of finite element simulations. Furthermore environmental tests were done to investigate the temperature influence of the comb drive. Now we are working on the ...
Reliability issues of an accelerometer under environmental stresses P. Schmitt1,2, F. Pressecq1, G. Perez1, X. Lafontan4, J. M. Nicot1, D. Estève2, J.Y. Fourniols2, H. Camon2, C. Oudea3 1
CNES, bpi 1414, 18, Av. E. Belin, 31401 Toulouse Cedex 9, France LAAS - CNRS, 7, Av. Col. Roche, 31077 Toulouse Cedex 4, France 3 EADS – ST, 66, route de Verneuil, 78133 Les Mureaux Cedex, France 4 Nova MEMS, 14, rue du Quai, 09700 Saverdun, France 2
ABSTRACT COTS (Commercial-off-the-shelf) MEMS components are very interesting for space applications because they are lightweight, small, economic in energy, cheap and available in short delays. The reliability of MEMS COTS that are used out of their intended domain of operation (such as a space application) might be assured by a reliability methodology derived from the Physics of Failure approach. In order to use this approach it is necessary to create models of MEMS components that take into consideration environmental stresses and thus can be used for lifetime prediction. Unfortunately, today MEMS failure mechanisms are not well understood today and therefore a preliminary work is necessary to determine influent factors and physical phenomena. The model development is based on a good knowledge of the process parameters (Young’s modulus, stress…), environmental tests and appropriated modelling approaches, such as finite element analysis (FEA) and behavioural modelling. In order to do the environmental tests and to analyse MEMS behaviour, we have developed the Environmental MEMS Analyzer EMA 3D. The described methodology has been applied to a Commercial-off-the-shelf (COTS) accelerometer, the ADXL150. A first level behavioral model was created and then refined in the following steps by the enrichment with experimental results and finite element simulations. Keywords: Reliability, Modelling, Physics of Failure, Environmental Characterization, Accelerometer
1. INTRODUCTION MEMS components are very interesting devices for space applications, because they are light, small and consume little energy. These properties enable cost reduction and mission time increase. In order to introduce MEMS in space applications, two ways seem quite promising. The first one is the replacement of conventional sensors, such as gyroscopes, accelerometers, pressure sensors by MEMS devices, the second one is the design of new space system based on MEMS.1 2 A MEMS device used in space applications has to operate under difficult conditions (reduced observability, impossibility to repair it, …) and in a hostile environment (temperature cycles, irradiation,…). One possibility is to use MEMS that have been designed for space applications. These devices are developed for one specific mission, all the space specific constraints have been taken into consideration since the first design step and their reliability is hence more or less ensured. Their disadvantage is that the development times are long and that they are expensive due to the reduced number of fabricated pieces. The second possibility is to use Commercial-off-the-shelf (COTS) MEMS. These MEMS are designed for mass applications, such as the automotive or information technology market and sold in big quantities. They are therefore cheap and due to their wide spread use a lot of empirical knowledge is available. But they are not designed and tested for space applications and their reliability under space specific constraints such as irradiation cannot be ensured without any further investigation. In the following we will present a methodology that serves to ensure the reliability of a COTS MEMS device in an operating environment that is different from the operating environment for which it has been designed. The methodology is derived from the Physics of Failure (PoF) approach and aims to estimate, by means of simulations, the component lifetime for a particular mission profile. The methodology will also be helpful to define adequate screening
22
Reliability, Testing, and Characterization of MEMS/MOEMS III, edited by Danelle M. Tanner, Rajeshuni Ramesham, Proceedings of SPIE Vol. 5343 (SPIE, Bellingham, WA, 2004) · 0277-786X/04/$15 · doi: 10.1117/12.524575
tests. However, MEMS failure mechanisms are not well understood today. A preliminary work is necessary to determine influent factors and to model physics of phenomena. The model development is therefore based on three principal columns: material characterization, environmental tests and the choice of an appropriated modeling approach. This methodology has been applied to a typical MEMS COTS component: an accelerometer for automotive applications, the ADXL150 from Analog Devices. A first level behavioral model was created and then refined in the following steps. Accelerometer data have been acquired by the use of characterization tools. Based on the obtained results a finite element model has been created and used to refine the behavioral model of its mechanical part. Furthermore we have done environmental tests on the accelerometer. In the first paragraph of this paper we will present the tool that we use to do environmental tests on MEMS devices. It has been developed by CNES and MEMSCAP and is able to couple static and dynamical studies of MEMS with environmental tests. In the second paragraph we will present our reliability approach and in the third paragraph we will present the application of this approach to one case study in the fourth paragraph: the ADXL150.
2. DESCRIPTION OF THE APPARATUS In order to study MEMS failure mechanisms, the CNES and MEMSCAP have developed the environmental MEMS Analyzer EMA 3D that is able to couple dynamical studies with environmental tests. This system is based on an interferometry microscope for optical profilometry and on environmental chambers.
Fig. 1: Environmental chamber coupled to optical profilometer
Optical Profilo/Vibrometer An optical profilometer acquires the 3D profile of a surface without contact. The CNES uses a ZoomSurf 3D developed by Fogale Nanotech.3 This tool couples both profilometry and vibrometry capabilities. The profilometry mode is based on the white light fringe scanning interferometry principle. In this mode, the interferometer produces a luminous focal plane that illuminates iso-altitudes of the device under test (DUT). The 3D profile is obtained by scanning vertically the device with the light focal plane and by plotting iso-altitudes curves as a function of altitude. Spatial resolution of the profilometer is lower than 1 µm in the XY directions and lower than 10 nm in the Z direction. In the vibrometry mode the optical profilometer is coupled to a stroboscope. It allows the acquisition of 3D profiles of periodic movements. Illumination peaks as low as 10 ns enable 3D profiles of MEMS vibration modes up to 1 MHz with a detection limit of 5 nm. For flat membranes, the monochromatic mode can measure automatically the vibration spectrum. For this, a mechanical frequency sweep is applied to the membrane and the vibrometer detects by real time FFT peaks of vibration.
Proc. of SPIE Vol. 5343
23
Environmental Chamber Environmental chambers are used to characterize the behavior of unpackaged MEMS exposed to various environments (temperature, humidity, pressure, vacuum, and gas nature). The environmental chamber, designed by SMS4 and modified by CNES and MEMSCAP, is conceived to control temperature and humidity. This can be useful to quantify the sensitivity of MEMS to humidity and to determine maximum humidity levels. Its temperature performances range from 15°C to 40°C and the humidity rate can be chosen from 8% to 95%. The chamber is equipped with electrical connectors and optical measurements are carried out through a dedicated window.
3. RELIABILITY APPROACH Our reliability approach is based on the Physics of Failure methodology. The Physics of Failure approach (PoF) has been used for decades in the civil engineering domain for the construction of unique products such as buildings or bridges. During the last years CALCE (University of Maryland) has applied it with some success to electronic modules. The principal advantage of this approach is that it incorporates reliability aspects in the design process and enables that way reliability enhancement or even assessment already during the development process of a product. 5 The PoF approach is based on the idea that failure is governed by physical and chemical processes. A good knowledge and understanding of these processes enables an identification of potential failure sites, modes or mechanisms and gives the designer the possibility to ameliorate these weak points of the design and to make the final product more reliable. The failure identification is done by the simulation of models suitable for the PoF approach. The simulation tools take as input parameters the environmental and operating conditions of the component. The simulation determines the various stresses (mechanical, thermal,…) that are acting on the component during operation in its environment. Based on the stress results, failure sites can be identified and a reliability assessment can be done.6 In the MEMS domain this methodology would introduce reliability early in the design process of a MEMS based system and enable us to increase reliability already at this stage. However, MEMS failure mechanisms are not well understood today. A preliminary work is necessary to determine influent factors and to model physics of phenomena. The model development is therefore based on three principal points: environmental tests, material characterization, and the choice of a suitable modeling approach. Environmental tests are necessary to determine the influence of environmental changes on the MEMS behavior. They can be done with specific tools, such as the environmental MEMS Analyzer EMA 3D presented in the second paragraph. 7 Material characterization is necessary to identify possible failure sources and to determine precisely the material properties that are used in the modeling and simulation tools. The material properties vary depending on the fabrication process and can be measured or extracted by conventional technological analysis tools (FIB-SEM), test structures or state of the art micro-characterization tools and methods (optical profilo-vibrometer & micro-indenter). The use of modeling and simulation tools is used for the rapid and cheap estimation of the impact of variations to the MEMS component and the system containing it. The variations can be intrinsic (i.e. technology parameters, geometries, etc.) and external (temperature, gazes, etc.) to the MEMS component. Moreover the use of modeling and simulation tools can increase the reliability of a MEMS by modeling failure mechanisms and developing tests by faults injection. 8 The modeling and simulation tools for MEMS have to deal with complex multidisciplinary systems. All the time, a compromise is necessary between models accuracy and the time of development and simulation. There are two main approaches: finite element analysis (FEA) and behavioral modeling in VHDL-AMS. FEA is used on the physical level for MEMS component simulation. It is able to couple different physical domains (e. g. mechanical, thermal, electrical, fluidic, etc.) and very exact, but unfortunately also very processor and time consuming. The behavioral modeling approach is used for MEMS system simulation and can be used to evaluate the reliability of a MEMS at system level by fault injection. Behavioral models are based on analytical equations and/or experimental results and can be refined by FEA simulations. 9
24
Proc. of SPIE Vol. 5343
In the following paragraph a case study of a MEMS COTS accelerometer, the ADXL150, is presented. A first level behavioral model was created and then refined by characterization results and finite element simulations. Moreover, environmental tests have been performed.
4. CASE STUDY: ACCELEROMETER The used accelerometer is an ADXL150 from Analog Devices. 10 It has been commercialized in 1996, is used in automotive applications and is a very successful MEMS COTS device. It is an uni-axial accelerometer, based on a capacitive measurement principle. The whole sensor is monolithically integrated on one die. Like all MEMS devices, it can be divided in a mechanical and an electronic part. The mechanical part is made of a comb drive located in the center of the die and detects the applied acceleration. The electronic part is arranged around the mechanical one, it treats the information and transforms it into a user friendly output.
Mechanical part
Electronic
Fig. 2: Optical microscope image of an ADXL150
In the following paragraphs we will present the models that we have developed for this accelerometer and the characterization and experiences that we have carried out. In a first step, we developed a first level behavioral model of the mechanical part of the accelerometer. The model development was based on the knowledge of the physical laws that describe its function, material properties coming from literature11 and some observations done with an optical microscope. Then this first level behavioral model was refined by further characterization, FEA simulations and environmental tests.
VHDL-AMS behavioral model
external acceleration
C1 mass
C2
mass motion
Fig. 3: Damped mass-spring model
The first level behavioral model is based on a spring-mass-damper model that describes the displacement x of the mobile part as a function of the external applied acceleration aext: 12
Proc. of SPIE Vol. 5343
25
m&x& + Dx& + Kx = ma ext + Fel1 + Fel 2 where x is the displacement of the mobile part, m is its mass, D the damping coefficient, K the spring stiffness and Fel1 and Fel2 the electrostatic forces applied to the self-test fingers. The U-spring stiffness can be calculated by considering the spring as an assembly of beams (Fig. 4) on which a force Fx is acting. By using Castigliano’s second theorem, we obtain k = 6 N/m.13
Fx Fig. 4: Modeling of the spring as an assembly of beams
The damping coefficient can be calculated as a combination between Couette flow, Stokes flow and Hagen-Poiseuille flow: 12
D = µ (A pm + 0.5 At + 0.5 Ab )
1 d f
+
1
δ
+ Nf
7.2µl
t g
3
where Apm is the area of the proof mass and the comb fingers, At the area of the spring and Ab the area of the combs. The spring and the combs travel on average at half the velocity of the proof mass. df is the air film thickness between proof mass and substrate, δ the penetration depth, µ the viscosity of air, l the finger length, t the finger thickness, g the gap between two fingers and Nf the number of fingers. As already said the model is based on a capacitive measurement principle, the mobile and the fixed combs form differential capacitances
C1 =
ε0 A g+x
C2 =
ε0 A g−x
where A is the surface of all combs and ε0 the dielectric constant. The parameters have been acquired by optical observation and from literature. The model has been programmed in VHDL-AMS. begin vel == x’dot; acc == vel’dot; force x)));
==
K*x+vel*D+m*acc+(eps0*A/2.0)*((vsup/(g+x))*(vsup/(g+x))-(vinf/(g-x)*(vinf/(g-
isup == (A*eps0/(g+x))*vsup’dot; iinf == (A*eps0/(g-x))*vinf’dot; end architecture behav; Fig. 5: Part of the VHDL-AMS model
26
Proc. of SPIE Vol. 5343
Characterization In order to refine the first order behavioral model, it was necessary to acquire some more information about the accelerometer. Several analysis techniques were used. First, the comb drive geometry was determined by SEM measurements (Fig. 6).
Fig. 6: SEM measurements
After µsection by polishing the comb drives thickness was also determined by SEM measurements. The thickness of the mechanical part was determined to be 1.66 µm and its distance from the substrate to be 1.52 µm. In a last and most complex step the Young’s Modulus of the comb drive was determined by nanoindentation.
Capacitor for nanoindentation
Fig. 7: Accelerometer before nanoindentation
Nanoindentation is a material characterization technique and used to determine Young’s Modulus, the hardness and the stiffness of a material or beams. In order to carry out a nanoindentation measurement, a probe tip is pushed into a material and the mechanical parameters can be calculated dependent on the required force and the indentation depth. One basic prerequisite for a successful nanoindentation is a test layer with a stable base.11 In the case of this accelerometer it was impossible to find such a point on the comb drive itself, because the connection to underlying layers is limited to some vias. Therefore we tried to find the same layer of polysilicon elsewhere on the die with a stable base and sufficient place to do some measurements. This same layer could be found on the capacitors of the ADXLs electronic. After a FIB microsection, the different layers were determined and the polysilicon layer was accessed by successive plasma and wet etching steps. The Young’s Modulus was determined on two capacitances by
Proc. of SPIE Vol. 5343
27
nanoindentation and found to be between 140 and 150 GPa (Fig. 8). This value is a little bit lower than the values found in literature. 160 150 140
Young's Modulus (GPa)
130 120 110 100 90 80 70 60 50
Capacitor one Capacitor two
40 30 20 10 0 0
10
20
30
40
50
60
70
80
90
100 110
120 130 140 150
160 170 180
190 200 210 220
230
Displacement Into Surface (nm)
Fig. 8: Nanoindentation’s results
Experiences a) Electronic/Mechanical decoupling The electronic of the ADXL150 is sophisticated and hampers us to evaluate directly the reaction of the mechanical part due to an applied external influence. This observation is only possible if we accomplish to access to the mechanical part without passing by the integrated electronics and by measuring the capacitance change with a known electronic.
Fig. 9: Accelerometer modified with FIB
In order to do the mechanical/electronic decoupling, we identified by optical inspection the signal lines connecting the comb drive to the electronics. There is a total of five connecting lines: two lines connected to the fixed and one to the mobile electrode, all three are used for capacitance change measurement. Two other connections access the self-test combs and can be used to apply a voltage. The separation of electronic and mechanical part was done by cutting the corresponding lines with a focused ion beam tool (FIB). The FIB is also used to remove the passivation from the signal lines and to enable thus the access to the comb drive.
28
Proc. of SPIE Vol. 5343
Self-test combs
Capacitance measurement
Fig. 10: Different accesses to the comb drive
We dispose of a capacitance detection that is able to determine very small capacitance changes and that we will connect to the three corresponding signal lines of the accelerometers’ comb drive. The self-test fingers will be connected to a voltage source in order to actuate the accelerometer.
b) Temperature Tests In order to investigate the influence of an environmental stresses on the accelerometer comb drive, it was exposed to some temperature variations and its behavior was observed with an optical profilometer. No variation could be observed.
Fig. 11: Optical profilometer image of a part of the combdrive
Finite element model / VHDL-AMS behavioral model The SEM measurements gave us the exact geometry of the comb drive. The spring constant analytically calculated was now found to be 4.8 N/m. The determined geometry, as well as the results from a µsection, were used to create a finite element analysis (FEA) model of the accelerometers’ comb drive and to carry out some simulations. In the mechanical simulation, the displacement of the movable part dependent on the applied acceleration was observed. The stiffness of the spring could be extracted and compared to the analytical result and was found to be 3.5 N/m.
Proc. of SPIE Vol. 5343
29
In other simulations the four first resonant frequencies were determined. For a Youngs’ Modulus of 140 GPa, we obtained the following resonant modes: 22.9 KHz - 29.7 KHz - 57.3 KHz -70.3 KHz. For a Youngs’ Modulus of 150 GPa, the resonant mode frequencies slightly change: 23.7 KHz - 30.75 KHz - 59.3 KHz -72.7 KHz. Work in progress We are still working on the connection between the COTS comb drive and the capacitance detection. The problem is to reduce the stray capacitances that appear due to the connecting lines. They can be reduced by small cable lengths and adapted connections methods.
5. CONCLUSIONS The successful use of MEMS COTS in space applications is closely linked to reliability aspects. In this paper we presented a reliability methodology derived from the Physics of Failure approach. It aims to estimate, by means of simulations, component lifetime for a particular mission profile. MEMS failure mechanisms are not well understood today and thus model development is based on material characterization, environmental tests and the use of appropriated modeling tools. In order to do the environmental tests CNES and MEMSCAP have developed a tool that couples static and dynamical studies of MEMS with environmental tests, the environmental MEMS Analyzer EMA 3D. The presented methodology has been applied to a case study, the MEMS COTS accelerometer ADXL150 from Analog Devices. A first-level behavioural model of its mechanical part has been programmed in VHDL-AMS and and was refined by the use of characterization techniques, such as SEM measurements and nanodindentation, as well as the use of finite element simulations. Furthermore environmental tests were done to investigate the temperature influence of the comb drive. Now we are working on the mechanical/electronic separation of our COTS accelerometer to be able to connect the mechanical COTS part to a different electronic.
ACKNOWLEDGEMENTS The authors would like to thank EADS-CRC (Ottobrunn, Germany), and especially Mr. Martin Kluge, for their cooperation.
REFERENCES 1. F. Pressecq et al., “MEMS and microtechnologies for space applications”, 1st CANEUS Workshop, Montréal, Canada, August 2002. 2. H. Helvajian (editor), Microengineering aerospace systems, Chapter 2, Aerospace Press, El Segundo (USA), 1999 3. Fogale Nanotech: http://www.fogale.fr 4. Surface measurement systems: www.smsuk.co.uk 5. R. Valentin et al., “Virtual Life assessment of electronic hardware used in the advanced amphibious assault vehicle (AAAV)”, Proceedings of the 2002 Winter Simulation Conference, Vol. 1, p. 948-953 6. M. Pecht et al., “Physics of Failure: an approach to reliable product development”, Integrated Reliability Workshop, p. 1- 4, 1995 7. X. Lafontan et al., “Sensitivity of RF MEMS to environmental stresses”, MEMSWAVE, Toulouse, France, July 2003 8. B. Courtois et al., “Fault Simulation of MEMS using HDLs”, Spie Symposium on Design, Test and Microfabrication of MEMS/MOEMS, Vol. 3680, p. 70 - 77, Paris, France, April 1999 9. P. Schwarz, “Physically oriented modeling of heterogeneous systems”, 3rd IMACS Symposium of Mathematical Modeling (Mathmod), Wien, 2. - 4. Feb. 2000, Vol. 1, p. 309 -318 10. www.analog.com 11. S. Rigo et al., “Correlation between X-ray microdiffraction and a developed analytical model to measure the residual stresses in suspended structures in MEMS”, Microelectronics Reliability, Vol. 43 , p. 1963-1968, 2003 12. Y. Zhou, “Layout Synthesis of accelerometers”, Master of Science Project Report, Carnegie Mellon University, 1998 13. G. K Fedder, “Simulation of Microelectromechanical Systems”, PhD Thesis, University of California at Berkeley, 1994
30
Proc. of SPIE Vol. 5343
