Hoffman D.M., Singh B., Thomas J.H. (Eds). Handbook of Vacuum Science and Technology

Подождите немного. Документ загружается.

614

Chapter

5.1:

High-Vacuum-Based Processes: Sputtering

Fig.

1600^

1200 4

800

400 4

PD (Torr-cm)

The Paschen curve for the breakdown voUage for a DC diode in Ar as a function of the prod-

uct of

the

chamber pressure P and the electrode spacing D (adapted from [4]).

current. The high pressure results in poor transport of the sputtered atoms away

from the cathode and subsequently a very low deposition rate. A second problem

arises when the cathode is insulating. Since there is little current flow, the sputter

etching and deposition process effectively stops. This would occur either if the

cathode were constructed from an insulator, or if a metallic cathode was operated

in the presence of sufficient oxygen. In general, DC diodes are rarely used in any

practical applications.

5.1.2.2 RF Diodes

A simple variation of the DC diode results in a solution to these two fundamental

problems. By applying an alternating instead of a DC voltage to the cathode, the

S.1.2 Sputter Deposition Technologies

6\5

ion current density can be increased and concerns with the charging of insulat-

ing cathodes are eliminated. The most conmion AC frequency used is 13.56 MHz,

which is why this technology is often known as RF diode sputtering. However,

other frequencies from 60 Hz up to 100 MHz have also been used. For practical

purposes, because most of the available hardware for high-power sputtering sys-

tems has been designed for 13.56 MHz, the remainder of this discussion focuses

on that frequency.

The applied RF voltage to the cathode, typically on the order of several kV, re-

sults in a plasma that essentially oscillates at the same frequency. Electrons in the

plasma pick up additional energy from the oscillation, in a way that has been

compared to "surfing" on the applied electric "waves" [6]. In this way, energy is

coupled into the electron population, which results in more ionization by the high-

energy tail of the electron distribution (roughly a Maxwell-Boltzmann distribu-

tion).

The higher ionization means a higher ion current at the same applied power

than in a simple DC diode.

The second key advantage of RF diode sputtering is that the cathode receives

no net current from the plasma. The incident ions from one part of the RF cycle

are compensated for by the incident electrons from the other part of the cycle. In

a sense, the cathode and anode switch places once each RF cycle, resulting in no

net current or charging.

Because the electrons in a plasma move much more rapidly than the ions, the

electron bombardment rate of the cathode during the positive part of the RF cycle

can greatly exceed the ion flux during the negative half-cycle. If the cathode is ca-

pacitively coupled to the power supply, the net negative charge will start to look

like a net negative potential on the cathode. Effectively, the DC bias of the cath-

ode begins to drift downward with each succeeding RF cycle. As this occurs, the

fraction of time that the cathode is positive with respect to the anode decreases

and the net number of collected electrons decreases.

Eventually this process reaches equilibrium after just a few cycles with a net

DC bias, which is just slightly less than half the applied peak-to-peak RF voltage

(Figure 5). Since the ions cannot respond to the 13.56 MHz voltages (their inertia

is such that they cannot respond to frequencies above a few hundred kHz), they

respond to the average DC bias of

the

cathode. In RF diode sputtering, the ion en-

ergy is generally described by the DC bias because the potential of the plasma is

usually close to zero.

RF diode sputtering, both etching and deposition, is most commonly used for

the sputtering of insulating materials such as oxides (silicon dioxide, titanium

dioxide, aluminum oxide, etc.) and also for the sputtering of polymers, such as

polyimide. Since sputtered material is mostly emitted in the form of

atoms,

rather

than compounds or molecules, it is often necessary to add additional reactive gas

such as oxygen during the sputter deposition of

oxides.

Reactive sputtering is de-

scribed in more detail later.

616

Chapter

5.1:

High-Vacuum-Based Processes: Sputtering

Fig.

:§

Approx. 1/2

peak-to-peak

voltage

Time

Cathode potential as a function of time showing the downward drift of

the

DC bias due to elec-

tron charging.

5.1.2.3 Magnetron Sputtering

A magnetron is a cathode in a diode plasma device that can be operated either in

a DC or AC mode, although the vast majority are operated DC. The magnetron

uses the basic effect that electrons respond to magnetic fields by the simple relation

F = qvXB

(1)

where q is the electron charge, v its velocity, and B the magnetic field. Since this

is a cross product, the force is at right angles to the initial velocity, which for a

free electron in the vicinity of a magnetic field results in the electron spiraling

around the magnetic field line. This spiraling results in a significantly longer path

length for the electron and if it is located in a plasma, a significantly higher prob-

ability that it will strike a background gas atom and perhaps cause ionization of

that atom. Magnetic fields have long been used to intensify plasmas in this way,

resulting in higher plasma densities and lower discharge voltages at constant ap-

plied power.

A magnetron uses a slight variation on this phenomena in which the electron

now must respond to both electric and magnetic fields. If these fields are at right

angles, the electron experiences a net

£"

X ^ drift, which is in a direction perpen-

dicular to the E and B fields. In solid-state physics, this is known as the Hall effect

and results in a slight voltage difference across a conductor that is carrying cur-

5.1.2 Sputter Deposition Technologies

617

Fig.

Top View

ExB drift path

Magnetic

field lines

Side View

Magnetic field

ExB drift patfi

.-. y

Pole piece assembly

Schematic of

magnetron cathode device.

rent in the presence of a large magnetic field. In a magnetron, though, the drift

path for the electrons is configured such that it closes on

itself,

often in the form

of a loop (Figure 6). The closed path allows electrons to travel around the drift

path several [3-10] times [7], which results in a high degree of efficiency of cou-

pling the electron energy to the plasma. This results in very low discharge volt-

ages,

high currents (amperes), and very little voltage increase as the current is in-

creased. Magnetrons generally follow an empirical current-voltage relationship

described as

/ = kr

(2)

where / is the discharge current, V the voltage, k is a system-dependent constant,

and the exponent n is typically in the range of 5-20.

The actual path of the secondary electron after it leaves the cathode is quite

complex. The electron spirals around the magnetic field lines while it oscillates

radially along the field lines, reflecting at each end of the field line as it intersects

618 Chapter

5.1:

High-Vacuum-Based Processes: Sputtering

the cathode surface. On top of these motions, the £ X 5 drift causes a generally

circular drift around the E X B drift path. In addition, since the electron is in

a plasma, it is subject to a lot of interactions (noise, turbulence) with the other

charged particles present.

Originally it was thought that the electron simply drifted around the drift path

of the cathode until it hit a gas atom, ionized it, and then continued around the

loop [8]. In reality this scenario is probably not accurate, and the electron has

many more effective collisions with other electrons, resulting in a general heating

of the electron distribution. Therefore, ionization by the high-energy tail of the

distribution is the primary means of ionization, not direct collisions between sec-

ondary electrons and gas atoms [9].

5.1.2.4 Aspects of Magnetron Sputtering

DEPOSITION RATES AND EFFICIENCIES

In sputter deposition, high-rate, controlled, uniform deposition is generally de-

sired. The nature of the magnetron, which is characterized by a dense plasma loop

at the cathode, results in a nonuniform emission, usually from a ring or rectangu-

lar geometry. In addition, because of the system geometry, the emission profile of

the sputtered atoms and the distance of the sample from the cathode, the deposi-

tion rate is difficult

predict

exactly.

Various studies have measured the deposition

rate and profile as a function of gas pressure and throw distance. Most of these

studies are only relatively quantitative in that they are system dependent, and

there is little effort to calibrate the results.

There are two approaches to quantifying the deposition process with a mag-

netron, both effectively equivalent. The first uses the intrinsic probability that a

sputtered atom emitted from the cathode will be deposited on the sample. This

leads to a number between 0 and 1, with 1 being the maximum, ideal transport

(all sputtered atoms deposit on the sample). An alternative approach, perhaps

more practical for the operation of systems, quantifies the deposition rate in terms

of "rate per watt." In this case, the general assumption is that the deposition rate

increases linearly with discharge power (discharge current times discharge volt-

age).

This assumption is fairly accurate in the region of operation of most mag-

netrons, where the sputter yield increases linearly with discharge voltage, and in

practice the discharge voltage changes very slowly.

DEPOSITION PROBABILITY

The probability can be determined by comparing the number of atoms sputtered

from the cathode with the number deposited at the sample. The number sputtered.

5.1.2 Sputter Deposition Technologies 619

Table 2

Deposition Probability for Magnetron Sputtering at lOOOW, 200-mm-

Diameter Planar Magnetron in Large System [10]

\ Throw (cm)

Ar Sputtering of Cu

5 cm

9.5 cm

14.5 cm

Ne Sputtering of Al

5 cm

9.5 cm

Ar Sputtering of Al

5 cm

9.5 cm

Kr Sputtering of Al

5 cm

9.5 cm

Pressure

(mtorr)

20, 30

20,

20, 30

20,

20, 30

20,

Deposition

Probability

0.63,

0.49, 0.54

0.48, 0.47, 0.45

0.39,0.35,0.31

0.80, 0.56, 0.52

0.40, 0.42, 0.40

0.60, 0.46, 0.42

0.44,

0.45,

0.35

0.52, 0.45, 0.38

0.35,

0.27, 0.22

to a first approximation, is simply the discharge current times the sputter yield

times the deposition time. This number might be accurate to 20% or so, and is

limited by knowledge of the sputter yield for that particular cathode and also by the

5-10% correction caused by the secondary electron yield to the discharge cur-

rent. Deposition probabilities are listed in Table 2 for a simple, 8-inch-diameter

planar magnetron in an open vacuum system. Every system is different, though,

and the presence of shielding, shutters, and even the chamber walls can alter these

probabilities slightly. It should be noted that the best probabilities (highest) occur

for low pressure, short throw distance, and a gas species that is lighter than the

sputtered species. In addition, sputtering with a heavy gas, such as Kr, is often

counterproductive. The sputter yield may be slightly increased, but the probabil-

ity of an atom getting to the sample and making a film is reduced significantly.

DEPOSITION RATES BY POWER

These numbers are also very system dependent, based on the exact configura-

tion of specific tools, operating conditions, etc. These rates are often shown as

angstroms/min/watt. An example of this type of data is shown for a batch-type

tool

[11].

In this case, however, the sample pallets (37-45 cm wide) were scanned

past the cathode such that the rates listed in Table 3 are less than the rates for a

fixed sample directly in front of the cathode. This is a systematic problem with

this type of rate measurement: The numbers apply to this system only.

It is extremely difficult to compare results such as these to another system. For

620 Chapter

5.1:

High-Vacuum-Based Processes: Sputtering

Table

Deposition Rate per Watt for MRC Magnetron Systems, Derived from

Data by Class [11]. Units are Ang/min/watt.

Cathode

(throw distance)

Ni-7%V

Focest Design

1.75 cm

.29

.55

.58

.36

.37

.83

.14

.15

Inset Design

5 cm

.16

.31

.32

.20

.21

.47

.07

.08

Planar

5 cm

.18

.34

.36

.22

.23

.52

.08

.09

example, present-day single-wafer semiconductor sputtering tools have a fixed

sample position, which would undoubtedly mean a higher deposition rate effi-

ciency than for a moving sample. Results for the sputter deposition of common

semiconductor materials are shown in Table 4. The throw distance in this case is

about 4 cm.

This general approach can be extended to specific systems in two ways. First,

ignoring gas rarefaction effects (next section), the deposition efficiency of any

particular system should be related to the numbers just listed by a simple, multi-

plicative factor. It is necessary to calibrate each system, but once done, calibration

should not change without major geometry changes. Second, the dependency of

the deposition rates just listed can be approximately coupled to the sputter yield,

particularly for short throw distances. Therefore, when changing from a cathode

included on the list to another target, a first estimate of the probable deposition

rate can be made by comparing the sputter yield of the new material to those

listed in the tables.

Table

Deposition Rates per Watt for an Applied Materials "Endura" Single-Wafer Sputter

Deposition System. Cathode is 32.7-cm diameter, samples are 200-mm wafers [12].

Material

AlCu (0.5)

TiN (nitride mode)

Ti (collimated,

1.5:1)

Power

12.7 kW

IkW

4kW

7kW

Rate (Ang/min/watt)

1 (new cathode)

0.75 (old cathode)

0.17

0.15

0.043

5.1.2 Sputter Deposition Technologies 621

GAS RAREFACTION

One effect that was not anticipated with magnetron sputtering was the significant

effect on the background gas density. This is a result of the thermalization or

slowing down of

the

sputtered atoms due to gas-phase collisions with the Ar in the

chamber. The heated Ar atoms then rarefy and reduce the local gas density in the

near-cathode region [13]. Reductions of 80% or more have been observed in

modest systems at pressures of 30 mtorr. This effect is strongly dependent on the

sputter yield, the emitted atom velocity, the thermal conductivity of the gas, and

the chamber walls. This effect may also be strongly coupled into the impedance

of the discharge and the different operating voltages observed for different cath-

ode species.

REACTIVE SPUTTERING

Reactive sputtering, and in particular reactive magnetron sputtering, is a process

in which the atoms sputtered from a metallic cathode combine with background

gas molecules at the sample surface to form a compound. Common examples are

the sputtering of Al with oxygen present to form aluminum oxide, or the sputter-

ing of Ti in the presence of nitrogen to form Ti-nitride. In general, the formation

of a compound film occurs at the sample surface, not in the gas phase or in flight.

However, this leads to a problem in that the metallic cathode is experiencing

roughly the same environment as the depositing film. If there is enough of the re-

active gas present to react with the metal on the film surface, generally the cath-

ode surface also reacts.

This process can be examined phenomenologically as follows (see Figure 7):

Consider an Al metal cathode being sputtered with only an inert gas species, such

as Ar. The films that are deposited are purely metallic and generally the deposi-

tion rate can be rather high. If a small amount of oxygen is then added to the

chamber, it is immediately incorporated into the depositing film. It is actually

very difficult to observe the effects of introducing small amounts of oxygen: The

deposition rate is unchanged, the discharge voltage on the cathode does not change,

and the pressure does not change: No oxygen is visible on a mass spectrometer.

One's first response is to check to make sure the flow controller is working, be-

cause it seems as if nothing is happening. The films, which are very reactive, are

effectively gettering all the reactive gas present.

If the reactive gas flow is increased, up to a critical flow virtually all the reac-

tive gas is still incorporated into the depositing films. In Figure 7, the flow can be

increased up to the point of the second vertical dashed line. The films are at this

point becoming saturated with the reactive species and are approaching their ter-

minal (and usually desired) composition. At the critical flow, however, the films

622

Rg.7.

Chapter

5.1:

High-Vacuum-Based

Processes:

Sputtering

€

Ik-

Reactive Gas Flow (arb. units)

Plots of the deposition rate (top curves), the discharge voltage (middle curves) and the cham-

ber pressure (lower curves) for the case of increased reactive gas flow during reactive sputter-

ing. The rightmost vertical dotted line is the critical flow point for the transition from the

metallic mode to the reacted mode. The leftmost vertical dotted line is the flow point for the

transition back to the metallic mode.

5.1.2 Sputter Deposition Technologies 623

become completely saturated and any additional reactive gas introduced into the

chamber serves to saturate the magnetron cathode, changing it from a metallic

mode to a compound mode, which generally has a much lower sputter yield. This

results in a rapid reduction of the sputter rate, and as a result a rapid reduction in

the ability of the deposited films to getter the reactive species. The system goes

through an irreversible transition at this critical flow: The deposition rate drops

sharply, the discharge voltage usually changes due to the changing surface of the

cathode, and the reactive species is now no longer totally absorbed by the de-

posited films because the supply rate of sputtered atoms has dropped significantly.

Increased reactive gas flow results in relatively little change to the system: The

cathode is fully reacted, and additional reactive gas atoms are irrelevant. Opera-

tion to the right side, or high-flow side, of this critical transition flow is known as

the oxide mode for reactive sputtering in oxygen, or the nitride mode for reactive

sputtering in nitrogen.

Reducing the flow of the reactive gas does not result in an immediate recovery

of the metallic mode at pressures just below the critical flow. That is because the

cathode is still fully reacted and the yield is still low. It is necessary to go to a

much lower flow (the leftmost dotted line in Figure 7) to revert back to the metal-

lic mode. These curves are often described as hysteresis curves due to their simi-

larity to curves describing the magnetization of materials.

5.1.2.5 Bias Sputtering and Unbalanced Magnetrons

The presence of a plasma during sputter deposition allows the use of ion bom-

bardment of the sample during deposition by ions present in the plasma. In con-

ventional DC or RF diodes, this effect is known as bias sputtering. The imposi-

tion of a sample bias, either DC or RF, results in energetic ion bombardment of

the growing film, which can cause several effects. From a materials point of view,

the depositing film can be compressed and densified by the ion bombardment, re-

sulting in densities close to bulk density, fine grain size, and modification of the

film

stress.

This technique has been described as "ion peening." Obviously, too

much bombardment can lead to negative effects on the film, ranging from high

compressive stress to actual resputtering of the depositing film. In this latter ef-

fect, due to the angular dependence of the sputter yield, jagged or pointed to-

pographies sputter at a faster rate than flat surfaces, and the eventual result can

be increased planarization of

the

deposit. Reactive sputtering can also be strongly

enhanced by the presence of ion bombardment, particularly for reactive chem-

istries such as the nitride system (AIN, TiN, etc.), which require elevated sample

temperature of another energy source to push the reaction along.

Because of the good confinement of die plasma close to the cathode in the

magnetron, there is usually very little ion flux available at the sample position