Thin Film Sensors for High Performance Silicon Nitride Ceramics

With the incorporation of more high strength ceramic components into the gas turbine engine, there is a subsequent increase in efficiency and thrustto-weight ratio since operating temperatures can be be significantly increased. Surface mount technology represents the best alternative to accurately measure the surface temperature of these ceramic components. Therefore, a thin film thermocouple system was developed and tested for hot isostatically pressed (HIP'ed) Si3N4 ceramic substrates. Oxidation studies were performed on the hot isostatically pressed Si3N4 substrates to determine the stability of these ceramics in a variety of oxidizing ambients at high temperature. These studies indicated that the growth kinetics of the oxide scale were strongly dependent on the oxygen partial pressure and the oxidation temperature. Parabolic growth kinetics were observed at temperatures >1300°C and oxygen partial pressures >0.02 atm, while logarithmic kinetics were observed below 1300°C and an oxygen partial pressures of 0.20 atm. An activation energy of 297 kJ/mol was determined for samples oxidized at temperatures in the range 1200°-1400°C in dry oxygen. The oxide scale generally consisted of amorphous Si02 and (101) a.-cristobalite with small amounts of yttrium silicate incorporated into the growing scale. Scanning electron microscopy show a cracked oxide scale due to the ~ to a.cristobalite transformation at 273°C. This as-oxidized surface was not suitable for sputter-deposited sensor elements. Therefore, a surface treatment was employed to improve stability and adhesion, which consisted of thermal and sputtered oxide interlayers to act as a diffusion barrier to high temperature oxidation. Formation of a stable surface prior to sensor element deposition was a critical step in the fabrication of a reliable, accurate sensor.

Thin film thermocouples consisted of 1 µm thick type S thermocouple sensor elements sputter-deposited onto the specially prepared substrates.
Lead wires were bonded to the films using parallel gap welding and an Al 2 0 3 overcoat was sputtered over the entire sensor pattern to prevent high temperature oxidation of the sensor elements. The thin film sensors were tested in a high temperature furnace where they were exposed to a variety of oxidizing conditions. Scanning electron microscopy of failed sensors revealed that emf drift and sensor failure was due to the oxidation of platinum and rhodium along with thermal sintering of the metal films. Aluminum oxide overcoats minimized the effects of both of these degradation mechanisms.
iii ACKNOWLEDGEMENT In the three years I have worked on this research project I have had a great deal of assistance from a large number of people. Therefore this research was, by no means, a solo effort and may not have been possible without their assistance.   However, pyrometry requires a knowledge of emittance of the surface, and with ceramics these emittances can vary widely. Another problem associated with pyrometry is is that the emitted radiant energy is a mixture of energy emitted at different depths in the material, along with reflected radiation from combustion gases, results in a significant bias in the measurement. Finally, optical access is needed for measurement, and with the limited space in gas turbine engines, this access can be difficult to accommodate. Therefore, thin film temperature sensors were determined to be the most practical alternative for surface temperature measurement of ceramic components.
The operation of thin film sensors are identical to conventional wire thermocouples in that two dissimilar metals are joined at a junction that produces an electromotive force (emf), otherwise known as the Seebeck voltage. The temperature at this junction is a function of the composition of the two metals and the measured emf. Instead of using wires as the thermocouple legs, thin film metal lines are formed on the desired component by a physical vapor deposition process, such as sputtering. The principles of rt sputtering has been previously described [17] and are included in Appendix A. The thin films are non-intrusive and add little weight to the ·component. However, thin film thermocouples do come with their own set of problems with adhesion of the metal films to the substrate material being the most significant. To address the problem of adhesion, it is important to know how the substrate material behaves at high temperatures, since the interaction of the ceramic substrate with the metal film can .lead to metallurgical interaction, such as silicide formation.
Hot isostatically pressed Si3N4 , more specifically NT154 1 , is the leading candidate material for use in gas turbine engines due to its strength retention at high temperatures and excellent oxidation resistance at temperatures below 1300°C. However, NT154 can still oxidize at temperatures below 1300°C but at significantly reduced rates. For a thin film sensor to remain adherent and provide a stable and reliable signal, a chemically stable surface is required. Significant outgassing of nitrogen from the reaction of NT154 with oxygen can cause spalling of the thin metal films and thereby, create an unstable surface. Therefore, a comprehensive oxidation study of NT154 was necessary to determine the high temperature behavior of the material in a variety of oxidizing ambients.
The overall goal of this research program was to develop techniques to fabricate a reliable, accurate thin film thermocouple on the surface of hot isostatically pressed silicon nitride engine components. The first manuscript in this thesis will discuss the oxidation of hot isostatically pressed silicon nitride.

ABSTRACT
Oxidation studies of hot isostatically pressed Si3N4 were performed in flowing dry oxygen and dry oxygen-nitrogen-argon gas mixtures at 1 atm pressure and at temperatures between 1100°C and 1400°C. Parabolic growth kinetics were observed in dry oxygen atmospheres over the entire temperature range studied. A single activation energy of 297 kJ/mol was established over the temperature range 1200°-1400°C and little weight gain was observed at temperatures below 1200°C. Reducing the oxygen partial pressure to 0.20 atm resulted in logarithmic growth kinetics at temperatures of 1200°C and 1300°C.
A further reduction in oxygen partial pressure to 0.02 atm yielded negligible weight gain, even at 1400°C. The oxide scale consisted primarily of amorphous Si0 2 , (101) a-cristobalite and small amounts of a, p, and 8-Y2S i201. The yttrium silicates form primarily from the solid state reaction between the Y 20 3 sintering aid and the silica oxidation product. X-Ray diffraction of the oxide scale showed an increase in intensity of the (101) a-cristobalite peaks were increased as temperature and oxygen partial pressure. The logarithmic growth rate observed for oxidations performed in ai.r are believed to be induced by a porous oxide scale formed at the outset of oxidation that continues to grow until a coherent, dense, oxide scale is formed, thereby creating a diffusion barrier that reduces further oxidation. The parabolic growth rates observed in dry oxygen are a direct result of the rapid formation of thin, dense oxide layer during the initial stages of oxidation which thickens with time.

INTRODUCTION
There has been considerable interest in the use of Si 3 N4 ceramics for high temperature structural applications. In particular, its high temperature strength, creep resistance, and thermal shock resistance, make it ideal for use in gas turbine engines. However, the stability of Si 3 N4 in oxidizing atmospheres at very high temperatures is limited. Since Si02 is a more stable phase than Si 3 N 4 , in oxidizing atmospheres, the ceramic readily oxidizes to the stable Si02 phase and forms a passive oxide scale that acts as a diffusion barrier to further oxidation. The extent of this oxidation and its growth rate (oxidation kinetics) should be established prior to the insertion of these materials into engine applications. There have been a number of studies on the passive oxidation of hot pressed Si3N4 (HPSN) [1][2][3][4][5][6][7], sintered Si 3N4 [8,9], CVD Si3N4 [10-14], and reaction bonded Si 3 N4 (RBSN) [15,16]. Although high temperature strength and creep resistance has been investigated for hot isostatically pressed Si 3 N4 (HIPSN) [17], no comprehensive study of the oxidation behavior has been performed on this form of hot isostatically pressed Si3N4 '.
Hot pressed Si3N4 is similar to hot isosta:tically pressed Si 3 N 4 , in that both heat and external pressure are applied to the powder compact to produce a fully dense ceramic. Hot pressed Si3N4 typically contains a small amount of an oxide sintering aid such as MgO or Y20 3 , which forms a liquid phase during to promote densification. Hot isostatic pressing utilizes a previously sintered product that is enclosed in a compressible, refractory metal can with a small amount of oxide sintering aid and subjected to pressurized gas. This process reduces the porosity in the product and thus, produces a dense ceramic product.
High temperature oxidation of hot pressed Si 3 N 4 in oxidizing Babini et. al. [7] found similar oxidation behavior of HPSN with Y 2 0 3 and Si02 additions as densification aids. Apparent activation energies ranged from 260-623 kJ/mol. with smaller energies corresponding to lower amounts of oxide additives, while higher energies corresponded to increased Y 2 0 3 and Si0 2 additions.
The main focus of this study was the passive oxidation behavior of HIPSN (NT154) over the temperature range 1100-1400°C in various oxygen partial pressures. Part of this objective was to establish the nature of the oxidation products formed on hot isostatically pressed Si3N4 and evaluate the oxide kinetics as a function of temperature and oxygen partial pressures. Once this was accomplished a comparison of the activation energies to other Si 3 N4 materials could be made and appropriate mechanisms identified. The apparatus used for all oxidations is shown in Figure 2. The Si 3N4 coupons were resistively heated with M0Si 2 ~lements in a high temperature alumina muffle furnace. The .hot zone of the furnace was calibrated over the range of oxidation temperatures to within ±4°C. A VTI corporation microbalance system, sensitive to 1 µg, was used to continuously monitor the weight change of the NT154 coupon. The coupons were supported on a sapphire fiber that extended to the furnace hot zone. The furnace tube was sealed with water cooled endcaps, containing Viton o-rings, and tested for leaks. MKS mass flow controllers were used to meter in furnace gases. The gas line was passed through a Neslab Instruments refrigeration unit at set at -20°C to condense any excess water vapor in the gas stream prior to entry in the furnace. The exiting 1 Bomas Machine Specialties Inc., Somerville, MA. gas stream, from the furnace, was passed through a check valve, followed by an oil bubbler to prevent backstreaming. High purity argon gas (99.999%) was used to purge into the system for at least 1 O hours prior to high temperature exposure. Isothermal oxidations were performed at temperatures between 1100°-1400°C, and at a total pressure of 1 atm. To determine the effect of oxygen partial pressure on the overall oxidation kinetics, oxidations were performed in atmospheres containing 1.0, 0.20, 0.02 and 0.004 atm of oxygen.
The furnace was ramped up to temperature in flowing argon until the oxidation temperature was achieved, at which time the oxidizing gas was leaked into the system and the weight gain data was recorded. The NT154 coupons were exposed to these different oxidizing atmospheres for 100 h.
X-Ray diffraction spectra were generated in a Scintag XRD system model #XDS-2000, to determine the phases of oxidation products in the scale. A Perkin Elmer multitechnique surface analyzer model # PHl-5500 was used to perform ESCA analysis (Electron Spectroscopy for Chemical Analysis) on various oxide scales to determine both chemistry and scale thickness. This analyzer was equipped with an argon ion sputtering gun to thin the sample as the chemical bonding information was being· collected. Morphological and microstructural features of the scale formed on the specimens after oxidation were analyzed with an ISl-SX40A scanning electron microscope.

Oxidation Kinetics
The oxidation behavior of hot isostatically pressedsilicon nitride was dependent on oxidation temperature, oxygen partial pressure and nitrogen partial pressure in the furnace. Figure 3 shows the weight change behavior of hot isostatically pressed coupons exposed to dry oxygen at temperatures between 1200°C and 1400°C. Parabolic growth kinetics were observed over the entire temperature range with negligible weight gains being observed at temperatures below 1200°c. Parabolic rate constants were calculated from the slopes straight lines replotted in Figure 4. These straight lines were based on the original weight gain data presented in Figure 3. All of the straight lines generated from this data intersected the y ordinate at slightly positive values, with the exception of the data at 1400°C. This suggests that these oxidation rates were slightly greater than those predicted by a perfect parabolic rate law, which concurs previous studies [6]. The rate data was fitted to the parabolic equation: where W is the sample weight gain, kp is the parabolic rate constant, t is the time, and c a constant which accounts for the initial reaction . When the logarithm of the parabolic rate constants were plotted as a function of reciprocal temperature, a typical Arrhenius relationship resulted ( Figure 5). The best straight line fit through these points corresponded to an activation energy of 297 kJ/mol. This value is comparable to that obtained by Singhal [1] and Tripp and Graham [2], who reported activation energies of 378 kJ/mol and 289 kJ/mol respectively, for their oxidation studies on HS-130 HPSN.
Oxidations performed under reduced oxygen partial pressures resulted in very different rate behavior. The most interesting result of these reduced oxygen partial pressure oxidations was that logarithmic rate behavior was observed at lower temperatures. The insert in Figure 6 shows the logarithmic rate behavior at 1200°c and 1300°C. Here, the majority of the weight gain occurred within the first 30 min, with only a slight weight gain observed for the remainder of the run. Weight gain measurements were repeated at these temperatures and the rate behavior again confirmed this result. To verify the reproducibility of the weight change measurements taken with this TGA apparatus, a piece of CVD Si 3N4 was oxidized at 1300°C in dry oxygen and the results were compared to those found in the literature [1 O]. This confirmed that the thickness of the oxide formed on the CVD coupon was within acceptable experimental error of the previously reported data for this material.
Phase Analysis In addition to the 13 and a-Si3N4 peaks observed in the as-received material, (101) a-cristobalite as well as a and o-Y 2 Si 2 0 7 peaks were observed in the spectra, regardless of temperature and atmosphere. Relatively small a and o-Y 2 Si 2 0 7 peaks were observed and were independent of oxygen partial pressure and temperature. The (101) a.:cristobalite peak intensity varied considerably with oxidizing conditions, with the most intense peaks found at high oxygen partial pressures and temperatures, whereas, only weak (101) acristobalite peaks were observed at low oxygen partial pressures (0.004 atm) regardless of oxidation temperature. Figure 7 shows the relative peak intensity of the (101) a-cristobalite peak relative to the (200) j3-Si 3 N 4 peak for coupons oxidized in pure oxygen at 1200, 1250, 1300, 1350, and 1400°C, while Figure 8 shows the relative peak intensities of the (101) a-cristobalite phase for coupons oxidized in air at 1200, 1300, and 1400°C. In addition the peak intensity was found to decrease as oxygen partial pressure decreased, as shown in Figure 9.
Once again the relative peak intensities of the (101) a-cristobalite was compared to the (200) 13-Si3N4 which did not vary during oxidation. In addition to these oxidations, a coupon was oxidized in dry oxygen at 1500°C to determine if other phases had formed at elevated temperatures. This coupon exhibited two additional silicate phases, a small peak associated with 13-Y 2 Si 2 0 7 , which results from the transformation of a-Y2Si201 to l3-Y2Si201 between 1350°-1400°C [19] and another strong peak identified as Y 4S i3012· Apparently, at very high temperature, (> 1500°C), transport of the Y 20 3 densification agent is accelerated along the grain boundaries to the external scale or a liquid phase is formed. However, no weight gain data was taken at this condition to substantiate this.
Morphological Analysis SEM micrographs of the as-received Si 3N4 surfaces were relatively featureless, with the exception being small residual grinding marks. All oxidized coupons exhibited a microcracked scale similar to the coupon oxidized at 1300°C in air, which is shown in Figure 9. The fissures in this scale were attributed to the 13->a cristobalite phase transformation at 273°C. Small acicular crystallites of Y 2S i20 7 were also observed on top of the cracked scale. At higher temperatures and oxygen partial pressures, these silicate crystallites were much more pronounced. Much larger needle-like crystals with larger aspect ratio were observed in pure oxygen at 1500°C as shown in Figure 10. Therefore, this oxynitride layer may be considerably thicker than 60A.

DISCUSSION
When Si 3N4 is exposed to oxidizing ambients at high temperatures, an outer scale of silica is formed on the surface. The formation of this silica scale occurs by one of two oxidation mechanisms, according to the recent literature.
The first oxidation mechanism involves the reaction of oxygen with Si3N4 to form Si0 2 . giving off nitrogen gas as a product according to equation (1 ). This reaction is a one step process in which the Si 3N4 substrate is directly oxidized to form Si02.
A second oxidation mechanism involves the formation of an intermediate silicon oxynitride phase, as illustrated by the sequential reactions below. The first reaction is the oxidation of the Si3N4 to a silicon oxynitride phase by the reaction : The silicon oxynitride phase then reacts with oxygen to produce Si02 and nitrogen gas: The stoichiometry of this layer does vary and is best represented as SiOxNy or SixOyNz [10, 18,20]. Furthermore this silicon oxynitride layer was found to be more structurally dense than the Si0 2 layer, leading to the conclusion that diffusion of molecular oxygen through this dense oxynitride layer is the rate limiting step for the oxidation of CVD Si3N4 [1 O]. This could not be confirmed by our results on NT154, since it was difficult to determine the exact thickness of this layer. However, it is sufficient to say that such a layer exists in the scale at the oxide-substrate interface.
For HPSN ceramics, it is the generally agreed that the outward diffusion of the cation species to the external scale is the rate limiting step in the overall oxidation process. The oxidation of higher purity forms of silicon nitride suggests that the inward diffusion of oxygen through the oxide scale is the rate limiting step. Since our hot isostatically pressed Si 3 N4 coupons contains approximately 4wt% Y20 3 it was anticipated that the rate limiting step would be the outward diffusion of yttrium along the grain boundaries in the nitride into the external oxide scale. However, observations during the initial phases of oxidation of this silicon nitride suggests that the behavior was more complicated than this.
Oxide growth rate hot isostatically pressed Si3N4 was dependent on both temperature and oxygen partial pressure. High temperature oxidation in dry oxygen showed classic parabolic rate kinetics as seen in Figure 3, which indicates a diffusion controlled growth mechanism. However logarithmic rate kinetics were observed when these same coupons were oxidized at an oxygen partial pressure of 0.20 atm over the same temperature range.
This logarithmic rate behavior has been previously observed for this material [21]. The insert in Figure 6 shows that in the first 5-1 O minutes of oxidation, very rapid oxidation of the Si3N4 occurred. During the early stages of oxidation it is suspected that a very thin, porous, non-protective film of Si0 2 forms on the surface of the nitride, as previously observed for silicon [22]. This allows the oxidant to remain in contact with the silicon nitride during the early stages of oxidation in dry air and thus, results in rapid initial growth. At higher oxygen partial pressures, it is believed that a dense, coherent, oxide scale is formed by the coalescence of this scale and creates an effective diffusion barrier. The growth kinetics change over much more quickly to parabolic rate behavior when the oxygen partial pressure in the furnace atmosphere is increased. Continued oxidation in the dry oxygen ambient showed a steady weight gain, while reduced oxygen partial pressures quickly reached a limiting weight gain. A previous study [21] on the oxidation behavior of NT154 attributed the non-parabolic rate behavior to the preferential oxidation within the grain boundaries. This preferential oxidation induced compressive stresses in the bulk grains and reduced lattice spacing, which retarded oxygen permeability in the grain boundary phase. This explanation does yield a viable mechanism for our oxidation studies at reduced oxygen partial pressures, but does not explain the observed parabolic rate kinetics for these oxidation studies performed in dry oxygen ambients. Evidence of a thin, nitrogen rich, silicon oxynitride layer is evident from the ESCA depth profile in Figure 12. However, the etching performed in this study may have removed some of the SiOxNy. making an accurate assessment of the layer difficult. Therefore, further evaluation of the oxynitride layer is necessary before its role in the oxidation process can be determined.
The oxide scale formed on the Si3N4 coupons consisted of both silica and yttrium silicate and corresponds to oxide scales typically found on a Si 3 N4 ceramics fabricated with various oxide densification aids. Specimens oxidized in pure oxygen showed a continual increase in intensity of the (101) acristobalite peak with increasing oxidation temperature, as shown in Figures 7 and 8. There is also a corresponding increase in the (101) a-cristobalite peak intensity with increasing oxygen partial pressure ( Figure 9). It appears that the presence of oxygen in the scale increases the stability of the cristobalite phase.
This assumption is supported by the fact that the oxidation of a coupon at 1400°C in dry oxygen had greater crystallinity and a lower weight gain than S i 3 N 4 coupons oxidized at 1400°C in dry air. The crystallization of the Si02 in the scale may also be partially responsible for the growth kinetics observed. It is suspected that the a-cristobalite phase does not form an effective diffusion barrier relative to that of an amorphous oxide. Literature values of the oxygen diffusivity [24] have found the diffusion of oxygen through amorphous Si02 to be approximately two orders of magnitude greater that crystalline Si02 due to its lower density. However, it is more difficult to create a coherent scale of crystalline Si0 2 on the substrate since the crystallization of this scale results in substantial microcracking. The micrograph in Figure 10 shows how the volume change promotes cracks in the scale, thereby creating short circuit diffusion paths for the ingress of oxygen through the scale. In the case of a totally amorphous scale a more continuous coating with less breaks and fissures would reduce the extent of these short circuit paths.
Therefore, oxidation in dry oxygen may have resulted in more rapid transport due to the transformation of the glassy phase to cristobalite. This most likely created a less protective oxide scale at temperatures below 1400°C. At reduced oxygen partial pressures, however, the glassy phase did not transform to ~-cristobalite as readily since ample oxygen was not present in the gas phase to stabilize the ~-cristobalite structure. This could have significantly reduced the diffusion of oxygen to the substrate by providing a more effective diffusion barrier. It is also apparent that the oxidation temperature contributes heavily to the extent of crystallinity of the oxide scale, as illustrated in were observed at all temperatures. The a and 8-Y2Si20 7 XRD peak intensities were very weak and appeared mainly as background in the spectra. In previous studies [1,6, 7 ,8] it was observed that the silicate phases were more prevalent at increasing temperatures. This observation, along with others lead us to the conclusion that the diffusion of the cation impurities in the bulk material was not the rate determining step in this form of Si 3 N4• Since we did not see an increase in the amount of silicate phase in the scale (over the temperature range 1100°-1400°C), we believe that the formation of this phase did not affect the rate kinetics. The very intense Y4Si30 12 peak observed at 1500°C, along with the growth of yttrium silicate crystallites shown in the micrograph of Figure   10, seem to support the conclusion that the diffusion of the sintering aid along the grain boundaries was enhanced at high temperatures. It is evident that at these higher temperatures (>1500°C), the oxidation mechanism and rate limiting step has changed.

CONCLUSIONS
The oxidation of hot isostatically pressed Si3N4 in dry oxygen follows a parabolic growth law with an activation energy of 297 kJ/mol. The resulting oxide scale consisted largely of a-cristobalite and amorphous Si02 with a small amount of yttrium silicate. However, the oxidation of hot isostatically pressed SisN 4 in reduced oxygen partial pressures resulted in a logarithmic growth law at temperatures <1400°C and a parabolic rate law at temperatures ~1400°C.
The logarithmic rate behavior observed at low temperatures consisted of an initial rapid oxidation followed by a very slow increase in oxide growth. XRD analysis of the oxidized hot isostatically pressed Si 3 N4 showed a decrease in crystallinity of the a-cristobalite phase present in the oxide scale as the oxygen partial pressure and oxidation temperature was reduced. ESCA analysis of the oxide scale formed by the oxidation of hot isostatically pressed Si3N4 in dry air showed the presence of a thin , nitrogen-rich silicon oxynitride layer at the scalesubstrate interface.
No single oxidation mechanism that has been considered in this investigation can fully explain . the observed oxidation kinetics over the entire range of temperatures and oxygen partial pressures studied. However, we believe that one or more of the following mechanisms may play a role in the overall oxidation process.
(1) Oxidation in dry oxygen resulted in the rapid formation of a thin oxide layer whose subsequent growth was controlled by the diffusion of oxygen through this layer. Oxidation in reduced oxygen partial pressures resulted in a porous oxide layer that allowed rapid oxidation of the hot isostatically pressed Si 3 N 4 substrate until complete coverage of an oxide film was achieved. Thereafter, this coherent oxide layer provided a barrier to further oxidation and yielded the observed logarithmic growth.
(2) The formation of a dense, oxidation resistant silicon oxynitride layer in reduced oxygen partial pressures was observed. This oxynitride layer impeded further oxidation under these more reducing conditions by forming an effective diffusion barrier. This oxynitride phase may have been stabilized by the higher nitrogen partial pressures associated with oxidation under these more reducing conditions.
(3) Oxidations performed at higher oxygen partial pressures and temperatures resulted in the formation of a more crystalline oxide scale, as indicated by SEM and XRD analysis. Crystalline or partially crystalline oxide scales are not as protective as amorphous oxide scales since grain boundary or interphase boundary transport is usually much more rapid than bulk transport.

INTRODUCTION
The development of gas turbine engines based on advanced ceramics has been extensively investigated over the last two decades. Lighter weight engine designs based on ceramics can significantly improve engine performance, in terms of improved fuel efficiency and reduced maintenance. These test bars were oxidized at 1300°C in air for 100 hours to produce a relatively thick outer scale of silica and a thin inner scale of silicon oxynitride.
The test bars were then etched in concentrated HF (48-50%) to remove the bulk Si0 2 scale that formed, thereby exposing the underlying oxynitride layer, which also acted as an etch stop. The presence of a silicon oxynitride (Si 2 0N 2 ) layer was confirmed by ESCA, as shown in Figure 1. This figure shows an ESCA profile of a silicon nitride surface that was etched in HF after 100 hours of exposure to air at 1300°C. This oxynitride layer is approximately 60 A thick and has a surface concentration that corresponds to a nonstoichiometric Si 2 0N 2 • Both metal and oxide sputtered films adhere well to this oxynitride.
The NT154 substrates were initially cleaned by sonicating in acetone followed by methanol, deionized water rinses, and blown dry in nitrogen.
Aluminum oxide was then sp_ uttered ( across. In addition, the sensor pattern had 1 µm thick elements and 3 µm thick bond pads. The bond pads were thickened from 1 to 3 µm to promote bonding of the 76 µm lead wires.
All of the thin films were prepared by rf sputtering in an MRC model 822 sputtering system equipped with a load lock and substrate heating. High purity (99.9%) platinum and platinum:1 Orhodium targets (12.7 cm in diameter) and semiconductor grade argon were used to deposit the thin film sensor elements. Prior to sputtering, the system was evacuated to a background pressure of 5x10-1 torr at which point argon gas was leaked into the system to establish a gas pressure of 8 mtorr. Intermittent in-situ heating cycles were employed without breaking vacuum at 300°C for 10 min. This was done to promote the outdiffusion of trapped argon, which is incorporated into the metal films during deposition. At elevated temperatures, the coalescence and sudden release of the trapped argon in the metallic films can induce blistering and ultimately result in failure of the thin film. No high temperature annealing step was performed after fabrication.
Lead wire attachment was accomplished via thermocompression welding. Cantilever loaded solid Al 2 0 3 bars, were placed over the test coupons to generate a compressive force between the thin film bond pads and the 76 µm diameter lead wires. One specimen had lead wires bonded to the lead film by parallel gap welding. Appendix C contains the procedures and optimal bonding parameters used for parallel gap welding. After attachment, a 1.6 µm thick Al 20 3 layer was sputtered over this particular sensor to provide protection to the Pt and Pt:1 ORh thin films.
EMF signals from the TFTC as well as the hot and cold reference thermocouples were recorded using Strawberry Tree software. The drift rate, estimated from the drift in temperature (Tref -Ttt) per unit time was used as a measure of device stability. Reference wire thermocouples were used to verify that there was virtually no change in furnace temperature during the tests.
Tests were performed in both oxidizing and inert gas atmospheres. The inert atmosphere was maintained by passing dry nitrogen gas through a P20 5 getter, followed by a titanium getter at 500°C to remove residual water and oxygen that may be present in the gas. Sensors were tested in a Deltech high temperature furnace with a 7 inch hot zone. The furnace was ramped up at a rate of 3°C/min to the desired temperature and held for a specified soak time.
The lifetime of the device corresponded to the time to catastrophic failure of the metal films. Temperature gradients between 0-600°C were established along the length of the test bars and scanning electron microscopy (ISi SX40A SEM) was utilized for all microstructural analysis. Figure 2 shows the results of TFTC's tested under different oxidizing conditions . The thin film sensors tested here did not have a protective overcoat, and consequently, they experienced significant drift, the rates of which depended on the atmosphere and imposed temperature gradient (~ T). To determine if the composition of the platinum :1 Orhodium thermoelement is changing due to oxidation, an ESCA depth profile was performed on a 1 µm platinum:1 Orhodium film that was exposed to air for 1 h at 1000°c. This depth profile is shown in Figure 4. This can lead to a significant variation in junction chemistry and thus, emf output. Previous studies studies on the oxidation of platinum and rhodium indicated a mass loss directly proportional to oxygen pressure, with larger losses due to greater volatility associated with platinum oxides compared to rhodium oxides at the temperatures investigated [5].

The sensors tested in air ambients, with a 300°C
An SEM micrograph of a platinum:1 Orhodium thin film after exposure to nitrogen gas at elevated temperature is shown in Figure 5. After investigating these sources of drift, thin film sensors were fabricated with Al 2 0 3 overcoats in an attempt to minimize this degradation. Figure 8 shows the measured temperature of an overcoated sensor, relative to that of a reference wire thermocouple. The output of the TFTC and reference wire thermocouple were almost identical, as shown by the deviation of the TFTC from the reference thermocouple after annealing. There was no measurable drift in the signal at temperatures of 1000, 1100, 1200, and 1300°C. In addition, the lifetime of these _ overcoated sensors was extended beyond those prepared without overcoats. Microstructural analysis of the thin film sensor elements showed little porosity within the metal film and no evidence of thermal sintering (Figure 9). Although these films showed some blistering along the edge of lines where step coverage was poor, the majority of the film experienced little degradation.
The results to date indicate that there are two possible sources of emf drift; oxidation of the precious metal films and thinning of the metal lines by mass transport. The corresponding decrease in drift rate with lower oxygen partial pressures is consistent with the hypothesis that volatile platinum and rhodium oxides were forming and the metal lines were thinning as a result of this oxidation. Studies have shown that both platinum and rhodium are unstable in oxidizing ambients and form a volatile oxide species (Pt02 and Rh0 2 gas) at elevated temperatures [5]. This can change the local concentration of metal in the film and affect the emf signal by forming a dissimilar metal junction of considerably different composition. The continual loss of rhodium and platinum (differential loss) generated an emf at the junction that was considerably different from that of the type S thermocouple junction originally formed. Other researchers have used relatively thick films (2-20 µm), in an attempt to minimize this differential loss of metal [6][7][8][9][10][11][12][13][14].
However, more Ar gas can be trapped in the as-deposited films and thus, lead to a greater density of blister. s after higher temperature exposure. When the metal films are exposed to high temperatures, trapped argon gas within the film can coalesce to form bubbles. These bubbles can rupture and cause ultimate failure of the metal lines. In addition, thicker sensor elements can make it more difficult to produce a conformal, protective overcoat due to poor step coverage at the edges of the thin films.
High temperature exposure of unannealed thin film lines produced sintered microstructures, as shown in Figure 3. This thermal sintering behavior of the films can be explained on the basis of conventional sintering models. For example, platinum crystallites grow via the incorporation of platinum atoms from the surrounding amorphous matrix, leaving behind a thin membrane or pore as shown in Figure 3. Effectively, the material withdraws to attain the equilibrium shape characteristic of these particles in which metalsubstrate and metal-gas interfacial energies are approximately equal [4].
These microstructures are typical of a non-wetting material and the facets evident in the micrograph suggest that surface driven mass transfer has occurred, similar to that formed when material is transported through a vapor phase. This will increase the thin film resistivity and ultimately affect the voltage signal generated. Dewetting of the platinum films leads to a receding contact line and pile up of material in the center of the metal line due to the non-wetting behavior of these platinum base materials on oxide substrates. Ostwald ripening process or coalescence of crystallites [2], when the film is made up almost entirely of small crystallites.
Overcoats of high purity alumina can effectively reduce degradation due to oxidation and thermal sintering. The SEM micrograph of a 1.6 µm alumina overcoat ( Figure 9) minimized both surface transport of platinum and vaporization of the precious metals as oxides. The overcoat reduced the extent to which these processes dominate by forming an effective diffusion barrier. Here, the only evidence of blistering was around the edges of the metal lines. Since this area of the Al203 overcoat has the poorest coverage, it was more prone to blistering. This evidence supports the hypothesis that a short circuit diffusion path was formed for the out-diffusion of trapped argon in the metal film. Gas migration was restricted along the interphase boundary to locally thin regions of the overcoat. The locally thin overcoat ruptured when sufficient pressure in these bubbles had developed. Although the overcoat did not completely eliminate the problems described above, it did minimize the adverse effects of sintering and oxidation.

CONCLUSIONS
Thin film thermocouples have been fabricated with lifetimes up to 50 hours at 1300°C with no measurable drift at temperature. However, the accuracy of the TFTC was affected by two different degradation processes: (1) the formation of volatile metal oxides and associated compositional changes at the TFTC junction and (2) the surface driven mass transport of the precious metal films via thermal sintering . Aluminum oxide overcoats have been successfully employed to reduce the extent of degradation by these processes and thus improve TFTC performance.

Physical Vapor Deposition
The thin film thermocouples being developed in this study are multilayered vapor deposited devices. Not only are the thermocouple legs deposited by sputtering, but an insulating layer, to electrically isolate the device from the metal components, a bond coat layer, to allow for adequate adhesion of the dielectric layer and an overcoat, to protect the device, have to be sputtered deposited onto each component. Due to the wide variety of materials that must be deposited and the need for very dense film, sputtering is a more practical method for this work than the two other forms of thin film deposition: vacuum evaporation and chemical vapor deposition.
Sputtering is an ejection of atoms from a target due to the bombardment of this target by ions. The process is a result of the transfer of momentum from the bombarding ions to the ejected atoms during collision. This transfer of momentum is analogous to the cue ball hitting the rack of balls in a game of billiards. Of these atoms a certain percentage will be deposited onto the particular substrate that is to be coated.
A typical schematic of a sputtering system is shown in Figure 1. This system is actually contained in a high vacuum chamber. The chamber is first evacuated then argon gas is introduced. Argon is generally used since it is an inert gas that has a fairly high mass to allow for a high sputtering yield since the process is dependent upon the transfer of momentum and is relatively cheap.
One component not represented in Figure 1 is a tungsten filament which is used to initiate the sputtering process. Electrons are emitted from the filament and then collide with the argon atoms, thereby ionizing some of the gas atoms creating a plasma. The ions are then accelerated toward the target, since it is biased negatively with respect to the plasma, the biasing is generated by an external power supply. Four different outcomes may result when an ion impacts the target material. The possibilities are that a target atom may be sputter ejected, the ion may also become implanted in the target or reflected as a neutral, the ion impact may result in an alteration of the target surface, or finally a secondary electron may be ejected. Of these four possibilities only the ejection of a target atom or a secondary electron are helpful to the sputtering process. Of the target atoms that are ejected only a certain percentage of these atoms will come to rest on the substrate where they can condense. The secondary electrons that are ejected maintain the plasma by creating additional ions when they collide with argon atoms.
The mobility of the cations in the plasma is much slower than the mobility of the electrons; therefore, cations accumulate at the surface of the target creating what is known as a dark space. It is across this dark space the entire voltage between the target and plasma is dropped. When a secondary electron is ejected from the target it is accelerated across the dark space before it makes an ionizing collision. For this reason, the ground shield ( Figure 1) is used so no sputtering will occur from the sides or back of the target since this material would be wasted. This shield must be kept within the dark space region to insure no ion bombardments will occur in this area.

RF Sputtering
In the previous explanation of sputtering the target was negatively biased resulting in the accumulation of ions at the surface of the target. However, if the target is an insulator, the positive charge can not be neutralized, thereby resulting in the accumulation of charge at the surface of the target. If an alternating current is applied instead, energy could be transmitted into the plasma through the form of displacement current through the insulator. If this AC supply was of a low frequency almost the entire applied voltage would be dropped across the insulator. Therefore, the ions that do arrive at the surface of the target would not have obtained a sufficient energy to result in sputtering.
Therefore, it is necessary to use a frequency in excess of 50-100 kilohertz. Due to Federal Communication Commission regulations the lowest frequency that can be used is 13.56 megahertz; this frequency is within the range known as radio frequency or rf.
In order to accomplish rf sputtering the target must be bonded to a metal backing plate that acts as an electrode. When an rf power supply is used and an alternating voltage is applied to the electrode, the target is alternately Another significant difference between rf and de sputtering is that rf sputtering can be performed at much lower pressures. This is due to the fact that rf discharge is much more efficient in producing ions and electrons within the gas and is not solely dependent upon the ejection of secondary electrons from the target to sustain a glow discharge. This occurs since electrons are able to oscillate in an rf field thereby absorbing energy until it is able to make an ionizing collision. By operating at lower pressures, fewer sputtered atoms will be scattered since there are fewer atoms to collide with.

Bias Sputtering
Two inherent problems associated with sputtering are detrimental to the success of this work. First, sputtering is for the most part a line of sight deposition process. For example, if a material were sputtered onto a sample that had slots machined into it, the side walls of these slots would have hardly any material deposited onto them. Secondly, gases, whether inert or reactive, may be incorporated into the films as they are deposited. This is harmful to the process when extremely pure or dense films are required. One process, known as bias sputtering, can help to alleviate both of these problems.
Bias sputtering is performed in a similar manner to standard sputtering except that a negative potential is also supplied to the substrate. This results in sputtering of the substrate surface since it will also be bombarded with ions. As long as the target voltage is greater than the substrate voltage a net deposition will occur. Generally the voltage applied to the substrate is significantly smaller than that supplied to the target. The ions that bombard the substrate result in selective removal of loosely bound or absorbed impurities.
If a substrate contains surfaces only partially or indirectly exposed to the target, the deposition of the target material over the substrate will vary. As shown in the work of Seeman, if a substrate has grooves machined into the surface, the grooves can be coated as long as some portion of the groove can be deposited by line of sight. The portion coated by line of sight then acts as a secondary target to coat the sidewalls of the slot through bias sputtering (see Figure 2). Continued biasing generally results in a leveling of the deposited film.
One extremely important result of bias sputtering that has not been mentioned yet is that improved adhesion is quite often obtained. This is a result of cleaning the surface of the substrate in the preliminary stage of sputtering.
Since there is little to no deposited material on the substrate, ionic bombardment would result in the sputter ejection of the substrate material. This exposes fresh material and cleans the substrate surface. This operation is similar to sputter etching. In etching, a negative bias is applied to the substrate, but no bias is applied to the target; therefore , the substrate acts as the cathode.
In order to prevent the contamination of the target material, a movable shutter is placed between the substrate and target during etching.

Patterning of Curved Ceramic Blades
The patterning sequence used to pattern the NT154 stator blades is shown in Figures 1 through 4. The patterning sequence is as follows: One side of a NT154 stator blade is immersed in an epoxy solution to form a mold that will replicate the curved surface. The epoxy mold is allowed to harden and is then cut down the middle in order form a split mold so that it can be removed from the blade. The resulting mold and blade are shown in Figure 1. Heavy duty aluminum foil is then formed around the blade by first loosely placing the aluminum foil around the blade and then forcing the aluminum against the blade with the epoxy mold ( Figure 2). Prior to this step the thermocouple legs are cut out of the foil (Figure 3). The formed aluminum is then removed and super glue is placed on those areas of the blade not occupied by the sensor elements. The aluminum mask is then applied to the blade and bonded. The pattern can then be transferred to the blade. After sensor deposition, the aluminum foil can be removed by heating above 300°C to completely decompose and ash the super glue. Little or no r~sidue is left on the blade after the ashing process. The final result is a sputtered stator blade that can be seen in Figure 4.

Introduction
Parallel gap welding was chosen for lead wire attachment to sensor films on the ceramic substrates. Other methods such a ultrasonic welding, oven brazing, flame-spray metallic coatings, and conductive pastes were not chosen since previous studies have found them to be unreliable and time consuming [1 ]. Thermocompression bonding was found to have promise for easily formed noble metals, such as gold, platinum, and rhodium. However, this method is quite time consuming and adequate fixturing was not available [2,3].
The advantages of parallel-gap welding process for thin-film-to-lead connections are: (1) the instantaneous formation of the bond; (2) no extraneous material is added into the thermocouple circuit; and (3) the ability of the weld to withstand a temperatures as high as the metals in the thin film thermocouple itself.
The welding equipment consisted of a Unitek, . Before the welding test$ coupons were sputtered with the appropriate precious metal, the Si 3 N 4 substrates were first oxidized at 1300°C in air for 100 hours. This was followed by a concentrated HF etch (49%) to planarize the surface. Precious metal films were then sputtered in argon gas at 8 mTorr and 200 W. Pt and Pt:1 ORh wire, 0.003" (76 µm) diameter, were flattened prior to wire bonding ( Figure 1) to prevent rolling and increase the contact area (interfacial area) between the wire and the precious metal film. Thereafter, a methanol wash and a nitrogen blow dry was used to remove any oils or impurities prior to wire bonding.
Prior to Pt metal deposition, a 1 oooA layer of Al 2 0 3 was presputtered for adhesion purposes followed by a 4 µm thick film of Pt. Since Pt:1 ORh already has acceptable adhesion properties, a presputtered layer of Al 20 3 was unnecessary. Again, a 4 µm thick Pt:1 ORh film is sputtered. A thickness of 4 µm was chosen for both Pt and Pt:1 ORh films since they had the best strength properties and any destructive blistering was minimized.
After lead wires were attached, a pull test was performed perpendicular to the metal film with measured weights to determine the bond strength in (lb 1 /in 2 ).

Platinum Lead Wire Attachment
The welding tip used was a constant gap, 0.004", Unitip electrode (model # UTM224C). This tip gave excellent welds when first used. However, over time (approximately 700 welds), the strengths of the bonds decreased.
Excessive polishing decrease the welding tips effectiveness, probably by rounding the tip. Earlier welding tests were able to support much more weight than the later tests, using identical weldir)g parameters.
The loss of effectiveness from the same welding tip over a large number of welds concurs with previous studies. 3 Despite this, strong welds were still obtained with an older tip.
The optimal parameters used for Pt wire bonding with a new electrode tip were as follows: mode-voltage + current; applied force-900 g ; pulse duration-27 ms.
The optimal parameters used for Pt wire bonding with a heavily used electrode were: mode-voltage + current; applied force-550 g ; pulse duration-70 ms.
The applied voltage was observed to have the largest influence on the weld strength. Sections 2. 1 and 2. 2 discuss the welding voltages used. Figure 2 shows an SEM micrograph of a low voltage (0.50 volts) weld of a Pt wire to a Pt film that used a new electrode tip. From this micrograph, a fusion zone on the wire cannot be seen . However, a slight heat affected zone is visible on the Pt film around the weld tip compression area. This weld had poor strength, since little fusion of the wire was observed. When a pull test was attempted on these bonds, the wire easily pulled away from the film, and therefore it was determined that no real join was formed between the wire and the film.   Figure 4 was one of a few, at 0.60 volts and above, that did not stick to the electrode tips, since a majority of these welds did stick. The welds formed at this setting caused too much damage to the film, wire, and the electrodes.   Figure 7 shows SEM micrographs of Pt wire welded with a 0.85 volt setting, that also used an older electrode tip. A slight heat affected zone can be seen on the low magnification micrograph, which indicates there is some wire-film interaction present.

Platinum weld parameters and SEM micrographs
However, there is no fusion zone present, only a slight necking of the wire between the electrode tip contact points.

Platinum weld strengths
Welds formed using voltages <0.50 and >0.60 volts and a new electrode tip did not have pull tests performed on them due to their poor quality. However, a normalized strength of 8200 lb/in2 at 0.55 volts was observed to give the best results from all welds. Figure 8 shows how the film strength varies with applied voltage when a tip has been used over 700 times and was cleaned with Al 2 0 3 polishing disks repeatedly. It is apparent that strength decrease drastically with over usage.
The plot here shows that the best voltage employed for a heavily used tip was 0.70 volts. However, this parameter can vary slightly and still give similar results. It is apparent that this tip was not well suited for heavy use. In addition to the weld parameters being different, the Pt:1 ORh films had to be annealed prior to bonding. The film was heated to 1200°c for 1 hour to relieve incorporated stresses and relax the films. To further aid in bonding the wire, a large AC current was passed through it to remove any organics and oils on the wire surface. In an attempt to further enhance welding, an aqua regia wet etch of the wire and film was utilized to · remove any oxides and other impurities that may be present on the surface. However, this additional etching step showed no significant increase in welding strength.
3.1 Platinum:1 ORhodium weld parameters and SEM micrographs Figure 9 is a SEM micrograph of a weld formed using 1.1 volt setting. It was apparent that little or no fusion occurred in the wire. This weld was insufficient to carry a load since there was little diffusion between the wire and the film. Figure 10 has SEM micrographs of welds that used 1.3 volts. The fusion zone can be clearly seen on the wire. This weld voltage was found to give good strengths and does not compromise the wire or film strength . With this weld, however, there was only a small amount of contact area between the film and the wire and thus slightly lower strengths based on this may be anticipated. Figure 11 shows SEM micrographs of welds formed at 1.5 volts. This setting caused considerable flowing of the Pt:1 ORh wire, but had the best strength. This large increase in contact area correspondingly produced the excellent strength. Figure 12 shows SEM micrographs of welds formed at 1.7 volts. The micrographs show how the wire dewetted from the substrate due to the damage done on the film. The film damage from this voltage can also be seen in Figure   12 (Note that the area inside the blown out wire is not the Pt:1 ORh film, but the charging from the Si 3 N 4 substrate).

Platinum:10Rhodium weld strengths
The pull test described in the introduction was also done for Pt:1 ORh weldments. The plot of weld strength-vs-applied voltage is shown in Figure 13.
Although 1.7 volts showed good weld strengths, the SEM micrographs show that excessive damage was done in the process. In the long run, this damage may effect the emf data and output of the sensor. From these results it was determined that the 1.5 volt setting was the best for Pt:1 ORh lead wire attachment.                                   Voltage (volts) Figure 13 Burner rig testing of a type S thin film thermocouple A thin film thermocouple was sputter-deposited onto a 1 /4" x 1 /2" x 4" bar of NT154 that was designed to fit in the high temperature burner rig located at AlliedSignal Engines, Phoenix AZ. The fabrication of this sensor was identical to that previously described in manuscript I I. This sensor had a 2.2 µm sputtered Al20 3 overcoat and was affixed to the burner rig apparatus shown in  Figure 3. The failure was located at the edge of the hot zone of the flame, as indicated by the bluing of the S i3N4 bar. The area adjacent to the break (on the bond pad side) had been covered by the clamping fixture and thus, was not directly exposed to the flame.
Since this area was not exposed to the high temperature exhaust gases from the burner nozzle, the region was significantly cooler than the area exposed to the flame. Unfortunately, there was no wire thermocouple to monitor the temperature at this location . It was believed that this very large temperature gradient accelerated the failure of the device in this region, due to the differential expansion of the metal/ceramic couple. Since the covered region of the bar restricted expansion of the metal film in this direction, the film buckled and was removed from the bar by the flame. Earlier furnace tests of sensors showed a similar type of failure associated with an imposed temperature gradient but the consequences were not nearly as rapid as those observed in this case.
SEM micrographs of the failed region ( Figure 4) show a complete delamination of the metal film. In addition, microcracking of the Al20 3 overcoat along the edge of the metal pattern was observed. This observation seems to support the hypothesis that that expansion of the metal film was restricted by the sputtered Al20 3 overcoat. In order to relieve this stress, the metal film expanded along its length and thus, produced cracks along the edge. This buckling of the thin film is apparent in the micrograph in Figure 5.
With the exception of some slight cracking of the oxide overcoat, the films were continuous on and around the area where the flame was directed. Little crystallization of the Al20 3 overcoat was observed ( Figure 6).