In-situ studies of silicide formation during growth of molybdenum-silicon interfaces Free

Many current day nanoscale devices for electronic, mechanic, and optical applications are based on nano-engineered structures of many layers with thicknesses in the order of a few nanometers or less. Interaction between these layers at the interfaces, including intermixing and formation of interface compounds, is generally detrimental to device performance. In particular, the growth stress at interfaces and related relaxation mechanisms play an important role in device performance as they determine mechanical properties such as the tensile strength and lifetime (layer delamination).^1,2

In order to understand the complex relationship between layer growth and growth stress, precise metrology is required that is able to properly characterize and understand the layer growth processes, with emphasis on the processes that take place at the interfaces between thin films. In particular, in situ metrology tools enable the analysis of complex systems as they evolve during deposition, rather than analyzing the resulting complex structure ex situ, where the individual effects of many synthesis steps may not be distinguishable anymore.

In this work, we use in situ stress measurements to measure the intrinsic stress during layer deposition of nanometer thick bilayer systems. As an example system, Mo/Si was chosen. Mo/Si is an extensively studied material combination relevant for various optical and mechanical applications, and as such it is well known that Mo/Si interfaces play an important role,^3–5 as well as the interface formation.⁶ Mo is also investigated due to its phase transition being accessible at room temperature⁷ and for its low mobility,⁸ which affects its growth mode and interface formation.⁹ $MoxSiy$ compounds are also of interest, as the interface reactivity¹⁰ and nucleation conditions¹¹ can be modified by this.

By using in vacuo Low Energy Ion Scattering (LEIS), we determine the composition of the outermost surface and correlate the results to the developing growth stresses, showing that the growth stresses that develop during deposition are intrinsically linked to the layered structure and, in particular, the formation of interfaces during growth.

All results presented here are obtained from thin films prepared in an UHV sputter deposition setup with a $10−8mbar$ base pressure, using DC magnetron sputtering with a $10−3mbar$ argon working pressure. Movable magnetron targets of 100 mm diameter are placed underneath the substrate for normal incidence deposition. The target to substrate distance is 300 mm, the substrate is grounded and stationary. Samples were deposited onto $⟨100⟩$ Si wafer cantilever substrates of $80×10mm$⁠, with deposition rates calibrated by ex situ x-ray reflectivity measurements on reference samples. A Mo deposition rate of 0.15 nm/s and a Si deposition rate of 0.27 nm/s were used. The magnetron voltages and powers were 340 V and 340 W for Mo and 470 V and 670 W for Si. All depositions were done at room temperature, the temperature rise of the cantilever during a separate deposition was measured by a thermocouple to be approximately 10$°$⁠.

Optical measurement techniques are regularly used to measure stress during deposition.^12–14 The in situ stress measurement is done by continuously measuring the curvature of a cantilever sample during the deposition process of the layers on the cantilever surface. The substrate curvature is measured using a dual laser beam deflectometer, where the beam position on the camera depends on the curvature of the cantilever, which is influenced by the added stress during growth. As the typical deposition speed is around 0.2 nm/s and the acquisition frequency is 10 Hz, a subangstrom thickness resolution is achieved. This is small enough to distinguish effects within a fraction of a monolayer of deposited material. Due to the low noise level of 0.012 N/m, these small features are still distinguishable.

Via Stoney’s equation as shown in (1), the cantilever curvature is related to the force per unit width

The cantilever used is cut from a silicon wafer with a thickness $hs$ of $150μm$⁠, this small thickness provides a high sensitivity with low noise. The Young’s modulus is denoted by $Es$ and the Poisson ratio by $ν$⁠. The stress $σ$ is related to the measured cantilever curvature $κ$ via the coating thickness $h$⁠. Processes such as compound formation and an amorphous-to-polycrystalline transition also induce a stress; however, as the thickness of the material involved in these processes can be difficult to determine, the force per unit width is used, defined as $σh$⁠. This quantity is directly proportional to the measured cantilever curvature.

The derivative of the force per unit width curve is often assigned to the incremental growth stress of the newly added material, expressed in units of Pascals. It should, however, be noted that this is strictly only possible in the absence of volumetric effects such as the occurrence of a phase transformation of the substrate layer during growth. In this work, the common procedure of taking the derivative of the force per unit width is applied, while specifically addressing volumetric changes where they occur. Intermittent starting and stopping of the deposition showed no relaxation effects.

The in situ stress measurement principle used is similar to Ref. 15. A fiber coupled laser with a 635 nm wavelength is used for easy visible alignment and a high beam quality, convenient in analyzing the camera images. The power per beam is below 1 mW, varying the power was found not to have an effect on the measurement. A beam splitter is used to generate two beams and combine them to capture their position on a single camera. The two laser beams are spaced 50 mm apart on the cantilever and reflect off the cantilever surface. The cantilever is clamped on only one side to avoid forces on the cantilever that would influence its curvature.

By taking the difference in deflection of the two beams, the measurement is sensitive to the change in curvature of the cantilever in-between the two beams, eliminating contributions from, e.g., clamping effects. The path length from the laser to the cantilever is approximately 1 m, after which the beams are reflected back close to the cantilever, reflecting off the cantilever a second time to double the sensitivity. The beam then follows the same 1 m long path back until the beamsplitter reflects them to the camera. The stress measured using the in situ deflectometer setup was calibrated ex situ by measuring the cantilever curvature before and after deposition using a white light interferometer.

LEIS measurements were performed in a $Qtac100$ instrument manufactured by IONTOF, in an UHV chamber with base pressure around $1×10−10mbar$⁠. The incident ion beam is 3 keV $He+$ with normal incidence on the sample surface. A double toroidal energy analyzer accepts ions with a scattering angle of $145°$⁠, using an azimuthal acceptance angle of $360°$ for an efficient collection. The ion energy after scattering from a given surface atom depends on the mass ratio of the ion and the surface atom. The high surface sensitivity of LEIS originates in the very high neutralization probability of scattered ions as the ion-surface interaction is much longer for ions scattered from buried atomic layers, and therefore, the neutralization probability for those ions is close to 1.^16–19 Projectiles scattered from buried layers lose additional energy proportional to the travel length and can be detected if they re-ionize upon escaping the sample surface. The signal from these ions is, therefore, separated in both intensity and energy from the signal from surface ions.¹⁶ 

The surface composition was quantified according to the method described in Ref. 16. First, integral peak areas were obtained by fitting a Gaussian to Si or O peaks. Only the high-energy side of the Mo peak was used for fitting due to interference of the intense tail at the low-energy side. Then, to calculate surface coverages of Mo and Si, integrated intensities $Si$ of their LEIS surface peaks for each spectrum were divided by intensities of reference samples, for which pure sputter-cleaned Mo and Si surfaces were chosen (⁠$SMoref=17280counts/nC$ and $SSiref=3740counts/nC$⁠). Next, surface atomic densities (SADs) $Ni$ of Mo and Si were calculated as

The surface atomic densities of the reference samples $Niref$ were calculated from the bulk mass densities $ρi$ in the following way:¹⁶ 

where $NAv$ is Avogadro’s number and $Mi$ is the molar mass. We obtain $NMoref=1.55×10−19$ and $NSiref=1.31×10−19atoms/m2$⁠. The accuracy of the values of SADs obtained this way depends on the validity of Eq. (3) for each reference sample, which means that values of SADs contain an unknown scaling factor, constant for each element. In this paper, we, therefore, only compare relative changes of SADs. The values of $SOref$ and $NOref$ for trace amounts of O contamination were taken from Ref. 20.

The SADs used in this work are included in Table I.

The LEIS measurements were performed for incremental thicknesses of top layers, which allows one to combine the results in a so-called deposition depth profile (in contrast to a sputter depth profile). To reduce the effect of surface contamination during LEIS analysis, for each film thickness, a dedicated sample was made in a separate magnetron sputter deposition chamber and transferred in-vacuum to the LEIS setup within 10–15 min after deposition. This procedure allowed one to keep oxygen content on the surface lower than 5% for most samples, with an exception of Mo films without Si, for which atomic fraction of O reaches 10%.

As LEIS measurements are not sensitive to the phase of the material (amorphous or polycrystalline), it cannot be concluded from them what the phase of the layers, interfaces, or compounds formed is. Only for Mo layers, the amorphous-to-polycrystalline phase transition was observed, using XRD.

Raw LEIS spectra for Mo on Si and Si on Mo growth are shown in Figs. 1(a) and 1(b). Figure 2 shows the stress that is measured during the growth of Si onto Mo, as measured in Newtons per meter (a) and the derivative in gigapascals (b). Analyzed LEIS data during Si on Mo growth are shown in the bottom graph (c).

In general, a predominantly strong compressive stress is observed during Si layer growth, consistent with earlier work.^21,22 The in situ metrology reveals several additional details as seen in Fig. 2(b). Several features can be observed in the stress development. A small tensile peak of 0.1 N/m is visible within the first 0.15 nm, indicated by the first vertical dotted line in Fig. 2. A compressive stress of $−5GPa$ is observed after the initial 0.15 nm. The compressive stress reduces with increasing Si growth, resulting in a tensile stress around 1.3 nm of deposited Si, indicated by the second vertical dotted line in Fig. 2(b). At 1.6 nm [indicated by the third vertical line in Fig. 2(b)], the stress has settled to $−1.1GPa$ (indicated by the lower horizontal line) and remains constant for the rest of the layer thickness.

The LEIS results in Fig. 2 show a monotonic increase of Si coverage and monotonic decrease of Mo coverage. The total SAD monotonically decreases from the bulk Mo value to the bulk Si value, which it reaches after 1.3 nm of Si deposited.

The XY plot of the Mo SAD and Si SAD in Fig. 3(a) shows how far the SADs deviate from a theoretical simple mixture of the two. The blue line indicates the convex combination (CC) of the Mo and Si SAD corresponding to the case of a simple mixed coverage. The red line is the experimental data and shows the combined Mo and Si SAD progressing from bulk Mo to bulk Si.

From the start of the Si deposition up to 0.5 nm of Si deposition, the combined SAD is increasing compared to the CC. After 0.5 nm, the combined SAD decreases and converges to the bulk Si value, reaching it at 1.3 nm Si deposited.

Figure 4 shows the stress that is measured during the growth of Mo onto Si, as measured in Newtons per meter (a) and the derivative in gigapascals (b). The LEIS data during Mo on Si growth are shown in the bottom graph (c).

The initial stress is strongly tensile at 5 GPa. The tensile stress levels off with the Mo growth toward a compressive growth, reaching 0 GPa around 1.8 nm. At 2.1 mn, a sharp tensile step of 0.7 N/m occurs, after which the growth is compressive at approximately 1 GPa.

The LEIS measurement shows the SAD of Si and Mo change slower than for the Si on Mo case. The Si SAD only reaches 0 after 3 nm of Mo deposited. The total SAD does not increase monotonically from the bulk Si value to the bulk Mo value.

Similar to Si on Mo, the combined SAD does not follow the CC from pure Mo to pure Si as can be observed in Fig. 3(b). For up to 2 nm of Mo deposited, the combined Mo and Si SAD is increasingly higher (indicated in red) than the CC (indicated in blue) of Mo and Si.

During magnetron sputter deposition, adatom energies are typically of the order of a few electron volts to a few tens of electron volts, sufficient to break bonds at the substrate surface and enabling the formation of Si–Mo bonds during interface growth. When the substrate is near room temperature, the low mobility per adatom results in a stochastic growth mode, leading to amorphous and/or polycrystalline film growth. The adatom energies during sputtering are sufficiently high to densify the growing layer and create a smooth adlayer with a low roughness. The roughnesses of Mo/Si systems in the deposition setup used are typically in the order of 0.2 nm RMS, determined by the AFM measurement. Finally, the large negative enthalpy of mixing of Mo and Si²³ shows that island formation due to segregation should not occur. This excludes the large scale island formation typically seen in Volmer Weber growth due to the low roughness observed. The low mobility and amorphous growth also excludes the Stranski-Krastanov growth.

Surface morphology and roughness affects the LEIS signal as well, reducing it by a certain roughness factor $R$ due to blocking and shadowing of ions by features located higher than them.¹⁶ This effect is nonconventional—the roughness factor does not depend on the absolute height of features, but instead on their angle, which means that even atomic scale disorder can play a role.¹⁸ Since exact atomic scale arrangement of atoms for our surfaces is unknown, the value of roughness factor also remains unknown but lower than unity (we can expect it to be around 0.8 for polycrystalline metal¹⁸). Without additional information, we have to assume it to be constant.

Even if the assumption is wrong, one can argue that the increase of the sum of the SAD of Mo and Si, as observed in Fig. 3, cannot be explained by a roughness factor. For example, in intermediate stages of growth of Mo on crystalline Si substrates roughness increases,²⁴ and the appearance of extra step edges should lead to a decrease of the LEIS signal. However, the opposite is observed in Fig. 3. As such, we explain this effect by compound formation, and potential changes in roughness would only mean the effect of compound formation is even stronger.

The initial tensile peak, indicated in Fig. 2(a) by the vertical line at 0.15 nm, corresponds to a submonolayer coverage, indicating that the tensile peak is due to a quickly saturating effect. According to the LEIS measurements, the SAD’s change only slowly, i.e., on a nanometer scale, suggesting strong intermixing. The submonolayer equivalent tensile stress is, therefore, not related to the nanometer scale intermixing. Surface stress effects due to island formation, such as treated in Ref. 25, do not occur due to the stochastic growth mode. In addition, any significant island formation would result in a substantial change in surface coverage, which is not observed in the submonolayer regime. The first submonolayer of Si atoms may actually diffuse into the Mo grain boundaries and may introduce a tensile stress as the Si atoms fill voids that are too small for Mo atoms and act as a cohesive force between grains.

The compressive stress following after the tensile peak is attributed to compound formation of the arriving Si with the Mo layer. The more mobile Si atoms diffuse into and expand the Mo layer when creating a compound, creating a compressive stress. This process depends on the accessibility of Mo, therefore, the compressive stress due to compound formation and the availability of Mo on the surface are proportional. Both the compressive stress and the Mo SAD decrease as visible in Figs. 2(b) and 2(c), where after 1.3 nm of deposited Si, the Mo SAD reaches zero and the Si incremental growth stress reaches a maximum.

Figure 3(a) shows the Mo SAD plotted as a function of the Si SAD. The covering of Mo with Si without any interaction of the two would be a convex combination of the Mo and Si SAD’s as indicated by the blue line. However, the combined SAD of Mo and Si is higher, up to a 20% increase at 0.5 nm. This is a clear indication of compound formation, where for all known molybdenum silicides, the compound density is much higher than a convex combination of the element densities. This compound formation apparently takes place, in particular, in the first 0.5 nm Si deposited. After 0.5 nm, there is a trend toward Si SAD, indicating reduced or no compound formation and additional elemental Si slowly covering the interface compound, until around 1.5 nm where no Mo is present at the surface and bulk Si growth is starting.

The compounds formed during Si on Mo growth depend mainly on the availability of Mo and Si.²⁶ According to the SAD’s as measured by LEIS, the effective stoichiometry at the surface in the first 0.5 nm, where an abundance of Mo is present, can be expected to be close to $Mo3Si2$ and/or $Mo5Si3$⁠. $Mo3Si2$ has not been reported in the literature for Mo/Si systems, so this compound is not considered. For surface coverages above 0.5 nm, much more Si rich stoichiometries are observed at the surface, suggesting either coexistence of Si and $Mo5Si3$ or the formation of $MoSi2$⁠. Since $MoSi2$ has an even higher SAD than $Mo5Si3$⁠, any significant formation of $MoSi2$ would have shown a “further” increase in the combined SAD past 0.5 nm. Since the interface formation occurs mainly in the first 0.5 nm, the abundance of Mo most likely gives rise to the $Mo5Si3$ compound formation as reported by Zoethout et al.⁶ Estimating the interface width from the LEIS measurement using the method also used by Coloma Ribera et al.²⁰ results in a $σ$ of 0.4 nm, similar to what is found in the literature.²⁷ 

Both the $Mo5Si3$ formation at the start of the Si deposition and the bulk Si growth after about 1.5 nm show compressive growth stresses. However, in-between these two growth regimes, around 1.3 nm of deposited Si (indicated in Fig. 2 by a vertical line), a tensile stress occurs. The SAD’s change monotonically, and therefore, the compressive-tensile-compressive behavior cannot be explained by processes only depending on the availability of the material at the surface, as this would result in a stress development that would be proportional to the SAD’s.

Instead, the compressive-tensile-compressive growth indicates that the transition from $Mo5Si3$ formation to bulk Si growth is not gradual but shows a tensile interface contribution from the $Mo5Si3/Si$ interface. As the Mo SAD decreases, much of the surface consists of $Mo5Si3$ and the Mo layer is no longer accessible, stopping the $Mo5Si3$ formation. The arriving Si atoms now intermix with $Mo5Si3$ with which the Si cannot form a covalent bond. This causes many dangling bonds in the Si due to this interface, resulting in a tensile stress. This is similar to porous Si, where the atoms sit further apart and, therefore, have more dangling bonds, which also results in a tensile stress. This tensile component is, therefore, proportional to the $Mo5Si3$ present on the surface which has a maximum after the $Mo5Si3$ interlayer has formed but before the bulk Si layer forms on top of it.

In other work, a subsurface interface formation around 1 nm of deposited Si in a Si on Mo system is suggested,⁶ which could also explain a tensile stress due to compaction of the subsurface layer. Such a process would not be visible in LEIS measurements due to the high surface sensitivity of LEIS. However, in this study, the subsurface interface formation is not considered as a cause for tensile stress due to the fact that a similar broad tensile peak was observed when depositing Si onto a $SiO2$ layer, suggesting that the tensile peak is not due to subsurface interface formation effects but due to formation of a Si/inert surface interface, e.g., Si on $Mo5Si3$ or Si on $SiO2$⁠.

After 1.3 nm of Si deposition, the LEIS data show a pure Si surface. At 1.6 nm (indicated in Fig. 2 by a vertical line), the Mo presence is no longer influencing the stress of the newly added material at the surface, clearly demonstrating that interface effects were the cause of the nonlinear stress behavior observed in the Si on Mo growth. The constant stress of $−1.1GPa$ from this point onward (indicated in Fig. 2 by a horizontal line) indicates that the bulk growth starts at this point. This compressive stress depends on the specific growth conditions, where in the case of magnetron sputtering adatoms arriving with significant energy densify the layer by a peening effect^28,29 and create compressive stress.

The initial stress is strongly tensile at 5 GPa, which is consistent with what is reported elsewhere in the literature.^8,10,21 The interface formed is reported to be $MoSi2$⁠,^6,21 agreeing with the abundance of Si on the surface favoring $MoSi2$⁠.²⁶ This initial tensile stress decreases strongly when the underlying Si layer is prevented to form $MoSi2$⁠, for example, due to the presence of a several angstrom interlayer or passivation (both with no significant contribution to the stress themselves), indicating that the tensile stress is mainly due to compound formation and not due to surface stress.

In contrast to $Mo5Si3$ formation for the case of Si on Mo, where Si atoms moving into the existing surface create a compressive stress, the addition of Mo on a surface allows the Si atoms to move out of the surface to form a compound.²⁴ The voids left by Si when moving to the Mo on top to form a compound introduce tensile stress.

The tensile stress reduces with Mo growth, effectively being limited to the availability of free Si at the surface to form $MoSi2$⁠, consistent with the changes in SAD’s as observed in the LEIS data. Note that the intermixing range is much larger for Mo on Si as compared to Si on Mo, which in the literature is often attributed to difference in crystallinity,⁹ cohesion³⁰ and adatom-substrate interaction strength.³¹ Estimating the interface width from the LEIS measurement using the method also used by Coloma Ribera et al.²⁰ results in a $σ$ of 1.0 nm, similar to what is found in the literature.²⁷ 

Figure 3(b) shows the Mo SAD plotted as a function of the Si SAD. The combined SAD is higher than what would be expected by the simple covering of the Si layer by a Mo layer (represented by the blue line). This indicates compound formation occurs up to 2 nm, which coincides with the amorphous-to-polycrystalline transition observed in the in situ stress measurement. After 2 nm, the combined SAD reduces, indicating the covering of the $MoxSiy$ compound with Mo, and reaching the value of a convex combination of Mo and Si at 2.7 nm. From 2.7 nm, the combined SAD remains a linear combination of Mo and Si, indicating no compound present at the surface, but there is still Si present at the surface according to LEIS. This may be due to Si which segregates onto the surface without actually forming a compound.

As the Si availability drops, as shown in the LEIS measurements, the stress slowly reduces to a stress free growth. Continued growth would actually become compressive as shown by Fillon et al.¹⁰ if crystallization would be prohibited. However, at 2.1 nm a sharp tensile step of 0.7 N/m is visible. XRD measurements done confirmed that this is the phase transformation of the Mo layer from amorphous to polycrystalline, in line with what is reported in the literature.^7,10 The phase transformation compacts the Mo layer as the crystallites have a denser packing of the atoms. This volume decrease can relax in the out of plane direction during growth, but still induces a strong tensile stress in the in-plane direction, causing the tensile step.

After the crystallization, the stress is expected to be mainly determined by the growth of the crystallites and the grain boundaries. The observed compressive stress may be due to overfilling of the grain boundaries. For increasing thicknesses, the roughness increases and the competition between the growing grains can lead to voids, which would introduce a tensile contribution. The evolution of the grains in size and grain boundary density depends strongly on the initial polycrystalline texture, which is also influenced by impurities.

The growth of Si on Mo and Mo on Si has been investigated with in situ stress and LEIS. The Si growth shows a compressive-tensile-compressive behavior, where added compressive stress from the initial, presumed $Mo5Si3$ interlayer formation is proportional with the Mo availability on the surface. This compressive stress is followed by a tensile $Mo5Si3/Si$ interface that has many Si dangling bonds due to the inert $Mo5Si3$ and this tensile stress component reduces with $Mo5Si3$ availability on the surface. Finally, Si bulk growth starts at 1.5 nm deposited Si and shows a compressive stress of 1.1 GPa due to peening.

For Mo, a $MoSi2$ interface is formed that shows strong tensile stress of 5 GPa, due to defects created in the Si surface upon $MoSi2$ formation. This tensile stress slowly reduces due to reduced availability of Si. A tensile stress increase of 0.7 N/m is observed at 2.1 nm, exactly at the amorphous-to-polycrystalline phase transition.

For both Si on Mo and Mo on Si, the LEIS data clearly show that the combined SAD’s during interface growth are much higher than a convex combination of the Mo and Si SAD’s, clear evidence of high density compound formation. The exact point where compound formation stops and additional deposited material remains elemental is clearly observed.

We acknowledge the support of the Industrial Focus Group XUV Optics at the MESA+ Institute for Nanotechnology at the University of Twente, notably the industrial partners ASML, Carl Zeiss SMT, and Malvern Panalytical, as well as the Province of Overijssel and the Foundation FOM (now part of the NWO, the Netherlands Organization for Scientific Research).