Growth-induced temperature changes during transition metal nitride epitaxy on transparent SiC substrates

Accurate noncontact substrate temperature monitoring and control during the MBE growth has been a topic of continued interest for more than 25 years.^1–3 While accurate temperature control is well understood for homoepitaxial growth, large substrate temperature changes that are not accurately sensed by the substrate heater thermocouple (TC) can be induced by high carrier densities,^4,5 reduction of the effective bandgap of the substrate/epitaxial layer stack,^1,2 and changes in the surface reflectivity [e.g., induced by the accumulation of gallium droplets in GaN (Ref. 6)] during MBE growth. In general, any change of the thermal cavity that partially encloses the substrate can also be expected to lead to a change in the substrate temperature.

The physical processes that manifest themselves in the thermal properties of a material can be modeled as an effective surface emissivity. The Stefan–Boltzmann law describes the net thermal energy radiation rate for surfaces that can be approximated as a blackbody,

where P is the net thermal radiation rate in watts, e is the surface emissivity (0–1, where 0 is a perfect reflector and 1 is a perfect blackbody), σ is the Stefan–Boltzmann constant (5.6704 × 10⁻⁸ W m⁻² K⁻⁴), A is the radiating area in m², T is the surface temperature in kelvin, and T_s is the temperature of the surroundings in kelvin. It is easy to see that changes in surface emissivity can lead to large temperature changes for fixed P or large changes in the thermal radiation rate for fixed T. In general, the physical absorption, reflection, and transmission properties of a material are a function of the details of the electronic band structure, surface properties, thickness, and free carrier density, and all these affect the resultant surface emissivity.

An experimental case that exemplifies one of the most extreme examples of growth-induced temperature changes is the MBE growth of metallic films on transparent semi-insulating SiC. (By “transparent,” we mean that the SiC is transparent to visible radiation. In general, the transparency range of a material is a function of the photon energy.) The integration of epitaxial metal layers is technologically important as they will enable substantial performance benefits, design flexibility, and novel device structures such as metal-base transistors⁷ and integrated epitaxial superconductor/semiconductor heterostructures.⁸ Metals are highly reflective in the visible through the near IR range,⁹ but thin metal layers can have strong absorption below the far IR range,¹⁰ have very low emissivity, have very high carrier densities, and have zero bandgap compared to the high emissivity, low reflectivity, low carrier density, and 3.26 eV room temperature (RT) energy gap of semi-insulating 4H-SiC. All of these factors mean that the growth of even very thin epitaxial metal layers can be expected to lead to large SiC substrate temperature changes because of the large change in the surface emissivity. Similarly, one would expect that the thermal radiation rate of a metal surface would change as a result of growing thin, high emissivity, epitaxial semiconductors on top, as a direct consequence of the changing emissivity of the system. Here, we report on measurements and modeling of the substrate temperature excursions observed during the MBE growth of metal and metal/semiconductor layers on transparent SiC substrates.

Metallic epitaxial niobium nitride (NbN_x) thin films were grown by MBE on 3-in.-diameter (0001)-oriented double-side-polished (DSP) semi-insulating 4H-SiC substrates that had been chemically mechanically polished to an epi-ready finish. No ex situ cleaning or backside metallization of the substrate was performed before loading the substrates into the MBE system. The epitaxial growth was performed in a customized Scienta-Omicron PRO-75 MBE growth chamber having a base pressure below 2 × 10⁻¹¹ Torr described previously.^11,12 This system is equipped with a substrate heater element machined from silicon carbide. The niobium flux was generated by an in situ 5 kW e-beam evaporator operated with an estimated melt temperature of 3000 K, which gave a resultant NbN_x growth rate of 1.2 nm/min. The incident reactive nitrogen flux, N*, was generated by a Veeco Uni-Bulb RF-plasma source using a 99.9999% pure N₂ gas that was further purified using an inline purifier. The RF-plasma power level was 250 W, and the N₂ flow rate was 1.2 SCCM, giving an N-limited GaN growth rate (measured in separate experiments) of 3.9 nm/min. The chamber pressure during growth was 1.7 × 10⁻⁵ Torr, well below the millitorr pressure range where the thermal conductivity of the vacuum starts to become significant.¹³ 

The SiC substrate was degassed at 700 °C for 30 min in the ultrahigh vacuum preparation chamber and then was transferred to the MBE growth chamber. After thermal cleaning of the substrate in the MBE growth chamber in vacuum at a substrate heater thermocouple temperature of 1000 °C, the temperature was adjusted to the desired value before starting the NbN_x growth. Additional information about the MBE-grown NbN_x films is reported in Ref. 14.

For experiments, where AlN was grown on NbN_x, a conventional dual-filament effusion cell operated at 1471 K was used to generate the Al flux corresponding to an AlN growth rate of 3.8 nm/min. After the NbN_x layer growth was complete, the e-beam shutter was closed, a growth interrupt was performed (with N* open) to retract the quartz crystal microbalance (QCM) used to measure the Nb flux (so that AlN was not deposited on the QCM crystal), and then AlN growth was started on the NbN_x surface by opening the Al effusion cell shutter.

An in situ single-port BandiT diffuse-reflectance system from K-Space Associates was used for band edge (BE) substrate temperature measurements.¹⁵ This system uses broadband light that is diffusely reflected (DR) from the substrate in conjunction with a separately generated calibrated lookup table to relate the measured optical BE to the actual wafer temperature.³ Note that with the DSP substrates used here, incident light is scattered from the surface of the rough substrate-heating element rather than from the backside of the 4H-SiC substrate itself. The optical fixture used for incident light positioning and DR light collection was mounted on a single 6-in. diameter viewport flange that was oriented 24° from the substrate normal. The incident broadband light was generated by a 300 W Xe lamp with an adjustable exit iris. The collected DR light was split via a bifurcated optical fiber and measured by two spectrometers—a visible-range spectrometer operating between 353.9 and 601.7 nm for SiC BE measurements, and a near-infrared-range spectrometer operating between 877.3 and 1675.8 nm for pyrometer temperature (PT) measurements. The PT measurements here assume a fixed emissivity of 0.7 at 910 nm. The measured optical spectra were collected in 1 s intervals and analyzed using the system's software routines.

While semi-insulating 4H-SiC is transparent and colorless to the eye, it has strong absorption for wavelengths shorter than the BE energy (387 nm at RT) and in the IR (absorption coefficient peaks at 12.6 μm at RT)¹⁶ resulting in efficient heating from IR radiation.

Figure 1 shows the measured BE temperature (BET), the PT measured in a 910 ± 10 nm window, along with the substrate heater TC temperature reading and applied electrical heater power (HP) for the case of slowly heating a 3-in. diameter 4H-SiC substrate from a starting TC temperature of 200 °C.

Several features are apparent in Fig. 1. The blue dot (black dots grayscale) illustrates the measured SiC BE temperature. Note that there is a small step increase in the measured BE temperature at 2 min in the figure—this is the point in time when the main substrate shutter is opened for the first time after loading the substrate. The BE temperature decays by ∼10 °C over 2 min to an equilibrium value of 180 °C. Note that the HP and TC readings remain constant as the BE temperature stabilizes.

On ramping the TC temperature from 200 to 1000 °C over 20 min, the HP, BE, TC, and PT readings behave as expected—they smoothly increase (except for the step in initial heating from 200 °C TC) and then stabilize. The measured TC temperature is well controlled with <1 °C overshoot or undershoot at the final setpoint and the HP stabilizes quickly. The PT readings are substantially higher than the TC and BE readings and are dominated by the direct emission from the substrate heater (the 4H-SiC substrate is transparent at 910 nm).

During the ramp-up to 1000 °C TC, there are regions where the SiC BE temperature is increasingly noisy, at least partially as a consequence of the details of the band structure near the band edge of the indirect bandgap SiC material. This noise can be mitigated to some extent by tweaking the software fitting parameters. The large difference between the heater TC temperature and the substrate BE temperature is common for radiatively heated substrates in MBE growth systems. The difference in temperature readings is a function of the particular heater and heater cavity design, the temperature range, and the substrate.

At 30 min, the main shutter was closed (as indicated by the red vertical arrow), stopping optical access to the substrate and the apparent BE temperature thus rapidly fell to its system minimum, and the HP quickly fell and stabilized. The apparent PT initially fell but then quickly increased to a steady-state value as a consequence of the shutter being heated by radiation passing through the substrate. The actual PT of the closed shutter is uncalibrated since the emissivity of the shutter at the 910 nm measurement wavelength is unknown, but it quickly reaches a steady-state value.

At 35 min, the main shutter was opened (indicated by the green vertical arrow), and it is immediately apparent that although the TC reading did not change (as a result of the fast heater TC control feedback loop), there was a very large, fast transient in the BE temperature and the HP.

At 40 min, the TC temperature was ramped to 1100 °C and similar behavior was observed as for the 1000 °C TC case. The TC temperature is very well controlled and the applied HP is stable. On closing the main shutter at 51 min, the HP quickly fell and stabilized, and the apparent PT quickly rose and stabilized. On opening the main shutter at 56 min, there is a large transient in the BE and HP traces, while the TC temperature remains fixed. While the TC temperature is 100 °C higher than the previous steady-state value, and the HP is 240 W higher, the BE temperature of the SiC is only 45 °C higher than for the 1000 °C TC case.

On ramping the TC reading to 1140 °C at 65 min, the heater current limit was reached when the main shutter was open. When the main shutter was closed and the temperatures allowed to stabilize, upon opening the shutter, the HP was below the current limit and the system behaved as expected. The BE temperature of ∼712 °C was only 25 °C higher than the previous value even though the applied heater power was 110 W higher than for the 1100 °C TC case.

Finally, the TC temperature was ramped to 900 °C and again the system behaved as expected. The large scatter in the BE temperature reading is similar to that observed in the initial ramp-up.

Figure 1 demonstrates that very large substrate temperature transients are created by opening and closing the main shutter for radiatively heated SiC substrates, and these transients are not detected unless one monitors the heater power or uses a technique to directly measure the substrate temperature.

Figure 2 shows the 4H-SiC BE temperature transients measured upon opening the main shutter for a TC temperature of 1140 °C.

In this experiment, Fig. 2(a) shows that the peak 4H-SiC BE temperature is clearly a function of the shutter closed duration, with longer shutter closed times (until the system reaches equilibrium) increasing the magnitude of the temperature transient. BE temperature transients as large as 135 °C are measured for closed durations exceeding 2 min. Figure 2(b) shows that the average temperature rise is ∼1.6 °C/s while the shutter is closed, and the BE temperature stabilizes in 70 s for the case of this static TC temperature of 1140 °C. In addition to the fast shutter-induced transient, there is a much longer BE temperature time constant that is apparent in the slow increase in the stable baseline temperature. The baseline BE temperature slowly increased 14 °C while the average HP with the shutter open only increased ∼5 W during the duration of this experiment. This longer heating time constant is apparently related to the heating of the rest of the heater cavity.

Figure 3 illustrates the measured temperature and HP characteristics as a function of epitaxial growth of a thin NbN_x layer at 1.2 nm/min on 4H-SiC. This film is metallic and consequently has a high density of free electrons and low emissivity and thus is very different from the semi-insulating, very low density of free electrons and high emissivity 4H-SiC substrate.

As with the earlier examples, there is a large BE temperature transient on opening the main shutter (at −2 min). The BE temperature was allowed to stabilize over 2 min and then the growth started by opening the N* and Nb e-beam shutters (at 0 min as indicated by the vertical green arrow). There is an ∼1 min nonlinear baseline in the temperature traces, and then the BE temperature begins to rise rapidly and the applied HP begins to fall rapidly to keep the TC temperature fixed at 1000 °C. The BE temperature quickly approaches an asymptote and is apparently stable by the time the BandiT fit becomes noisy after ∼20 min of the NbN_x growth (24 nm NbN_x film thickness). The increasing thickness of the NbN_x film means that the front surface becomes increasingly reflective, and any light that does specularly reflect from the heater and pass through the SiC substrate is increasingly reflected at the SiC/metal interface, thereby reducing the intensity of the optical signal that reaches the BandiT detectors. Since the BE temperature trace during the NbN_x growth appears to have a reciprocal-time functional form, a simple polynomial expression was used to fit the BET trace as a function of growth time,

where b = 903.9 °C, m = −1.09 °C min⁻¹, c = −226.3 °C min, and t₀ = 0.07 min between 1 and 19 min (1079 datapoints) with a small error in the fitted parameters (reduced χ² = 9.81). If this expression holds beyond 19 min, then the expected BE temperature at the end of the 56 nm thick NbN_x film would be ∼30 °C lower than if the temperature remained constant at ∼870 °C. At this time, it is not clear whether the actual temperature remained fixed over the remainder of the growth run, but the heater power is changing slowly so we expect any additional temperature changes from the last reliable BE measurement to be relatively small (<±30 °C).

Figure 4 shows the temperature and power traces for an NbN_x growth experiment, where a single TC temperature ramp was programed early in the growth sequence to minimize the temperature excursion. In this example, the NbN_x growth on SiC was initiated as soon as the main shutter was opened and then the TC temperature was ramped from 1000 to 850 °C over 6 min beginning after 3 min of growth (∼3.6 nm NbN_x). The BE temperature remained constant for the next ∼15 min, and the maximum BE temperature excursion was reduced from 240 °C for a fixed TC temperature case to 115 °C here.

This figure illustrates that if the temperature rise is known in advance for the type of metallic structure to be grown on transparent SiC substrates, then the appropriate temperature ramp can be programed into the MBE growth control software and the temperature transient can be reduced. Alternatively, if the MBE growth control software can accept multiple temperature measurement inputs for the substrate HP control, then the BE temperature or other measurements can be used directly to control the heater appropriately to reduce the temperature excursion even further. However, note the increasing noise in the BE temperature measurement beyond ∼20 min (NbN_x thickness ∼24 nm). Any TC and HP control algorithm based on feedback from the BE temperature measurement must take noise into account.

Up to this point, we have seen the effect of changing the thermal cavity (via opening and closing the main substrate shutter) on the temperature of a transparent SiC substrate and the effect that growing a metal film has on the transparent substrate temperature. Changing the state of the main shutter changes the thermal cavity, which changes the T_s term in the Stefan–Boltzman law and thus increases or decreases the net radiation from the substrate surface. The decreasing emissivity of the system as a result of metal growth leads to a substantial rise in the substrate temperature unless the heater power is reduced appropriately.

The inverse structure case—growing an ultrawide-bandgap semiconductor on a metal—is considered next. Figure 5 shows the temperature and heater power curves measured for the growth of 100 nm AlN on NbN_x on 4H-SiC. As in the previous example shown in Fig. 3, the BE temperature rapidly rises and stabilizes during the first ∼5 min of the NbN_x growth. In this case, the heater power continues to fall during the NbN_x growth, and the BE temperature becomes noisy after ∼12 min. Equation (2) was used to fit the early part of the BE temperature rise (shown in orange) and the fit parameters between 1 and 13 min were found to be b = 897.7 °C, m = −0.17 °C min⁻¹, c = −133.5 °C min, and t₀ = 0.33 min with a small error in the fitted parameters (reduced χ² = 9.40). The asymptotic BE temperature of the fit was 885 °C. After ∼50 nm of NbN_x growth, the Nb e-beam shutter was closed and a 190 s interrupt was used to retract the QCM and prepare for the AlN growth. No significant change in the HP or PT was observed during the interrupt, implying that the hot e-beam source is not a major source of the substrate heating in this case.

Upon initiating the AlN growth at ∼48 min, there is a short delay in the PT and HP traces, but then they begin to increase rapidly in the first 5–10 min of AlN growth. The HP trace then begins to flatten out (but still has a slight positive slope) through the remainder of the AlN layer, while the PT temperature continues to increase. Throughout this growth run, the TC temperature has remained fixed at 1000 °C.

The simplest explanation for the need to increase the heater power, and the observed increase in PT (which is mostly sensitive to the heater temperature in this thickness and wavelength range), is that the emissivity of the surface of the NbN_x/SiC sample was substantially increased by the AlN surface coating, and thus the power lost from the sample and heater cavity increased, necessitating an increase in HP to keep the TC temperature fixed. While no direct measurement of the wafer temperature is available for the AlN/NbN_x/SiC sample at this time, simple scaling arguments indicate that the sample temperature has fallen tens of degrees, perhaps as much as 70 °C if the partitioning of the sample temperature with HP shows the same trend as for the NbN_x/SiC case.

As a baseline check of the validity of this physical picture, we can simulate the thermal behavior of a system with layers of different emissivities using the wptherml package¹⁷ that runs on Python. This software uses a transfer matrix method to solve Maxwell's equations for layered isotropic media. It computes the emissivity and other thermal properties for a structure via

where A is the absorptivity, R is the reflectivity, and T is the transmissivity of the structure as a function of wavelength and angle of incidence. The thermal radiation rate from the surface is the product of the emissivity spectrum and the radiation spectrum given by Planck's blackbody law. Using tabulated values for the dielectric function for the materials, the emissivity and other thermal properties can thereby be calculated for the structures of interest.

Figure 6 shows a baseline simulation of an analogous system consisting of a 50 nm thick layer of W on a sapphire (Al₂O₃) substrate at 1150 K. The red (light gray) trace shows the calculated thermal radiation rate of the system with no AlN layer on top of the W, while the blue (dark gray) trace shows the substantially increased thermal radiation rate when 100 nm of AlN is on top of the W layer. The enhanced thermal emission resulting from the AlN film on the wafer, and thus increased heat loss from the heater cavity, means that more heater power would be required to maintain the same heater TC temperature and agrees with our explanation for the observations in Fig. 5. Modeling work is continuing to gain a better understanding of the temperature evolution during the MBE growth of heterostructures with widely varying emissivities.

There are several immediate implications from our observations of these temperature transients during the MBE growth of materials with widely varying emissivities on radiatively heated transparent SiC substrates. First, we confirm earlier observations that the heater TC temperature is approximately proportional to the actual substrate temperature, but that proportionality changes depending on the temperature range, the main shutter state, and the substrate. The main substrate shutter should be kept open if the substrate temperature control is critical. Closing the main shutter decreases the amount of heat lost from the heater cavity, increases the T_s term in the Stefan–Boltzmann equation, and can lead to dramatic increases in the substrate temperature. The poor coupling between the substrate heater TC and the wafer temperature means that even with the reduced heater power when the shutter is closed, the substrate temperature can still increase dramatically. Note that the substrate shutter effect is seen even at very low substrate temperatures (∼+10 °C at 180 °C substrate temperature in Fig. 1).

Second, coating the backside of transparent substrates with a highly reflective metal is counterproductive in terms of coupling heat into the radiatively heated substrate. Metals have low emissivity and thus will reflect more of the incident IR radiation from the substrate heater compared to the uncoated substrate case. In addition, note that if DR measurements were contemplated for BE temperature measurements, such a metal film would need to be applied to a substrate with a rough backside surface to ensure sufficient DR signal.

Third, in our MBE system, the vast majority of the heating of the substrate is from radiation from the substrate heater. The radiation from the ∼3000 K e-beam evaporator source has little effect on the substrate temperature. While the BE and TC temperatures are not affected by radiation from the e-beam, small effects are noted in the 910 nm PT trace on closing the e-beam shutter. A common approach to increase the accuracy of PT measurements is to use a dual or multiwavelength pyrometer to better fit the blackbodylike radiation curve and extract an effective emissivity. However, this approach assumes that the radiation measured can be described by a single blackbodylike curve and that assumption will not hold if intense radiation from sources other than the substrate (e.g., hot effusion cells, hot e-beam evaporator) is detected (because their emission curves peak at different temperatures).¹⁸ Thus, optical pyrometry remains problematic for measuring MBE substrate temperatures during the growth of materials with changing emissivity, especially when high-temperature sources are required to generate the growth fluxes.

Fourth, BE temperature measurements can be performed on substrates with thin metal coatings. The vast majority of the observed temperature rise for NbN_x grown on 4H-SiC occurs in the first ∼5 nm of growth. BE temperature measurements begin to fail after 20–25 nm NbN_x thickness as a consequence of reduced light reaching the detector. It appears that the substrate temperature stabilizes by ∼50 nm NbN_x thickness and potentially for even thinner layers. Slow changes in the heater power over time after the substrate temperature stabilizes may be the result of small changes in the optical properties of the heater cavity (wafer mount and heat shielding) during growth.

Finally, as predicted by the Stefan–Boltzmann law, the growth of ultrawide-bandgap AlN films on NbN_x metal layers leads to a decrease in the sample temperature through an increase in radiation emitted from the surface. Most of the HP increase, and temperature decrease, during AlN on the NbN_x growth happens in the first ∼10 min of AlN growth (∼35–40 nm). The effects of emissivity changes on the SiC substrate temperature happen in relatively thin layers, and changing emissivity can have a dramatic impact on the actual temperature during the MBE growth. Emissivity changes will change the thermal radiation rate, and thus the heater power, consequently, changes in substrate temperature must be suspected whenever there are substantial changes in the heater power. These temperature transients can be controlled through proper ramping of the TC temperature during growth if the excursion is known in advance, or via direct feedback of the BE temperature if the control algorithm is sophisticated enough to handle the case when the BE fit does not converge properly.

A band edge temperature measurement system was used in the MBE growth of NbN_x metallic layers on transparent 4H-SiC substrates. Large temperature transients were observed and quantified for the cases of opening and closing the main substrate shutter, in growing epitaxial NbN_x, and in growing epitaxial AlN on NbN_x when fixed substrate heater TC temperatures were used. The substrate temperature transients were explained using the Stefan–Boltzmann law and a baseline model of thermal emission. Strategies for reducing these temperature transients, which will occur to some extent in any heteroepitaxial growth system with large changes in emissivity, were discussed. Direct measurement of the substrate temperature during large emissivity changes remains a challenging problem in the MBE growth.

This work was supported by the Office of Naval Research.