- Email: [email protected]
A hollow stiffening structure for low-pressure sensors
Sensors and Actuators A 160 (2010) 35–41
Contents lists available at ScienceDirect
Sensors and Actuators A: Physical journal homepage: www.elsevier.com/locate/sna
A hollow stiffening structure for low-pressure sensors P.K. Kinnell ∗ , J. King, M. Lester, R. Craddock GE Sensing & Inspection Technologies, Fir Tree Lane, Groby, Leicestershire, UK
a r t i c l e
i n f o
Article history: Received 15 December 2009 Received in revised form 18 February 2010 Accepted 12 March 2010 Available online 18 March 2010 Keywords: Pressure sensor Electrochemical etch Hollow structure Boss Diaphragm Silicon
a b s t r a c t This paper presents a novel process for producing thin-walled hollow stiffening structures on thin silicon diaphragms using an electrochemical etch-stop process. Examples of structures produced using the method are presented together with focused ion beam (FIB) analysis of critical areas within the structure. These demonstrate the integrity of the structures and show that the process is suitable for use in MEMS sensor applications. Using this process a 30 mbar full-scale differential pressure sensor has been demonstrated, and used to verify the suitability of these hollow structures for use in MEMS sensors. The novel process allows for increased sensor performance, with reduced die size. Details of the pressure sensor design and characterization are presented, showing a device with 18 mV/V full-scale output with linearity <0.4% (terminal base non-linearity). © 2010 Elsevier B.V. All rights reserved.
1. Introduction Piezoresistive silicon pressure sensors typically consist of a wheatstone bridge of piezoresistors fabricated on a square silicon diaphragm. The sensitivity of these sensors will be inversely proportional to the square of the diaphragm thickness, and directly proportional to diaphragm area [1]. To fabricate higher sensitivity devices the designer is forced to increase die size or reduce diaphragm thickness. Cost constraints often lead the designer towards reducing die size. Therefore the approach taken is generally to reduce diaphragm thickness in order to meet the required sensitivity. Silicon diaphragms for MEMS pressure sensors are typically fabricated by wet anisotropic etching of a cavity into bulk silicon, and to achieve a required diaphragm thickness requires careful timing of this etch. Even if great care is taken to continually monitor the etch in order to achieve the desired thickness, process variation from wafer to wafer, and from die to die within a wafer will ultimately dictate the minimum practicable thickness. Typically diaphragms thinner than 40 m prove challenging for many manufacturing environments [2]. To overcome this problem there are a number of etch-stop methods that may be employed to automatically stop or limit the anisotropic etching once the required thickness has been reached. Examples of such techniques are, the use of silicon on insulator wafers [3], the boron etch-stop process [4], or an electrochemical etch-stop [5,2]. In this work the electro-
∗ Corresponding author at: GE Sensing, Silicon Engineering, Fir Tree Lane, Groby, Leicestershire, LE60FH, UK. Tel.: +44 1162317507. E-mail address: [email protected] (P.K. Kinnell). 0924-4247/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.sna.2010.03.024
chemical etch-stop technique is used to form the basis of a novel process that firstly addresses the issue of increased sensitivity by allowing thinner diaphragms, but also allows for additional structuring of the diaphragms to improve other device performance characteristics such as output linearity. As high sensor linearity is often a very attractive sensor characteristic, flat silicon diaphragms are generally modified with the addition of a lump or ‘boss’ structure to stiffen the centre of the diaphragm. Fig. 1 shows a micrograph of a typical pressure sensor diaphragm with a central stiffening boss, the cross-section marked on the micrograph can be seen as a schematic in Fig. 3b. These features improve linearity by limiting strain stiffening of the diaphragm, which is a significant cause of output non-linearity [6]. The addition of such stiffening bosses is often done at the expense of other device parameters such as die size. Fabricating the boss on the diaphragm generally increases die size, and the additional mass that is suspended will tend to cause increased acceleration sensitivity. This acceleration sensitivity becomes especially critical as higher sensitivity pressure sensors are fabricated. This is because the diaphragm that is supporting the boss becomes increasingly flexible, thus any inertial loads imposed on the boss will result in a greater deflection of the diaphragm and thus be seen as a more significant proportion of the sensor output. To address these issues this work presents a novel fabrication route that allows for a wide variety of hollow structures to be fabricated on the surface of a thin silicon diaphragm. These structures may be used, in effect, as hollow bosses that have considerable stiffness relative to the diaphragm, yet due to their hollow construction add very little mass to the diaphragm (see Fig. 3c). The diaphragm and hollow bosses are produced using an electrochemical etch-
36
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
Fig. 1. Micrograph of an etched pressure sensor diaphragm with a solid lump at the centre, the diaphragm is approximately 1.5 mm square.
Fig. 3. Cross-sectional schematic view of a boss etched into a wafer follow (1 1 1) crystal planes.
stop technique and therefore this allows for controlled etching of thin diaphragms that are <10 m in thickness. The combination of these features allows the advantages gained by a stiffening boss in terms of sensor linearity, without either the negative affects of inertial sensitivity, or unnecessary increase in die size that may result from traditional fabrication techniques that produce solid bosses. For example the increase in die size comes about due to the constraints imposed by standard processing methods. These typically require that the boss is patterned on the surface of a wafer then etched down following (1 1 1) crystal planes to reveal pyramid shaped bosses with sloping sidewalls of 54.7◦ as shown in Fig. 2. Etching bosses in this way means the designer is constrained to use bosses that may be taller or wider than required. This has the adverse affect of removing active diaphragm area and therefore leads to an increased die size, as the designer has to compensate by increasing the diaphragm area. The hollow boss process is not limited by these constraints and allows boss heights independent of wafer thickness. In the following sections the innovative process that was developed to fabricate hollow bosses will be described, demonstrating that in principle the technique may be used to create a wide range of generic hollow structures suitable for many MEMS applications. A specific example of the technology is then given in the form of a piezoresistive pressure sensor. A description of the design and fabrication work carried out to develop test samples of a prototype pressure sensor are detailed. Finally, analysis and characterization are presented that verify the performance of the pressure sensor and validate the feasibility of the hollow structures for use as mechanical elements in MEMS sensors.
Fig. 2. Cross-sections of three types of pressure sensor diaphragm, type a is the most basic form of diaphragm that is completely flat, type b is the traditional form of diaphragm that contains a solid lump, and type c is a hollow version of the solid lump.
2. The hollow boss process In order to create thin-walled hollow structures using wet etching in a KOH solution the hollow boss process has been developed. An overview of the process is shown in Fig. 4 and it consists of five main process steps. Firstly a p-type silicon wafer is prepared and suitably patterned for etching (step 1); then a negative of the required boss is etched into a silicon wafer (step 2); the etched feature is implanted to render a layer of n-type silicon in the p-type substrate (step 3); a second n-type layer is then fusion bonded over the previously etched and implanted features (step 4); finally the p-type substrate material is etched away using an electrochemical etch-stop process to reveal a hollow structure. The electrochemical etch-stop, which is a key part of the process, results from the application of an anodic potential to silicon in OHcontaining solution that causes the formation of silicon oxide on the silicon surface, known as passivation. The process demonstrated here utilises the difference in passivation potentials of n-type and p-type silicon to generate an automatic etch-stop process that can preserve a hollow boss structure [5,7]. A fixed voltage, between the n-type and p-type passivation potentials, allows the p-type to etch, but causes passivation on reaching the n-type layer, marked by the cessation of bubble evolution at the etch surface. Since silicon oxide etches at a rate of approximately 100th that of silicon n-type and is self-sustaining under the cell potential, the integrity of the etch-stopped layer may be preserved for many times the lifetime of a standard sacrificial oxide. The electrochemical etch-stop was set-up as shown in Fig. 5. To calculate the correct passivation voltage, data showing the relationship between current and voltage was collected for wafers and electrodes being used, see Fig. 6. The maximum current reached on the chart shown in Fig. 6 corresponds to the passivation potential, which was 1.05 V. A power supply (indicated by PS in Fig. 5) with fixed anodic potential of at least 1.2 V was connected to an aluminum contact on the n-type layer of the silicon wafer, such that the wafer potential was always above the n-type passivation potential. The n-type surface was wax-bonded onto glass for protection of the sensing structure. A nickel plate was inserted into the solution and formed the cathode. The p-type silicon etched in normal anisotropic fashion in the aqueous KOH solution, as it was well below the p-type passivation potential. The electrochemical etch-stop ensured passivation of the hollow structure and allowed it to be fully revealed as the etch progressed.
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
37
Fig. 4. Process flow to create hollow boss structure.
Fig. 5. Diagram of electrochemical etch-stop equipment set up.
In this way n-type structures were preserved to full integrity during a wet anisotropic etch, evidenced by the fabrication of the hollow boss structure in Fig. 7. This micrograph shows that complete removal of the p-type silicon was obtained, leaving the n-type layer with the buried hollow boss structures bonded to it. The hol-
low boss structure shown in Fig. 7, is approximately 400 m square, 80 m deep with a wall thickness of 3 m. These relatively large very thin-walled structures were seen as a good test of the ability of the process to produce well-formed hollow bosses. Of particular interest was the area of intersection between the diaphragm and the walls of the hollow boss as shown in Fig. 7a, an arrow indicates the point of interface between the boss and diaphragm noting that the wall thickness at this interface is approximately 3 m. If these areas are not fully bonded together then the use of the hollow bosses as structural elements in sensors may be prohibited due to poor mechanical stability. The structural integrity of the hollow bosses were analysed using a combination of focus ion beam (FIB) and scanning electron microscopy (SEM) techniques. Using the FIB system a section of the hollow boss was removed by ion milling such that the interface between the walls of the boss and the diaphragm could be observed. A magnified image of this cut section is shown in Fig. 7b, from which it can be seen that there are no voids or other structural faults that may lead to poor performance. The hole that was cut in the flat section at the top of the boss was used to measure the top layer thickness. This was found to be approximately 2.4 m indicating the sidewalls at the base of the boss are slightly thicker than the flat top of the boss. It is expected that the top of the boss will be slightly thinner that the base due to the etch rate of the silicon under etch-stop conditions, which is set by the passivation oxide etch rate. This structure demonstrates that large thin-walled structures may be created with this process. In this work the negatives used to create the hollow bosses were created using anisotropic etching, hence the pyramidal shape. These negative structures can also be created using other etching techniques. For example, isotropic etching or deep reactive ion etches may be used to create bosses with either rounded or vertical sidewalls, depending on the required application. 3. Test sensor design
Fig. 6. Chart showing relationship between current and voltage for the electrochemical etch-stop process.
The hollow structures detailed above demonstrate that hollow bosses may be manufactured, however, to determine if these structures may be used as part of MEMS sensors a low-pressure sensor was designed and fabricated using hollow bosses and the process described above. A low-pressure sensor was specifically chosen as this type of device benefits from the controlled thin diaphragms given by the hollow boss process, as well as optimal sensitivity and linearity due to the addition of hollow bosses. The low-pressure sensor was designed to have a target full-scale pressure in the range of 20–30 mbar. This is based on achieving an electrical output of approximately 20 mV/V output at full-scale pressure. A lowpressure sensor was chosen to demonstrate this process, as this is a challenging area of design space for pressure sensors. It would benefit from the combination of an electrochemical etch-stop to control diaphragm thickness with the hollow boss process to optimise sensor performance.
38
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
Fig. 7. (a) Focused ion beam section cut from a hollow boss and (b) close-up of cut section that is indicated by the arrow shown in (a).
The pressure sensor was designed with a diaphragm of 1550 m square and 5 m thick, onto which three hollow boss structures 80 m high were fixed. The wall thickness of the hollow bosses was designed to be approximately 10 m; the overall geometry of the diaphragm can be seen in Fig. 8. As mentioned above, the most basic pressure sensor would comprise of a simple flat diaphragm (see the cross-section shown in Fig. 3a). In this work three bosses have been added to the diaphragm to improve sensor performance. Adding the three bosses to the diaphragm is effective because it stiffens the diaphragm in the regions of the bosses but keeps the diaphragm flexible elsewhere. This means that when pressure is applied to the diaphragm stress is concentrated in the areas that are kept flexible between the bosses. To illustrate this Fig. 9 shows an exaggerated deflection plot resulting from pressure applied to the lower side of the diaphragm (created using Ansys FEA software). From the plot it can be seen that the bosses remain rigid with the areas of diaphragm between the bosses and the frame taking up all the bending and therefore developing maximum tensile and compressive stresses. The arrows in Fig. 9 indicate the areas of maximum and minimum stress situated between the lumps. A close-up of this area is also shown in Fig. 10, in this plot the areas of uniform stress that have been created between the bosses can be seen. This is a key advantage of this type of diaphragm over a conventional flat diaphragm. Uniform stresses have thus been created between the bosses aiding the positioning of piezoresistors to sense pressure-induced stress. The resistors would therefore be positioned in these regions as indicated on the diagram shown in Fig. 11. In this diagram the positions of the resistors are shown relative to the bosses, and the corresponding position of each resistor in the sensing bridge is also shown.
Fig. 8. Geometry of low-pressure test sensor with three hollow bosses.
Fig. 9. Stress plot from Ansys 10.0 finite element software after a pressure is applied to the lower side of the diaphragm, the exaggerated deflection plot illustrates how the bosses remain rigid and stress is concentrated in the areas of diaphragm between the bosses.
As well as ease of positioning of the piezoresistors, the three bosses also concentrate stress such that for a given diaphragm area an increased sensor output is achieved with a reduced nonlinearity. The exact size of the three bosses was arrived at following
Fig. 10. Close-up stress plot of the areas of maximum and minimum stress situated between the hollow mesas. The view is from below the diaphragm looking up with the hollow bosses on the far side.
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
39
Fig. 11. Schematic top view of the test sensor showing the three hollow bosses on a diaphragm with the position of the piezoresistive elements shown relative to the bosses, along with a diagram of the wheatstone bridge used and the corresponding position of the resistors in the bridge.
an iterative optimisation process carried out using finite element analysis methods. As a basic comparison the estimated output for two comparative devices was considered, these were a flat diaphragm, and a diaphragm with a solid boss of 380 m, see Fig. 12. The solid boss was constrained to be 380 m high, which corresponds to the thickness of a typically available silicon wafer. The overall size and thickness of the diaphragms was kept the same at 1550 m square and 5 m thick. For the solid boss die it must be noted that only one boss is used that has the same base length as the central boss on the hollow boss device. The side bosses have not been added because these bosses are constrained to be 380 m tall, which as a result of the anisotropic etching process means they become to wide to fit them on the diaphragm. The comparison was done using Ansys FEA software to run a non-linear static analysis of the diaphragms under the influence of a pressure load, and a linear static analysis under the influence of an acceleration load. The pressure load was specifically a pressure of 1000 mbar applied to both sides of the diaphragm to simulate line pressure, with an additional pressure of 16 mbar applied to the flat side of the diaphragm to simulate the small differential pressure being sensed. For the case of the diaphragm with a hollow boss the inside of the boss was modeled as being at 0 pressure. This simulates the vacuum inside the boss that is expected due to the fusion bonding process used to create it. The results shown in Table 1 demonstrate that introducing hollow bosses results in the highest sensitivity and the best linearity as compared to the alternatives. Sensitivity is approximately a factor of three better than the alternatives, with similar non-linearity to the solid boss diaphragm. The results for acceleration sensitivity indicate that while the flat diaphragm is the least sensitive as expected, the hollow boss device is only three times more sensitive and is a factor of 10 less sensitive than the solid boss device. With these performance advantages in mind a three boss design was chosen to demonstrate the feasibility of the hollow boss process. A micrograph of the fabricated sensor, Table 1 Comparison of flat diaphragm versus a diaphragm with hollow bosses. Device
Sensitivity (mV/V/mbar)
Linearity (%f.s.TBNL)
G-sensitivity (ppm of f.s./g)
Hollow boss 80 m height Solid boss 380 m height No boss flat diaphragm
1.16 0.37 0.39
0.48 0.50 4 64
11 114 3
which is based on the dimensions detailed above, is shown in Fig. 13. 4. Characterization of hollow boss sensor performance To fully determine whether the hollow structures were stable enough for use in a MEMS sensor the performance of the fabricated low-pressure die was assessed. Thermal stability at 125 ◦ C, pressure sensitivity and linearity, and pressure hysteresis were all characterized using the fabricated die. Packaging stress is a well-known source of sensor instability, therefore to assess the stability of the die only, special care was taken to remove these affects. Die were solely attached to test electronics by means of the gold wire-bonds made to the chip such that the die were essentially floating on the 25 m gold wires to minimise the affects of packaging induced stress. The measured stability would therefore be due solely to the influence of the silicon die with hollow bosses and therefore be a good indicator of the ultimate performance that may be achieved using hollow bosses. Die were heated to 125 ◦ C in an oven, after approximately 8 h a fully stable temperature was reached. The variation in 0 mbar offset from this point was measured in percentage change of the sensor full-scale. The full-scale output was not measured for these units, as the floating construction did not allow for a differential pressure to be applied to the diaphragm. Therefore this was estimated to be 18 mV/V, which is a value of full-scale output that is consistent with the measured sensitivity for other die. Two die were mounted in this way and held at 125 ◦ C for a 450-h period and total drift was approximately 0.02% of full-scale, as shown by the chart in Fig. 14. This level of drift was deemed to be acceptable and in line with the performance of other MEMS pressure sensors. To test for pressure sensitivity, linearity, and pressure hysteresis a die was bonded to a stainless steel packaged such that a differential pressure ranging from 0 to 30 mbar could be applied to the device using a low-pressure Ruska pressure controller. The sensitivity of the device was determined to be 0.5 mV/V/mbar (see Fig. 15), with a terminal base non-linearity of less that 0.4% of full-scale (see Fig. 16). There was no indication of pressure hysteresis on return to 0 mbar after pressure cycling, demonstrating the good structural nature of the hollow boss to diaphragm interface.
40
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
Fig. 14. Zero offset drift for two sensors over a 450 h period spent at 125 ◦ C.
Fig. 15. Plot of device output, with sensitivity indicated.
Fig. 12. Quarter sections of the three hollow bosses die, a single solid boss die, and a simple flat diaphragm die.
Fig. 16. Plot of terminal based non-linearity calculated for two hollow boss devices.
5. Conclusion
Fig. 13. Micrograph of the hollow bosses shown on a silicon diaphragm.
A novel process for the fabrication of hollow boss structures mounted to thin silicon diaphragms has been developed. These structures have been employed to create hollow stiffening bosses for use in a 30 mbar full-scale pressure sensor. The feasibility of the process to create structural elements that may form part of a micro-electro-mechanical systems (MEMS) device has been
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
demonstrated through the fabrication and testing a hollow test structure that spanned a width of 400 m with a height of 80 m and a wall thickness of only 3 m (see Fig. 7). This type of structure may be used in a variety of MEMS applications for example as thin-walled encapsulations layers, thermal isolation structures, or as in this example for producing hollow stiffening structures on silicon diaphragms. The fabricated pressure sensor showed the suitability of these structures for use as structural elements in MEMS sensors. This was demonstrated by the excellent thermal stability and absence of any pressure hysteresis that would not be possible if the hollow structure was not a fully integral part of the sensor. References [1] W.C. Young, R.G. Budynas, Roark’s Formulas for Stress and Strain, 7th ed., McGraw-Hill, 2002. [2] S. Franssila, Introduction to Micro Fabrication, Wiley, 2004, pp. 205–216. [3] M.J. Madou, Fundamentals of Microfabrication the Science of Miniaturization, 2nd ed., CRC Press, 2002. [4] J.C. Greenwood, Etched silicon vibrating sensor, J. Phys. E: Sci. Instrum. 17 (1984) 650–652. [5] M. Hirata, K. Suzuki, H. Tanigawa, Silicon diaphragm pressure sensor fabricated by anodic oxidation etch-stop, Sens. Actuators A: Phys. 13 (1) (1988) 63–70. [6] Y. Kanda, A. Yasukawa, Optimum design considerations for silicon piezoresistive pressure sensors, Sens. Actuators A: Phys. 62 (1997) 539–542. [7] J. Gardner, V. Varadan, O. Awadelkarim, Microsensors, MEMS, and Smart Devices, John Wiley & Sons, 2001, pp. 126–134.
41
Biographies Peter Kinnell Peter Kinnell gained a MEng in Mechanical Engineering from the University of Birmingham (UK), and The Danish Technical University in 2001. After graduating he continued his studies at the University of Birmingham undertaking a PhD in MEMS sensor design, specializing in advanced packaging for resonant strain gauges. Since completing his PhD he has worked as a senior design engineer at GE Sensing. His work has included a range of products from large volume medical and automotive sensor applications to high performance resonant pressure sensors. Russell Craddock Russell Craddock graduated with a BSc in Chemistry before undertaking an MSc in Semiconductor Devices. He joined the Lucas Research Centre to investigate silicon pressure sensor and accelerometer design before transferring to Lucas NovaSensor in the USA to work on the development of automotive accelerometers. In 1992 Russell joined Druck Ltd. – now GE Sensing, leading piezoresistive and resonant pressure sensors development for GE Druck and automotive pressure and accelerometer products for GE NovaSensor. Jim King Jim King graduated with a BSc (Open) with a Physics Diploma from the Open University. His semiconductor engineering background includes working for Agilent Technologies fabricating lasers, photodetectors and diodes, Corning Research developing semiconductor optical amplifiers and electro-absorption modulators, Plastic Logic working on the development of polymer transistor arrays and flexible displays. He has worked for GE Sensing since 2005 as a Senior Process Engineer developing and aiding manufacture of piezoresistive and resonant pressure sensors. Mandy Lester Mandy Lester is currently studying for an MPhys degree at the University of Manchester, where her research interests have ranged from ozone measurement by Brewer spectrophotometry to radio thin layer chromatography for medical PET imaging. She is due to graduate July 2010 and plans to pursue a career in nuclear or renewable energy.
Contents lists available at ScienceDirect
Sensors and Actuators A: Physical journal homepage: www.elsevier.com/locate/sna
A hollow stiffening structure for low-pressure sensors P.K. Kinnell ∗ , J. King, M. Lester, R. Craddock GE Sensing & Inspection Technologies, Fir Tree Lane, Groby, Leicestershire, UK
a r t i c l e
i n f o
Article history: Received 15 December 2009 Received in revised form 18 February 2010 Accepted 12 March 2010 Available online 18 March 2010 Keywords: Pressure sensor Electrochemical etch Hollow structure Boss Diaphragm Silicon
a b s t r a c t This paper presents a novel process for producing thin-walled hollow stiffening structures on thin silicon diaphragms using an electrochemical etch-stop process. Examples of structures produced using the method are presented together with focused ion beam (FIB) analysis of critical areas within the structure. These demonstrate the integrity of the structures and show that the process is suitable for use in MEMS sensor applications. Using this process a 30 mbar full-scale differential pressure sensor has been demonstrated, and used to verify the suitability of these hollow structures for use in MEMS sensors. The novel process allows for increased sensor performance, with reduced die size. Details of the pressure sensor design and characterization are presented, showing a device with 18 mV/V full-scale output with linearity <0.4% (terminal base non-linearity). © 2010 Elsevier B.V. All rights reserved.
1. Introduction Piezoresistive silicon pressure sensors typically consist of a wheatstone bridge of piezoresistors fabricated on a square silicon diaphragm. The sensitivity of these sensors will be inversely proportional to the square of the diaphragm thickness, and directly proportional to diaphragm area [1]. To fabricate higher sensitivity devices the designer is forced to increase die size or reduce diaphragm thickness. Cost constraints often lead the designer towards reducing die size. Therefore the approach taken is generally to reduce diaphragm thickness in order to meet the required sensitivity. Silicon diaphragms for MEMS pressure sensors are typically fabricated by wet anisotropic etching of a cavity into bulk silicon, and to achieve a required diaphragm thickness requires careful timing of this etch. Even if great care is taken to continually monitor the etch in order to achieve the desired thickness, process variation from wafer to wafer, and from die to die within a wafer will ultimately dictate the minimum practicable thickness. Typically diaphragms thinner than 40 m prove challenging for many manufacturing environments [2]. To overcome this problem there are a number of etch-stop methods that may be employed to automatically stop or limit the anisotropic etching once the required thickness has been reached. Examples of such techniques are, the use of silicon on insulator wafers [3], the boron etch-stop process [4], or an electrochemical etch-stop [5,2]. In this work the electro-
∗ Corresponding author at: GE Sensing, Silicon Engineering, Fir Tree Lane, Groby, Leicestershire, LE60FH, UK. Tel.: +44 1162317507. E-mail address: [email protected] (P.K. Kinnell). 0924-4247/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.sna.2010.03.024
chemical etch-stop technique is used to form the basis of a novel process that firstly addresses the issue of increased sensitivity by allowing thinner diaphragms, but also allows for additional structuring of the diaphragms to improve other device performance characteristics such as output linearity. As high sensor linearity is often a very attractive sensor characteristic, flat silicon diaphragms are generally modified with the addition of a lump or ‘boss’ structure to stiffen the centre of the diaphragm. Fig. 1 shows a micrograph of a typical pressure sensor diaphragm with a central stiffening boss, the cross-section marked on the micrograph can be seen as a schematic in Fig. 3b. These features improve linearity by limiting strain stiffening of the diaphragm, which is a significant cause of output non-linearity [6]. The addition of such stiffening bosses is often done at the expense of other device parameters such as die size. Fabricating the boss on the diaphragm generally increases die size, and the additional mass that is suspended will tend to cause increased acceleration sensitivity. This acceleration sensitivity becomes especially critical as higher sensitivity pressure sensors are fabricated. This is because the diaphragm that is supporting the boss becomes increasingly flexible, thus any inertial loads imposed on the boss will result in a greater deflection of the diaphragm and thus be seen as a more significant proportion of the sensor output. To address these issues this work presents a novel fabrication route that allows for a wide variety of hollow structures to be fabricated on the surface of a thin silicon diaphragm. These structures may be used, in effect, as hollow bosses that have considerable stiffness relative to the diaphragm, yet due to their hollow construction add very little mass to the diaphragm (see Fig. 3c). The diaphragm and hollow bosses are produced using an electrochemical etch-
36
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
Fig. 1. Micrograph of an etched pressure sensor diaphragm with a solid lump at the centre, the diaphragm is approximately 1.5 mm square.
Fig. 3. Cross-sectional schematic view of a boss etched into a wafer follow (1 1 1) crystal planes.
stop technique and therefore this allows for controlled etching of thin diaphragms that are <10 m in thickness. The combination of these features allows the advantages gained by a stiffening boss in terms of sensor linearity, without either the negative affects of inertial sensitivity, or unnecessary increase in die size that may result from traditional fabrication techniques that produce solid bosses. For example the increase in die size comes about due to the constraints imposed by standard processing methods. These typically require that the boss is patterned on the surface of a wafer then etched down following (1 1 1) crystal planes to reveal pyramid shaped bosses with sloping sidewalls of 54.7◦ as shown in Fig. 2. Etching bosses in this way means the designer is constrained to use bosses that may be taller or wider than required. This has the adverse affect of removing active diaphragm area and therefore leads to an increased die size, as the designer has to compensate by increasing the diaphragm area. The hollow boss process is not limited by these constraints and allows boss heights independent of wafer thickness. In the following sections the innovative process that was developed to fabricate hollow bosses will be described, demonstrating that in principle the technique may be used to create a wide range of generic hollow structures suitable for many MEMS applications. A specific example of the technology is then given in the form of a piezoresistive pressure sensor. A description of the design and fabrication work carried out to develop test samples of a prototype pressure sensor are detailed. Finally, analysis and characterization are presented that verify the performance of the pressure sensor and validate the feasibility of the hollow structures for use as mechanical elements in MEMS sensors.
Fig. 2. Cross-sections of three types of pressure sensor diaphragm, type a is the most basic form of diaphragm that is completely flat, type b is the traditional form of diaphragm that contains a solid lump, and type c is a hollow version of the solid lump.
2. The hollow boss process In order to create thin-walled hollow structures using wet etching in a KOH solution the hollow boss process has been developed. An overview of the process is shown in Fig. 4 and it consists of five main process steps. Firstly a p-type silicon wafer is prepared and suitably patterned for etching (step 1); then a negative of the required boss is etched into a silicon wafer (step 2); the etched feature is implanted to render a layer of n-type silicon in the p-type substrate (step 3); a second n-type layer is then fusion bonded over the previously etched and implanted features (step 4); finally the p-type substrate material is etched away using an electrochemical etch-stop process to reveal a hollow structure. The electrochemical etch-stop, which is a key part of the process, results from the application of an anodic potential to silicon in OHcontaining solution that causes the formation of silicon oxide on the silicon surface, known as passivation. The process demonstrated here utilises the difference in passivation potentials of n-type and p-type silicon to generate an automatic etch-stop process that can preserve a hollow boss structure [5,7]. A fixed voltage, between the n-type and p-type passivation potentials, allows the p-type to etch, but causes passivation on reaching the n-type layer, marked by the cessation of bubble evolution at the etch surface. Since silicon oxide etches at a rate of approximately 100th that of silicon n-type and is self-sustaining under the cell potential, the integrity of the etch-stopped layer may be preserved for many times the lifetime of a standard sacrificial oxide. The electrochemical etch-stop was set-up as shown in Fig. 5. To calculate the correct passivation voltage, data showing the relationship between current and voltage was collected for wafers and electrodes being used, see Fig. 6. The maximum current reached on the chart shown in Fig. 6 corresponds to the passivation potential, which was 1.05 V. A power supply (indicated by PS in Fig. 5) with fixed anodic potential of at least 1.2 V was connected to an aluminum contact on the n-type layer of the silicon wafer, such that the wafer potential was always above the n-type passivation potential. The n-type surface was wax-bonded onto glass for protection of the sensing structure. A nickel plate was inserted into the solution and formed the cathode. The p-type silicon etched in normal anisotropic fashion in the aqueous KOH solution, as it was well below the p-type passivation potential. The electrochemical etch-stop ensured passivation of the hollow structure and allowed it to be fully revealed as the etch progressed.
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
37
Fig. 4. Process flow to create hollow boss structure.
Fig. 5. Diagram of electrochemical etch-stop equipment set up.
In this way n-type structures were preserved to full integrity during a wet anisotropic etch, evidenced by the fabrication of the hollow boss structure in Fig. 7. This micrograph shows that complete removal of the p-type silicon was obtained, leaving the n-type layer with the buried hollow boss structures bonded to it. The hol-
low boss structure shown in Fig. 7, is approximately 400 m square, 80 m deep with a wall thickness of 3 m. These relatively large very thin-walled structures were seen as a good test of the ability of the process to produce well-formed hollow bosses. Of particular interest was the area of intersection between the diaphragm and the walls of the hollow boss as shown in Fig. 7a, an arrow indicates the point of interface between the boss and diaphragm noting that the wall thickness at this interface is approximately 3 m. If these areas are not fully bonded together then the use of the hollow bosses as structural elements in sensors may be prohibited due to poor mechanical stability. The structural integrity of the hollow bosses were analysed using a combination of focus ion beam (FIB) and scanning electron microscopy (SEM) techniques. Using the FIB system a section of the hollow boss was removed by ion milling such that the interface between the walls of the boss and the diaphragm could be observed. A magnified image of this cut section is shown in Fig. 7b, from which it can be seen that there are no voids or other structural faults that may lead to poor performance. The hole that was cut in the flat section at the top of the boss was used to measure the top layer thickness. This was found to be approximately 2.4 m indicating the sidewalls at the base of the boss are slightly thicker than the flat top of the boss. It is expected that the top of the boss will be slightly thinner that the base due to the etch rate of the silicon under etch-stop conditions, which is set by the passivation oxide etch rate. This structure demonstrates that large thin-walled structures may be created with this process. In this work the negatives used to create the hollow bosses were created using anisotropic etching, hence the pyramidal shape. These negative structures can also be created using other etching techniques. For example, isotropic etching or deep reactive ion etches may be used to create bosses with either rounded or vertical sidewalls, depending on the required application. 3. Test sensor design
Fig. 6. Chart showing relationship between current and voltage for the electrochemical etch-stop process.
The hollow structures detailed above demonstrate that hollow bosses may be manufactured, however, to determine if these structures may be used as part of MEMS sensors a low-pressure sensor was designed and fabricated using hollow bosses and the process described above. A low-pressure sensor was specifically chosen as this type of device benefits from the controlled thin diaphragms given by the hollow boss process, as well as optimal sensitivity and linearity due to the addition of hollow bosses. The low-pressure sensor was designed to have a target full-scale pressure in the range of 20–30 mbar. This is based on achieving an electrical output of approximately 20 mV/V output at full-scale pressure. A lowpressure sensor was chosen to demonstrate this process, as this is a challenging area of design space for pressure sensors. It would benefit from the combination of an electrochemical etch-stop to control diaphragm thickness with the hollow boss process to optimise sensor performance.
38
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
Fig. 7. (a) Focused ion beam section cut from a hollow boss and (b) close-up of cut section that is indicated by the arrow shown in (a).
The pressure sensor was designed with a diaphragm of 1550 m square and 5 m thick, onto which three hollow boss structures 80 m high were fixed. The wall thickness of the hollow bosses was designed to be approximately 10 m; the overall geometry of the diaphragm can be seen in Fig. 8. As mentioned above, the most basic pressure sensor would comprise of a simple flat diaphragm (see the cross-section shown in Fig. 3a). In this work three bosses have been added to the diaphragm to improve sensor performance. Adding the three bosses to the diaphragm is effective because it stiffens the diaphragm in the regions of the bosses but keeps the diaphragm flexible elsewhere. This means that when pressure is applied to the diaphragm stress is concentrated in the areas that are kept flexible between the bosses. To illustrate this Fig. 9 shows an exaggerated deflection plot resulting from pressure applied to the lower side of the diaphragm (created using Ansys FEA software). From the plot it can be seen that the bosses remain rigid with the areas of diaphragm between the bosses and the frame taking up all the bending and therefore developing maximum tensile and compressive stresses. The arrows in Fig. 9 indicate the areas of maximum and minimum stress situated between the lumps. A close-up of this area is also shown in Fig. 10, in this plot the areas of uniform stress that have been created between the bosses can be seen. This is a key advantage of this type of diaphragm over a conventional flat diaphragm. Uniform stresses have thus been created between the bosses aiding the positioning of piezoresistors to sense pressure-induced stress. The resistors would therefore be positioned in these regions as indicated on the diagram shown in Fig. 11. In this diagram the positions of the resistors are shown relative to the bosses, and the corresponding position of each resistor in the sensing bridge is also shown.
Fig. 8. Geometry of low-pressure test sensor with three hollow bosses.
Fig. 9. Stress plot from Ansys 10.0 finite element software after a pressure is applied to the lower side of the diaphragm, the exaggerated deflection plot illustrates how the bosses remain rigid and stress is concentrated in the areas of diaphragm between the bosses.
As well as ease of positioning of the piezoresistors, the three bosses also concentrate stress such that for a given diaphragm area an increased sensor output is achieved with a reduced nonlinearity. The exact size of the three bosses was arrived at following
Fig. 10. Close-up stress plot of the areas of maximum and minimum stress situated between the hollow mesas. The view is from below the diaphragm looking up with the hollow bosses on the far side.
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
39
Fig. 11. Schematic top view of the test sensor showing the three hollow bosses on a diaphragm with the position of the piezoresistive elements shown relative to the bosses, along with a diagram of the wheatstone bridge used and the corresponding position of the resistors in the bridge.
an iterative optimisation process carried out using finite element analysis methods. As a basic comparison the estimated output for two comparative devices was considered, these were a flat diaphragm, and a diaphragm with a solid boss of 380 m, see Fig. 12. The solid boss was constrained to be 380 m high, which corresponds to the thickness of a typically available silicon wafer. The overall size and thickness of the diaphragms was kept the same at 1550 m square and 5 m thick. For the solid boss die it must be noted that only one boss is used that has the same base length as the central boss on the hollow boss device. The side bosses have not been added because these bosses are constrained to be 380 m tall, which as a result of the anisotropic etching process means they become to wide to fit them on the diaphragm. The comparison was done using Ansys FEA software to run a non-linear static analysis of the diaphragms under the influence of a pressure load, and a linear static analysis under the influence of an acceleration load. The pressure load was specifically a pressure of 1000 mbar applied to both sides of the diaphragm to simulate line pressure, with an additional pressure of 16 mbar applied to the flat side of the diaphragm to simulate the small differential pressure being sensed. For the case of the diaphragm with a hollow boss the inside of the boss was modeled as being at 0 pressure. This simulates the vacuum inside the boss that is expected due to the fusion bonding process used to create it. The results shown in Table 1 demonstrate that introducing hollow bosses results in the highest sensitivity and the best linearity as compared to the alternatives. Sensitivity is approximately a factor of three better than the alternatives, with similar non-linearity to the solid boss diaphragm. The results for acceleration sensitivity indicate that while the flat diaphragm is the least sensitive as expected, the hollow boss device is only three times more sensitive and is a factor of 10 less sensitive than the solid boss device. With these performance advantages in mind a three boss design was chosen to demonstrate the feasibility of the hollow boss process. A micrograph of the fabricated sensor, Table 1 Comparison of flat diaphragm versus a diaphragm with hollow bosses. Device
Sensitivity (mV/V/mbar)
Linearity (%f.s.TBNL)
G-sensitivity (ppm of f.s./g)
Hollow boss 80 m height Solid boss 380 m height No boss flat diaphragm
1.16 0.37 0.39
0.48 0.50 4 64
11 114 3
which is based on the dimensions detailed above, is shown in Fig. 13. 4. Characterization of hollow boss sensor performance To fully determine whether the hollow structures were stable enough for use in a MEMS sensor the performance of the fabricated low-pressure die was assessed. Thermal stability at 125 ◦ C, pressure sensitivity and linearity, and pressure hysteresis were all characterized using the fabricated die. Packaging stress is a well-known source of sensor instability, therefore to assess the stability of the die only, special care was taken to remove these affects. Die were solely attached to test electronics by means of the gold wire-bonds made to the chip such that the die were essentially floating on the 25 m gold wires to minimise the affects of packaging induced stress. The measured stability would therefore be due solely to the influence of the silicon die with hollow bosses and therefore be a good indicator of the ultimate performance that may be achieved using hollow bosses. Die were heated to 125 ◦ C in an oven, after approximately 8 h a fully stable temperature was reached. The variation in 0 mbar offset from this point was measured in percentage change of the sensor full-scale. The full-scale output was not measured for these units, as the floating construction did not allow for a differential pressure to be applied to the diaphragm. Therefore this was estimated to be 18 mV/V, which is a value of full-scale output that is consistent with the measured sensitivity for other die. Two die were mounted in this way and held at 125 ◦ C for a 450-h period and total drift was approximately 0.02% of full-scale, as shown by the chart in Fig. 14. This level of drift was deemed to be acceptable and in line with the performance of other MEMS pressure sensors. To test for pressure sensitivity, linearity, and pressure hysteresis a die was bonded to a stainless steel packaged such that a differential pressure ranging from 0 to 30 mbar could be applied to the device using a low-pressure Ruska pressure controller. The sensitivity of the device was determined to be 0.5 mV/V/mbar (see Fig. 15), with a terminal base non-linearity of less that 0.4% of full-scale (see Fig. 16). There was no indication of pressure hysteresis on return to 0 mbar after pressure cycling, demonstrating the good structural nature of the hollow boss to diaphragm interface.
40
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
Fig. 14. Zero offset drift for two sensors over a 450 h period spent at 125 ◦ C.
Fig. 15. Plot of device output, with sensitivity indicated.
Fig. 12. Quarter sections of the three hollow bosses die, a single solid boss die, and a simple flat diaphragm die.
Fig. 16. Plot of terminal based non-linearity calculated for two hollow boss devices.
5. Conclusion
Fig. 13. Micrograph of the hollow bosses shown on a silicon diaphragm.
A novel process for the fabrication of hollow boss structures mounted to thin silicon diaphragms has been developed. These structures have been employed to create hollow stiffening bosses for use in a 30 mbar full-scale pressure sensor. The feasibility of the process to create structural elements that may form part of a micro-electro-mechanical systems (MEMS) device has been
P.K. Kinnell et al. / Sensors and Actuators A 160 (2010) 35–41
demonstrated through the fabrication and testing a hollow test structure that spanned a width of 400 m with a height of 80 m and a wall thickness of only 3 m (see Fig. 7). This type of structure may be used in a variety of MEMS applications for example as thin-walled encapsulations layers, thermal isolation structures, or as in this example for producing hollow stiffening structures on silicon diaphragms. The fabricated pressure sensor showed the suitability of these structures for use as structural elements in MEMS sensors. This was demonstrated by the excellent thermal stability and absence of any pressure hysteresis that would not be possible if the hollow structure was not a fully integral part of the sensor. References [1] W.C. Young, R.G. Budynas, Roark’s Formulas for Stress and Strain, 7th ed., McGraw-Hill, 2002. [2] S. Franssila, Introduction to Micro Fabrication, Wiley, 2004, pp. 205–216. [3] M.J. Madou, Fundamentals of Microfabrication the Science of Miniaturization, 2nd ed., CRC Press, 2002. [4] J.C. Greenwood, Etched silicon vibrating sensor, J. Phys. E: Sci. Instrum. 17 (1984) 650–652. [5] M. Hirata, K. Suzuki, H. Tanigawa, Silicon diaphragm pressure sensor fabricated by anodic oxidation etch-stop, Sens. Actuators A: Phys. 13 (1) (1988) 63–70. [6] Y. Kanda, A. Yasukawa, Optimum design considerations for silicon piezoresistive pressure sensors, Sens. Actuators A: Phys. 62 (1997) 539–542. [7] J. Gardner, V. Varadan, O. Awadelkarim, Microsensors, MEMS, and Smart Devices, John Wiley & Sons, 2001, pp. 126–134.
41
Biographies Peter Kinnell Peter Kinnell gained a MEng in Mechanical Engineering from the University of Birmingham (UK), and The Danish Technical University in 2001. After graduating he continued his studies at the University of Birmingham undertaking a PhD in MEMS sensor design, specializing in advanced packaging for resonant strain gauges. Since completing his PhD he has worked as a senior design engineer at GE Sensing. His work has included a range of products from large volume medical and automotive sensor applications to high performance resonant pressure sensors. Russell Craddock Russell Craddock graduated with a BSc in Chemistry before undertaking an MSc in Semiconductor Devices. He joined the Lucas Research Centre to investigate silicon pressure sensor and accelerometer design before transferring to Lucas NovaSensor in the USA to work on the development of automotive accelerometers. In 1992 Russell joined Druck Ltd. – now GE Sensing, leading piezoresistive and resonant pressure sensors development for GE Druck and automotive pressure and accelerometer products for GE NovaSensor. Jim King Jim King graduated with a BSc (Open) with a Physics Diploma from the Open University. His semiconductor engineering background includes working for Agilent Technologies fabricating lasers, photodetectors and diodes, Corning Research developing semiconductor optical amplifiers and electro-absorption modulators, Plastic Logic working on the development of polymer transistor arrays and flexible displays. He has worked for GE Sensing since 2005 as a Senior Process Engineer developing and aiding manufacture of piezoresistive and resonant pressure sensors. Mandy Lester Mandy Lester is currently studying for an MPhys degree at the University of Manchester, where her research interests have ranged from ozone measurement by Brewer spectrophotometry to radio thin layer chromatography for medical PET imaging. She is due to graduate July 2010 and plans to pursue a career in nuclear or renewable energy.