MICROFABRICATED SILICON CARBIDE MICROENGINE STRUCTURES Kevin A. Lohner, Kuo-Shen Chen, Arturo A. Ayon, S. Mark Spearing Massachusetts Institute of Technology, Cambridge, MA 02139 ABSTRACT A research and development program is underway to develop technology for a MEMS-based microgas turbine engine. The thermodynamic requirements of power-generating turbomachinery drive the design towards high rotational speeds and high temperatures. To achieve the specified performance requires materials with high specific strength and creep resistance at elevated temperatures. The thermal and mechanical properties of silicon carbide make it an attractive candidate for such an application. Silicon carbide as well as silicon-silicon carbide hybrid structures are being designed and fabricated utilizing chemical vapor deposition of relatively thick silicon carbide layers (10-100 µm) over time multiplexed deep etched silicon molds. The silicon can be selectively dissolved away to yield high aspect ratio silicon carbide structures with features that are hundreds of microns tall. Research has been performed to characterize the capabilities of this process. Specimens obtained to date show very good conformality and step coverage with a fine (≈0.1 µm dia.) columnar microstructure. Surface roughness (Rq) of the films is on the order of 100 nm, becoming rougher with thicker deposition. Residual stress limits the achievable thickness, as the strain energy contained within the compressive film increases its susceptibility to cracking. Room temperature biaxial mechanical testing of CVD silicon carbide exhibits a reference strength of 724 MPa with a Weibull modulus, m =16.0. Keywords: CVD silicon carbide, microstructure, MEMS, residual stress INTRODUCTION A MEMS-based micro-gas turbine engine is being developed at MIT for use as a power-generation and propulsion source. The ‘microengine’ (Fig. 1) is a 2.1 cm diameter by 3 mm thick heat engine designed to produce 10-20 Watts of electric power or 0.05-0.1 Newton of thrust while consuming under 10 grams/hour of H2. Later versions may produce up to 100 W using hydrocarbon fuels [1].
Figure 1: Micro gas turbine generator cross-section [1]. Currently, the engine is being fabricated out of deep reactive ion etched (DRIE) silicon. This yields the high level of dimensional precision that is necessary for the high speed rotating turbomachinery to function successfully. However the thermo-mechanical properties of single-crystal silicon are insufficient for optimal engine performance. The yield strength at temperature limits the speed at which the spool can spin, while the engine’s longevity is limited by the high creep rates exhibited by silicon at temperatures above 1100 K. This dictates the need for developing a comparable micro-fabrication process for a more refractory ceramic material. For an alternate material to be incorporated into the process flow it must meet several criteria. First, the material must be able to be formed into high aspect ratio structures. The current design for the Materials Research Society Symposium Proceedings Series, Volume 546 (Fall 1998).
Page 1 of 6
microengine uses a 6-wafer bonded stack of individually etched wafers. Many of these wafers contain fine structures etched to depths of hundreds of microns. Tolerances of less than ± 1 µm are required in some critical areas of the engine. Second, the material must be compatible with the wafer-bonding sequence. Third, the material must have a homogenous microstructure free of mechanically significant flaws and voids. Inhomogeneities may lead to balancing and stress concentration problems during the high-speed operation of the engine. Finally, the material must offer considerable advantages over silicon in the areas of high-temperature specific-strength, creep and oxidation resistance. Silicon carbide is one such high-performance refractory ceramic that offers desirable thermal and mechanical properties (E=466 GPa, ν=0.21, ρ=3210 kg/m3) [2]. Its chemical stability and inertness, which help make it an excellent candidate, unfortunately also currently limit its ability to be etched to the same accuracy and depths as silicon. However, chemical vapor deposition (CVD) of relatively thick silicon carbide layers (10-100 µm) over etched silicon molds is a potential way to circumvent this limitation. The silicon can be selectively dissolved away to yield high aspect ratio silicon carbide structures with features that are hundreds of microns tall. Similar approaches have been proposed elsewhere [2,3]. Silicon carbide as well as silicon-silicon carbide hybrid structures are being designed and fabricated using the CVD process. Research is being performed to determine its feasibility as a potential candidate for the microengine –including its compatibility with the established processing criteria. Specific areas of focus include processing concepts, residual stress issues, conformality and microstructure. EXPERIMENT CVD Silicon Carbide Processing Experiments with CVD SiC began in February 1997. Preliminary characterization has been performed on CVD SiC produced by three different vendors. From considerations of surface quality and microstructure, CVD SiC produced by Hyper-Therm, Inc., was chosen as the most suitable candidate for further characterization and processing trials. The CVD SiC was produced by the thermal decomposition of vaporized methyltrichlorosilane (MTS) using hydrogen as a carrier gas at elevated temperature and reduced pressure according to the following chemical reaction: CH3SiCl3 + αH2 Æ SiC + 3HCl + αH2
(1)
where α is defined as the molar ratio of H2 to CH3SiCl3. MTS is the preferred precursor for depositing stoichiometric SiC because of the 1:1 atomic ratio of silicon to carbon. The deposition temperature was between 1000°C and 1300°C, which typically produces crystalline β-SiC with a cubic crystal structure and a theoretical density of 3.21 g/cm3. The deposition rate was between 2 and 10µm per hour [4]. Silicon carbide has been deposited on over one hundred 4” silicon wafer substrates as part of this study. They represent a range in Si wafer thicknesses (300-1000µm), substrate types (etched vs. plain), deposition thicknesses (10-100µm), and coverage (single sided vs. double sided). SiC Characterization The SiC-coated plain wafers have been used mainly to characterize residual stress, microstructure, and mechanical strength. A laser profilometer (Tencor Flexus 2320) was used to measure radius of curvature as a function of film thickness for the wafers with single-sided SiC coverage. Readings were taken over the central 50mm of the wafers. From these measurements, plane stress calculations were used to determine the stress state of the system. The surface roughness of the as-deposited carbide as a function of thickness was determined using atomic force microscopy (AFM). Scanning electron microscopy (SEM) was used to characterize the cross-sectional deposition thickness profile of representative wafers as well as to analyze the microstructure in terms of grain size and flaws.
Materials Research Society Symposium Proceedings Series, Volume 546 (Fall 1998).
Page 2 of 6
Mechanical strength under biaxial stress was tested on 1 cm square dies of 250µm thick CVD SiC. The dies were held in a biaxial fixture, and a force was applied in the center with a 1/64” steel ball. The fracture force was measured with a load cell and subsequently converted to a failure stress [5]. Strawman concepts In addition to the plain wafers that were SiC-coated and used for material characterization, several Si wafer substrates were etched with various microengine patterns prior to CVD SiC processing. This was for the purpose of demonstrating potential processing concepts that may be compatible with the microengine process flow. Concepts are being evaluated in terms of conformality, process integration, residual stress and the absence of cracking. Three generic process concepts are currently being investigated: the positive mold, negative mold, and hybrid configurations (see Fig. 2). The positive mold concept consists of depositing a thin (<10µm) layer of CVD SiC over an existing silicon microengine component (i.e. the turbine disc) with the aim of increasing its stiffness and resistance to creep and oxidation. The negative mold concept involves fabricating the inverse pattern of the desired component into the silicon substrate and then wholly or partially filling the mold cavity with thick CVD SiC. A solution of HF/ HNO3 can be used to selectively dissolve away the silicon mold to yield free standing silicon carbide structures. The negative mold technique has the advantage of being able to control the outer component dimensions and surface roughness while obtaining a shell or solid SiC structure. The hybrid concept entails using a combination of deposition and post-processing techniques to yield a microengine component that has SiC reinforcements in strategic locations to increase component strength, stiffness, and life above that which could be achieved with monolithic Si.
Positive Hub
Negative
Hybrid
Blades
Disc Key:
= SiC
= Si
Figure 2: Illustrations of positive, negative, and hybrid rotor configurations. RESULTS CVD Silicon Carbide Microstructure Hyper-Therm’s CVD SiC processing produces SiC with desirable microstructure from a mechanical property standpoint. The average grain diameter was measured to be 0.1-0.5µm growing in a columnar fashion radially outward from the substrate surface (Fig. 3). No voids or obvious defects/inhomogeneities were observed under SEM examination. Root mean squared surface roughness (Rq) for the films varied between 70nm for 15µm thick SiC to 115nm for 100µm thick SiC. Figure 4 shows a typical AFM output for a 65µm SiC film (Rq=87nm). The step coverage of the SiC in and over etched trenches and features was very conformal, as can be seen in Fig. 3. Both the positive and negative mold concepts yielded promising results (Fig. 5, Fig. 6, Fig. 7). The main issue that arose was residual stress-induced cracking as the SiC was deposited at thicknesses greater than 70µm. CVD SiC Residual stress Wafers that received single-sided deposition coverage exhibited a convex bow on the film side. This indicates that the film is under compression at room temperature. For 500µm wafers with nominal 15µm
Materials Research Society Symposium Proceedings Series, Volume 546 (Fall 1998).
Page 3 of 6
µm
Figure 3: Cleaved cross section of thick shell SiC blades showing conformal coverage.
Figure 5: 16µm SiC on silicon micro-turbine rotor, produced from a positive mold.
Figure 4: 3-D AFM plot of 65µm thick SiC.
Figure 6: Thick shell SiC micro-turbine rotor, produced in a negative mold.
single-sided SiC coverage, the average radius of curvature, R, was measured to be 3.2 m. For 500µm wafers with nominal 65µm thick SiC the average radius of curvature was 1.9 m. The residual stress state in a bilayer of isotropic materials can be considered as a superposition of an average stress in each layer and a bending stress gradient. The average stress in layer 1 is given by: _ _ E2 t23 + E1 t13 (2) σ1 = 6 R (t1 + t2) t1 and the bending stress gradient, assuming plane stress conditions, is given by: _ dσ (3) = -E dz R _ where E is the biaxial modulus for the layer (= E / 1-ν), and t1,2 are the thicknesses. It should be noted that (2) reduces to the well-known Stoney formula for cases where one layer thickness is much greater than the other. Equations (2) and (3) were used to estimate the stress state in the SiC / Si bilayers from the curvature measurements. These are shown graphically in Figure 8. It is apparent that the stress in the SiC decreases with increasing film thickness, however, the stress in the Si at the interface becomes increasingly tensile as the SiC thickness increases. At SiC thicknesses approaching 100µm it was observed that the silicon substrate generally cracked, presumably due to exceeding the critical strain energy release rate in the Si near the interface.
Materials Research Society Symposium Proceedings Series, Volume 546 (Fall 1998).
Page 4 of 6
200µm
200µm
Figure 7: Negative mold processing –Si being dissolved away from SiC blades. Insert shows a cross section through the blades. Calculated Residual Stress Profile [15µm CVD SiC]
Calculated Residual Stress Profile [65µm CVD SiC] deposited on 500µm Si wafer
40
40
0
0
-40
Stress (MPa)
Stress (MPa)
deposited on 500µm Si wafer
Si Stress (MPa) SiC Stress (MPa)
-80
-40
Si Stress (MPa) SiC Stress (MPa)
-80
-120
-120
-160
-160
-200
-200 0
100 200 300 400 Position Through Wafer (µm)
500
0
100 200 300 400 500 Position Through Wafer (µm)
(a)
(b)
Figure 8: Calculated residual stress profile for (a) 15µm and (b) 65µm SiC on a 500µm Si wafer. Mechanical Properties Room temperature biaxial mechanical testing of CVD SiC shows a reference strength of 724 MPa with Weibull modulus, m =16.0 (Fig. 9). This material was produced by a different vendor that deposits SiC at a higher temperature and faster rate –producing a coarser microstructure. It is expected that HyperTherm SiC will have even higher strength than this due to the finer grain size. In addition, reduction of the surface roughness is likely to further increase the strength. In order to estimate the effect of creep on the microengine operation, finite element calculations were performed using a power law creep model fitted to experimental data from several forms of silicon carbide [6]. Figure 10 shows the calculated results for an un-bladed disk of CVD SiC with a peripheral speed of 500 m/s at several operating temperatures. This shows that at 1600K, the radial growth is about 0.1µm after one day of operation. This should not seriously affect the sustained operation of the microengine.
Materials Research Society Symposium Proceedings Series, Volume 546 (Fall 1998).
Page 5 of 6
10 0 Radius = 2 mm
Estimated m = 16.0 σ0 = 724 MPa
20 10 5
Radial Growth (µm)
Failure Probability (%)
99 90 80 50
95% confidence
1
300
400
500
600
700 800 9000
Fracture Stress (MPa) Figure 9: CVD SiC Biaxial Strength Results.
Thickness = 300 µm Tip Speed = 500 m/s K 00 19 0 K = K T 00 180 T = T = 17
10 -1
10 -2
1 T=
10 -3 3 10
600
K
10 4
10 5
Operation Time (s) Figure 10: Calculated radial creep of a CVD SiC rotating disk at several operating temperatures.
CONCLUSIONS Positive, negative, and hybrid configurations appear to be feasible processing routes. They may be limited, however, to deposition thicknesses of less than 100µm, due to residual stress issues. The CVD SiC that Hyper-Therm produces conforms well to etched surfaces. The mechanical properties and microstructure are suitable for microengine applications. Surface roughness (Rq) of the films is on the order of 100nm, becoming rougher with thicker deposition. Room temperature biaxial mechanical testing of CVD silicon carbide shows a reference strength of 724 MPa with Weibull modulus, m=16.0. However, in order to implement SiC in devices as complex as the microengine, significant process development is still required. REFERENCES 1. A. H. Epstein, et al, “Micro-Heat Engines, Gas Turbines, and Rocket Engines –the MIT Microengine Project”, 28th AIAA Fluid Dynamics and 4th AIAA Shear Flow Control Conference, (June 1997). 2. T. Rogers, “Vapour Deposited Silicon Carbide”, IEE Colloquium on Extremely Hard Materials for Micromechanics, (April 1997). 3. N. Rajan, M. Mehregany, C.A. Zorman, and T.P Kicher, “Fabrication and Testing of Micromachined Silicon Carbide and Nickel Fuel Atomizers for Gas Turbine Engines”, Solid State Sensor and Actuator Workshop, (June 1998). 4. W. Steffier, Hyper-Therm, Inc., Huntington Beach, CA, (private communication). 5. K-S. Chen, A. Ayon, S. M. Spearing, “Tailoring and Testing the Fracture Strength of Silicon at the Mesoscale”, J. American Ceramic Society, (June 1997). 6. C. H. Carter Jr., R. F. Davis, and J. Bentley, “Kinetics and Mechanisms of High Temperature Creep in Silicon Carbide: II, Chemically Vapor Deposited,” J. American Ceramic Society, Vol. 67, pp. 732740, (1984).
Materials Research Society Symposium Proceedings Series, Volume 546 (Fall 1998).
Page 6 of 6