UPTEC F 15035
Examensarbete 30 hp
Juni 2015
Simulation of how pressure influences
the reactive sputtering process
Erik Strandberg
Teknisk- naturvetenskaplig fakultet UTH-enheten
Besöksadress:
Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0
Postadress:
Box 536 751 21 Uppsala
Telefon:
018 – 471 30 03
Telefax:
018 – 471 30 00
Hemsida:
http://www.teknat.uu.se/student
Abstract
Simulation of how pressure influences the reactive
sputtering process
Erik Strandberg
Sputtering is a physical vapor deposition (PVD) process used to create thin films, i.e very thin layers of material. To form compounds, such as oxides and nitrides, it may be beneficial to add a reactive gas to the process which is known as reactive sputtering. This thesis focuses on the simulation of the reactive sputtering process and, more specifically, the effect of the process pressure.
Two models have been developed. A Monte Carlo model simulates the distribution of sputtered material throughout the chamber. It is based on the binary collision model with initial conditions acquired from simulations in TRIM. The hard-sphere potential is used as interaction potential in the scattering calculations. The effect of the process pressure is studied for two different elements, sulfur and tungsten. It is found that the distribution of material is heavily influenced by the pressure. A high pressure gives a more diffusion-like distribution compared to a low pressure. As the pressure is increased the deposited material’s energy distribution is found to be shifted towards lower energies until it reaches the energy of thermalized atoms.
The second model developed is an extended Berg model that incorporates the effect of redeposition of sputtered material on the target, implantation of reactive ions in the target and preferential sputtering. Using simulations the effect of these extensions is discussed. It is found that an increased pressure may eliminate the hysteresis region which has been observed experimentally. Finally an outline is presented on how the two models can be unified into a Berg-model that takes the non-uniform distribution of sputtered material into account.
Ämnesgranskare: Sören Berg Handledare: Erik Särhammar
Populärvetenskaplig sammanfattning
Tunnfilmer, väldigt tunna lager av material, har en stor spridning i vår moderna värld. Exempel på användningsområden inkluderar metallskiktet i CDskivor, transistorer i elektriska kretsar, solceller och hårda beläggningar på borrar. För att skapa väldigt tunna lager finns det flera tekniker varav den som studerats i detta examensarbete heter sputtering. Vid sputtering
placeras det objekt som skall beläggas (substratet) tillsammans med materialet som skall utgöra beläggningen (targeten) i en vakuumkammare. Genom att beskjuta targeten med energirika partiklar så kommer det att “slås ut” atomer från targeten. Dessa kommer att färdas genom kammaren för att fastna på substratet och därigenom skapa tunnfilmen. För vissa beläggningar, exempelvis oxider, är det även önskvärt att tillsätta en reaktiv gas, såsom syre, till kammaren i en process som kallas reaktiv sputtering.
För att modellera reaktiv sputtering finns det olika modeller varav en är Bergmodellen.
Bergmodellen inkluderar dock inte processtrycket som i experiment visat sig påverka processen. Det här examensarbetet har därför fokuserat på att utöka modelleringen för att inkluderar processtrycket. Utökningen består av två separata modeller som utvecklats och implementerats som simuleringsprogram.
Den första modellen beräknar hur atomer som lämnar targeten sprider sig i kammaren. Ungefär tio miljoner partiklar simuleras individuellt. Varje partikel följs från det att den lämnar targeten tills dess att den träffar någon yta i kammaren. Partiklen kommer under sin färd att krocka med gasmolekyler i kammaren vilket medför att partikelns riktning och energi förändras. När alla partiklar har simulerats så aggregeras informationen och resultatet blir en distribution över hur partiklarna sprider sig i kammaren. Simuleringar visar att processtrycket spelar en stor roll för spridningen där ett högt tryck medför att partiklarna sprider sig mera än vid ett lågt tryck. Vidare medför ett högt tryck att partiklarnas energi är lägre då de träffar kammarytan.
Den andra modellen är en utökning av Bergmodellen med två fysikaliska effekter som är beroende på processtrycket. Det är dels joniserad reaktiv gas som inplanteras under ytan på targetet samt även atomer som, på grund av kollissioner med gasmolekyler, återdeponeras på targeten. Utifrån simuleringar har effekten av dessa mekanismer studerats. Resultaten visar att den utökade modellen kan förutsäga de trender som observerats experimentellt. Slutligen presenteras hur de två modellerna kan kombineras till en Bergmodellen som även tar spridningen av atomerna i kammaren i beaktande.
Contents
1. Introduction ………1
2. Sputtering process……… 3
3. Material distribution……….. 6
3.1 Initial conditions……… 6
3.2 Scattering throughout the gas phase……… 11
3.3 Aggregating the results………18
3.4 Interaction potentials……… 19
3.5 Simulations and analysis……… 22
4. Reactive sputtering simulation……… 26
4.1 Berg model……… 26
4.2 Dissociative sputtering……… 29
4.3 Ion implantation……… 32
4.4 Redeposition………. 34
4.5 Nonuniform distribution of sputtered material ………35
5. Conclusion………. 37
Acknowledgements………... 38
References………. 39
1. Introduction
Thin film technology is an important part of our modern lifestyle. Without it there would be no CDs or electronic devices such as computers. So what is a thin film? A thin film is a very thin layer of material. A thin film may be applied to give a surface certain characteristics that the bulk material does not exhibit. Examples include extremely hard coatings on drill bits, which enables to drill into hard materials, or low friction coatings in valves, to achieve better performance in engines. Or the product may be so small that thin films are part of the product itself. Examples include the transistors found in modern electronics and thin film solar cells. There are many different techniques for creating a thin film and most of them belong to one of two major groups, physical vapor deposition and chemical deposition. [1]
There are a wide range of different chemical processes with different advantages and drawbacks. What they do have in common is that they are based on chemical reaction.
Processes that uses a precursor gas that reacts with the surface of the object to coat are known as chemical vapor deposition (CVD) processes. CVD processes generally have good uniformity of the deposited film, i.e the thickness of the thin film is almost the same regardless of the structure of the coated surface. Another advantage of the CVD processes is that they have a high throughput which is critical in an industrial application. However CVD processes usually involve adding energy, in the form of heat, in order to stimulate the chemical reaction. This makes it hard to grow thin films on materials that are sensitive to heating, such as plastics. The chemicals involved in a CVD process may be toxic, requiring extra precautions to be taken in production, and they can also be very expensive. [2]
The other category of thin films deposition processes are physical vapor deposition (PVD) processes. In a PVD process atoms are removed from the source material by either vaporizing it or bombarding it with energetic particles. The former is known as an evaporation process and the latter as a sputtering process. Both processes requires that the system is operated in a lowpressure environment. In contrast to the chemical processes the physical deposition processes are characterised by the directionality of the deposition [2]. It is hence hard to achieve a uniform film on a substrate with a complex geometry such as holes. However most coatings can be produced by PVD processes, the deposition rate is good and films can be grown on heat sensitive substrates [1]. Examples of thin films that are produced by physical deposition processes are the metallic layer in a compact disc and the diffusion barrier on the inside of plastic food containers.
This paper focuses on the sputtering process and, more specifically, on reactive sputtering.
Reactive sputtering is a process where a reactive gas, such as O2, is added to the sputtering chamber. The gas will react with the deposited metal atoms and aid in the formation of compounds such as oxides. To get a better insight and to be able to predict the behaviour of these complex processes it is important to have a model that described them. A well used model for reactive sputtering is Berg’s model [3]. It is a simple, yet powerful, model that
described the process at steady state. One property that is omitted in Berg’s model is the partial
pressure of the sputtering gas. There are however experiments that indicate that it influences the sputtering process in a nonneglectable way. Hence the purpose of this paper is to extend Berg’s model so that it incorporates the partial pressure of the sputtering gas and this is done in two parts.
In the first part of this paper a new model, based on the binary collision model, is presented [4].
The model describes how the sputtered material is distributed throughout the chamber. Using simulations the effects of different process parameters, such as the pressure and choice of material, are studied. The second part of the paper describes the enhancements made to Berg’s model. The physical mechanisms that have been considered are the effect of dissociative sputtering, redeposition of sputtered material on the target and ion implantation. Results from simulations are presented as well as an analysis of how these mechanisms influence the sputtering process. Finally an outline is presented on how the two models can be unified into a Berg model that takes the nonuniform distribution of sputtered material into account. The implementation as a simulation software was however considered to be outside the scope of this project due to the limited amount of time available. Hence no simulation, or analysis, for the unified model is presented in this paper.
2. Sputtering process
There are many different variations of the sputtering process but they all share the same principle. The material to deposit, the target, and the object that is to be deposited, the
substrate, are put in a sputtering chamber. The system is operated at a low pressure, usually in the range of a few pascal, and hence the chamber must be sealed and connected to a pumping system. To grow the film the target is bombarded by energetic particles. The energetic particle will start a cascade of collisions between the atoms in the target. During these collisions some of the target atoms will be given such an kinetic energy that they are ejected from the target i.e they are sputtered. The average number of sputtered particles per incident energetic particle is defined as the sputtering yield. The sputtered atoms will then traverse the vacuum chamber and stick to the substrate where the film is grown. [1]
To generate the energetic particles the sputtering gas is ionized and the ions are accelerated by an electric field. To achieve this there are different techniques, direct current (DC) sputtering and radio frequency (RF) sputtering. In the case of DC sputtering a DC power supply is connected to the chamber in such a way that the target is held at a negative voltage, in the order of 1 kV, and the chamber is grounded. The substrate may also be grounded or it can be under floating potential. Due to the difference in potential between the target and the chamber there is an electric field. The free charge carriers (electrons and ions) will be accelerated by it and, if accelerated for a long enough distance without colliding, gain a kinetic energy that is greater than the ionization energy of the sputtering gas. The ions, due to their much larger size, have a higher probability of colliding and they will thus reach a significantly lower kinetic energy compared to the electrons. Hence the ionization of the sputtering gas is mainly accounted for by the electrons. When ionizing the sputtering gas an electron is broken free from its atom and the number of free electrons will increase. Since the electrons are negatively charged they will always be accelerated towards the chamber wall. So if an electron is formed close to the target number of electrons are increased as it is accelerated towards the chamber wall but the newly created electrons will be formed closer and closer to the chamber wall. To be able to sustain the ionization process additional electrons must be created close to the target. This is achieved by the ions, accelerated towards the target, since they will have a great chance of producing new free electrons upon the impact with the target. Due to this effect the ionization process can be selfsustained. [1]
RF sputtering works in a similar way to DC sputtering but due to the very fast oscillation of the power supply, typically in the range of 107 Hz, the electric field in the chamber will quickly alternate. The ions, due to their heavy mass, are not affected by this to any greater extent but the electrons are. They will travel back and forth ionizing sputtering gas atoms in the process.
This reduces the need of the free electrons created when the ions collide with the target’s surface. As a consequence it is possible to sputtering insulating materials with RF sputtering which is not possible with DC sputtering. This however comes at the cost of a decreased deposition rate. [1]
In conventional DC and RF Sputtering the formation rate of ions is rather low resulting in a low deposition rate. To be able to sustain the ionization process a rather high voltage must be applied and the process pressure must be kept high as well. In order to mitigate these effects magnets can be placed behind the target and this is known as magnetron sputtering. The magnets will result in a strong parallel magnetic field close to the target’s surface. Due to the Lorentz force the electrons emitted from the target’s surface will be “stuck” in the magnetic field contributing to the ionization of sputtering gas atoms close to the target’s surface. This results in a higher density plasma that is closer to the target. A drawback of magnetron sputtering is that the sputtering will not be uniformly across the target’s surface but there will be areas with a higher erosion rate. This results in a nonflat surface and since the flux of sputtered particles are highest at the angle perpendicular to the target’s surface a larger fraction of the sputtered material will not be deposited on the substrate but on the chamber walls. This results in a less efficient use of target material. [1]
When the film to grow consist of a compound (oxides, nitrides etc) a reactive gas, such as O2 or N2, is usually added to the chamber. The gas will react with the deposited metal atoms and form compound. Depending on the partial pressure of the reactive gas more or less compound is formed. For a very high partial pressure the film will most likely be stoichiometric and for a low pressure it will not. However the dependence of the composition of the deposited film on the partial pressure of the reactive gas is not trivial. And the partial pressure of the reactive gas does not only affect the composition of the deposited thin film. If the partial pressure is high enough then the target will be mostly compounded resulting in a significantly lower deposition rate. To make matter worse it is hard to control the partial pressure of the reactive gas in the chamber in practice. A more convenient process parameter to adjust is the flow of reactive gas that is let into the chamber. However the dependence between the flow of the reactive gas and the composition of the thin film is even more complicated than that between the composition and the partial pressure and it may even included hysteresis effect. [1]
Figure 1; Hysteresis behaviour with marked transitions (dashed) in a simulated deposition rate
Hysteresis effects are not unheard of and occurs in several different fields of physics such as ferroelectric materials [5]. The deposition rate versus the flow of reactive gas for a typical
sputtering process is shown in Figure 1. Assume that the process is operating stably at point M1 but it is sought to operate at point P. If the flow of reactive gas is increased, even so slightly, the process will however not follow the curve to P but instead go directly to C1. If the process is to return to M1 it is not sufficient to simply decrease the reactive gas flow to the previous setting.
Instead the flow must be decreased so the process passes C0 at which point it will go directly to Mo. It is hence impossible to operate in stable mode in the part of the process curve that lies between M1 and C0. This is known as the hysteresis effect in the context of reactive sputtering.
[1]
3. Material distribution
To gain further insight into how the particles involved in the sputtering process are distributed across the chamber a MonteCarlo simulation model was developed. A MonteCarlo process is a process that includes randomness, i.e two consecutive executions of the process will not generate the exact same result. However the result of the two executions should only differ by an ‘statistical error’ [6]. The model developed in this paper supports simulating sputtered
particles, reflected sputtering gas and negative ions formed close to the target. The simulation is carried out by simulating a large number of particles independently and then aggregating the results. Each particle is followed until it hits a chamber wall and when it does the particle’s energy, incident angle and position is saved. Hence each particle undergoes three major steps during the simulation; obtaining its initial condition, scattering through the gas phase and finally hitting the chamber wall.
3.1 Initial conditions
For each sputtered particle its initial position, direction and energy must be defined. The position is assumed to be uniformly distributed across the target’s racetrack. This is a simplification since a typical racetrack profile, as seen in Figure 2, shows that there are more particles being
sputtered from the center of this circular track than from the edges of it. However the effect of this simplification is minor since the width of the racetrack is small in relation to the dimensions of the chamber. Supporting a nonuniform distribution would require a more complex function and hence more processing power resulting in less particles being simulated per unit of time.
Figure 2 (top); A typical racetrack shown in profile.
Figure 3 (left); Model of racetrack with inner radius and outer radius .r0 r1
Let the racetrack be modelled as a circle of radius with a circular hole of radius . This isr1 r0 illustrated in Figure 3. To calculate the initial position of a sputtered particle the following equation is used
x = r cos β
y = r sin β (3.1)
where is the azimuthal angle and the radial distance from the center of the target. Since theβ r azimuthal angle is uniformly distributed on the interval [ 20, π[ it can be calculated using the equation
πR
β = 2 01 (3.2)
where R01 is uniformly distributed random number on the interval [ 10, [. The radial distance isr however not uniformly distributed on the interval [r , r ]0 1 . To see this consider the area of the segment bounded by two circles centered at ( 00, ) and with radiuses ρ − 0.5dρ and ρ + 0.5dρ where dρ is infinitesimal. The area dA of this segment is given by
A (ρ .5dρ) (ρ .5dρ) πρdρ
d = π + 0 2− π − 0 2= 2 (3.3)
The total area of the racetrack is given as
r r
A = π 12− π 02 (3.4)
Define the cumulative area fraction function CAF F (r )′ as the fraction of the total area of the racetrack that is bounded by a circle of radius . By using (3.3) and (3.4) r′ CAF F (r )′ can be determined using the following function
AF F (r ) πρ dρ
C ′ =πr −πr1
12 02
∫
r′r02 =r −rr −r′2 02
12 02 (3.5)
To randomly pick a value of on the interval r [r , ]0 r1 such that all points on the racetrack have an equal probability of being picked is equivalent to solving the equation
AF F (r)
C = R′01 (3.6)
where R′01 is a random number on the interval [ 0, 1]. This since points randomly spread across the racetrack will have a uniformly distributed amount of area on its inside. Substituting (3.5) into (3.6) yields
r −r2 02
r −r12 02 = R′01
which can be rewritten as
r =
√
R (r′01 12− r02)+ r02 (3.7)
To calculate the initial position of the sputtered particle (3.7) and (3.2) are now substituted into (3.1) which gives
x =
√
R (r′01 12− r02)+ r02cos 2πR 01 y =√
R (r′01 12− r02)+ r02sin 2πR 01where the values of R′01 and R01 are randomized once respectively and then used in both equations.
In addition to an initial position each particle is also given an initial direction and energy. These depends on the type of particle that is being simulated. Negative ions are assumed to form close to the target surface. They are then immediately accelerated by the electric field and hence obtains an energy corresponding to the target potential and a direction that is
perpendicular to the target’s surface. The particle will now be a small distance away from the target but in order to simplify the simulation model the particle is assumed to be infinitely close to the target’s surface. It is also assumed that the particle will not undergo any collisions while being accelerated. The effect of this simplification should be pretty minor since the chance of a particle colliding is proportional to the distance that it travels during the acceleration which is small.
For sputtered and reflected particles the sputtering angle and initial energy are obtained from TRIM simulations. TRIM (Transport of Ions in Matter) is a software used to simulate the effect of ion bombardment [7]. It is based on the binary collision model where a flat surface of a user defined composition is bombarded with ions. The ions have a predefined energy and they hit the surface at a specified incident angle. The output of the simulation is, among other, a list of the particles that are sputtered from the surface along with their energy and direction. Since the sputtering angle and initial energy are correlated they must be saved together and hence the result of the TRIM simulation is a list of pairs. To obtain the initial sputtering angle and energy of a particle a pair from the TRIM simulation is picked randomly. The initial direction and energy, as reported by TRIM, are those of the particle when it is at an infinitesimal distance away from the target’s surface. However to leave the surface the particle will lose energy equal to the surface binding energyEsb. So the energy of the sputtered particle is calculated using the equation
Esput= Etrim− Esb
where Esput is the energy of the particle being sputtered and Etrim is the energy as reported by TRIM. The energy loss will also affect the sputtering angle. Lets start by noting that the velocity
of the particle is parallel to its direction and if is normalized it follows that
vˉ Dˉ Dˉ
vˉ = | ˉv|D (3.8)
Combining equation (3.8) with the equation for kinetic energy
E =mv22 (3.9)
yields the equation
vˉ =
√
2Emtrim D ˉ (3.10)
In TRIM the coordinate system is defined as the zaxis being perpendicular to the target surface and the xyplane coincides with the target surface. Using (3.10) the velocity, before leaving the surface, along the zaxis is given by
D
vtrim,z=
√
2Emtrim trim,z (3.11)
Using (3.9) and (3.11) the kinetic energy, in the direction of the zaxis, before leaving the surface is given by
D
Etrim,z=mvtrim,z2 2 = Etrim trim,z2 (3.12)
Since the surface binding energy is lost in the direction that is perpendicular to the surface the kinetic energy of the particle after leaving the surface, Esput,z, is calculated as
D
Esput,z= Etrim,z− Esb= Etrim trim,z2− Esb (3.13)
where (3.12) has been used. Using (3.9) and (3.13) the velocity of the particle in the direction of the zaxis after leaving the surface, vsput,z, is given by
vsput,z=
√
2Emsput, z =√
2(Etrim trim,zDm 2−E )sb (3.14)
The velocity of the particle after leaving the surface, vsput is given by
vsput=
√
2Emsput=√
2(Etrimm−E )sb (3.15)
Let the sputtering angle be defined as the angle between the particle’s velocity vector, afterα leaving the surface, and the the zaxis. It can be calculated using the equation
cos α = v
sput
vsput,z
(3.16)
By substituting (3.14) and (3.15) into (3.16) the equation
cos α =
√
Etrim trim,zEDtrim−E2sb−Esb (3.17)
is obtained. Figure 4 shows the initial energy distribution for tungsten and sulfur along with the theoretical distribution for a target potential of 400 V [4]. It can be seen that the values from TRIM match the theoretical distribution very well. The sputtering angle distribution is shown in Figure 5 for the same materials and target potential. This distribution also match the theoretical value very well. However it should be noted that these calculations are made under the
assumption that the target is a perfectly flat surface. In practice this is not the case since the erosions is not uniform and hence the formation of a racetrack.
Figure 4; The energy distribution for sulfur and tungsten.
Figure 5; The sputtering angle distribution for sulfur and tungsten.
To get the particle’s initial direction the sputtering angle must be combined with an azimuthalα angle . The azimuthal angle is uniformly distributed on the interval β [ 0, 2π[ and given by the function
πR
β = 2 01 (3.18)
where R01 is a random number on the interval [ 0, 1[. If the target’s surface is placed at the xyplane, i.e its normal vector is ( 0 10, , ), then the definition of spherical coordinates
x = r sin α cos β y = r sin α sin β z = r cos α
together with (3.17) and (3.18) yields the following function to calculate the normalized initial direction of the particle
πR x = sin
(
arccos√
Etrim trim,zEDtrim−E2sb−Esb
)
cos 2 01πR y = sin
(
arccos√
Etrim trim,zEDtrim−E2sb−Esb
)
sin 2 01z =
√
Etrim trim,zEDtrim−E2sb−Esb
3.2 Scattering throughout the gas phase
Once the initial conditions of the sputtered particle are determined the simulation enters its next step which is to simulate how the particle is scattered throughout the gas phase. This is done by repeating the following steps for each particle:
1. Calculate the distance that the particle will travel before its next potential collision with a gas particle. The distance is based on the particle’s mean free path and a random number.
2. Check if particle will hit a surface before reaching the point of its next collision. If so, calculate the incident angle and save this together with the energy of the particle. Move on to the next particle.
3. Determine the amount that the particle is scattered and the energy that it will lose due to the collision. Using this information the particle’s new direction and energy is calculated.
A schematic overview of the particle’s traversal is shown in Figure 6.
Figure 6; The traversal of a sputtered particle (red) through the chamber where it collides with gas particles (blue).
Distance between collisions
The distance that a particle will travel between two successive collisions is distributed nonuniformly. To obtain a value from this distribution the equationλ
−
λ = λ0ln R01 (3.19)
is used where λ0 is the mean free path and R01is a random number on the interval ] 10, ][4]. If the gas is thermalized and the velocity of the sputtered particle is much larger than that of the gas molecules then the mean free path is given by
λ0=nσ1 (3.20)
where is the concentration of gas molecules and is the scattering crosssection of a gasn σ molecule. If on the other hand the velocity of the sputtered particle and the gas molecules are roughly the same, i.e the sputtered particle has been thermalized, then the mean free path is given by
λ0=√21nσ
The scattering crosssection can be seen as the area that a particle must be inside to scatter against another particle. Its value depends upon what type of potential that is used to simulate the scattering. The simulation uses the hardsphere potential so the scattering crosssection is given by
(r )
σ = π gas+ rparticle 2 (3.21)
where rgas and rparticle are the radiuses of the gas molecule and sputtered particle respectively [4].
Using the ideal gas law
V k T
P = N B
the concentration of gas molecules can be calculated asn
n = NV =k TP
B (3.22)
where is the gas pressure, P kB is Boltzmann’s constant and is the temperature of the gas.T
−
λ = cπP(r +rk TB )
gas particle2ln R01 (3.23)
where c = √2 when the energy of the sputtered particle is low, i.e the particle is thermalized, and when the energy is high. The molecule’s energy in a gas is distributed according to the
c = 1
MaxwellBoltzmann distribution. The mean energy Eavg for a gas molecule is thus given by [8]
k T
Eavg=32 B (3.24)
From (3.24) it follows that the mean energy of a molecule in a thermalized gas with a
temperature of 300K is about 0.26 eV. Based on this the sputtered particle is said to have a high energy, thus using c = 1, if its energy is above 1 eV.
Collision with wall
Let P be the point where the prior collision occurred or, if the particle has not yet collided, the position on the target where it was sputtered. Let P’ be the position where the next collision will occur. Let Fˉ(t)= (F (t), F (t), F (t))x y z define the particle’s trajectory from P to P’. is given by theFˉ functions
(t) (P )t
Fx = Px′ + x− Px′
(t) (P )t
Fy = Py′ + y− Py′
(t) (P )t
Fz = Pz′ + z− Pz′
where is a parameter on the interval t [ 1 0, ].
The chamber is defined by a number of surfaces that are either bounded or unbounded. An unbounded surface is a surface that extends infinitely and is only defined by its equation. A bounded surface on the other hand is defined by its equation but also by at least one constraint that defines the boundary. Consider one of the surfaces that make up chamber. Let the surface be described by the equation S. If there exists a solution to the equation
(t) S
Fˉ = (3.25)
such that is on the interval t [ 10, ] it means that the particle’s trajectory intersects the surface.
Solutions that yield a value of t outside the interval [0, 1] are rejected since the intersection is outside the particle’s trajectory i.e they start before P or end after P’. Using the value of thet position of the impact is calculated. If the surface is unbounded the particle will collide
regardless of the point of intersection. If the surface on other hand is bounded it must be checked whether the point lies on inside or outside of the surface’s boundary. If the point lies within the boundary then it will collide, otherwise it will not. It is possible that the particle’s
trajectory will intersect with more than one surface that defines the chamber. If that’s the case then the solution with the lowest value of is the surface that the particle will hit since it is thet first surface that the particle will intersect with on its path from P to P’.
If the particle does collide with a surface then the particle’s incident angle is calculated. The incident angle is defined as the positive angle between the surface’s normal and the particle’sβ trajectory, i.e an angle of 0 indicates that the particle will hit the surface perpendicular to the surface. Let the surface normal be Nˉ = (N ,z Ny,Nz) and the normalized vector describing the direction of the particle by Dˉ = (D ,x Dy,Dz). Using the dot product definition [9]
|A|| ||B|| (α) Aˉ ∙ Bˉ = | cos
the incident angle can be calculated as
β = arccos
(
||N|| ||D||−N∙Dˉˉ ˉˉ)
= π − arccos(
||N|| ||D||ˉ ˉN D +N D +N Dx x y y z z
)
(3.26)
Using the fact that both and are normalized, i.e have a length of 1, equation (3.26) can beNˉ Dˉ rewritten as
β = π − arccos (N Dx x+ Ny yD + Nz zD )
which is used to calculate the incident angle. The calculated incident angle is then saved together with the particle’s energy and the point of the impact.
Scattering against a gas particle
Since the energy of the gas molecule is distributed according to the MaxwellBoltzmann distribution it is given an initial energy of
k T
E =32 B (3.27)
where kB is Boltzmann’s constant and is the temperature of the gas. The reasons to notT assume that the gas particle is at rest is that the energy of the scattered particle would then approach zero for systems where it would undergo multiple scattering collisions. This since a scattering collision will always decrease the energy of the scattered particle since some of it will be transferred to the gas atom. In practice the sputtered particle will, after undergoing multiple collisions, be thermalized and have an energy according to the MaxwellBoltzmann distribution.
Using the equation for kinetic energy and (3.27) the velocity of the gas molecule vgas is calculated as
vgas=
√
3k TmB (3.28)
The direction dˉ = (gas dgas,x, d gas,y, d gas,z) of the gas molecule is calculated using
dgax,x= sin α cos β
dgas,y= sin α sin β (3.29)
dgas,z= cos α
where is randomly picked on the uniform interval β [ 0, 2π] and is randomly picked on theα uniform interval [ 0, π]. Scattering calculation is performed in a xyplane where the gas molecule is at rest at origo prior to the collision. However since the gas molecule is moving lets introduce a new cartesian coordinate system that is fixed in the gas molecule, i.e it moves with the velocity
that is calculated as
vcoordˉ
v vcoordˉ = dˉgas gas
The velocity of the sputtered particle in the new coordinate system is calculated using (3.28) and (3.29) together with the equation
v
vpartˉ = v′partˉ − vcoordˉ = v′partˉ − dˉgas gas (3.30)
where v′partˉ is the particle's velocity in the initial frame of reference, vpartˉ is its velocity in the new coordinate system. Since there exists infinitely many planes that contains v′partˉ there exists an infinite amount of possible scattering planes. To uniquely fix the scattering plane in the initial frame of reference an azimuthal angle is randomly picked on the uniform interval ω [ 0, 2π]. It is now possible to convert the velocity of the particle, and thus its energy and direction, between the scattering plane and the initial frame of reference.
To calculate the scattering angle the sputtered particle is initially placed at ϕ ( L − , p) where p is the impact parameter and is sufficiently large so that there exists no significant interactionL between the sputtered particle and the gas molecule at this distance. The impact parameter is uniformly distributed on the interval [ 0, pmax] where pmax is the maximum distance at which there is any significant scattering. This is typically defined as the cutoff where the scattering is less than a certain threshold such as 0.01π. The reason for this definition is that there are interaction potentials that interact at an infinitely long distance and hence there is no value of
such that the scattering equals zero. An impact parameter of is thus a head on collision
pmax 0
and an impact parameter of pmax represents a collision where the particles barely touch each other. To choose an impact parameter the following equation is used
R p = pmax 01
where R01 is a random number on the interval [ 0, 1]. Lets define asA
A =mmgas
part (3.31)
where mgas is the mass of the gas molecule and mpart is the mass of the sputtered particle.
When calculating the scattering angle a centreofmass coordinate system is usually introduced to simplify the calculation. To transform the particle’s energy into this new centreofmass system the following equation is used
E
Ecom=1+A partA (3.32)
where Epart is the energy of the particle [4]. By using the equation for kinetic energy (3.32) can be rewritten as
v Ecom=m1+Agas part2
where vpart is calculated using equation (3.30). If the collision is assumed to be inelastic then the scattering angle ϕcom in the centreofmass system is given by the equation
ϕcom= π − 2∞
∫
drR 1
r2
√
1−Ecom r2V (r)−p2 (3.33)
where V(r) is the interaction potential as discussed in Section 3.4 [4]. For the hard sphere potential (3.33) reduces to
ϕcom= π − 2 arcsinpp
max (3.34)
To transform ϕcom into the scattering angle the following equation is usedϕ
ϕ = arctan
(
1+A cosA sinϕcomϕcom)
(3.35)
where it is assumed, as before, that the collision is inelastic [4]. Using (3.35), (3.34) and (3.31) can be calculated as function of the impact parameter . In Figure 7 a plot of this is shown
ϕ p
for the collision between an argon gas atom and a sulfur, a tungsten and an argon atom. The elements are chosen such that the mass of the scattered particle is greater than, less than and equal to that of the gas particle. From Figure 7 it can be seen that element that have a mass greater than that of the gas particle will always scatter less than . More generally it can beπ2 seen that the scattering angle decreases with increased mass where a heavy elements such as
atom the scattering angle ranges from to . In other words it is possible that a particle lighter0 π than the gas particle will completely backscatter which is not possible for elements heavier than the gas particle.
Figure 7; The scattering angle for sulfur (red), tungsten (blue) and argon (green) when scattered against an argon atom.
Figure 8; The energy loss for sulfur (red), tungsten (blue) and argon (green) when scattered against an argon atom.
The energy of the sputtered particle after it has been scattered is calculated as
Eafter= 1
(
−4A sin((1+A)ϕcom22 )2)
Epart
where Eafter is the energy of the particle after it’s been scattered [4]. A plot of the energy loss of a scattered particle as function of the impact parameter is found in Figure 8. The mostp important conclusion that can be drawn from this plot is that elements that are heavier than the gas atom will never lose all of it’s energy in the scattering whereas elements that are lighter can.
Also if one correlates the plots in Figure 7 and Figure 8 it can be seen that tungsten will not lose its maximal amount of energy when it scattered the most. Instead the highest scattering angle corresponds to a relatively low amount of energy being lost.
3.3 Aggregating the results
The result of the simulation so far is a list of impact points along with the incident angle and the particle’s energy at the time of impact. To be able to interpret this as a meaningful result the list must be
aggregated. First each surface that makes up a part of the chamber is assigned at least one unique mesh of userdefined size. For areas of interest, such as the substrate, the mesh can be very fine and for areas that are not of great interest it can be more sparse.
An example of a chamber with the mesh defined is found in Figure 9 where the red part indicates the target and each gray area is a mesh element. Once the mesh is defined the list of particles is iterated and each particle is assigned to the mesh that it contains its impact point.
Figure 9; A sputtering chamber with a
userdefined mesh (different gray areas) and the target (red).
For each mesh it is now possible to calculate a distribution of the energy and incident angle that the incoming particle will have. The number of particles that must be simulated in order to achieve a statistically significant distribution is heavily dependent upon the mesh resolution but usually at least 106 particles are needed. The number of particles that end up on the target is counted in order to obtain the redeposition factor , i.e the fraction of the particles that aref sputtered but ends up at the target, which is calculated as
f = N
total
Ntarget
(3.36)
where Ntotal is the number of particles simulated and Ntarget the number of particles that are deposited at the target. The fraction of the sputtered particles that are deposited on mesh gi i is given by
gi= (N N−Ni )
total target (3.37)
where Ni is the number of particles that are deposited on mesh . Let the flux be defined asi F the number of sputtered particles per unit area and time. Using (3.37) the flux Fi for mesh i can be calculated as
R Fi= giAAi
total
where is the number of sputtered particles per unit time, is the area of segment andR Ai i is the total area of all mesh segments excluding the target.
Atotal
3.4 Potentials
The simulated scattering of a particle in the gas phase will depend on which interatomic potential V(r) that is used. There are a number of difference potentials to choose from but in this paper the following has been given some extra attention; the hardsphere potential, the screened Coulomb potential and the LennardJones potential. The potential chosen in the simulation model is the hardsphere potential. Since the scattering involves two species where at least one of them is a neutral the hardsphere potential should be good enough
approximation. Using one of the other potentials would greatly decrease the performance of the simulation since it is computationally much easier to implement and evaluate the hardsphere potential than the other potentials.
Hard-sphere potential
The hardsphere potential models each particle as a hard sphere. The potential is defined as
(r) r r
V = 0 > gas+ rpart (r) r r
V = ∞ < gas+ rpart
where rgas is the radius of the gas particle and rpart is the radius of the sputtered particle [4].
There exists multiple different definitions of the radius of a particle such as the van der Waal radius and the covalent radius but they are all in the order of 0.1 nm and increases with the number of nucleons in the atom. Using the hardsphere potential it is possible to solve the scattering equation (3.33) analytically which simplifies the computational implementation.
Another unique property of the hardsphere potential is that the scattering angle will be independent upon the energy of the particle.
Screened Coulomb potential
The nonscreened Coulomb potential is defined as
(r)
V = Z Zgas partr q2
where Zgas is the number of protons in the gas particle, Zpart is the number of protons in the sputtered particle and is the elementary charge [4]. However this yields a too large interactionq at large interatom distances due to neglecting the effect of electrons. To cope with this the potential is usually screened by a screening function Φ( )ra where is the screening length [4].a There exists a wide variety of screening functions but one of the more commonly used ones is the Molière screening function defined as
( ) .35e .55e .1e
Φ ra = 0 −0.3ra+ 0 −1.2ra+ 0 −6ra
where the screening length is defined asa
.885a (Z )
a = 0 0 gas0.5+ Zpart0.5 −32
with a0 being the Bohr radius [10]. In contrast to the hardsphere potential the scattering angle depends on the energy of the particle when the screened Coulomb potential is used. It is not
ϕ
possible to analytically solve the scattering equation (3.33) for this potential but there exists a number of numerical methods that can be used [4].
Figure 10; The scattering angle for a tungsten atom scattered against an argon atom with aϕ screened Coulomb potential used as the interaction potential.
In Figure 10 the scattering angle as function of the particle’s energy and the impactϕ
parameter is plotted for a collision between a tungsten atom and an argon atom. Figure 11p shows the same plot but for a collision between a sulfur atom and an argon atom. Note that the maximal amount of scattering is much less for the tungsten collision and that a headon collision (corresponding to a very small value of the impact parameter) will backscatter the sulfur but not the tungsten.
Figure 11; The scattering angle for a tungsten atom scattered against an argon atom with aϕ screened Coulomb potential used as the interaction potential.
Lennard-Jones potential
The LennardJones potential, also known as the 612 potential, is defined as
(r) ε
V = 4
( (
rσ)
12−(
rσ)
6)
where is the depth of the attractive well and is the distance at which the repulsive andε σ attractive parts are equal, i.e the potential is zero [11]. The constants and are dependentε σ upon the material and have been determined experimentally. The potential is loosely based on quantum mechanics perturbation theory where the positive term represents the strong repulsive force and the negative term the weak attractive force [11]. In contrast with the hardsphere potential and the screened Coulomb potential the LennardJones potential can be either repulsive or attractive depending on the scattering. This manifest itself in that the scattering angle may be negative for certain combinations of the impact parameter and the particle’sϕ energy. However since the scattering plane, as described in the previous section, is symmetrical and fixed in the initial frame of reference by a azimuthal angle the negative scattering angle isω equivalent to a positive scattering angle of the same size in the scattering plane with azimuthal angle ω + π. A plot of the scattering angle is found in Figure 12 where it is clearly visible that the scattering angle may be negative. If the LennardJones potential, as seen in Figure 12 is
compared to the screened Coulomb potential, as seen in Figure 11, it can be seen that they resemble each other but there are some differences. The scattering angle approaches faster0
for the LennardJones potential when the impact parameter decreases. It can also be seen that the impact parameter must be slightly larger to achieve the same amount of scattering when the LennardJones potential is used.
Figure 12; The scattering angle for a tungsten atom scattered against an argon atom withϕ the LennardJones potential used as interaction potential.
3.5 Simulations and analysis
A series of simulations have been performed to investigate how the distribution of the particles across the chamber is affected by different process parameters. The chamber used is cylindrical in shape with a diameter of 0.32 m and a height of 0.25 m. The substrate is 11.6 cm in diameter and is placed on a substrate holder at the bottom of the chamber. The target is specified to have a diameter of 4 cm and is placed at the center in the top of the chamber. It consist of WS2 with the radius of tungsten set to 193 pm and the radius of sulfur to 88 pm [12]. The masses used in the simulation are 183 u for tungsten and 32 u for sulfur. The target potential is 400 V, the sputtering gas is argon and at least 106 particles have been simulated in each simulation.
Difference depending on sputtered element
Different elements will scatter differently due to the difference in mass as discussed in Section 3.3. The flux of sputtered tungsten and sulfur is shown In Figure 13 and Figure 14 respectively.
The gas pressure is 4 mTorr in both simulations. It can be seen that the flux of tungsten at the substrate is slightly higher than the flux of sulfur. At the top of the chamber however the flux of sulfur is much larger than the flux of tungsten. This since the mass of the tungsten will make it