A simpliﬁed representation of the chemistry typically involved

in GaAs etching is given by the following equations. To initiate

118

Wet etching and photolithography of GaAs and related alloys

the etching process, an oxide is formed on the surface of the GaAs

by reaction with an oxidising agent, typically H

GaAs + (oxidiser) → GaO

/AsO

(4.1)

The rate can be calculated from

rate = k

[oxidiser]

{GaAs} exp(−E

∗

/kT) (4.2)

Here, [oxidiser] represents the concentration of oxidiser, m is

related to the number of oxidiser molecules involved in oxidising

one Ga and one As, {GaAs} is the effective surface “concentra-

tion” of exposed Ga and As atoms, k

is the reaction prefactor,

∗

is the activation energy for the oxidation reaction, k is the

Boltzmann factor and T is temperature in Kelvin. While this reac-

tion may affect the overall rate, with rates being faster with higher

oxidiser concentration, it is the second step, the dissolution of the

oxide, that generally controls proﬁles.

The surface oxide is then attacked by either an acid or a base

and is dissolved to give Ga

and AsO

2−

ions in solution.

GaO

/AsO

+ (acid or base) → Ga

+ AsO

2−

(4.3)

It is this second step in the total etching process that is usually key

in determining whether the etching will be reaction-rate limited or

diffusion controlled.

rate = k

diss

[acid or base]

{oxide} exp(−E

∗

diss

/kT) (4.4)

Here, [acid or base] represents the concentration of acid or base,

n is related to the number of H

or OH

−

involved up to the slow-

est step in the dissolution reaction, {oxide} represents the effective

concentration of the surface oxide species that is slowest at dis-

solving, k

diss

is the reaction prefactor and E

∗

diss

is the activation

energy for the dissolution reaction.

Factors affecting diffusion versus

reaction-rate control

Reactant concentration

Solution mixing

Solution viscosity

Temperature

Assuming an adequate supply of oxidiser, whether the rate of

the dissolution process will be limited by the supply of acid or base

(diffusion controlled) or by the activation energy of the dissolution

reaction (reaction-rate limited) will be determined by the relative

magnitudes of their respective terms. Any reaction has the poten-

tial to be either reaction-rate limited or diffusion controlled. It is

merely a question of which term dominates under the conditions

that are being employed to etch GaAs. If an activation energy for an

etching process is less than 5 kcal/mol, it will probably be diffusion

119

Wet etching and photolithography of GaAs and related alloys

controlled, while a value greater than 10 kcal/mol probably means

reaction-rate control.

Rates can depend on:

Doping type and level

Crystallographic facet

Heterojunctions

pn junctions

Nearby exposed metal

It is easy to see from these equations that changing concen-

tration, solution mixing or temperature might shift the etching

reaction from one dominant mode to the other. In fact, some

common etching solutions perform well rather close to the switch-

over point. In general, low concentrations of reactants favour

reaction-rate control of etching while high concentrations favour

diffusion control. Higher viscosities also favour diffusion control.

Something as simple as diluting a solution to slow an etch rate can

switch the dominant mode from diffusion to reaction-rate control.

Consequently, caution must be used in altering “standard” recipes.

Unintentional switching from one type of control to the other can

produce very different results from those expected and destroy

your devices.

Rate dependence on doping

> p > n > n

n > SI

The activation energy for the initial oxidation of the surface

∗

) can vary appreciably depending on the crystallographic face

that is exposed and on the doping type of the material. Con-

sequently, it will always be important to determine the etch rate

for a particular wafer structure rather than assuming that a lit-

erature value will apply in any speciﬁc case. The presence of

heterojunctions and pn junctions in a particular device structure

can profoundly affect the energetics of the oxidation and, there-

fore, the etch rate. Consequently, it can affect the proﬁle, with

rates generally being p

> p > n > n

and n > SI. This pro-

duces greater undercut for p-type material than n-type and a

sloped proﬁle across the depletion region near the pn junction.

Doping effects are discussed in greater detail in Section 4.6 and

heterojunction issues will be discussed with selective etching in

Section 4.5.

Care must also be taken in aligning the mask with respect to the

crystallographic axes of the wafer since there is a difference in the

activation energy for oxidation of different crystal faces. This can

produce totally different etch proﬁles for devices that are aligned

differently with respect to the crystal axes. Crystallographic

dependences will be discussed in Section 4.3.2.

Excessively fast rates will cause poor surface quality. Oxida-

tion rate is generally controlled by the strength of the oxidiser so

increasing the H

fraction increases the rate; however surfaces

tend to roughen with etchants that are >15% H

. Dissolution of

the oxide is controlled by solution pH; surface roughness increases

at pH > 7.5. At higher pHs, there is less efﬁcient oxide removal so

the thickening oxide can crack and ﬂoat away in small segments.

This exposes the underlying surface for further oxidation. This

cracking and ﬂaking away is inherently a non-uniform process

and roughening is the result.

120

Wet etching and photolithography of GaAs and related alloys

4.3.1 Diffusion control

The rate of diffusion-controlled reactions can be limited by the

diffusion of reactants to the surface or of products away from

the surface. For relatively narrow features, this rate limitation by

diffusion can lead to decreases in the reaction rate as features are

etched deeper and diffusion times into and out of the deepening

trench increase. The exponential term (exp[−E

∗

T]) is of less

importance so these rates are relatively insensitive to temperature.

However, they are highly sensitive to the degree of agitation so

controlled stirring or rotation is important to achieve consistent

and predictable results.

Slow rotation (<100 rpm) permits very uniform thinning of a

wafer. Fast rotation in a centrifuge can decrease lateral under-

cutting. In practical terms, “rotation” often consists of manually

swirling the sample holder in solution, which may not be very

reproducible. Stirring at a reproducible rate with a magnetic

stirrer or an over-head propeller design can produce more predict-

able results. Automated equipment is available for higher volume

applications.

The classic proﬁle of a diffusion-controlled reaction is shown in

FIGURE 4.1. Lateral undercutting proceeds at a rate comparable to

the vertical etch rate in the ﬁeld region away from the mask. How-

ever, a shallow trench or downward bulge forms near the edge of

the mask and slopes back away from the edge. This proﬁle is well

understood and has been quantitatively modelled [1] assuming no

closely adjacent additional mask edges. The degree of trenching

can be quite dependent on the areal ratio of exposed and masked

areas. The classic example of a diffusion-controlled etchant for

GaAs is the HCl/H

O system. Under “ideal” diffusion con-

ditions, i.e. with no convection, the ratio of the bulge depth to

the lateral undercut beneath the mask is about 1.33 and the pro-

ﬁle would remain constant with time (FIGURE 4.1(a)). However,

some convection is always present during etching due to temper-

ature and/or concentration gradients generated near the surface

by the etching reaction itself. For short etch times, the bulge-to-

undercut ratio is essentially 1.33, but as the etch proceeds in time,

the mask-edge region appears more and more like the ﬁeld region

more remote from the mask edge and the downwardbulge becomes

less prominent with time (FIGURE 4.1(b)).

The potential problem of trenching should be considered before

using these etches. For example, a trench-prone etch would be

totally unsuitable for gate-recess etching. For acid/peroxide ratios

of 1 HCl : y H

:1 H

O, negligible trenching is obtained for

y ≤ 10 and up to 10–20% deeper etching near mask edges is

obtained than at the centre between two masked areas for y = 40

121