Enhanced desorption of surface methyl groups in time-resolved chemical vapour deposition of aluminium nitride from trimethylaluminum

Aluminium nitride (AlN) is a semiconductor with a wide range of applications from light emitting diodes to high frequency transistors. Electronic grade AlN is routinely deposited at 1000 °C by chemical vapour deposition (CVD) using trimethylaluminium (TMA) and NH 3 while low temperature CVD routes to high quality AlN are scares and suffer from high levels of carbon impurities in the film. We report on an ALD-like CVD approach with time-resolved precursor supply where thermally induced desorption of methyl groups from the AlN surface is enhanced by the addition of an extra pulse, H 2 , N 2 or Ar between the TMA and NH 3 pulses. The enhanced desorption allowed deposition of AlN films with carbon content of 1 at. % at 480  C. Mass spectrometry combined with kinetic- and quantum chemical modelling show that the extra pulse between TMA and NH 3 enhances the desorption and prevents re-adsorption of the methyl groups, terminating the AlN surface after the TMA pulse. The surface methyl groups are found to desorb as CH 3 , CH 4 and C 2 H x A 1550 (SEM) an acceleration of 10-20 to study the morphology of the Kratos AXIS Ultra DLD X-ray photoelectron spectroscopy (XPS) equipped Ar analyse (5% error and in the The composition of the films was obtained after with a sputtering power of 0.3 for 600 CasaXPS software was used the data. Gaussian-Laurentius functions and Shirley background were used to fit the experimental XPS data.

high temperatures. ALD of AlN have previously been reported using TMA with NH3 both via thermal4,5,6 , NH3 plasma7, 8,9,10 and N2 plasma11 routes. Plasma processes can lead to crystalline and conformal AlN films at temperatures <300 °C.9,10 To obtain a crystalline AlN film via the thermal route, a temperature >375 °C is needed. 5,6 The TMA molecule is reported to decompose above 330 °C12, with the breaking of one of the Al-C bonds, forming dimethyl aluminium (DMA) and a methyl group at temperatures 500 °C.13 Thermal ALD of AlN typically contain 5-10 at.% C depending on the deposition temperature,4,6 making thermal ALD routes not ideal for depositing electronic grade AlN.
Plasma ALD is also associated with high levels of carbon impurities as atomic hydrogen, produced in the plasma discharge, can induce a chemistry trapping carbon impurities in the film by abstracting H2 from surface methyl groups.14 Deposition of crystalline AlN with low levels of carbon is thus dependent on an efficient carbon cleaning surface chemistry.
Herein, we report a low temperature, ALD-like, CVD approach depositing AlN using TMA and NH3 delivered in separate pulses at 480 °C. We show that adding an extra gas pulse, with the gas flow perpendicular to the substrate surface between the TMA and NH3 pulses, leads to a drastic decrease in C content and increases the crystalline quality of the films. Mass spectrometry, quantum-chemical density functional theory and kinetic modelling show that the lower carbon content in the films is attributed to enhanced desorption which prevents readsorption of methyl groups to the AlN surface.

Film deposition
The AlN films were deposited in a Picosun R-200 ALD system with a base pressure of 400 Pa and continuous N2 (99.999 %, further purified with a getter filter to remove moisture) flow through the deposition chamber. The chamber walls and the substrate holder were heated using separate heating systems. Si (100) wafers without further cleaning were cut into 15x15 mm2 pieces and used as substrates. The pieces were loaded into the deposition chamber without using a load-lock. Commercially available TMA (Pegasus Ltd, Alpha grade) in a stainless-steel bubbler was used at room temperature with 100 sccm N2 as carrier gas. NH3 (AGA/Linde, 99.999 %) was used as the nitrogen source in the process. The TMA pulse time was set at 0.1s with 6s purge and the NH3 pulse time was set to 12s with 6s purge if nothing else is stated. H2 (99.999 %, further purified with a getter filter to remove moisture), N2 or Ar (99.999 %, further purified with a getter filter to remove moisture) was used as a cleaning pulse between the TMA and NH3 pulse. It should be noted that the flow of these cleaning pulses (150 sccm) was perpendicular to the substrate surface compared to horizontal for the purge pulses. The difference between the cleaning pulse and the purge, beside the flow direction, is also the flow rate were the purge gas (N2) had a total flow of 500 sccm (continuous flow into the reaction chamber) distributed over the six gas lines, were all the six gas lines had the same flow expect one which is the main N2 line into the chamber with a flow of 200 sccm. While the cleaning pulse had a flow of 150 sccm independent of which gas was used.
A second, homebuilt ALD reactor with an integrated quadrupole mass spectrometer, QMS, (MKS, Spectra Microvision Plus) was used to analyse the composition of the gas phase from the studied surface reactions. The small ALD chamber consists of 10 75x25 mm2 quartz slides, stacked in a sandwich structure, allowing the precursors to pass between the substrates and thereby increasing the total surface area for the chemical reactions. The QMS cell is differentially pumped and samples the gas phase through an orifice placed a few cm downstream of the quartz slides. Glass slides coated with AlN by the ALD process described above were used as substrates for this study. The ALD chamber was heated to 350 °C and the total pressure was set to 1 Torr by using 10 sccm Ar and a regulating valve. The corresponding pressure in the QMS cell was measured to be 5x10-5 Torr. For the experiment, TMA (SAFC, electronic grade) was introduced into the reaction chamber to saturate all surfaces, where after it was turned off and changes in the gas composition was monitored by the mass spectrometer (MS).

Characterisation
The crystallinity of deposited films was studied using a PANalytical EMPYREAN MRD XRD with a Cu-anode x-ray tube and 5-axis (x-y-z-v-u) sample stage in grazing incidence (GIXRD) configuration with a 0.5o incident angle. PANalytical X'Pert PRO with a Cu-anode tube and Bragg-Brentano HD optics was used for X-ray Reflectivity (XRR) mode to measure the thickness of the films. From the XRR measurements the software PANalytical X'Pert reflectivity was used to fit the data using a two-layer model, AlN/substrate. A LEO 1550 Scanning electron microscopy (SEM) with an acceleration energy of 10-20 kV was used to study the morphology of the films. Kratos AXIS Ultra DLD X-ray photoelectron spectroscopy (XPS) equipped with Ar sputtering was used to analyse the composition (5% error margin) and chemical environments in the films. The composition of the films was obtained after clean sputtering the surface with a sputtering power of 0.3 eV for 600 s. CasaXPS software was used to evaluate the data. Gaussian-Laurentius functions and Shirley background were used to fit the experimental XPS data. were geometry optimized using a 331 k-point grid. Free energies were calculated from the computed vibrational mode frequencies of the optimized structures23 (Table S1).
The kinetics of the surface adsorption and desorption processes were simulated based on the quantum-chemical results. The adsorption rate was approximated from the impingement rate24,25 as no tight transition state is present along the internal reaction coordinate. The reverse reaction rates were derived based on the forward reaction rates to ensure that the thermodynamic equilibria can be reached as enough time has passed. The DFT-based surface reaction rates together with the gas phase reaction mechanism and rate for C2H6 formation from CH3 from literature26,27,28,29,30 were employed in the kinetic model. The simulation of the kinetics was performed using the MATLAB SimBiology module31,32. The simulations of the amount of surface species as a function of time start from the surface fully covered by methyl groups (4CH3(s)) at the pressure 100 pascal and the temperature 773 Kelvin. The pressure was held constant by the presence of non-interacting gas molecules (e.g. N2) plus molecules formed after desorption. The number of gas molecules was assumed to be much larger than the number of surface sites, i.e. the initial molar ratio of gas molecules/CH3(s) was set to 50. Two kinetic models are considered here. In the first model, the methyl radicals produced from the reactions are assumed to be completely purged away from the system at all times similar to when H2, N2 and Ar is introduced between the TMA and NH3 pulse in the ALD cycle, while in the second model no extra pulse is used, i.e. no extra pulse between TMA and NH3, is assumed.

Results and discussion
In initial experiments, H2 pulses of different lengths were added between the TMA and NH3 pulses. Without addition of H2 pulse, the C content in an AlN film deposited at 480 °C was measured to be 3.1 at. % from XPS analysis. Addition of an H2 pulse decreased the C content, Figure 1a. A short H2 pulse of 3.3s decreases the C content to around 1 at. %, which is the detection limit of carbon by XPS. It should be noted that simply doubling the purge length from 6s to 12s had no measurable effect on the carbon content in the film. The carbon content is at this low level also for longer durations of the H2 pulse. The same trend is also observed for a deposition temperature of 400 C (Fig. 1a). For depositions made at lower temperature, 400 °C, the carbon content is slightly lower without the H2 pulse (2.6 at %) as compared to 480 °C and the amount of carbon decreases to about 1.5 at. % with the addition of a H2 pulse. The growth per cycle of the AlN films is stable upon increasing the length of the H2 pulse, Figure 1b  GIXRD measurement shows that the films consist of polycrystalline, hexagonal AlN and that the crystallinity of the films increases with the time for the H2 pulse, as seen that the intensity of the (100) plane more than triples when the H2 is added (black line) compared to without a H2 (red line) (Fig. 2a). The full width at half maximum value (FWHM) for the (100) plane also decreases from 0.657 to 0.503 upon addition of the H2 pulse, indicating a higher degree of crystallinity. This shows that adding a H2 pulse not only decreases the carbon content in the film, but also increases the crystallinity. Figure 2b shows the morphology of the fine-grained polycrystalline nature of the AlN film with 19.5 s H2 pulse deposited at 480 °C, which was also seen from the GIXRD measurement, Fig. 2a.   To test if the reduction in carbon content was due to a surface chemical reaction between surface methyl groups and the hydrogen gas, N2 and Ar were also tested as purge gas between the TMA and NH3 pulse. The pulse time for H2, N2 and Ar was kept constant at 19.5 s. XPS measurements show that the carbon content in the films is 1±0.4 at % independent of which gas that is used.
This suggests that carbon reduction by the added pulse between TMA and NH3 is not a chemical cleaning pulse, but rather acts to assist desorption and prevent re-adsorption of carbon containing species from the surface. Interestingly, the different gases used influence the crystallinity of the films. GIXRD measurements of films deposited with H2, N2 and Ar show that the intensity of the (100) plane decreases when going from H2 to N2 and to Ar (Fig 3).
Theta-2theta showed the same trend, Fig. S3. To study which carbon containing species are cleaned away by the extra pulse between the TMA and NH3 pulse, the AlN films were exposed to TMA vapor in an ALD chamber and the exhaust gases were analysed by mass spectrometry. Fig. S4 shows the MS graph for the background from the system without TMA supply into the chamber. From these experiments, it was noted that the intensity ratios (the region at a molecular weight of 15 and 28) between the detected masses differ between experiments with a continuous TMA flow and a TMA flow that is stopped and the chamber is left to purge, allowing the Ar purge gas more time to interact with the surface to mimic the extra pulse (Fig 4). when the TMA supply into the chamber is stopped and the Ar purge gas has more time to interact with the surface (Fig. 4). The overall intensities decrease when the TMA flow is stopped for all regions except for the second region at a M.W. of 27 (Fig. 4), which increases upon longer Ar exposure time. Interestingly, the intensities between the peaks changes when the TMA flow is stopped and Ar purge gas have more time to interact with the methyl terminated surface (Fig. 4), compared to the case with a continuous TMA flow. The M.W. of 15 has a higher intensity compared to 16, which is the opposite to when continuous TMA flow into the chamber. This indicates a higher probability to form •CH3 radicals when the Ar purge gas is given more time to interact with the surface, Fig. 4 and Fig. S5. The intensity for M.W. 27 increases by approximately 67% for experiments when the TMA flow is stopped for 15 min compared to continuous flow of TMA into the chamber, Fig. 4. This suggests favoured formation of xHC-CHx over formation of •CH3 radicals with increasing Ar purge time. Kinetics simulations were used to obtain an atomistic level understanding of the surface chemistry (Fig. 5). Starting from a fully saturated surface (i.e. in the model 4 CH3 groups at 4 surface sites), the first methyl group is seen to quickly desorb from the fully occupied surface.
This could be expected since the desorption of a CH3 from a fully occupied surface is actually exergonic (and with a negligible reaction barrier), see Table S1, which is likely due to the repulsion between the closely-packed surface-adsorbed methyl groups. The desorption rates become significantly slower with increasing number of empty sites on the surface. This is to be expected since the repulsion between the adsorbed groups decreases as the number of empty sites increases for the most densely packed surfaces. The desorption rates were also found to increase with increasing temperature which is also observed experimentally (Fig. 1a). Figure   5a shows the kinetics when a pulse of inert gas is added between TMA and NH3. Adding the extra pulse between the TMA and NH3 pulse will remove the methyl radicals from above the surface, which facilitates a more accelerated desorption of the carbon groups by suppressing the re-adsorption process. With no added pulse of inert gas between TMA and NH3 (Fig. 5b), the methyl groups previously released from the surface can either become re-adsorbed on the empty sites or converted to C2H6 molecules, which is manifested by the persistence of the dominantly CH3-terminated surface in Figure 5b. This was also seen in the mass spectra (Fig.   4); more xHC-CHx species are formed from the desorbed methyl groups when the TMA supply is stopped. The C incorporation in the film versus time as estimated from the computed surface coverage is shown in Figure. 6. The re-adsorption process leads to a significant increase in the carbon incorporation in the film (Fig. 6b). On the other hand, when a pulse is introduced between TMA and NH3, the re-adsorption process is suppressed, and the desorption dominates, resulting in the reduction of the carbon incorporation in the film (Fig. 6a). This correlates with the experimental results in Figure 1a. We believe that the extra pulse between TMA and NH3 gives surface methyl groups enough time to desorb from the surface and prevents their re-adsorption. In the kinetic simulations ( Without the extra gas pulse the desorbed surface methyl groups could re-adsorb before the NH3 is introduced into the chamber which could lead to them being trapped in the film. Carbon impurities in semiconductor graded materials have always been a problem when deposited at high temperature and we believe that this ABC-type pulsed process (were A is TMA pulse, B is the extra pulse and C is NH3 pulse) could possibly be the solution to accomplish high temperature ALD depositions with low C impurities.

Summary and Conclusions
A high temperature thermal ALD-like CVD approach with time-resolved precursor supply for AlN was explored by adding an extra pulse of H2, N2 or Ar between the TMA and NH3 pulses to investigate if that could change the surface chemistry and lower the C content in the film.
We show that by adding an extra purge gas pulse perpendicular to the substrate surface, the carbon impurities in the film decreased drastically to approximately 1 at.% giving a Al/N ratio of 1.07. The surface chemistry and the role of the extra pulse between TMA and NH3 was investigated with mass spectrometry, quantum chemical and kinetic modelling. It was found from the modelling that the most loosely bound surface methyl groups can desorb quickly with negligible reaction barrier. The added pulse between TMA and NH3 is believed to clean the region just above the surface from desorbed methyl groups, promoting additional desorption and suppressing any re-adsorption of the desorbed methyl groups. It was found both experimentally and from the calculations that the desorption of the surface methyl groups increases with increasing temperature. Mass spectrometry measurements show that more xHC-CHx species, compared to CHx species, are formed with longer purge times. The growth of the AlN film was somewhat dependent of the extra pulse time were longer pulses gave smaller growth rate which were attributed to less C impurities in the film. The type of gas used as the pulse between TMA and NH3 was found to have a stronger impact on the crystallinity were the crystallinity of the films increased when going from Ar, N2 to H2.