Reactive magnetron sputter deposition: a journey from...

Copyright D. Depla 1

Reactive magnetron

sputter deposition:

a journey from target to substrate

D. Depla

Dedicated Research on Advanced Films and Targets

Ghent University

D. Depla Plasma diagnostics and modelling, Mons, 2018 www.DRAFT.ugent.be

Setting the scene

Who ?Some keywordsMain features of the model

Setting the scene Question 1 Question 2 Question 3 Question 4 Question 5 Final question

1



Who ?

Further development of the RSD modelSimulations

Koen Strijckmans

Oxide formation on the target

Roeland Schelfhout


2


RSD2018

December 6th and 7th.

Short course : December 5th www.rsd2018.be


3



A few keywords

The RSD model aims to get a betterinsight in the reactive sputtering processKey features are

• Analytical approach• Click on/off approach• GUI to assist the user• Main focus on the “hysteresis”

D. Depla, S. Heirwegh, S. Mahieu, and R. De Gryse, J. Phys. D-Appl. Phys. 40404040 (7), 1957 (2007).D. Depla, X. Y. Li, S. Mahieu, K. Van Aeken, W. P. Leroy, J. Haemers, R. De Gryse, and A. Bogaerts, J. Appl. Phys. 107107107107 (11) (2010).K. Strijckmans, W. P. Leroy, R. De Gryse, and D. Depla, Surf. Coat. Technol. 206206206206 (17), 3666 (2012)K Strijckmans and D Depla, Journal of Physics D: Applied Physics 47474747 (23), 235302 (2014)K. Strijckmans and D. Depla, Appl. Surf. Sci. 331331331331 (0), 185 (2015)K. Strijckmans, Ghent University, 2015


4


Target processes

As we know ions getimplanted, we describe theion implantation in the bulk

Initially the implantedatoms are not chemicallybonded. We describeexplicitly the chemicalreaction

knrnmSurface processes are described as abalance between removal and addition

The transport is defined bysputtering and redeposition

K. Strijckmans and D. DeplaJournal of Physics D: Applied Physics 47 (2014) 235302


5



An example

Target poisoning occurs not onlyat the surface, but also in thesubmonolayers.The major contribution is reactiveion implantation.The RSD model accounts for thechemical reaction between theimplanted species and the targetmaterial.


6


Question 1 : Can we predict the first critical point ?


7



An experiment

Target material : aluminiumCircular planar target (2 inch diameter)Pumping speed : 35 l/s

The first critical point linearlydepends on the discharge current

Before the first critical point hardly anincrease of the oxygen pressure

Origin : the target and thedeposited material react with theintroduced oxygen.Getter pump action.


8


A few experiments more …

Different conditions show a similar behaviour

Can we describe this more quantitative ?


9



Some equations, sorry !We assume the target is much smaller than the deposition area, orstated differently before the first critical point the gas is onlyconsumed by the deposited material

The condition of the deposited material is given by the following balance equation

chemisorption compound deposition metal deposition

But the target is hardly poisoned :

Metal deposition

The maximum getter is just given by the product of the discharge current, andthe metal sputter yield. Or the first critical point scales linearly with thecurrent. And what about your experiments ?


10


Database

Your data is stored in a database(currently 120 entries, 32 differenttarget-gas combinations)

Slope between experimental flow at the1st critical point, and maximum getter rateis 0.541.

Engineering approach

�� sccm �� 3.75 sccm/A ��


11



How good is our model ?

Perform simulations (53248 in total)

Most common value for �̅�̅� : 0.575

Simulations for different substrateareas, sticking coefficients, sputteryields, pumping speed (colour coderefers to combinations) But we also deviations from the fit …


12


DeviationsLet us return to assumptions …

Low reactive gas pressure before critical point

Not true for low reactivity materials

Only consumption by the deposited material

Not true at high pumping speeds

Only consumption by the deposited material

Not true at low discharge currents


13



Question 1 : Can we predict the first critical point ?

Answer 1 : Yes, reasonable well…


14


Question 2 : Can we predict the discharge voltage behaviour ?


15



Experimental observation

M = metalX = nitrogen/oxygen

M = metalX = nitrogen/oxygen in the bulk

at the surface


16


Why do we have this relation ?For high values of R (>1), we have a reduced oxide (nitride) target surface.Reduced oxides (and most nitrides) are (semi)-conductors with a low electronyield or high discharge voltage during poisoning. The opposite holds for lowvalues of R (<1).

Another interpretation of R : it is a measure for the difficulty to form a fullystoichiometric compound during ion bombardment.

But, despite the impressive amount of experimental data supporting this idea,we need first to proof that during reactive sputtering, we indeed have areduced oxide surface.


17



XPS to confirm …reactive sputtering of Ta

target

1” magnetron

A new design of a 1’’magnetron always toremove the target invacuo and transfer itto the XPS analysischamber.

For Ta…

For Ta, suboxides are formed, but athigh fractions, the stoichiometricoxide is observed.


18


XPS to confirm it … reactive sputtering of Al

For Aluminium only theoxide and the metal aredetected. No suboxides areobserved.


19



Relation discharge voltage-oxide fraction

Al

Ta

The discharge voltage can be correlated with the presence of thestoichiometric oxide.

The oxygen fraction during reactive sputtering, even in poisoned mode, isseldom larger than 0.2


20


Answer 2 : Yes, and we understand the mechanism.

Question 2 : can we predict the discharge voltage behaviour ?


21



Question 3 : Do we have data on the oxide sputter yield ?


22


Is data on the sputter yield important … ?

The ratio between the metal sputteryield Ym and the compound sputter yieldYtot defines to a large extend thehysteresis phenomenon.

So, a simple conclusion is that hysteresisis caused by the lower sputter yield ofcompounds.

“…the authors come to the conclusion that the sputtering yield of oxides are in most cases similar to oreven higher than those of the corresponding metals.", and further in the same overview “For comparison,measured data for Si and Ta are included. For SiO2, the sputter yields are about the same magnitude orslightly lower than Si, but for Ta2O5, generally higher yields than for Ta are observed."

G. Betz and G. K. Wehner, \Sputtering of multicomponent materials," in Sputtering by Particle Bombardment II: Sputtering of Alloys and Compounds, Electron andNeutron Sputtering, Surface Topography, edited by R. Behrish (Springer Berlin Heidelberg, Berlin, Heidelberg, 1983) pp. 11-90.


23



This is the data we have …

Tantalum oxide

In agreement with theexpectation, no hysteresisobserved.

J. M. Ngaruiya, S. Venkataraj, R. Drese, O. Kappertz, T. P. L. Pedersen, and M. Wuttig, Physica Status Solidi a-Applications and Materials Science 198198198198 (1), 99 (2003).K. Juškevičius, M. Audronis, A.s Subačius, S.. Kičas, T. Tolenis, R. Buzelis, R. Drazdys, M. Gaspariūnas, V. Kovalevskij, A. Matthews, and A. Leyland, Thin Solid Films 589589589589, 95 (2015).

Niobium oxide


24


But there is a problem …

Tantalum oxide12 times lower

Niobium oxide5 times lower

Why is there a huge drop in deposition rate ? Most points are addressed.

Bombardment by O2+ instead of Ar+ ?

Not always a valid argument

Ta Nb

Yield (Ar) 0.876 0.688

Yield (O2+) 0.384 0.502

��

��

� � �� !!

��!!

��

��

"��#$%�

�$�

"�#$%

Deposition profileswill be similar.

Voltage are different.�$�

�

�#$%�

��, '(

�#$% ��, '(


25



Can we trust the sputter yields ?Measuring the mass difference and discharge current at constant discharge voltage

Difficulties1. Electron yield2. Neutrals3. Ion energy

NeutralsCharge exchange mechanismWindow based on literature

I.Y. Burmakinskii et al.Tech. Phys., 49 (1) (2004), pp. 119–122

E. Bultinck, et al., New J. Phys., 11 (2009)

Consistent with advancedsimulations

E. Bultinck, et al., New J. Phys., 11 (2009)

M.J. Goeckner, J.A. Goree, T.E. SheridanIEEE Trans. Plasma Sci. 19 (1991), 301–308


26


But for oxides …

The weight loss, and so thesputter yield, as a function ofthe discharge current duringreactive sputtering (poisonedmode) of an aluminiumtarget.

Re-deposition on the target was excluded by shielding the re-deposition areas.

Can we proof it in another way ?


27



An alternative way to measure the sputter yield

)#% � )� � )� � )�

Balance equation

) $%� � )� � )� ��

2�#$%

In poisoned mode, theconsumption rate isdirectly related with thecompound sputter yield.

Measure with a mass spectrometer the oxygenpressure (simple explanation)


28


Comparison

This new method confirmsthe current dependency ofthe sputter yield

Reason : we now work onthe hypothesis that thetarget contains non-reactedimplanted oxygen atoms. Atlow discharge current theseatoms can diffuse from thetarget. At high dischargecurrent their presencereduces the sputter yield.


29



Overview sputter yields

Several observations

• Not related to the targetcondition (see sputter cleaningexperiments)

• Large difference between ratiobased on ion beam experimentsand deposition rate experiments


30


Correlation between compound sputter yieldsA remarkable plot because R ensuresthat the flux from the target equalsthe compound composition.There is a big difference betweenion beam (or sputter cleaning)experiments, and sputtering inpoisoned mode

Ar+ Ar+ O2+

O2

The removal of an oxygen atom increases the probability to sputter a metal.

A removed oxygen atom, can be replaced by another (implanted) oxygen atom.

Outliners are probably related to texturing.


31



Answer 3 : No, we have not. Moreover the mechanisms are not fully clear.

Question 3 : Do we have data on the oxide sputter yield ?

“…the authors come to the conclusion that the sputtering yield of oxides are in most cases similar to oreven higher than those of the corresponding metals.", and further in the same overview “For comparison,measured data for Si and Ta are included. For SiO2, the sputter yields are about the same magnitude orslightly lower than Si, but for Ta2O5, generally higher yields than for Ta are observed."

G. Betz and G. K. Wehner, \Sputtering of multicomponent materials," in Sputtering by Particle Bombardment II: Sputtering of Alloys and Compounds, Electron andNeutron Sputtering, Surface Topography, edited by R. Behrish (Springer Berlin Heidelberg, Berlin, Heidelberg, 1983) pp. 11-90.

This is 35 years ago !


32


Question 4 : And what about the time dependency of the process ?


33



Consumption/sputter yield as a function of time (x2)

Just wait …

Again … can we trust this sputter yield behaviour ?


34


Sputter cleaning to proof

Sputter cleaning experiments give information about initial target behavior

Measure cleaning time

Calculate Y:

(With fixed thickness d = 2 nm)

+ (,'%#%- ∝/

Confirmation of our experiments.Origin is probably due todiffusion of the non-reactedoxygen from the target.


35



Desorption

An abrupt change as a function of the reactive gas fraction is noticed.Slow outdiffusion of nitrogen is detected.


36


The reaction rate constant

So, the reaction rate constant is an effective reaction rate constant

The grey regions are based on adedicated fitting of hysteresisexperiments for Y and Al,published before this work andusing other data.

knrnm


37



Answer 4 : Diffusion (radiation enhanced, thermal, pressure driven) is the processwe need to describe to quantify in detail our experiments. Moreover, for pulsedexperiments, it will become a necessity.

Question 4 : And what about the time dependency of the process ?


38


Question 5: Does our traditional view show all details ?

P (or V)

Q

P

Q


39



Our “traditional” view on reactive sputtering

In literature different ways are used to study the hysteresis phenomenonFlow control (Q)Stepwise change of the reactive gas flowWorking at constant current (I), voltage (V) or power

Measuring the change in voltage, current and pressure (P)

P (or V)

QPressure control (P)Stepwise change of the reactive gas pressureWorking at constant current (I), voltage (V) or power

Measuring the change in voltage, current and pressure (P)

P

Q

Of course, voltage control is also possible. It has several advantages.It is fast, easy and permits to work in a “random” way to exclude any long termeffects such as target erosion.


40


PVIQ plot : metallic to poisoned mode

high pressure, low voltagePoisoned mode

low pressure, high voltagemetallic mode


41



Comparison with common plots (projections from (P,V,I,Q)

voltage (V)Pressure(Pa)

I-V characteristic for the metallic mode

I-V characteristic for the poisoned mode

flow (sccm)

Current(A)


42


PVIQ plot : metallic to poisoned mode

high pressure, low voltagePoisoned mode

low pressure, high voltagemetallic mode


43



PVIQ plot : from poisoned to metallic mode

When don’t get the same result when the (P,Q,I,V) space isaccessed in a different direction.


44


So forward and back is different

From metallic to poisoning

From poisoning to metallic


45



Double hysteresis

Origin :

1. An avalanche due the getteringcapacity of the deposited material which cause a single S (see first critical point)2. An avalanche due to the reaction kinetics of the implanted reactive ions which causes the double S .

The existence of a double S-shaped process curve during reactive magnetron sputteringR. Schelfhout, K. Strijckmans, D. DeplaApplied Physics Letters 109 (2016) 111605


46


Origin of the double hysteresis : a possible answerSetting the scene Question 1 Question 2 Question 3 Question 4 Question 5 Final question

47



Question 5 : Does our traditional view show all details ?

Answer 5 : No. The new approach confirms the model, but adds new challenges.

We observe a double hysteresis in the simulations,but the fit with the experiment is not completely successful.


48

Reactive magnetron sputter deposition: a journey from...

Documents

Transcript of Reactive magnetron sputter deposition: a journey from...