Koster G., Rijnders G. (Eds.) In situ Characterization of Thin Film Growth

Подождите немного. Документ загружается.

214 In situ characterization of thin ﬁlm growth

measurements from the lm, discussed in Chapter 5, can be used in situ to

calibrate the thickness of the deposited lm and hence the sticking coefcient.

A prolometer and transmission electron microscopy (TEM) cross-sections

are among ex situ tools that can also be used to deduce lm thickness.

Measurement of the deposition rate can be either intrusive, by placing

the sensor inside the vapor and obstructing deposition in the line from the

source, or non-intrusive by performing a remote sensing measurement such

as optical absorption through the vapor; see Fig. 8.1.

A deposition source has a deposition distribution that can typically be

modeled as a cos

(q) dependence where q is the angle from the source

direction (Smith, 1995). The exponent n is often found to be between 1 and

4, depending on the type of source and source condition. Intrusive vapor

sensing techniques suffer the disadvantage that the ux is measured at a

different position, i.e. angle, than the substrate and hence need a correction

factor, commonly called the ‘tooling factor’, for proper calibration to the vapor

ux at the substrate. More signicantly, the ux prole can change during

deposition time and thus the calibration may need to be done periodically or

even continuously. It is of course desirable to measure the vapor ux directly

under the substrate, i.e. non-intrusively. Although non-intrusive techniques,

such as optical sensing, are preferred, the eld has suffered from a lack of

commercially available sensors.

We differentiate also between techniques that measure mass uxes and/or

mass accumulation directly and ones that measure the density of the vapor.

Typically a quartz crystal microbalance (QCM) is used for a mass ux

measurement. Most other vapor sensors measure only the vapor density. The

other division of sensors is whether they are atomic species specic or not.

QCM is not species specic, but atomic absorption (AA), for example, is.

Techniques that can measure atomic species in the vapor quickly, accurately,

and non-invasively are the best ones from a rate control perspective. This

includes several atomic absorption techniques, which we discuss in more

detail. The bandwidth requirement for control depends on the type of source

and rate that is used. Typical effusion cells where temperature is kept

constant require ux sampling times that are tens of seconds or longer, and

similarly for sputtering sources where the power is kept constant. On the

other hand electron-beam evaporators may require fast feedback in the scale

of milliseconds. Thus different feedback bandwidths are required depending

on the system, but in each case the sensor signal can be fed back in a loop

to keep the vapor ux constant.

8.3 Quartz crystal microbalance (QCM)

QCM for mass ux measurement is the most powerful and widely used

technique in vapor sensing. QCMs are used in most industrial processes for



215In situ deposition vapor monitoring

rate control where rate control is implemented. QCM is also unique among

the techniques discussed in this chapter in that it directly measures mass

per unit area deposited from a vapor, assuming that one knows the elastic

property (acoustic impedance) of the material. Unlike other techniques,

it needs no further calibrations. Furthermore if the  lm density and the

sticking coef cient on the sample are known, then the deposition rate can

be precisely calculated.

A QCM consists of an oscillating piezoelectric quartz crystal that is

sensitive to the added mass, as shown in Fig. 8.2. When a voltage is applied

across the sides of a piezo crystal it will resonate at discrete frequencies.

Then when mass is added to the face of a resonating crystal, the resonant

frequency is reduced. This change in frequency is very repeatable and well

understood for speci c oscillations of the quartz, such as for the quartz AT

cut. The AT-cut quartz, which vibrates in its thickness shear mode, is the

most commonly used crystal monitor because of its superior temperature

stability and good mass sensitivity. Typically, one surface of the crystal is

made in a spherical shape to contain the vibrational energy within the center

of the crystal and to reduce the coupling to unwanted modes at the edges

(Shockley et al., 1963; Wajid, 1997).

In 1959 Sauerbrey  rst described the equation for the change in frequency,

D f, of an oscillating quartz crystal after coating with a certain added mass

–

where M

is the mass of the uncoated quartz crystal and f

is its uncoated

resonant frequency. The  rst frequency measurement instruments used the

~ 5–6 MHz

Electrodes

Quartz

Film

Vapor ﬂ ux, G

8.2 Schematic of a quartz cystal monitor.



216 In situ characterization of thin ﬁ lm growth

following simple equation, known as the Sauerbrey equation, for the deposited

 lm thickness on a QCM, d

= –

where r

and r

are the densities of the  lm and quartz crystal (2.649 g/cm

respectively, and N

is the frequency constant of an AT cut quartz crystal

(166,100 Hz cm).

If the change in resonant frequency is greater than 2% of f

the above

equation is not valid and one needs to take into account the acoustic impedance

mismatch between the quartz and the  lm. The resulting equation that treats

the problem as a one-dimensional composite resonator was developed by

Lu and Lewis (1972):

= –

Zn Z

–

˜˜˜

where Z is the acoustic impedance ratio Z = (d

)

1/2

and u

are the

shear moduli of the quartz and  lm, respectively. Advances in microprocessors

taking place at the same time in the 1970s made it practical to solve the full

equation in real time with the instrument. Most QCM deposition process

controllers sold today use this sophisticated equation for calculating the

 lm thickness.

Typical instrumentation prior to the mid-1990s relied on the use of an

active oscillator circuit that actively keeps the crystal in resonance. Oscillation

in the crystal is then sustained as long as the gain provided by the ampli ers

is suf cient to offset losses in the crystal and the circuit. As the crystal is

loaded with more mass the electrical characteristics of the oscillator circuit

degrade, the impedance Z rises, and the system becomes unstable to mode

hopping.

A new type of system developed by Wajid (1996), using a so-called

‘mode-lock technology,’ eliminates the crystal from the active oscillator

and instead has a separate resonating circuit that allows the instrument to

constantly test the crystal response to an applied frequency. In this way the

instrument can determine the resonant frequency and at the same time verify

that the crystal is oscillating in the desired mode. The new system is immune

to mode hopping and resulting inaccuracies. The crystal’s frequency can be

determined to less than 0.005 Hz at a rate of 10 times per second. Furthermore

since this system can measure several modes as it sweeps the frequency, other

features are available. In particular one can measure the frequency shift of

the lower frequency anharmonic that is close to the fundamental mode. This

mode has a slightly different mass sensitivity from the fundamental mode

and this difference can be used to estimate the Z-ratio of the material. This



217In situ deposition vapor monitoring

is the so-called automatic Z-ratio instrument calculation performed in the

advanced QCM controllers today.

Among the most common applications of QCM technology are in the

production of optical coatings. For these dielectric materials the useful

crystal life is much shorter than for deposition of metals. Unlike typical

metal lms where the crystal failure is due to the viscous losses in the bulk

of the material, in this case stress in the thin lm causes cracking and the

rough surface breaks up the oscillatory wave of the quartz crystal which

causes loss of resonance. Wajid (1993) reported signicant improvements in

the life of the quartz crystal by adding a compliant buffer layer between the

gold electrode and the crystal. This buffer layer needs to have good adhesion

and low acoustic damping and one such material is zinc. Observed crystal

life enhancements were two to six-fold.

Still one of the main drawbacks of the QCM is the nite lifetime of the

quartz oscillator. The crystal is continuously being coated with a damping

material and the resonant Q decreases continuously; eventually the resonant

mode cannot be measured. To help alleviate this problem, manufacturers of

QCM systems have developed 6-, 12- and even 24-crystal carriers where

the crystals in reserve can be automatically switched in during the process.

For long continuous deposition processes, as in web coating, however, the

nite QCM life can still be a signicant limitation.

Another signicant disadvantage is that the QCM is not species specic,

i.e. it cannot differentiate between different atoms deposited or potential

reactions taking place on the QCM such as oxidation of the atoms. On the

other hand, when properly calibrated, the QCM is a very powerful tool and

can also be used to measure reactions, such as lm oxidation on the crystal

or etch rates, and consequently can also be used to measure reactive gas

uxes; see for example Matijasevic et al. (1990).

8.4 Vapor ionization techniques

Most vapor ux sensing techniques other than the QCM utilize excitation of

the vapor molecules to detect their response. Vapor ionization techniques,

for example, ionize the atoms in the vapor plume in order to measure the

atomic density. The ionization rate and the resulting collected ion current

are proportional to the atomic density of the vapor in the sensor, N. Figure

8.3 shows a schematic of a generic device of this type.

8.4.1 Ion gauge and chopped ion gauge (CIG)

The simplest type of an ionization sensor is just an ion gauge. In an ion-gauge

ux monitor the lament should be shielded so the vapor does not react with

the lament and other components should be shielded from accumulating



218 In situ characterization of thin ﬁlm growth

deposits. Simultaneous monitoring during co-deposition can be achieved by

shielding each monitor from the other sources, as shown in Fig. 8.4.

In order for the ion gauge sensor to be sensitive to the deposition ux only

and not to the background gas, a mechanical chopper wheel can be used;

see Fig. 8.4. This is the so-called ‘chopped ion gauge’ (CIG) monitor. The

deposition ux that enters the gauge is then pulsed by the chopper and can

be measured with a phase-sensitive lock-in detector.

Electrons

Filament

Vapor

Light

emission or

ion collection

8.3 Schematic of a generic vapor ionization sensor. The ionized

atoms can be collected or mass ﬁltered, or light emission can be

analyzed.

Monitor

of A

Monitor

of B

Shield

Chopper

Source

Shield

Substrate

8.4 Schematic of ion gauges used in co-deposition.



219In situ deposition vapor monitoring

The CIG has seen extensive use in basic research over the past 45 years,

particularly for superconducting compounds (A-15 and high-temperature

superconducting cuprates) and for industrial optical coating on a large scale

(Hammond, 1975, 1978). As large-scale processes requiring electron beam

sources for their high rates become attractive, the knowledge and experience

of the unique control features needed for high rate are becoming important.

Initially after the introduction of the CIG the bearings in the motor and

chopper made the unit unreliable. However, the advent of modern high-

vacuum high-temperature greases have eliminated this issue.

The lessons learned from the use of the CIG can be valuable for controlling

processes involving high-rate multiple source co-evaporation, especially in

cases where at least one of the elements is a high melting point metal. This

knowledge can be applied to other rate monitors that sample individual

evaporation sources, either by collimation or by element specicity. Niobium

and similar transition metals are such an example where suitable crucibles

or liners are not available. The evaporation of such materials from a water-

cooled bare hearth by the electron beam heating of the surface results in a

rate, as measured at one point in space, having random uctuations in time.

The amplitude of the uctuations may be 50% of the rate, with frequency

components up to tens of Hz. The source of this variation is easily seen

upon looking at the molten source material using an optical lter (welding

glass): the large gradient in the temperature between the impact point of

the electron beam and the region in contact with the water-cooled copper

hearth produces violent convection turbulence in the molten material. The

brightest regions, which correspond to the regions with the highest rates of

evaporation, are seen to move around and change shape.

Co-condensation of atoms from two sources would, under the conditions

just described, produce an inhomogeneous composition. The CIG evaporation

rate monitors that were developed to solve the above problem are based

on the ionization gauge, with features to permit its use with electron beam

evaporation. The sensor has the Bayard–Alpert conguration, i.e., the

lament outside of the helical grid, and the collector inside. This permits

the majority of the evaporant to pass through the gauge without condensing

onto the elements of the gauge (the collector is offset from the center of

the grid axis, as is the collimated evaporant ux). Charged particles of both

signs are present in electron beam systems, and can cause erroneous signals

on the collector. Electric and magnetic stripping are used to mitigate this

issue. Small bar or horseshoe magnets usually are sufcient to deect the

electrons that are present at the energy corresponding to the high voltage

at the gun. The secondary electrons produced are captured by the electric

strippers with ±30 V. In order that the sensor be sensitive to the evaporant

ux only, and not to the background gas also, a mechanical wheel chopper

is used. The evaporant ux that enters the gauge is thus pulsed and phase



220 In situ characterization of thin ﬁlm growth

locked to the chopper. The AC signal, which is proportional to the rate, can

be measured with a phase-sensitive lock-in detector. This is compared with

a pre-set DC level, which is set proportional to the desired steady rate of

evaporation. The difference signal, an error signal of positive or negative

voltage, is fed back to the electron beam power supply.

The error signal is used in two ways at the electron beam power supply

to provide a feedback to the evaporation rate. The rst and most commonly

used method is to use the error signal to vary the electron gun emission by

changing the temperature of the lament by controlling the emitter current.

This is satisfactory for error signals whose main frequency components are

less than 1 to ½ Hz. The electron gun lament temperature response begins

to get out of phase with the change in the heating current at about 1 Hz.

This results in the feedback changing from negative to positive, and wild

oscillations result.

A second feedback loop has been devised to deal with the higher frequency

uctuations. Lateral deection magnets (as part of the e-gun source) are

used as follows: a triangular, or other, current wave-form at about 50 Hz

is used to move the electron beam from side-to-side at the point of impact

on the molten metal being evaporated. The effect of moving the beam is to

decrease the average power density, thereby reducing the evaporation rate.

The amplitude of the beam motion is adjusted such as to reduce the rate

by 5 to 20%. The error signal from the CIG is then used to modulate the

amplitude of the sweep, thus changing the evaporation rate. The advantage

of this method is that it is capable of responding to changes in the rate at

frequencies in the 1–10 Hz range. A similar method can be used by controlling

the sweep amplitude on the Wehnelt grid in the electron emitter of sources

so equipped. In that case one can extend the feedback bandwidth to 100s of

Hz. The long-term stability of the CIG-controlled evaporation rate can be

better than 2%.

8.4.2 Quadrupole mass spectroscopy

In addition to the ion-gauge monitors described, it is even better, especially

in co-deposition, to use a species-specic monitor. One such device is

a quadrupole mass spectrometer (QMS), which is simply an ion gauge

followed by a mass/charge-selective lter. The QMS has the advantage of

high sensitivity and large bandwidth, down to 0.1 pm/s at 1 Hz. Typically

the QMS is equipped with an electrometer as a current multiplier and its

bandwidth can be 100s of Hz. The key issue with a QMS as a rate sensor

is the long-term stability.

Schellingerhout et al. (1989) described a control system using a Balzers

QMS for e-beam evaporators. They passed the control signal through a low-

pass lter to control the lament current and a high-pass lter to obtain a



221In situ deposition vapor monitoring

control for the fast Wehnelt grid voltage. Jaccard et al. (1994) used a Hiden

QMS to precisely calibrate several monolayer deposition of each species

in a sequential deposition molecular beam epitaxy (MBE) process. Newer

versions of the QMS allow for multiplexing of up to 16 outputs for different

masses in feedback control.

The QMS method is becoming especially useful for molecular vapors

that need to be cracked to the monoatomic state in order to prepare lms

of high-quality materials, such as, for example, the topological insulators

(such as Bi

), and in commercial production of the solar photovoltaic

CIGS (CuInGaSe) and other materials containing the molecular elements

(As, Sb, S, Se, and Te). These materials as molecules (As

, se

, etc.) in

their normal evaporative state do not stick readily or combine in compounds,

and require orders of magnitude larger ux than if the atomic species were

supplied. Insufcient and inconsistent molecular cracking results in non-

stoichiometric compounds, difculty in achieving the best combination of

ux and substrate growth temperature, as well as an enormous consumption

of material increasing the cost of production, as well loading the chamber.

Methods of cracking the molecules are either thermal or RF plasma. The

thermal cracking often results in the dimer (As

, se

, etc.) and does achieve

better quality and utilization. On the other hand RF plasma cracking results in

the monoatomic state and has much superior sticking capacity and reactivity

to form high-quality compounds.

For the consideration of the usefulness of the QMS for monitoring these

cracked molecules one needs to understand what the QMS does to the

molecules: the ionizer uses electron impact to ionize the elements, typically at

70 V, because this is close to the maximum in the ionization cross-section, in

general across the periodic table. In the case of molecules there are additional

pathways of producing an ionized particle that is extracted and displayed as

the charge/mass ratio in the operation of the spectrometer. These include a

two-step process, disassociation followed by ionization, and also a one-step

ionization during disassociation. Since the latter involves only one step it

appears to have the larger probability. This becomes important because the

energy dependence of the cross-section as the energy is lowered close to the

threshold (typically 10 eV) allows the ionization during disassociation to be

removed at roughly 15 V in the case of As, leaving only the molecule visible

in the spectrum without the false impression that cracking has produced

atomic species. Thus As without cracking and an electron energy of 70 V

will show disassociation-ionization for all molecular sub-units from 4 down

to 1. At the lower ionization energy of 15 V only the As

ion is found. With

thermal cracking As

and as

appear at 70 V, while at 15 V only the As

seen (Campion et al., 2010). This allows for a good evaluation of the cracking

efciency. The same possibility is indicated by the cross-sectional data for

the molecules O

, N

, H

, and carbon oxide (Kieffer and Dunn, 1966).



222 In situ characterization of thin ﬁlm growth

It is expected that in the case of radio frequency (RF) plasma cracking

of the other molecules (in addition to As) the same methods will show the

efciency of the cracking, using the lowered electron energy (soft ionization),

and can be used in monitoring and controlling the rates. There are unpublished

data indicating that this is the case.

8.4.3 Electron impact emission spectrometry (EIES)

Another species-specic ionization monitor is the electron impact emission

spectrometer (EIES). EIES uses the characteristic atomic emission to identify

the element being monitored. As in a QMS, an electron beam, in this case of

180 V, is oriented perpendicular to the ux, and upon impact with the atoms

excites them to higher levels, which then decay to lower or ground state,

emitting photons of characteristic wavelengths (see Fig. 8.3). The intensity

of the particular emission summed over all the atoms is linearly proportional

to the vapor density of the corresponding species. The following equation

relates the light intensity measured for a specic transition I

to the vapor

number density N:

= KNs

I /e

where K is a calibration constant relating all the optical system losses, s

is the transition cross-section at constant excitation energy, I is the electron

current density, and e is the electronic charge.

The light emitted is guided through a vacuum window to a lter or

spectrometer, and then the intensity is measured by a photomultiplier tube

(PMt) (Lu et al., 1977; Gogol and Reagan, 1983). To avoid interference

when evaporating a number of elements, the light path can be divided using

split optical ber bundles and individually chosen optical lters are inserted

before independent PMTs. The gain for each PMT can be controlled to

distinguish the elements and control the power to the evaporation sources.

In practice as many as four channels have been used.

There can be interference among the spectra of several elements when

co-depositing elements. This mainly depends on the specications of the

wavelength selection device (lter or monochromator) and the elements

of interest. However, in most of the co-deposition cases, one can select

appropriate lines to avoid interference, although sometimes this prevents

the use of the strongest emission line for certain elements.

The broadband emission from residual gases may also be a problem in

certain cases, particularly for low rate deposition in a relatively high background

pressure (>10

–5

torr). However, for non-reactive MBE processes, the residual

gas interference is not a serious problem, nor for some industrial processes

running at a very high rate with non-ultra-high vacuum (UHV) conditions.

In those cases where gas interference does become a problem, particularly



223In situ deposition vapor monitoring

in reactive depositions, a dual-chamber EIES sensor can be used to solve the

problem (Lu et al., 2008). One chamber has the vapor ux passing through,

while the other is open only to background gases. The emitted spectra are

added and using appropriate signal processing technique, the interfering

signals from the residual gases can be completely eliminated from the output

signal.

All the above electron-beam ionization monitoring techniques can be used

only when the electron mean free path is larger than the monitor diameter,

i.e., in high vacuum with a pressure of less than 10

–4

torr. The sensors can be

differentially pumped and thus could support an order of magnitude higher

pressure in the chamber. However, scattering in the chamber at pressures

of mid-10

–5

torr range will change the sensitivity for these instruments and

render them useless (Appelboom et al., 1993). The sensors are also not

usable in a plasma environment because of interference. For higher pressure

processes or in plasmas, optical excitation and detection techniques must be

used to monitor vapor concentration.

The ionization monitors typically have a lifetime that is equivalent to

hundreds of QCMs. They still need to be cleaned out periodically, but they

are effective when depositing large amounts of material continuously. QMS

and EIES are species specic as well.

8.5 Optical absorption spectroscopy techniques

Use of optical methods for measuring AA in the vapor was rst demonstrated

for barium (553.6 nm) by Gunter Wessel (1949). It was as early as the

late 1960s that Stirling and Westwood (1969) and Fazekas et al. (1972)

demonstrated the use of AA in vapor deposition processes. The technique was

thus demonstrated more than 40 years ago, but instrumentation development

had a long and complicated path.

Theoretically the fraction of an optical beam remaining after propagating

through a medium with absorption coefcient a is given by Beer’s law:

I (z) = I (0) exp (– az)

where a has units of 1/cm and is a function of light frequency. The absorption

coefcient is proportional to the density of absorbing species N (1/cm

) and

can be expressed as a = Ns, and the effective absorption path length per pass

of length l is Nsl. s is the absorption cross-section in units of area (cm

Infrared radiation usually probes molecular vibrations, while visible/

UV light probes transitions between different electronic states. Electronic

transitions can be to predissociative or dissociative states, leading to a mild

or extensive broadening of the absorption lineshape. Moreover, excitation

to a level that dissociates strongly perturbs the deposition process. Typical

UV absorption cross-sections are s ~ 10

–19

–10

–16

. For s = 10

–17

