Design of Capacitive Displacement Sensors for Nanoalingment A Thesis Presented by Jose Luis Medina To The Department of Electrical and Computer Engineering in partial fulfillment of the requirements for the degree of Master of Science in the field of Electrical and Computer Engineering Northeastern University Boston, Massachusetts December 17, 2007 ABSTRACT Nanotechnology is a very promising technology that could potentially have a great impact in fields such as medicine, electronics, material science, and energy storage. Improvements in microscopy, as the invention of the Scanning Tunneling Microscope in 1981 and the Atomic Force Microscope in 1986, enabled the observation of molecules at nanoscale and helped obtain a better understanding of the extraordinary dominant effects at nanoscale, and later enabled the manipulation of single atoms or molecules. Furthermore, the discovery of carbon nanotubes brought a novel material with excellent properties at nanoscale. Fabrication of nanodevices is already a reality. However, in order for this technology to become cost-effective and available for society, scalable fabrication processes that would enable massive production of nanodevices must be developed. Towards this goal, the ‘Center for High Rate Nanomanufacturing’ is developing a scalable nanomanufacturing process flow based on transferring nanoelements from a template to a substrate. For one of the step of this process flow, a alignment of template and substrate at nanoscale is required. This work aims at designing a capacitive displacement sensor for alignment of two chips or wafers. First, the alignment process is analyzed and different sensing techniques studied in relation to this specific project. After that, modeling and analysis of the sensor is carried out, followed by the design of the sensor, including its readout circuit. ii ACKNOWLEDGMENTS From my experience studying and researching in several institutions in three different countries, I know usually advisors care more about the final goals of the project than about the students who are doing that research. I would specially like to thank professor McGruer for, not only giving me the opportunity to work with him on this project, but also for his interest in my academic and personal development, in addition to my research. I appreciate his guidance through my studies at Northeastern and beginning of my career. I also want to thank professors George Adams, Demetrios Papageorgiou, Ahmed Busnaina, Sun, and Sinan Muftu for their help and guidance during my thesis. In the Northeastern Microfabrication Lab, I want to thank the students Prashanth Makaram, Andy Pamp, Peter Ryan, Juan Aceros, and Siva Somu for helping with the tools and equipment in the lab. I also want to thank Jen Dawson, for copyediting my thesis, and for so much more. iii TABLE OF CONTENTS Abstract Acknowledgments ii iii 1. Introduction 1.1 Nanotechnology and Nanomanufacturing 1.2 Alignment and Nanopositioning 1.3 Displacement Sensors 1.4 Sensor requirements 1.5 Comparison sensors literature and nanoaligner sensor 1 1 2 3 5 2. Sensor design 2.1 Arrangement for connections-free substrate 2.2 Arrangement for large gap variations 2.2.1 Fractal arrangement 2.2.2 Switchable variable width 2.2.3 Central fractal 2.3 Fabrication 8 10 10 12 14 15 3. Modeling 3.1 Capacitance and FEM field solvers 3.2 Modeling scenarios 3.3 Equivalent capacitance circuit 3.4 FEM simulations 3.4.1 Algorithm 3.4.2 Fringing field effect 3.4.3 Influence of plate conductivity 3.5 Convergence 17 22 24 27 27 28 32 45 4. Capacitance-to-voltage converter 4.1 Introduction 4.2 Converters 4.3 Oscillators 4.4 Capacitive Bridge 4.5 Transimpedance amplifier 4.5.1 Gain and noise 4.5.2 Stability 4.5.3 Demodulator 4.6 Switched-Capacitor Converter 4.6.1 Switched-Capacitor Amplifier 4.6.2 Parasitic Capacitance 4.6.3 Correlated Double Sampling 49 50 51 52 52 52 55 59 60 60 63 64 iv 4.6.4 Bandwidth 4.6.5 Noise Analysis 4.6.6 Simulations 4.7 Sigma-Delta modulator 68 70 73 76 5. Experimental verification 5.1 Scaled model 5.2 Setup 5.3 Readout circuit 5.4 Designs tested 5.4.1 Large features test board 5.4.2 Central fractal test board 81 84 85 86 87 90 6. Conclusions and future work 6.1 Experimental results 6.2 Recommendations for design 6.3 Future work 93 95 97 Appendices A. Calculation of equivalent output capacitance B. APDL macro for capacitance extraction simplified model table 3.2 C. APDL macro model for large gap and displacement range table D. MATLAB filter for ANSYS capacitance results E. APDL macro to calculate capacitance versus number of nodes per electrode’s horizontal surface F. Input equivalent capacitance for connected plates. G. Hspice code readout circuit Figure 4.24 and RC LPF v 101 vi 1. Introduction 1.1 Nanotechnology and Nanomanufacturing Nanotechnology can be defined as the engineering of functional systems at nanometer scale, i.e. at the molecular scale. Currently there is technology available that enables manipulation single molecules and building of nanodevices [1.5]. However, in order for nanotechnology to become commercially available for the general public, instead or manipulating single molecules at a time, billions of molecules have to be manipulated at a time. A scalable nanomanufacturing process that would enable the fabrication of nanodevices at high rates is one of the main keys towards the commercialization of nanotechnology. 1.2 Alignment and Nanopositioning Towards the successful commercialization of nanotechnology, the ‘Center for High Rate Nanomanufacturing’ (CHN) aims to develop a process flow that would enable scalable fabrication of nanodevices. A novel process studied at this center involves the transfer of nanotubes and or nanoparticles from a template to a substrate. In order for the transfer to be successful, an extraordinarily accurate alignment between substrate plate and template plate must be achieved. This alignment would have two phases. A first phase would align the plates up to a few nanometers by means of a displacement sensor. A nanopositioner, like the one in Figure 1.1, would be used to move the plates, while a displacement sensor 1 would sense the relative position and close the feedback loop. The nanopositioner would move one of the plates, while the other one would remain static. Figure 1.1: Model of nanopositioner used for high rate nanomanufacturing [1.6]. Then, the second phase of the alignment would take over, where molecular forces of chemical and physical nature would further align the plates. Once alignment is achieved, the nanotubes or nanoparticles would be transferred from template to substrate. 1.3 Displacement Sensors Displacement sensors are widely used in industry. Most displacement sensors are based on transducers that convert displacement to another signal such as temperature, light, or electrical resistance. The main transducer principles are given in the following table. Criteria Probe-based Optical Capacitive Thermal Accuracy ++ ++ +/++ +/++ Range -/+ ++ +/++ +/++ Speed + + + Fabrication -++ ++ Electronics integration + ++ ++ Parasitic forces -/+ ++ --/+ Power consumption +/++ -/+ ++ + Table 1.1: Main displacement sensor transducers that enable integration within a MEMS device, with accuracy in the nm-range, and range over 100 μm [1.1]. Optical transducers achieve high accuracy, but are complex to fabricate and integrate. Capacitive transducers are easier and inexpensive to fabricate, can easily be integrated in 2 any system, and do not need complicated electronics. On the other hand, capacitive transducers are highly affected by any surrounding capacitances different from the capacitance to be measured. Thermal transducers are simple and inexpensive as well, although the measurement is slower. 1.4 Sensor requirements To align two chips or wafers, a displacement sensor can be integrated in the chips to align, so that the relative position of both chips can be sensed. A displacement sensor for plate alignment must meet several characteristics from fabrication and economics point of view. First, its fabrication should be compatible with the nanomanufacturing process flow, and enable integration. Second, it must be flexible and compatible with general nanopositioner equipment. Third, it should be as inexpensive and robust as possible. A probe-based converter may not be compatible with the alignment process and nanopositioner. An optical transducer would require a complex circuitry to obtain an electrical signal for the feedback. A thermal actuator may be too slow, which would require a more complex controller for the feedback and may result in instability. A capacitive transducer is easy to integrate in the feedback loop with rather simple circuitry, and inexpensive to fabricate. Furthermore, built-in capacitive sensors can easily be fabricated on top of the same plates to align, which results in easier integration in the alignment process and compatibility with the nanopositioner. For all these reasons, capacitive transducers are used for this project’s purposes. As illustrated in the following figure, a group of metal strips can be fabricated on the chips to align, in a way that the capacitance between the strips in the substrate and the 3 strips in the template would depend on the relative position of the plates. Four independent sensors are incorporated on both plates, two sensors oriented in one direction and the other two perpendicular. In principle, these four sensors should be able to determine the relative position of both plates in six degrees of freedom. Figure 1.2: substrate and template plates (grey) with built-in electrodes (yellow) for capacitive sensing. Figure 1.3 shows the strips on substrate and template. Figure 1.3: Detail of substrate and template strips. The figure on the left shows the pads for external connections. The main drawback of this approach is the parasitic capacitances in the sensor, which deteriorates the sensor performance. Care should be taken in the design of the electronic readout circuit to minimize the influence of parasitic capacitances. 4 1.5 Comparison sensors literature and nanoaligner sensor There is extensive research literature in the field of micromachined capacitive sensors for applications such as Bioengineering [1.2], positioners [1.3], temperature and humidity sensor [1.7], and automotive [1.4]. Most of these sensors are based on two parallel electrodes, where the overlapping area of the electrodes changes with the lateral displacement of the plates, while the gap remains constant. Figure 1.4: Schematic model of a capacitive position sensor [1.2]. Humidity and temperature capacitive sensors have fixed electrodes, so the structure usually used for these sensors is quite different from the structure needed for alignment with a nanoaligner [1.7]. Something common in all displacement sensors reviewed during the literature research done is that in all sensors only one dimension varies, while the other dimension is meant to stay as constant as possible. Indeed, in most displacement sensors, the gap between the electrodes is kept constant, while the overlap electrode area changes with displacement. The sensor in Figure 1.4 follows such approach. Typically capacitive sensors are based on the capacitance from the strips on one plate to the strips on the other plate. 5 ~ Figure 1.5: typical scheme used in capacitive displacement sensors. The top strips and the bottom strips are attached to two different plates. An excitation is applied to the strips on one plate, while the strips on the other plate are connected to a readout circuit. The gap remains constant, while the plates move laterally in parallel to each other. The sensor detects the relative plate displacement. Unlike these sensors, a sensor aimed at alignment with a nanopositioner like the one shown in Figure 1.1 would have two novel main requirements. First, in order for the nanopositioner to move one of the plates, this plate must be free of connections. Therefore all the connections to a readout circuit have to be done to the static plate. This means that a scheme such as the one in Figure 1.5 cannot be used, since this scheme requires connections to both top and bottom plates. Second, capacitive sensors are usually designed to work under either a constant gap or an overlapping area constant. In this way the electrode geometry can be optimized for given range. However, in this project it will be assumed that the nanopositioner starts the alignment at a long gap. Once the best possible alignment is achieved at that gap, the gap is decreased, so the sensitivity of capacitance to displacement increases, and a new finer realignment can be achieved. This process is repeated several times until the minimum gap before contact is reached. As a first assumption, the gap at the beginning of alignment process would be 0.5 mm, while the last smallest gap could be as short as 50 nm. A widely variable gap introduces an important issue to overcome in the sensor 6 design. Parallel plate capacitors such as the ones shown in Figure 1.5 become insensitive to lateral displacement if the ratio of strip width to gap is large. A large number of strips with small width would give high sensitivity for small gaps, but it would be almost insensible to lateral displacement for long gaps. However, a smaller number of strips with larger width would result in greater sensitivity for large gaps, but less sensitivity for small gaps. Potential solutions and trade-offs to this issue, such as fractal geometries with sets of electrodes with different width, will be addressed during the sensor design. References: [1.1] A. A. Kuijpers, ‘Micromachined Capacitive Ling-Range Displacement Sensor for Nano-positioning of Microactuator Systems’, PhD thesis, Universiteit Twente. [1.2] A. Pedrocchi, S. Hoen, G. Ferrignl, A. Pedotti, ‘Perspectives on MEMS in Bioengineering: A Novel Capacitive Position Microsensor’, IEEE Vol. 47, No. 1, January 2000 [1.3] H. U. Meyer, ‘An Integrated Capacitive Position Sensor’, IEEE Vol. 45, No. 2, April 1996 [1.4] G. Brasseur, B. Brandstatter, H. Zangl, ‘State of the Art of Robust Capacitive Sensors’, IEEE 2003. [1.5] A. Busnaina, ‘Nanomanufacturing Handbook’, Northeastern University, Boston, CRC Press [1.6] A. Busnaina, ’High-rate Directed Assembly of Nanoelements’. NSF presentation [1.7] A. Tetelin, C. Pellet, ‘Modeling and Optimization of a Fast Response Capacitive Humidity Sensor’, IEEE Sensors Journal, Vol. 6, No. 3, June 2006. 7 2. Sensor design 2.1 Arrangement for connections-free substrate As explained before, the scheme of Figure 1.4 would be incompatible with the alignment process, since the nanopositioner needs the moving plate to be free of connections. Instead, a scheme such as the one shown in Figure 2.1 could be used. Figure 2.1: scheme of capacitive sensor with connections (the black arrow on the left and the voltage source on the right) on only one side and equivalent capacitances. The capacitance symbols represent the main capacitances between electrodes. In the scheme of Figure 2.1 the two bottom electrodes 1 and 2 would be attached to plate 4, while top electrode 3 would be attached to plate 5. As in Figure 1.5, several groups of strips could be placed in parallel to increase the capacitance. The resulting capacitance that the detection circuit would sense when connected to electrodes 1 and 2 neglecting capacitances to plates 4 and 5 would be Ctotal C12 C13C 23 C13 C 23 (2.1) A scheme similar to the one in Figure 2.1 is used for differential capacitive sensors [2.1]. This is shown in the following figure. 8 Figure 2.2: differential capacitor. The figures show the bottom plate, the bottom electrodes and the top electrodes. In the figure on the left, three electrode groups are set in parallel to increment the capacitance. The figure on the right shows a simplified model of the capacitances [2.1]. The differential sensor of Figure 2.2 uses excitation applied at electrode 3 (the top electrode), while electrodes 1 and 2 are connected to a differential amplifier. Figure 2.3 shows a differential readout circuit and the electrical signal through its stages when there is a displacement [2.2]. Figure 2.3: differential capacitor sensor and readout circuit [2.2]. The circuit is based on a differential transimpedance amplifier as explained in section 4.5. The circuit of Figure 2.3 models the sensor capacitance C1 and C2 of Figure 2.2 as an initial capacitance Cx0 plus an increment due to a lateral displacement. This increment is the same in absolute value for both capacitances C1 and C2, but of opposite signs. The output voltages are processed by the circuit independently and subtracted at a final stage. 9 Therefore the sensitivity of capacitance to displacement compared to single capacitive sensors is twice as much. This scheme is useful for alignment if a connection to the plate moved by the nanopositioner is possible. Otherwise, the scheme of Figure 2.1 should be considered. 2.2 Arrangement for large gap variations Capacitive sensors composed of parallel strips or plates as in Figure 1.4 are designed to have a gap around 10 times smaller than the strip width. As will be shown in section 3.4.2, larger gaps than one third of the width would give poor sensitivity of capacitance to lateral displacement. Therefore, hypothetically, a parallel plate capacitor for lateral displacement sensing would need to be optimized for a specific gap or a short range around it. In the following sections several approaches are proposed to overcome this issue. These approaches will be analyzed by means of a FEM simulator in Chapter 3. 2.2.1 Fractal arrangement Fractals can be defined as geometrical objects that are self-similar under a change of scale, for example, magnification [2.3]. In principle, the strips of a capacitor could be arranged in fractal geometry, in a way that the different levels of the fractals would be used for different gaps. The simplest and first discovered fractal is the Cantor set [2.4]. This fractal is constructed be deleting the open middle thirds of a set of line segments. The first step starts by deleting the open middle third from the interval [0, 1], leaving two line segments: [0, 1/3] , [2/3, 1]. Next, the open middle third of each of these remaining segments is 10 deleted. This process is continued ad infinitum. Figure 2.4 shows the first six steps of the process [2.4]. Figure 2.4: From the top to the bottom, first six steps of Cantor set [2.4]. A capacitor arranged in such way would be expected to give peaks with displacement that would be divided in several new peaks for shorter gaps. These new peaks would be again divided in new peaks for even shorter gaps, and so on. Instead of dividing a line into three segments, for the sensor design the line is divided in five segments, where the second and fourth segment would be removed. The following figures show the process. For small gaps, a couple of strips is used to sense lateral displacement. Figure 2.5: first fractal level When the gap increases, the association of the first level unit with the strips around it takes over, sensing as it was one unit. Figure 2.6: second fractal level In a similar way, for longer gaps, three second level units form a larger unit, which gives the third level. 11 Figure 2.7: third fractal level And finally, for larger gaps, the fourth level takes over Figure 2.8: fourth fractal level For a three electrodes scheme as in Figure 2.1, hypothetically the fractal arrangement could be implemented as well: Figure 2.9: first fractal level Figure 2.10: second fractal level Figure 2.11: third fractal level Figure 2.12: fourth fractal level 12 2.2.2 Switchable variable width By connecting and disconnecting different strips, the sensor configuration can be changed and adapted to the most convenient arrangement for a specific gap. With the threeelectrode configuration of Figure 2.1, for small gaps the electrodes can be arranged as in Figure 2.13. Figure 2.13: first level arrangement for short gaps. The colors represent the three groups of electrodes that are connected. For larger gaps, two consecutive plates can be grouped as one plate, obtaining a larger plate, as shown in Figure 2.14. Figure 2.14: second level arrangement for short gaps. And further on, more consecutive plates can be grouped for larger gaps as in Figure 2.15 and Figure 2.16 13 Figure 2.15: third level arrangement for short gaps. Figure 2.16: fourth level arrangement for short gaps. The sensor configuration could be changed by means of switches that would connect the different parts. 2.2.3 Central fractal The central fractal geometry is meant to obtain high resolution for short gaps while still using electrodes with large width. Starting with three groups, instead of continuing the fractal in the three groups, only the central part is replaced by three identical parts. The process can be repeated ad infinitum with the new three central parts. The geometry starts with three identical parts as shown in Figure 2.17. Figure 2.17: capacitor geometry arranged in three identical groups Then the central part is replaced by three more parts as in Figure 2.18. Figure 2.18: capacitor geometry of Figure 2.17 replacing the central group by three identical groups 14 And further on, the central part is replaced by three more parts as in Figure 2.19. Figure 2.19: capacitor geometry of Figure 2.18 replacing the central group by three identical groups To decrease the change in capacitance with the gap, the lateral groups could be disconnected and left floating once the alignment at that level has been achieved and the gap is reduced. This would probably increase the sensitivity of the sensor as well, since the capacitance from the larger groups may overlap the changes in the smaller groups for short gaps. 2.3 Fabrication The fabrication of the capacitor sensor is rather simple, with a three steps process. First, to isolate the chip substrate from the capacitor electrodes, a layer of oxide is grown. Figure 2.20: first step fabrication process. Substrate (grey), silicon dioxide (brown) Second, metal is deposited to form the electrodes. A thin layer of zinc is deposited, followed by a layer of gold. Figure 2.21: second step fabrication process. Deposition of Zn (green), and gold (yellow) 15 The third step is pattern of the metal. Photoresist is spunned on top, and exposed by optical lithography. Figure 2.22: clear mask for metal pattern. The green line delimits the section showed in following figures Figure 2.23: chip section after exposure. From the bottom to the top: substrate (grey), oxide (brown), Zn (green), gold (yellow), photoresist (blue). Once the photoresist is exposed, the metal is etched, and the photoresist removed. Figure 2.24: chip section 16 Reference: [2.1] Z. Liu, Q. Huang, W. Li, ‘A differential Capacitive Mini-Displacement Sensor’, Key Lab. Of MEMS of the Minist. Of Educ., Southeast Univ., Nanjing, China. Sensors, 2004. Proceedings of IEEE. [2.2] J. C. Lotters, W. Olthuis, P. H Veltink, ‘A sensitive differential capacitance to voltage converter for sensor applications’. IEEE Vol. 48, No. 1, February 1999 [2.3] McGraw-Hill Encyclopedia of Science and Technology. Copyright © 2005 by The McGraw-Hill Companies, Inc. [2.4] http://en.wikipedia.org/wiki/Cantor%27s_fractal_set 17 3. Modeling 3.1 Capacitance and FEM field solver When a battery is connected to two isolated conductors, the battery extracts negative charges, i.e. electrons, from one conductor to the other, until a determined difference of potential is achieved. This is shown in Figure 3.1. Conductor a e- + + + + + + + + Q + + + + + ++++++++++++++++++++++++++++ E Vs e- - - - - - - - - - - - - - - - - - - - -Q Conductor b Figure 3.1: two conductors being charged by a battery Capacitance is defined as the ratio of the charge stored to the voltage applied. C Q Vs (3.1) The capacitance of a system only depends upon the geometry of its elements and the material properties of the surrounding medium. The capacitance of simplified models, such as uniformly charged parallel plate capacitor, can be obtained analytically by resolving Maxwell’s equations. Although analytical solutions to more complex geometries are sometimes available in transmission line and microstrips theory literature [3.1]-[3.4], numerical methods are needed to obtain accurate solutions for systems composed by more than two electrodes. 18 Finite Element simulations calculate ground capacitance, which relates the charge on each conductor with the conductor’s voltage drop to ground [3.5]. The following figure illustrates a system with four conductors, where the fourth one is ground. Figure 3.2: Multielectrode system with three conductors and ground The ground capacitances are given by Eqns. 3.2-3.4 Q1=Cg11U1+Cg12U2+Cg13U3 (3.2) Q2=Cg21U1+Cg22U2+Cg23U3 (3.3) Q3=Cg31U1+Cg32U2+Cg33U3 (3.4) Where Ui is the potential of conductor i, Qi is the charge of conductor i, and Cgij is the ground capacitance of electrode i to electrode j. A matrix of ground capacitance is defined from the coefficients of Eqns. 3.2-3.4 C g11 C g12 C g C g 21 C g 22 C g 31 C g 32 C g13 C g 23 C g 33 (3.5) For a linear system Cgij=Cgji, and therefore Cg becomes symmetric. Because ground capacitance does not relate the capacitance between conductors, circuit simulators such as Spice use lumped element capacitance models. A lumped model is defined as a model in which the dependant variables of interest are function of time only, as opposed to 19 distributed models, whose variables are function of time and more spatial variables [3.6]. The lumped circuit model of the system of Figure 3.2 is shown in Figure 3.3. Figure 3.3: Lumped capacitor equivalence of four conductor system The lumped matrix capacitance is extracted from the ground capacitance equations. Q1=(Cg11+Cg12+Cg13)(U1-0)+(-C12)(U1-U2)+ (-C13)(U1-U3)=Cl11U1+Cl12(U1-U2) +Cl13(U1-U3) Q2=(-C21)(U2-U1)+ (Cg21+Cg22+Cg23)(U2-0)+ (-C23)(U2-U3)=Cl21(U2-U1)+Cl22U2 +Cl23(U2-U3) Q3= (-C31)(U3-U1)+ (-C32)(U3-U2)+ (Cg31+Cg32+Cg33)(U3-0)=Cl31(U3-U1)+Cl32(U3-U2)+Cl33U3 (3.6), (3.7), (3.8) The lumped matrix capacitance is therefore obtained from the ground capacitance with the following relation. C g11 C g12 C g13 C g12 C g13 Cl C g12 C g12 C g 22 C g 23 C g 23 C C C C C g 13 g 23 g 13 g 23 g 33 (3.9) To illustrate an example of matrix capacitance, the results from a capacitance extraction simulation of a five electrode system with ANSYS are shown below 20 ________________ CMATRIX SOLUTION SUMMARY ___________________ *** Ground Capacitance Matrix *** Self Capacitance of conductor 1. = Self Capacitance of conductor 2. = Self Capacitance of conductor 3. = Self Capacitance of conductor 4. = Self Capacitance of conductor 5. = Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Ground capacitance matrix is stored in *** Lumped Capacitance Matrix *** Self Capacitance of conductor 1. = Self Capacitance of conductor 2. = Self Capacitance of conductor 3. = Self Capacitance of conductor 4. = Self Capacitance of conductor 5. = Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Mutual Capacitance between conductors Lumped capacitance matrix is stored in 0.30142E-03 0.30118E-03 0.54565E-03 0.35634E-02 0.35719E-02 1. and 2. = -0.45432E-06 1. and 3. = -0.19700E-03 1. and 4. = -0.24790E-04 1. and 5. = -0.79176E-04 2. and 3. = -0.19686E-03 2. and 4. = -0.24780E-04 2. and 5. = -0.79081E-04 3. and 4. = -0.12600E-03 3. and 5. = -0.25787E-04 4. and 5. = -0.33717E-02 3d array parameter cmatrix ( 5., 5.,1) 0.22795E-16 0.33339E-16 0.39858E-16 0.16123E-04 0.16122E-04 1. and 2. = 0.45432E-06 1. and 3. = 0.19700E-03 1. and 4. = 0.24790E-04 1. and 5. = 0.79176E-04 2. and 3. = 0.19686E-03 2. and 4. = 0.24780E-04 2. and 5. = 0.79081E-04 3. and 4. = 0.12600E-03 3. and 5. = 0.25787E-04 4. and 5. = 0.33717E-02 3d array parameter cmatrix ( 5., 5.,2) Capacitance values are per unit length Capacitance matricies are stored in file cmatrix .txt _____________________________________________________________ Figure 3.4: capacitance extraction simulation output with ANSYS. The solution given by ANSYS includes both the ground capacitance matrix from the simulations and the lumped capacitance matrix for a lumped circuit model. In order to obtain the matrix capacitance of a system, the following equation, known as Gauss’s law, can be used. Qi o r E (r )dAi (3.10) Ai where Qi is the charge of electrode -i, E (r ) is the electric field at point r , εo and εr the vacuum permittivity and medium relative permittivity, and Ai is the area of integration around electrode -i. Due to numerical reasons, a more efficient way to calculate capacitance is by using the principle of energy conservation [3.7]. 21 W 2 1 1 o r (r ) E (r ) dV E (r ) D(r )dV 2 V 2V (3.11) Instead of using Eqn. (3.1), the following relation of energy and capacitance is used. W 1 CU 2 2 (3.12) Eqn. (3.12) is used to set up a system of equations that relates voltage differences to energy stored. C W U1,12 2 U1,132 U1,14 2 U1, 232 U1, 24 2 U1,34 2 12 1 2 2 2 2 2 2 C13 W2 U 2,12 U 2,13 U 2,14 U 2, 23 U 2, 24 U 2,34 C W 1 14 3 . . . . W4 C23 2 . . U 2 U 2 U 2 U 2 U 2 U 2 C24 W5 6 ,13 6 ,14 6 , 23 6 , 24 6 , 34 6,12 C34 W6 (3.13) Eqn (3.13) represents a system of 4 electrodes, which has 6 unknown capacitances. Even though different electrode potentials would give a completely different electric field distributions, the capacitance must remain the same because it is exclusively a geometric and material factor. Using this principle, any electrode potential Uk,ij can be used as boundary condition. Once the electric field is obtained for a particular boundary condition, the energy is calculated with the numerical equivalent of Eqn. 3.11 Wk 1 l n 2 2 2 2 2 2 E xk ,l E yk ,l E zk ,l Dxk ,l D yk ,l Dzk ,l velem,l 2 l 1 (3.14) The process is repeated until the number of required independent equations is obtained. Once mutual capacitances are calculated, self capacitances can easily be calculated with the following expression Cii 22 n C j 1, j i ij (3.15) It can easily be verified that the ground matrix capacitance given in the example above matches Eqn. (3.15). 3.2 Modeling scenarios In electrostatic, magnetic, and thermal domain, a system model can be classified as open scenario or closed scenario [3.5]. A closed scenario is a model whose field is restricted to a determined area by its boundary conditions. Figure 3.4: Three conductor system modeled as closed scenario On the other hand, an open scenario assumes that the field is not restricted and extends indefinitely. An open scenario can be seen as a closed scenario with boundary conditions at the infinity. There are three main approaches to model the far-field. The simplest one is to assume a natural boundary condition that the field has to match. This is illustrated in Figure 3.5. Figure 3.5: Three conductor system. Natural boundary condition to model far-field A more accurate approach to model far-fields is by defining Far-field elements around the area to model, as illustrated in Figure 3.6 23 Figure 3.6: Three conductor system. Far-field elements (infinite elements) to model far-field The most accurate approach, although restricted to 3D models, is to define a Trefftz domain. This method models the far-field with hybrid infinite elements and Trefftz nodes. This is illustrated in Figure 3.6. Figure 3.6: Three conductor system. Trefftz method to model far-field If the plates to align were conductive and could be connected to ground, a closed model scenario could be used to simulate the system. However, for the most general case, no connection to either plate is assumed. Therefore an open model scenario must be used. Due to the large number of strips that compose the sensor, and the longitudinal geometrical symmetry, 2D simulations are accurate and far more computationally efficient. Thereby, infinite elements were used as modeling scenario for the project’s sensor. 3.3 Equivalent circuit In Figure 2.1 the capacitances between electrodes 1, 2, and 3 are illustrated. For the most general case, however, the plates which will be aligned are conductors as well. This 24 means that the system is composed of 5 electrodes instead of 3, with mutual capacitances from every electrode to every electrode. Figure 3.7: Sensor capacitance model The scheme of Figure 3.7 can be simplified to a circuit with lumped capacitances. This circuit is shown in Figure 3.8. Figure 3.8: Sensor equivalent circuit The equivalent circuit of Figure 3.8 can be simplified to an equivalent input impedance from nodes 1-2. To do so, a test voltage is applied between nodes 1 and 2. By applying the loop-current method, the test current is obtained in function with the equivalent impedance. In order to resolve a circuit with b branches and n nodes, b-n+1 equations are sufficient to obtain the current through every branch [3.8]. The following figure shows the six branches whose equations are to be obtained, and the current references. 25 Figure 3.9: Circuit graph with selected tree and branch current references The selected branches are a, b, c, d, e, and f. The current in the remainder branches g, h, j, and k can be obtained as a function of the selected branches. ig=ia+ib+if (3.16) ih=ig -id -ie=ia+ib+if -id - ie (3.17) ij=if+ik-id=ib+ic+if -id -ie (3.18) ik=ib+ic –ie (3.19) Assuming an impedance Zij at every branch, and applying a test voltage in branch g, the following circuit is obtained Figure 3.10: Circuit to obtain equivalent impedance To resolve the circuit, the current equations of the six fundamental branches are obtained as in Table 3.1. 26 Loop Branches Equation a a, h, g 0=Z23 ia + Z13 ih - Vt b b, k, j, h, g 0=Z24 ib + Z45 ik – Z35 ij + Z13 ih -Vt c c, k, j 0=Z34 ic + Z45 ik – Z35 ij d d, h, j 0=Z15 id – Z13 ih + Z35 ij e e, h, j, k 0=Z14 ie – Z13 ih + Z35 ij –Z45 ik f f, j, h, g 0=Z25 if – Z35 ij + Z13 ih - Vt Table 3.1: Fundamental branches, loops’ branches, and loop equations Substituting the currents ig, ih, ij, and ik from Eqns. (3.16)-(3.19) into the equations in Table 3.1, a system of six equations with six unknown currents is obtained. Calculated the fundamental branch currents, the test current it is obtained with Eqn. (3.16). The equivalent impedance is then calculated as, Z equiv it Z12 . Vt (3.20) To resolve the system of equations for generic capacitance Cij , the equations from Table 3.1 were implemented in MATLAB. The code and results are given in appendix A. A first interesting result that can be extracted from the output capacitance expression is the weight of every capacitance over the total output capacitance. This can be obtained as the partial derivative of the output capacitance Cout Cout (Cij Cij ) Cout (Cij ) Cij Cij Cij (3.21) Table 3.2 shows the derivative of the output capacitance respect to each lumped capacitance. The lumped capacitance values were extracted from one of the simulation of Section 3.4. 27 Cij Set up point [pF/μm] dC/dCij C12 3.8079e-5 1 C13 1.108e-5 0.2148 C14 7.677e-4 0.2488 C15 5.518e-4 0.2453 C23 7.263e-6 -0.2596 C24 7.692e-4 0.2444 C25 5.540e-4 0.2393 C34 9.059e-4 -0.0078 C35 1.265e-4 -0.0056 C45 0.2314 -3.01e-5 Table 3.2: Derivative of output capacitance respect to each lumped capacitance. Results from one of the simulation of section 3.4. The capacitance between the two plates (C 45), and from electrode 3 to both plates have almost no influence in the output capacitance 3.4 Simulations 3.4.1 Algorithm A finite element simulation software (ANSYS), was used to model and simulate the sensor. As illustrated in Figure 3.4, ANSYS calculates the matrix capacitance for a given model. A model is defined basically by its geometry, material properties, element properties, electrode’s nodes, and mesh. To obtain the sensor response to lateral displacement, a series of simulations was performed, in which the plates' relative position was varied. This resulted in vectors of capacitance and displacement, which can be plotted as capacitance versus displacement. Once the lumped matrix capacitance is obtained, the equivalent capacitance is calculated as explained in Section 3.1. This gives the input equivalent capacitance that a readout circuit connected to the sensor would ‘see’. Since a large number of simulations had to be carried out for every geometry to obtain curves capacitance-displacement, macros in APDL language were written. The macro 28 codes for the simulations are given in different Appendices. The diagram in Figure 3.11 shows the algorithm on which the macros were based. k=1 Plate displacement δ=[δ1, …, δN] k=k+1 δk k = 1? no Remove model yes Define geometry material properties and element type Define number of nodes Mesh Set up external flag and electrodes solve Store results Cijk k=N? no yes Cij=[Cij1,…,CijN] Figure 3.11: Flow diagram of FEM simulations with ANSYS 29 3.4.2 Fringing field effect The variation or sensitivity of capacitance per unit length to displacement with the infinite parallel plate model can be calculated as the derivative of capacitance with respect to lateral displacement. By using the well-known capacitance of infinite parallel plates, the sensitivity is dC d ( w x) n n dx dx g g (3.21) where g is the gap, ε the air permittivity, w the electrode width, and n the number of parallel capacitors. In actuality, the capacitance displacement sensitivity is lower than in Eqn. 3.21 due to fringing field capacitance. Eqn. 3.21 does not take into account the fringing field capacitance, which is much less sensitive to lateral displacement than the ideal ‘overlap’ capacitance of Eqn. 3.21. C real Cideal C fringing nwL C fringing g (3.22) Figure 3.12 shows the variation of dimensionless capacitance to lateral displacement for different gaps for a simple parallel plate geometry. 2 electrodes 1 um width capacitor dimensionless capacitance 1 0.9 0.8 0.7 gap 0.1 um gap 0.2 um gap 0.3 um gap 0.5um gap 1 um gap 2 um gap 5 um gap 10 um gap 100 um 0.6 0.5 0.4 0.3 0.2 0.1 0.5 1 1.5 2 2.5 3 3.5 4 4.5 displacement [um] Fig 3.12: Dimensionless capacitance ( C(x)/C(x=0)) versus displacement for different gaps. The system modeled was two parallel plates of 1μm width and negligible thickness. 30 Figure 3.12 shows how capacitance becomes less sensitive to displacement for larger gaps. Eqn. 3.21 does not hold for large gaps, in which case the fringing field capacitance is no longer negligible. The percentage of fringing field capacitance to total capacitance is shown in Figure 3.13 for different gaps. Fig 3.13: Ratio of fringing field capacitance (Cnumerical-Canalytical) to total capacitance (Cnumerical). Analytical capacitance is calculated with the ideal parallel plate capacitance equation. Numerical capacitance is calculated with a FEM simulator and includes both the ideal term and fringing term of Eqn. 3.22. The system modeled was two parallel plates of 1μm width and negligible thickness. Although Figure 3.13 shows fringing field for the case of two electrodes capacitor of Figure 1.4, the results can be extrapolated to the 3 electrode capacitor of Figure 2.1. When capacitance is dominated by fringing field capacitance, the capacitance sensitivity to lateral displacement becomes negligible because the fringing effect sensitivity to lateral displacement is very low. This can be extracted from Figure 3.13 and Figure 3.12. For 100 μm gap, Figure 3.13 shows that capacitance is mostly made of fringing field effect. For the same gap, Figure 3.12 shows that this capacitance, which is basically fringing field, has very little variation with lateral displacement. According to Figure 3.13, if fringing fields want to be kept under 30% of the total field when the plates are aligned, the ratio of gap to width should be in the order of 1/3. In this way, a minimum capacitor displacement sensitivity is guaranteed. 31 3.4.3 Influence of plates conductivity The influence of the plates’ conductivity was analyzed by using different models that include the plates as conductors or dielectrics. -Simplified model, 5 electrodes To reduce the number of nodes and make the simulations more time efficient, simplified models with just a small fraction of the plates and reduced number of strips were used. Figures 3.14-3.16 show the FEM model. Figure 3.14: FEM model of capacitive sensor The outer ring with square elements encloses the infinite elements modeling the far-field. The triangular elements model the surrounding medium, which is air. And the two white squares are sections of bottom and top plate. Between the two plates, the thin area models the sensor’s electrodes, air and oxide. Figures 3.15-3.16 focus on this area. Figure 3.15: Space between plates from Figure 3.14 32 Figure 3.16: Zoom in figure 3.14 showing electrodes, gap air, oxide layers, and plates Figure 3.16 shows a part of the two plates, the three electrodes that compose the basic sensor unit, two meshed layers of oxide separating the electrodes and the plates, and the meshed space between the plates. A description of the model’s geometry is summarized in Table 3.3. gap electrode thickness width bottom electrodes width top electrode thickness oxide layer thickness Si substrate plates width plate spacing between two bottom electrodes spacing bottom electrode different groups electrode length plate lenght Number of groups Electrodes per group Total number of electrodes 0.1 μm 0.1 μm 1 μm 2 μm 0.5 μm 381 μm 500 μm 1 μm 3 μm 500 μm 15 mm 3 3 (2 bottom, 1 top) 9 Table 3.3: model geometry By changing the relative position of bottom and top plate, the change of capacitance matrix versus displacement is obtained. Figures 3.17-3.21 show the results obtained. The code is given in Appendix B. 33 -3 3.5 capacitance simulations g=0.1 um x 10 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 3 capacitance [pF/um] 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.17: Lumped mutual capacitances per unit length. The electrodes are numbered as in Figure 2.1 and 3.7 As expected, the largest capacitance is C45, which is the capacitance between the two plates. Figure 3.18 focuses on the other capacitances. -4 capacitance simulations g=0.1 um x 10 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 capacitance [pF/um] 3 2 1 0 -3 -2 -1 0 1 displacement [um] 2 3 Figure 3.18: Lumped mutual capacitances per unit length. Figure 3.16 shows that the capacitances C13 and C23 change drastically with displacement. Capacitance C12 can be neglected compared to C13 and C23, in agreement to an assumption commonly used in differential capacitance sensors [2.1]. However, the rest of capacitances are in the same order of C13 and C23 and must be taken into account. Figures 3.19-3.25 show the total capacitance in picoFarads. 34 total capacitance one group g=0.1 um 60 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 capacitance [pF] 50 40 30 20 10 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.19: Lumped mutual capacitances. total capacitance one group g=0.1 um c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 0.16 0.14 capacitance [pF] 0.12 0.1 0.08 0.06 0.04 0.02 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.20: Lumped mutual capacitances. Plugging the values from the capacitance matrix in the expression of equivalent input capacitance deduced in Appendix B, input capacitance versus displacement curves are obtained. Figures 3.21-3.22 show these results. 35 total capacitance one group g=0.1um c12 c13 c23 cout 0.16 0.14 capacitance [pF] 0.12 0.1 0.08 0.06 0.04 0.02 0 -3 -2 -1 0 1 displacement [um] 2 3 Figure 3.21: C12, C13, and C23 from the lumped matrix capacitance, and equivalent input capacitance (Cout). total capacitance one group g=0.1um c12 c13 c23 cout 0.075 0.07 capacitance [pF] 0.065 0.06 0.055 0.05 0.045 0.04 0.035 -3 -2 -1 0 1 displacement [um] 2 3 Figure 3.22: C13 and C23 from the lumped matrix capacitance, and equivalent input capacitance (Cout). Scaling up the results obtained with the simplified model, results for a model in scale can be obtained. Figures 3.23-3.25 show the result of the simplified model scaled-up by assuming 15 mm x 15 mm plates, and 27 groups of strips, across the 500 μm sensor width. 36 total capacitance 27 groups g=0.1 um 1400 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 1200 capacitance [pF] 1000 800 600 400 200 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.23: Lumped mutual capacitances. total capacitance 27 groups g=0.1 um 4.5 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 4 capacitance [pF] 3.5 3 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 Figure 3.24: Lumped mutual capacitances. total capacitance 27 groups g=0.1um 4.5 c12 c13 c23 cout 4 capacitance [pF] 3.5 3 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.25: C13 and C23 from the lumped matrix capacitance, and equivalent input capacitance (Cout). 37 -Model for large gap and displacement range, 5 electrodes For larger gaps and displacement ranges, a more precise model was developed. This model is summarized in Table 3.4. gap electrode thickness width bottom electrodes width top electrode thickness oxide layer thickness Si substrate plates width plate spacing between two bottom electrodes spacing bottom electrode different level 1 spacing bottom electrode different level 2 spacing bottom electrode different level 3 electrode length plate lenght Number of electrodes group level 1 Number of groups level 1 Number of groups level 2 Number of groups level 3 Total number of electrodes 20 μm 0.1 μm 1 μm 2 μm 0.5 μm 381 μm 500 μm 1 μm 2 μm 13 μm 65 μm 500 μm 15 mm 3 3 3 3 81 Table 3.4: characteristics model for large gap and deflections The APDL macro is given in Appendix C. Figures 3.26-3.27 show the meshed model. Figure 3.26: FEM model of sensor from table 3.4. Detail gap and 9 group of electrodes 38 Figure 3.27: FEM model of sensor from table 3.4. Detail of gap and 3 groups of electrodes The curves equivalent input capacitance versus displacement are shown in Figure 3.283.29. -4 1.64 x 10 output capacitance, 3x3x3x3 model, g=20 um 1.63 capacitance [pF/um] 1.62 1.61 1.6 1.59 1.58 1.57 1.56 -100 -80 -60 -40 -20 0 20 displacement [um] 40 60 80 100 Figure 3.28: Input equivalent capacitance per unit length versus displacement Due to the large number of electrodes, the number of nodes per electrode had to be kept low to avoid an excessive number of elements in the model. Because of the lower number of nodes per electrode, the results for different displacements showed less consistency than in other simulations. Figure 3.29 filters the results to minimize the error. 39 -4 filtered output capacitance, 3x3x3x3 model, g=20 um x 10 1.62 capacitance [pF/um] 1.61 1.6 1.59 1.58 1.57 -80 -60 -40 -20 0 20 displacement [um] 40 60 80 100 Figure 3.29: Input equivalent capacitance per unit length versus displacement. Results from ANSYS filtered with MATLAB (MATLAB code given in Appendix D) -Model for connected plates, 4 electrodes During the alignment, the nanopositioner may connect electrically both plates. For this case, a model was developed to obtain the input equivalent capacitance. In this case, instead of 5 electrodes, there would be 4 electrodes. The equivalent input impedance in function with the lumped capacitances was obtained as well. Its deduction is given in Appendix F. The model has the characteristics given in Table 3.4. The results from the FEM simulator are shown in Figure 3.30-3.31. -4 3.5 x 10 substrates wired,three groups, per unit lenght g=0.1 um c12 c13 c14 c23 c24 c34 3 capacitance [pF/um] 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 Figure 3.30: Lumped capacitance matrix 40 4 -4 3.5 x 10 substrate wired, three groups, per unit lenght, g=0.1um c12 c13 c23 cout 3 capacitance [pF/um] 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.31: C12, C13, and C23 from lumped capacitance matrix and input equivalent capacitance (cout) -Model for glass top electrode Another model was developed to analyze the influence of non-conductive plates. In this case, the plate on top was modeled as Silicon dioxide. Figures 3.32-3.33 show the model meshed. Figure 3.32: Glass top plate meshed model. It can be seen in the figure that the space for the top plate has been meshed 41 Figure 3.33: Glass top plate meshed model. The detail of the gap shows, from the bottom to the top, air not meshsed, a layer of silicon dioxed, the electrodes, air between the electrodes, a top electrode, and a large surface meshed which is the top plate The results from the model with the characteristics of Table 3.4 are shown in Figure 3.343.35. Figure 3.34: Lumped capacitance matrix model glass top plate -4 1.5 three groups, top substrate glass, g=0.1um x 10 1.4 capacitance [pF/um] 1.3 1.2 1.1 1 0.9 0.8 0.7 0.6 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.35: Input equivalent capacitance model glass top plate 42 -Model for glass top electrode In Figures 3.36-3.38, simulations for a model with glass bottom plate are shown. Figure 3.36: Meshed model with glass bottom plate. Meshed model with glass bottom plate. Detail of gap on the left -4 3.5 three groups, bottom substrate glass, g=0.1um x 10 c12 c13 c14 c23 c24 c34 3 capacitance [pF/um] 2.5 2 1.5 1 0.5 0 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.37: Lumped capacitance matrix model glass bottom plate 43 -4 1.4 three groups, bottom substrate glass, g=0.1um x 10 1.3 capacitance [pF/um] 1.2 1.1 1 0.9 0.8 0.7 0.6 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.38: Input equivalent capacitance model glass bottom plate -Four models compared To analyze the influence of the plates in the equivalent input capacitance, the following figure compares the results from the four models above. simulations 4 simplified models 2.2 top subs glass subs wired bottom subs glass 5 electrodes 2 capacitance [pF] 1.8 1.6 1.4 1.2 1 0.8 -4 -3 -2 -1 0 1 displacement [um] 2 3 4 Figure 3.39: Equivalent input capacitance from models with different plate characteristics. Although whether the plates are conductive or dielectric has an influence on the equivalent input capacitance, the sensor would work for all the cases studied. 44 3.5 Convergence To obtain the optimal mesh in terms of accuracy and computation time, the algorithm convergence was analyzed. Because the greatest field gradient is at the space between electrodes 1, 2, and 3 (see Figure 2.1), the mesh must be finer around them. The macros used for capacitance extraction specify the number of nodes on the surface of electrodes 1, 2, and 3 surfaces, and leave the rest as auto mesh. This results in a much denser mesh around the electrodes. To find the optimal number of nodes on the electrodes’ surface, simulations changing the number of nodes were carried out. Figures 3.40-3.41 show a mesh with 5 nodes per electrode and a mesh with 105 nodes per electrode. The APDL macro is given in Appendix E. Figure 3.40: Detail of meshed model with 5 nodes per electrode’s horizontal surface Figure 3.41: Detail of meshed model with 105 nodes per electrode’s horizontal surface Figures 3.42-3.44 show the dimensionless lumped matrix capacitance versus the number of nodes per electrode surface for different gaps. 45 convergency, 3x3 model, g=0.1 um capacitance(nodes)/capacitance(5 nodes) 1.03 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 cout 1.02 1.01 1 0.99 0.98 0.97 0.96 0.95 0.94 0 20 40 60 80 nodes per surface electrode 100 120 Figure 3.42: Dimensionless lumped matrix capacitance. Gap 0.1 μm. The capacitance is divided by the capacitance obtained with 5 nodes per electrode surface convergency, 3x3 model, g=1 um capacitance(nodes)/capacitance(5 nodes) 1.015 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 cout 1.01 1.005 1 0.995 0.99 0.985 0.98 0 20 40 60 80 nodes per surface electrode 100 120 Figure 3.43: Dimensionless lumped matrix capacitance. Gap 1 μm. convergency, 3x3 model, g=10 um capacitance(nodes)/capacitance(5 nodes) 1.004 c12 c13 c14 c15 c23 c24 c25 c34 c35 c45 cout 1.002 1 0.998 0.996 0.994 0.992 0.99 0.988 0.986 0.984 0 20 40 60 80 nodes per surface electrode 100 120 Figure 3.44: Dimensionless lumped matrix capacitance. Gap 10 μm. 46 Figures 3.42--3.44 show that reaching convergence requires less number or nodes for larger gaps. This was expected, since shorter gaps induce greater potential gradients. A large gradient leads to greater errors in the integration, so to keep the error low the number of nodes has to be increased. Reference: [3.1] F. Stellari, A. L. Lacaita, ‘New Formulas of Interconnect Capacitances Based on Results of Conformal Mapping Method’, IEEE Vol. 47, No. 1, January 2000 [3.2] N. D. Arora, K. V. Raol, R. Schumann, L. M. Richardson, ‘Modeling and Extraction of Interconnect Capacitances for Multilayer VLSI Circuits’, IEEE Vol. 15, No.1 January 1996. [3.3] A. Toulouse, D. Bernard, C. Landrault, P. Nouet, ‘Efficient 3D Modelling for Extraction of Interconnect Capacitance in Deep Submicron Dense Layouts’. [3.4] U. Choudhury, A. Sangiovanni-Vincentelli, ‘Automatic Generation of Analytical Models for Interconnect Capacitances’, IEEE 1994. [3.5] ‘Ansys Low-Frequency Electromagnetic Analysis Guide’, Ansys Release 10.0, August 2005 [3.6] http://ccrma.stanford.edu/~jos/NumericalInt/Lumped_vs_Distributed_Systems.html [3.7] A. Hieke, ‘Simple APDL implementation of a 3D FEM simulator for mutual capacitances of arbitrarily shaped objects like interconnects’, SIEMENS Microelectronics, c/o IBM Semiconductor Research & Development Center. [3.8] J, W. Nilsson, ‘Electric Circuits’, Addison-Wesley Publishing Company, Fourth Edition. 47 4. Capacitance-to-voltage converters 4.1 Introduction In chapter 2 the sensor was modeled by a lumped capacitor connected to a circuit. To measure the impedance of a capacitor, an excitation is applied to one of the terminals, as illustrated in Figure 4.1 Figure 4.1: Capacitance in function with displacement and voltage supplied The output current through the capacitor is the derivative of stored charge with respect to time. By using the definition of capacitance C=Q/V, and the definition of current i=dQ/dt, the following expression is obtained iC dQ dV C C V dt dt t (4.1) where V is the voltage applied, and δ the displacement. Assuming that the excitation voltage is a sinusoidal function whose frequency is much higher that the rate of capacitance change with displacement, Eqn. (4.1) is simplified to iC C dV d C A sin( t ) CA cos(t ) dt dt 48 (4.2) where A is the applied voltage amplitude, and ω the frequency. According to Eqn. (4.2), a capacitor with a sinusoidal voltage excitation can be modeled as a sinusoidal current source modulated by the capacitance itself. This description can easily be extended to any periodic excitation as long as the excitation frequency is much larger than the variation of capacitance over time. 4.2 Converters A converter is an electrical circuit aimed at transforming an electrical signal to another more convenient electrical signal. As explained in Section 4.1, the output of a capacitor is a periodic current modulated by the capacitance. To electrically process this signal, it must be converted to a different electrical signal easier to process, such as a bias voltage linear with the capacitance or a digital signal. Because the purpose of a capacitance-to-voltage converter is to detect changes in capacitance, the performance will be mainly evaluated by the circuit’s ability to read the minimum possible change in capacitance, which is limited by the circuit’s noise. The root-mean-square (RMS) of a circuit’s noise waveform vn(t) is defined as [4.13] tˆ / 2 v n rms v lim ˆ 2 n t (v n (t )) 2 dt (4.5) tˆ / 2 Similar to the definition of RMS voltage noise, the RMS minimum detectable capacitance is defined as [4.14] C min rms vnrms dvo / d (C ) 49 (4.6) where vo is the output voltage and ΔC is a change in the capacitance to measure over the initial capacitance. Extensive literature research on capacitance converters was carried out during this project to obtain a wide understanding of the different approaches and select the best readout circuit. The most common converters are described and analyzed throughout the following sections. 4.3 Oscillators Capacitance-controlled oscillators have been widely analyzed in the instrumentation literature. However, currently other types of converters are replacing this technique. A capacitance-controlled oscillator is based on an oscillator whose time period is modulated by a capacitance. A modification of Martin oscillator is shown in Figure 4.2 [4.2] Figure 4.2: Modified Martin oscillator for multielectrode capacitive sensors [4.2]. The period of the circuit in Figure 4.2 is T=4R(Coffset-Cos+Cx,i) (4.5) which is proportional to the sensor capacitance Cx,i. Although large accuracy and linearity can be achieved with this technique, the output would require complex signal processing. Furthermore, the oscillator performance is highly dependant on parasitic capacitances. 50 4.4 Capacitive bridge Capacitive bridges are based on the well-known Wheatstone bridge [4.3]. A possible implementation is shown in the Figure 4.3. Figure 4.3: Capacitive bridge The output voltage of such circuit is Vo V C1 C 2 Cf (4.6) In the scheme of Figure 4.3, the capacitance C1 and would be replaced by the sensor equivalent capacitance, and C2 by a reference capacitor. The main disadvantage of this circuit is its high sensitivity to parasitic capacitances. 4.5 Transimpedance amplifier 4.5.1 Gain and noise The transimpedance amplifier, or charge amplifier, is basically a current to voltage converter, i.e. a circuit that converts a current modulated signal to a voltage modulated signal. The following figure shows the schematic of such circuit, modeling the capacitor and its voltage supply as a current source 51 Figure 4.4: Transimpedance amplifier The output voltage of the circuit is Vo = -Rf Is (4.7) Modifying the circuit of Figure 4.4 for differential input, and including parasitic capacitances, the circuit results in that shown in Figure 4.5 Figure 4.5: Transimpedance amplifier, sensor capacitance, and excitation. Cref can be used to cancel an offset of Cs. Assuming that Cs=Coff+ΔC, if Cref=Coff, the circuit will only measure ΔC. where Cr is a reference capacitance, and Cp models any parasitic capacitance at the inverting input of the op-amp. Applying Eqn. (4.7), the output of this circuit is │Vo│=2 π fdrive Vs Rf ΔC (4.8) being Vs the input voltage amplitude, fdrive the input excitation frequency, Rf the feedback resistor, and ΔC the change in capacitance over Cs. The feedback resistor, with the parasitic and sensor capacitance, creates a low pass filter, which limits the bandwidth. The pole of this filter is [4.6] f pole 1 1 2 R f (2C s C p ) 52 (4.9) The optimal circuit performance is obtained at resonance frequency, which is given by the pole frequency given in Eqn. (4.9) and the amplifier gain bandwidth. Therefore, the optimal drive frequency is f driveopt GBWamp 1 2R f (2C s C p ) (4.10) where GBWamp is the amplifier gain bandwidth. By substituting Eqn. (4.10) in Eqn. (4.8), the output voltage for an optimal drive frequency is obtained Vo 2 (GBWamp ) R f (2C s C p ) Vm (C ) (4.11) The amplifier noise can be modeled by ideal components and current/voltage sources dependent on noise. The following circuit shows the amplifier with ideal components and the noise sources. Figure 4.6: Transimpedance amplifier with noise sources By applying superposition, the output voltage noise is v n ,rms en ( R f in ) 2 4 K B TR f 4 K B TR f 2 (4.12) The approximation of Eqn. 4.12 assumes that an optimal ultra-low noise op-amp is used, so that the feedback resistor is the dominant noise source. The minimum detectable capacitance as defined in Eqn. 4.6 is: 53 C min rms 2 K BT (2C s C p ) 1 Vm (GBWamp ) BW (4.13) Even though the feedback resistor is the main source of error, the minimum capacitance detectable does not depend on the feedback resistor value. While the feedback resistor increases the noise linearly with its resistance value, the capacitance measurement is also amplified linearly with the feedback resistance. Therefore the feedback resistance value is cancelled out for the signal to noise ratio, and the minimum capacitance detectable becomes independent of the feedback resistance. Figures 4.7-4.8 show the minimum detectable capacitance for different cases. minimum detectable capacitance transimpedance amplifier -3 1.5 x 10 Cmin-rms [aF/sqrt(BW)] 1.45 1.4 1.35 1.3 1.25 1.2 1.15 1.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 parasitic capacitance [pF] Figure 4.7: Minimum detectable capacitance by transimpedance amplifier assuming Vm=1V, T=300 K, GBWamp=1 MHz, Cs=1pF 54 minimum detectable capacitance transimpedance amplifier 3 10 Csensor=1nF Csensor=1pF Csensor=1fF 2 Cmin-rms [aF] 10 1 10 0 10 -1 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 parasitic capacitance/sensor capacitance Figure 4.8: Minimum detectable capacitance by transimpedance amplifier assuming Vm=1V, T=300 K, GBWamp=1 MHz, BW=100 KHz. 4.5.2 Stability The amplifier model of Figure 4.4 assumes no parasitic capacitances and infinite resistance. A more realistic model is shown in Figure 4.9. Figure 4.9: Transimpedance amplifier with parasitic capacitance Cs and sensor finite resistance Rs Cs introduces two drawbacks in the circuit. First, as the frequency increases, the impedance of Cs decreases. Therefore, at high frequencies Cs acts as a short circuit, and so the current from the source Is flows through Cs instead of through Rf. This leads to a reduction of bandwidth. Second, Rf and Cs create a low pass filter, which introduces negative phase into the system and reduces the bandwidth. Figures 4.10-4.11 show the 55 frequency and time response of the circuit in Figure 4.9 under different stray capacitances Cs. Figure 4.10: Output voltage Vo for Cs=0, Cs=50 pF, and Cs=200 pF. Figure 4.11: Frequency response for Cs=0, Cs=50 pF, and Cs=200 pF. Figure 4.10 shows how the output voltage oscillation increases with larger values of stray capacitance. The reduction of bandwidth due to stray capacitance is illustrated in Figure 4.11. To improve the stability, a resistance is included in parallel in the feedback as in Figure 4.12. 56 Figure 4.12: Transimpedance amplifier with capacitive feedback to improve stability. The values used for the simulations are indicated in the figure The feedback capacitor Cf creates a high pass filter with Rs, which adds positive phase to the system and pushes the circuit into stability. Figure 4.13 shows simulation of the circuit in Figure 4.12 under different values of Cf. Figure 4.13: Simulations circuit Figure 4.12 for Cf= 0, Cf=2 pF, Cf=4 pF The drawback of this approach is the reduction of bandwidth due to the low pass filter composed by Cf and Rf.. 57 Figure 4.14: Frequency response simulations circuit Figure 4.12 for Cf= 0, Cf=2 pF, Cf=4 pF 4.5.3 Synchronous demodulator The outputs of the transimpedance amplifier and capacitive bridge are both sinusoidal voltages modulated by the sensor capacitance. To convert this signal to a more convenient DC signal proportional to the capacitance, a synchronous demodulator can be used. This circuit ideally implements the following equations Vo = Vin if Vin>0 Vo= -Vin if Vin<0 (4.14) The following figure shows the diagram of this circuit Figure 4.15: synchronous demodulator circuit. An amplifier obtains the negative of the input signal. A periodic excitation, usually a square wave, controls two switches that close and open at opposite times. 58 An accurate and flexible converter can be implemented by applying a high frequency voltage to the unknown impedance and another known reference capacitance, amplifying the signal, applying it to a demodulator, and low-pass filtering it. A scheme is shown in the Figure. Figure 4.16: capacitance to voltage converter based on synchronous demodulation The following figure shows the output waveform of the different blocks Figure 4.17: output waveforms from the different blocks of a capacitance to voltage amplifier. From left to right, amplifier, synchronous demodulator, and low-pass filter 4.6 Switched-capacitor Converter 4.6.1 Switched-capacitor Amplifier Switched-capacitor circuits are based on replacing resistances by capacitors and switches. Figure 4.17 shows a simple circuit with a resistance 59 Figure 4.18: circuit with resistance The current flowing through the resistor is I=R1(V2-V1) (4.15) If the resistor is replaced by a capacitor and two switches, the circuit in figure 4.19 is obtained Figure 4.19: Circuit with switched-capacitor equivalent resistance The switches are controlled by two non-overlapping complementary clocks like in Figure 4.20. phi1 t phi2 t Figure 4.20: Non-overlapping clocks When clock φ1 is high, the capacitor charges up to V1, whereas when φ2 is high, the capacitor charges/discharges up to V2. Therefore, from one phase to the next one there is a transfer of charge of 60 Q = C (V1 – V2) (4.16) With a switch rate fclk, the amount of current transferred per unit time is Q fclk = C fclk (V1 – V2) (4.17) Eqn. 4.17 has the unit of currents, so that an average current can be defined as Iavg = C fclk (V1-V2) (4.18) Assuming that the clock frequency is much higher than the voltage waveforms frequency, the switches and capacitor can be modeled as an equivalent resistance [4.4] Req V1 V2 1 I avg Cf clk (4.19) Based on Eqn. 4.19, an inverting amplifier with capacitors instead of resistors can be implemented. In principle, an inverting amplifier implemented with capacitances would not provide a dc bias for the op-amp input. This is shown in Figure 4.21. Figure 4.21: Inverting amplifier with capacitive feedback. This configuration does not provide dc bias at the op-amp input However, by using switches to control the charge transfer of the capacitors, a dc path can be provided [4.5]. The following figure shows a schematic of such a circuit 61 Figure 4.22: Switched-capacitor integrator Circuit in Figure 4.21 can be modeled as an integrator if C1 and the switches are replaced by the equivalent resistor by using Eqn. 4.19. Two phases can be identified in this circuit. During the sample phase φ1 is high and its switch is closed, while φ2 is low and its switch open. Therefore the charge stored in the capacitors during this phase is Qsample = (0-Vs)C1 + (0-0)C2 (4.20) During the transfer phase φ2 is high and φ1 is low, so that the charge stored is Qtransfer = (0-0)C1 + (0-Vo)C2 (4.21) By conservation of charge, Eqns. 4.20 and 4.21 must be equal. Therefore Vo C1 Vs C 2 (4.22) 4.6.2 Parasitic capacitances In the circuit of Figure 4.22, the main source of parasitic capacitance appears from the switch nodes to ground. This circuit can be modified to become more insensible to parasitics by arranging the switches in a way that the sensor capacitor Cs is never in parallel with the parasitic capacitance [4.7]. This is shown in Figure 4.23. 62 Figure 4.23: Inverting switched-capacitor parasitic insensitive amplifier Adding a parasitic capacitance at the input of the op-amp Cp, and assuming a non-infinite op-amp gain a, the amplifier gain is [4.5] Vo C1 1 Vs C 2 1 C1 C 2 C p 1 C2 a (4.23) The amplifier gain, as Eqn. 4.23 shows, is parasitic insensitive to an extent that depends on the op-amp closed-loop gain. 4.6.3 Correlated Double Sampling The main function of the op-amp in the circuit of Figure 4.23 is, by means of a closedloop feedback, to create a virtual ground at the input without sinking any current. However, due to op-amp non-idealities, this voltage is larger than zero. The main nonideal effects are noise (mainly 1/f noise and thermal noise), and dc offset voltage. Two basic techniques are mainly used to attenuate such non-idealities, known as autozeroing (AZ) and chopper stabilization (CHS) [4.8] Chopper stabilization technique attenuates 1/f noise by transposing the signal to a higher frequency where there is almost no 1/f noise. At a higher frequency, it amplifies the 63 signal and demodulates it back to the original baseband. This technique is well suited in continuous systems, but more difficult to implement for sequential. Autozeroing (AZ) operates sequentially. During a first phase it disconnects the amplifier from the signal path, and samples and stores the unwanted quantities. i.e. the offset and 1/f noise, forcing the output voltage to a small value. Then, the offset-free op-amp is connected back to the signal source for amplification. Therefore any noise constant over time (like dc offset or low frequency 1/f noise) is in principal cancelled. AZ requires the amplifier to be disconnected during the sample phase, so that the signal is only available for amplification during the second phase. Although this requirement would make it incompatible for a continuous system, it is easily suited in a switchedcapacitor circuit. Correlated Double Sampling takes advantage of the switched-capacitor transfer phase to perform an AZ phase when the voltages are held. Therefore in one period, sampling is performed twice, a first one to obtain the unwanted part of the signal, and a second to sample the complete signal. An efficient implementation of CDS is known as Gregorian SCA [4.9], whose schematic is shown in Figure 4.24. Figure 4.24: Gregorian switched-capacitor amplifier 64 The op-amp offset, modeled as a voltage source Vos, is sampled with C1 and C2 in one phase and subtracted in the next one from the signal [4.10]. Figure 4.23 shows the designed amplifier suitable for this project. This circuit is based on the Gregorian amplifier shown in Figure 4.19 with some modification to adapt it to the purpose of the capacitive sensor. To avoid voltage ‘sparks’ at the output when both clocks are down and the op-amp is in open loop, the Gregorian amplifier was modified by placing a small capacitance between input and output [4.11], as shown in Figure 4.25. Figure 4.25: SC amplifier based on the Gregorian amplifier, with CDS and ‘anti-spike’ capacitor Cdg. The circuit of Figure 4.25 is composed of five switches, which could be MOSFETs, a constant voltage source, an operational amplifier, the sensor capacitance Cs, the antispike capacitor Cdg, the feedback capacitor Cf, and two nonoverlaping clock signals φ1 and φ2, whose clock circuits are not shown in the schematic. Voffset models nonidealities from the operational amplifier. To analyze this circuit, the two different phases will be studied independently. During the sampling phase (φ1 high and φ2 low), the circuit can be simplified to the circuit in Figure 4.26. 65 Figure 4.26: Simplified converter during sampling phase Assuming that the capacitance of both differential capacitors changes from C to C+ΔC and C-ΔC, the charge stored in the capacitors is: Q1 = (C+ΔC)(Vin-Vos)+(C-ΔC)(-Vin-Vos)+Cf(0-Vos) (4.24) During the charge transfer phase, the circuit can be simplified to circuit in Figure 4.27. Figure 4.27: Simplified Gregorian amplifier during charge transfer phase The charge storage during this phase is Q2 = (C+ΔC)(0-Vos)+(C-ΔC)(0-Vos)+Cf(Vout-Vos) (4.25) Applying conservation of charge, the offset voltage Vos is cancelled out in both Eqn. 4.24 and 4.25, and the output voltage is obtained as function of the change in capacitance ΔC Vout 66 C Vin Cf (4.26) 4.6.4 Bandwidth The transfer of charge of Eqn. 4.21 assumes infinite op-amp gain and no parasitic capacitances. Including terms for both nonidealities, the equation becomes V V V Q2 o C1 o C p o Vo C 2 A A A (4.27) The circuit gain with parasitic capacitance and finite gain A is Vo C1 1 Vs C 2 1 1 C1 C 2 C p A Cp (4.28) Assuming an op-amp with a single pole response A( s ) Ao s o Ao s s so 1 so (4.29) Substituting the op-amp gain into the amplifier gain Vo C1 Ao s o Vs C 2 C1 C p 1 A s ( s s ) o o o C2 C1 K C2 1 s p (4.30) where K Ao Ao C1 C 2 C p C2 67 (4.31) Ao s o p C1 C 2 C p C2 C1 C 2 C p so (4.32) C2 K can be simplified to one assuming that the op-amp open loop gain Ao is much larger than the ratio of capacitances (C1+C2+Cp)/C2. Under the same assumption, Eqn. (4.32) shows that the closed-loop bandwidth is Ao times larger than the open-loop bandwidth. This means that the product gain-bandwidth is constant for a given op-amp. This is shown in Figures 4.28-4.29 Bode Diagram 0 Magnitude (dB) -10 -20 -30 -40 -50 Ao=1e3, Cp=Cf/10 Ao=1e3, Cp=Cf Ao=1e3, Cp=Cf*10 Ao=1e5, Cp=Cf/10 Ao=1e5, Cp=Cf Ao=1e5, Cp=Cf*10 Phase (deg) -60 0 -45 -90 9 10 10 10 11 10 12 10 13 10 14 10 15 10 Frequency (rad/sec) Figure 4.28: Closed-loop frequency response of Gregorian converter for different op-amp open-loop gains (Ao) and different parasitic capacitances (Cp) 68 Bode Diagram Magnitude (dB) 0 -0.05 -0.1 Phase (deg) 0 Ao=1e3, Cp=Cf/10 Ao=1e3, Cp=Cf Ao=1e3, Cp=Cf*10 Ao=1e5, Cp=Cf/10 Ao=1e5, Cp=Cf Ao=1e5, Cp=Cf*10 -45 -90 9 10 10 11 10 12 10 10 13 10 Frequency (rad/sec) Figure 4.29: Closed-loop frequency response of Gregorian converter for different op-amp open-loop gains (Ao) and different parasitic capacitances (Cp). Zoom at lower frequencies From Figure 4.28 it can be extracted that the amplifier is more sensitive to parasitics for lower open-loop op-amp gains. The closed-loop gain remains constant at low frequencies with different parasitic capacitances values for a high open-loop op-amp gain. However, for a low open-loop op-amp gain, both the bandwidth and the closed-loop gain performance decrease at high frequencies. 4.6.5 Noise analysis The switched-capacitor amplifier has two dominant sources of error. The input referred KT/C noise is given by the following expression 2 v KT / C ,input KT CT CT = CS + Cf + Cp + CGS 69 (4.33) (4.34) where K is the Boltzmann constant, T temperature, and CT is the sum of the feedback capacitance Cf, sensor capacitance Cs and parasitic capacitance Cp, and the gate capacitance of the input stage of the op-amp. The output referred KT/C noise is v 2 KT / C C T C f 2 KT C T (4.35) The thermal noise is given by the voltage op-amp noise vn2 . The output referred thermal noise is [4.14] v 2 opamp C T C f 2 2 GBW amp vn fs 2 (4.36) where GBWamp is the amplifier gain bandwidth, and fs is the sampling frequent. As explained, the CDS technique cancels out any noise that is constant over a longer period than the sampling time. Therefore, KT/C noise and offset are removed. With these noise sources removed, the amplifier thermal noise becomes the dominating noise source. Using the gain given by Eqn. (4.26) and the noise from Eqns. (4.35) and (4.36) above, the minimum detectable signal without CDS is C min rms 1 Vin 2 2 KTCT CT v n (GBWamp )( 2) BW fs fs (4.37) while the noise with CDS applied is C min rms CDS 1 Vin CT2 v n2 (GBWamp ) fs 70 BW (4.38) CDS removes the term due to KT/C noise. However, the thermal noise is twice as large due to the double sampling. Figure 4.30 shows the performance under different sampling frequencies minimum detectable capacitance variation Cmin-rms [aF/sqrt(BW)] 0.25 SC fs=0.1 MHz SC CDS fs=0.1MHz SC fs=1MHz SC CDS fs=1MHz 0.2 0.15 0.1 0.05 0 0 2 4 6 8 10 12 14 16 18 20 Parasitic Capacitance [pF] Figure 4.30: Performance of switched-capacitor with and without correlated double sampling for 0.1 MHz and 1 MHZ sampling frequency. Sensor capacitance 1 pF, amplifier gain bandwidth 2 MHz. Figure 4.31 compares the performance of the transimpedance amplifier and the switchedcapacitor amplifier minimum detectable capacitance variation 0.12 switched-capacitor Cs=1pF switched-capacitor CDS Cs=1pF transimpedance amplifier Cs=1pF switched-capacitor Cs=1fF switched-capacitor CDS Cs=1fF transimpedance amplifier Cs=1fF Cmin-rms [aF/sqrt(BW)] 0.1 0.08 0.06 0.04 0.02 0 0 2 4 6 8 10 12 14 16 18 20 Parasitic Capacitance [pF] Figure 4.31: Minimum readable capacitance of switched-capacitor, switched-capacitor with CDS, and transimpedance amplifier. Amplifier gain bandwidth 1 MHz, sampling frequency 0.5 MHz 71 4.6.6 Simulations Since the sensor capacitance swings within a wide range, the converter of Figure 4.24 with fixed components values may not be appropriate for the complete sensor capacitance range. Assuming a fixed voltage supply of ±5 volts, which is a typical value for most low-noise op-amps [4.19], the feedback capacitor value Cf is a trade-off between capacitance range and accuracy. According to the noise model given by Eqn. 4.38, a smaller feedback capacitor results in lower minimum detectable capacitance. However, a smaller feedback capacitor also results in larger output voltage, since the output is given by the following equation Vout Vb C Cf (4.39) Therefore, Cf may have to be changed to obtain an optimal Vout according to Eqn. (4.39). The minimum detectable capacitance can be calculated with the model of Eqn. (4.38) assuming typical values for the amplifier gain bandwidth (GBW), signal bandwidth (BW), sampling frequency, and capacitances. C min rms / BW 1 (2 pF 15 pF 20 pF ) 2 (10MHz ) 0.041477aF / Hz 5 1MHz (4.40) This means that for a bandwidth of, say 1KHz, the minimum change in capacitance that the circuit could read would be as small as: C min 0.04 aF Hz 14 KHz 4.9aF (4.41) The circuit of Figure 4.24 was simulated with SPICE. Figures 4.32-4.33 show the input capacitances and the output voltage. The code is given in appendix G. 72 Figure 4.32: Voltage of voltage controlled capacitance (Vcap, red line), and output voltage from circuit in Figure 4.24 (Vout, blue line). The sensor capacitance (Cs) value is given as C=Vcap*10-9 Figure 4.33: Zoom in from Figure 4.27. Voltage of voltage controlled capacitance (red line), and output voltage (blue line) circuit Figure 4.20. A continuous RC low-pass filter (LPF) can be used to obtain a DC voltage proportional to the change of capacitance as in the following figure. Figure 4.34: RC low-pass filter 73 A LPF filters higher frequencies than the cutoff frequency and passes lower frequency, which results to an output proportional to the input average. The selection of values for R and C depends on the rate of change of capacitance. Figure 4.35 shows Vout from Figure 4.28 filtered by the RC-LPF of Figure 4.34 for C=1 nF and three different values for R. The hspice code is given in appendix G. Figure 4.35: output from 4.27 filtered with RC LPF versus time for different values of R. C=1 nF The ‘antispike’ capacitor Cdg value is a trade-off between losing gain for large Cdg and loosing stability for small Cdg. The simulations in Figure 4.36 show the same output of Figure 4.32 with the same circuit but different values for Cdg Figure 4.36: Output voltage converter versus time with Cdg = 1μF. The gain highly decreases 74 Figure 4.37: Output voltage converter versus time with Cdg = 0.1nF. Voltage spikes would lead circuit to instability. 4.7 Sigma-Delta modulators Sigma-Delta modulation is a widely used technique in Signal Conditioning. Extensive literature has been published on the topic. For the sake of clarity, only an overview is given here. A Sigma-Delta modulator converts a signal from analogical to digital. General Analogto-Digital Converters are based on a sampling and quantization process as shown in Figure 4.38 Figure 4.38: General Analog-to-Digital Conversion Process [4.20]. The figure on the left shows the block diagram, and the figure on the right shows the relation between the discrete-time signal x*(t) and digital signal x(n), which is represented by the multilevel quantization block on the left. 75 Delta modulation, instead, is based on quantizing the change in the signal from sample to sample rather than the absolute value of the signal at each sample, as shown in Figure 4.39 Figure 4.39: Delta modulation [4.20]. The output signal is integrated to obtain an average or expected output, and then subtracted from the input analog signal so that the input to the quantizer is the difference between the expected value and the input. Sigma-Delta modulation is based on Delta modulation-demodulation. Demodulating a modulated signal is done by integrating and lowpass-filtering the output from modulation. Figure 4.40: Delta modulation and demodulation [4.20]. When demodulation is implemented after modulation, because the integration is a linear operation, the integrator of demodulation can be moved and implemented before the modulation without altering the output. Furthermore, the two integrators can be combined and placed before the quantizer, which gives block diagram in Figure 4.41. 76 Figure 4.41: Block diagram of Sigma-Delta Modulation based on Delta Modulation-Demodulation. In fact, the name Sigma-Delta Modulation is due to the fact that there is an integrator (Sigma) before Delta Modulation. The integrator highpass-filters the quantization noise and pushes it to a higher frequency. With the block diagram of figure 4.41, a first-order oversampled Sigma-Delta A/D converter can easily be implemented by changing the lowpass filter with a digital decimation filter. Figure 4.42: Block diagram of First-Order Sigma-Delta A/D Converter. Using this technique Capacitance-to-Digital converters can be implemented based on Sigma-Delta modulation. There is extensive literature on this topic, including the implemention techniques such as switched-capacitor, bandpass, and correlated double sampling [4.20-4.23]. Figure 4.43 shows the implementation of switched-capacitor Sigma-Delta modulator that can be used as capacitive sensor interface [4.23] Figure 4.43: Switched-capacitor Sigma-Delta modulator. By replacing Cin by the capacitance to sense, the output produces a digital signal proportional to the sensor capacitor [4.23] 77 References: [4.1] Guru and Hiziro, ‘Electromagnetic Field Theory Fundamentals’ [4.2] F. N. Toth, Gerard Meijer, Harry Kerkvliet, ‘A Very Accurate Measurement system for Multielectrode Capacitive Sensors’, IEEE, Vol. 45, No. 2, April 1996 [4.3] L. K. Baxter, ‘Capacitive Sensors, Design and Applications’, IEEE Press [4.4] W. R Grise, ‘Applications of Switched-Capacitor Circuits in Active Filters and Instrumentation Amplifiers’, the Technology Interface, Vol 3 No 3, Fall 1999 [4.5] Gray, Hurst, Lewis, Meyer, ‘Analysis and Design of Analog Integrated Circuits’, Wiley, fourth edition. [4.6] N. Yazdi, H. Kulah, K. Najafi, ‘Precision Readout Circuits for Capacitive Microaccelerometers’, University of Michigan, Ann Arbor, MI, USA [4.7] www.ecircuitcenter.com [4.8] Ch. C. Enz, G. C. Temes, ‘Circuit Techniques for Reducing the Effects of Op-Amp Imperfections: Autozeroing, Correlated Double Sampling, and Chopper Stabilization’, IEEE Vol. 84, No 11, November 1996 [4.9] R. Gregorian, ‘High Resolution Switched-Capacitor D/A Converter’, Microelectronics J., Vol 12, No. 2, pp 10-13, 1981 [4.10] D. J. Willis, R. J. Jost, ‘Zero CVF Input Current Switched-Capacitor Instrumentation Amplifier’, IEEE 2005 [4.11] H. Matsumoto, K. Watanabe, ‘Spike-free SC circuits’, Electronic Letters, vol. 8, 1987 78 [4.12] N. Wonkgomet, B. E. Boser, “Correlated Double Sampling in Capacitive Position Sensing Circuits for Micromachined Applications”, 1998 IEEE [4.13] S. D. Senturia, ‘Microsystem Design’, Kluwer Academic Publishers. [4.14] N. Wongkomet, ‘Position Sensing for Electrostatic Micropositioners’, PhD dissertation, University of California, Berkeley. [4.15] J. Wu, G. K. Fedder, L. R. Carley, ‘A Low-Noise Low-Offset Chopper-Stabilized Capacitive-Readout Amplifier for CMOS MEMS Accelerometers’, 2002 IEEE International Solid-State Circuits Conference. [4.16] J. Wu, G. K. Fedder, L. R. Carley, ‘A Low Noise Low Offset Capacitive Sensing Amplifier for a 50-μg/Hz-1/2 Monolithic CMOS MEMS Accelerometer’, 2004 IEEE. [4.17] H., Luo, G. Fedder, L. R. Carley, ‘Integrated Multiple-Device IMU System with Continuos-Time Sensing Circuitry’, ECE Dept. Carnegie Mellon University, Pittsburgh, PA. [4.18] J. Chae, H. Kulah, K. Najafi, ‘An In-Plane High Sensitivity, Low-Noise Micro-g Silicon Accelerometer with CMOS Readout Circuitry’, 2004 IEEE, Vol 13, No 4 [4.19] www.analog.com [4.20] S. Park, ‘Principles of Sigma-Delta Modulation for Analog-to-Digital Converters’, Motorola. [4.21] N. A. Fraser, B. Nowrouzian, ‘A Novel highly Stable Hihg-Resolution Oversampled Sigma-Delta A/D Converter Configuration’, ECE, University of Alberta, Canada. [4.22] M. Saukoski, L. Aaltonen, T. Salo, K. Halonen, ‘Readout Electronics With Bandpass Delta-Sigma A/D Converter for a Bulk Micromachined Capacitive Gyroscope’, 79 Electronic, Circuit Design Laboratory, Helsinki University of Technology, Finland. [4.23] J. O’Dowd, A. Callanan, G. Banarie, E. Company-Bosch, ‘Capacitive sensor interfacing using sigma-delta techniques’, IAC group, Analog Devices BV, Limerick, Ireland. 80 5. Experimental verification 5.1 Scaled models The sensor studied in this project is meant to be fabricated by optical lithography, which would result in a minimum feature size of approximately 1 μm. Assuming that the sensor was used for 15x15 mm2 chips, a reasonable maximum sensor size would be 0.5 x 0.5 mm2. To test the resolution of such a sensor, a nanopositioner as shown in Fig. 1.1 would need to be used as actuator. Nevertheless, a ‘scaled’ sensor, acting in the micrometer range instead of the nanometer range, would simply need a microactuator, which is more often available at any microfabrication lab. In fact, with a scaled model a simple mechanical micropositioner can be used, which highly reduces the set-up complexity and cost. According to the infinite plate capacitance model, scaling-up both width and gap by the factor does not change the capacitance per unit length. This is shown in the following equations. C w ( wk ) n n L g ( gk ) (5.1) where C is capacitance, L capacitor length, n number of strips, ε the medium electrical permittivity, w the strip width, g the gap between strips, and k a real constant. As shown in Eqn. 5.1 if gap, width, and length are scaled, the resultant capacitance would be proportional to the original one. This is expressed in the following equation. C l ( w' , g ' , n' ) C l ( w, g , n) 81 n n' (5.2) where Cl is capacitance per unit length. The sensor accuracy, i.e. the minimum detectable displacement variation, can also be estimated by manipulating simple linear equations. The minimum detectable displacement can be expressed as x min dC C min dx (5.3) ΔCmin is defined in Eqn 4.6 and does not depend on the sensor geometry. In the case of two parallel plates, the derivative of capacitance C with respect to lateral displacement x is dC d w x 1 nL nL dx dx g g (5.4) From Eqn 5.4 it can be extracted that for parallel plates the capacitance sensitivity to displacement scales proportional to length, number of strips, and inverse of gap. dC dC L n g' ( w' , n' , L' , g ' ) ( w, n, L, g ) dx dx L ' n' g (5.5) In the case of 5 electrodes, the output capacitance function with displacement is more complex to handle algebraically. To be able to obtain a simpler analytical expression, the system of 5 electrodes can be simplified to capacitances C12 and C13. With such simplification, the total output capacitance is expressed as Cout f (Ciji , j 1..5 ) C13C12 C12 C13 (5.6) To calculate C13 and C12 as function of displacement x, the scheme in Figure 5.1 is used. 82 Figure 5.1: geometry used for capacitance analysis. Only the capacitances C 12 and C13 are considered According to the sign and geometry criteria used in Fig 5.1, Eqn 5.6 can be expressed as C out C13C12 C13 C12 w2 w w x 2 wx x x 4 nL 2 nL 2 w g w g w x x 2 2 (5.7) Eqn 5.6 hold as long as –w/2<x<w/2. The derivative of capacitance with respect to displacement is dC out nL 2 x 1 dx g w (5.8) Because the sensor aims at aligning at the center, i.e. when x equals 0, to obtain the minimum displacement Eqn 5.8 is evaluated at x=0 dC out ( x 0) nL dx g (5.9) Eqn 5.9 gives the same result than Eqn 5.4.Therefore, for both 3 plate capacitor and 2 plate capacitor, the minimum detectable lateral displacement is x min g C min nL (5.10) Obtained the accuracy of a sensor, the accuracy of a scaled version of the same sensor can be estimated as 83 xmin ( w' , L' , g ' , n' ) xmin ( w, L, g , n) g' L n g L ' n' (5.11) Interesting consequences can be extracted from Eqn 5.10. According to this result, the sensor accuracy is independent of electrode width, and varies proportional to the inverse of the number of electrodes. This means that a large number of thin electrodes result in greater sensor accuracy and one wide electrode. This makes sense with the approach commonly used in capacitive sensors as shown in Figure 1.5. A simple means of creating a scaled model for the sensor is by designing a printed circuit board whose traces on one face are routed as the sensor electrodes; the board dielectric would be equivalent to the chip oxide, and a layer of metal on the other face of the board would be the conductive substrate. 5.2 Setup The actuator used for the experimental verification, as shown in Figure 5.1, has the advantage that two samples, or boards for this project’s experiments, can be placed on it. The bottom board can be placed on top of a stage that can be moved with a 2X micropositioner. The actuator also allows separating the plates while still keeping them parallel by moving the rails where the top plate is placed. Figures 5.1-5.2 show the setup in the lab. 84 Figure 5.1: Experiment setup. The picture shows the stage with the micropositioner, one board on the stage, a second board on the table, and the computer connected to the readout circuit. 0 Figure 5.2: Detail of stage. In the center of the picture the two plates can be seen. The upper plate is fixed in the structure, while the plate below is placed on a stage controlled by a 2D micropositioner. Also, in the left there is a bar to move the fixed plate in the Z axis. In the right, two cables connected to the bottom plate are plugged into the readout circuit 5.3 Readout circuit To obtain the best performance, the readout circuit was purchased from Analog Devices for the experiments. The model used was the AD7745, which is a Capacitance-to-Digital Converter based on a 24-bit Σ-Δ modulator. The functional block diagram is shown in Figure 5.3. 85 Fig 5.3: Analog Devices AD7745 block diagram [5.1] The features of this circuit are as in Table 5.1. Resolution Accuracy Full-scale range Parasitic capacitance to ground tolerance Update rate 4 aF 4 fF ±4 pF 60 pF 10 Hz to 90 Hz Table 5.1: AD7745 main features [5.1] The evaluation board of this circuit includes the software needed for the computer interface. This interface shows the instantaneous capacitance sensed, as well as a plot of capacitance versus time. Figure 5.4 shows the interface window Figure 5.4: Capacitance-to-voltage converter interface. The capacitance sensed oscillates due to noise. The capacitance considered for the measurement is the mean of this oscillation. 86 5.4 Designs tested Two boards were designed to carry out two different experiments. First, to test the sensor performance when the gap between the plates is the minimum achievable and a final accurate alignment is needed, a board with large features was used to obtain a small width/gap ratio. This experiment aims at estimating the maximum alignment accuracy achievable by the sensor. The second board designed has larger feature size and is meant to test the performance of the fractal geometry, as well as to verify whether both simulations and capacitance analysis agree with the experimental results. 5.4.1 Large feature test board To reach small width/gap ratios, the designed board with wider traces is as shown in Figure 5.5. Figure 5.5: Board layouts. The red color shows the traces on the top layer, while the green color shows the traces on the bottom layer. The board dimension is 3.8’’x 2.5’’ (9.65 x 6.35 cm2) The board is composed of three groups of electrodes. The dimensions are shown in Table 5.2. 87 left board 0.12’’ 0.12’’ 0.36’’ trace width separation traces separation groups 3.048 cm 3.048 cm 9.144 cm right board 0.25’’ 0.47’’ 6.350 cm 11.938 cm Table 5.2: Geometry boards Figure 5.5 The smallest gap that could be reached was in the order of 1 mm. For this gap the results and comparative simulations are shown in Figure 5.6. simulation and experiments gap 1 mm 13 capacitance [pF] 12 11 10 experiment a experiment b sim 1mm sim1.5 mm 9 8 7 6 5 4 0 5 10 15 20 25 30 displacement [mm] Figure 5.6: Capacitance [pF] versus lateral displacement [mm].. The measurement was repeated twice (red and green in figure). The blue and black lines show simulation results for gap of 1 mm and 1.5 mm. The experimental results and the simulations show similar variations of capacitance with displacement. However the experimental curves have a larger ‘dc’ component than the simulations. To compare the results, Figure 5.7 shows the curves without the ‘dc’ component. 88 simulation and experiments gap 1 mm experiment a experiment b sim 1mm sim1.5 mm capacitance - C(displ=0) [pF] -0.2 -0.4 -0.6 -0.8 -1 -1.2 -1.4 -1.6 -1.8 0 5 10 15 20 25 30 35 displacement [mm] Figure 5.7: Capacitance versus displacement. The initial capacitance for displacement zero was subtracted to the capacitance, so that the curve profiles could be more easily compared. Although the experimental results do not fit any of the simulation results, they show a similar variation of capacitance with displacement Figure 5.8 shows the capacitance from another experiment, where the displacement step was reduced and the boards were near to the alignment centered position. experimental results, gap 1 mm 16.01 capacitance [pF] 16.005 16 15.995 15.99 15.985 15.98 15.975 1.6 1.8 2 2.2 2.4 2.6 2.8 displacement [mm] Figure 5.8: Capacitance [pF] versus lateral displacement [mm]. Gap 1 mm. The readout circuit shows to be able to read changes in capacitance up to the order of 5 fF, which results with this geometry in an accuracy of 0.1 mm For longer gaps, the results were as in Figure 5.9. 89 simulation and experiments 0 capacitance -C(displ=0) [pF] -0.05 -0.1 -0.15 -0.2 -0.25 -0.3 experiment 3 mm sim 3 mm experiment 6 mm sim 6 mm experiment 12 mm sim 12 mm -0.35 -0.4 -0.45 -0.5 0 5 10 15 20 25 30 35 40 displacement [mm] Figure 5.9: Capacitance [pF] versus lateral displacement [mm]. The results show that even for a large gap such as 12 mm the readout circuit is able to detect changes in capacitance up to 3 mm within alignment. 5.4.2 Central fractal geometry test board The second board was designed with the traces following the central fractal geometry explained in Section 2.2.3. The boards are shown in Figure 5.10. Figure 5.10: Test board layout implementing central fractal geometry. The red color shows the traces on the top layer, while the green color shows the traces on the bottom layer. The board dimension is 3.8’’x 2.5’’ (9.65 x 6.35 cm2) Figure 5.11 zooms on the left board of Figure 5.6 and gives the geometry as function of the minimum feature size u. 90 Figure 5.11: geometry of left board of Figure 5.6. The dimensions are given in function of the minimum feature size u, which in this case is 0.03’’ The dimensions are also gathered in Table 5.3. central group lateral group width spacing traces spacing subgroups total width group width spacing traces spacing groups left board units (u) 1 1 3 15 5 5 15 inches 0.03’’ 0.03’’ 0.09’’ 0.45’’ 0.15’’ 0.15’’ 0.45’’ mm 0.762 0.762 2.286 11.43 3.81 3.81 11.43 Table 5.3: dimensions board Figure 5.13 The experimental results and simulations are as in Figure 5.12. 91 right board units inches 2 0.06’’ 1 0.03’’ 4 0.12’’ 14 0.42’’ 10 0.3’’ mm 1.524 0.762 3.048 10.668 7.62 0.54’’ 13.716 18 central fractal test board 0.2 exp g 1 mm exp g 2 mm exp g 3mm exp g 4mm sim g 1mm sim g 2 mm sim g 3 mm sim g 4 mm 0 capacitance [pF] -0.2 -0.4 -0.6 -0.8 -1 -1.2 0 5 10 15 20 25 30 35 40 displacement [mm] Figure 5.12: capacitance [pF] versus displacement [mm]. As in Figure 5.7, the initial capacitance for displacement zero was subtracted to the capacitance. Here again, the simulations and experiments give curves with similar profile. Reference: [5.1] http://www.analog.com/en/prod/0,2877,AD7745,00.html 92 6. Conclusions and future work 6.1 Experimental results The experiments with the test boards showed in general that the models of Section 3.3 and simulations have sufficient accuracy for design purposes. The simulations, nevertheless, result in lower capacitance values than the experiments. This is most likely caused by the assumption that there is no capacitance from the electrodes to the ground. Including this capacitance to ground, the equivalent circuit model of Figure 3.8 would look like in Figure 6.1. Figure 6.1: Sensor equivalent circuit. The circuit includes the capacitance from the electrodes to ground (green). Also the capacitance from the plates (nodes 4 and 5) to the stage (S), and the capacitance from the stage to ground are included. The difference between the simulations and the experiments increases with larger displacement, according to Figure 5.9. For large displacement the capacitance obtained in the experiments is less sensitive to displacement than the capacitance from simulations. This is believed to occur because the models do not include the capacitance from the 93 plates to the stage, as shown in Figure 6.1. This large capacitance becomes dominant for large gaps when the capacitance C12 decreases. Since this capacitance to stage is independent of displacement, the output capacitance sensitivity to displacement decreases. Another important difference between simulations and experiments is the assumption in the simulations that the plates are parallel. Figure 5.7 shows two different curves from experiments that are supposed to have the same gap. The difference between the two experiments is most likely due to plate tilting, which can make the plates to get closer of further with displacement depending on the tilting orientation. The chips to align with the nanoaligner would have four sensors built-in on the four chip corners. To minimize tilting, the nanoaligner would have to move the plates keeping the capacitance from the four sensor as equal as possible. As shown in Figure 5.8, the best alignment obtained with the large feature test board was in the order of 0.1 millimeters. If optical lithography is to be used in the sensor fabrication, the sensor width could be in the order of 1 μm. For this width, assuming that the nanoaligner can reach a gap in the order of 100 nm, by using Eqns 5.3 and 5.4 the expectable achievable alignment can be calculated in principle. x min x min PCB LPCB n PCB g 75mm 3 100nm 0.1mm 13.63nm ~ 10nm L n g PCB 15mm (3 4 2) 1mm (6.1) The sub-indexes μ and PCB refer to the sensor fabricated by means of optical lithography, and the printed circuit board designed for the experiments respectively. The test board with central fractal geometry gave results as expected and obtained in the simulations. The central thinner electrodes give three peaks when the gap is in the order of the electrodes width, and one wider peak when the gap is larger. 94 6.2 Design recommendations Three possible sensor designs were pointed in section 2.2. To choose one of the three approaches the nanoaligner’s capabilities have to be considered. From the simulations and experiment results, it can be extracted that the central fractal geometry showed to obtain the best performance. This geometry is a good trade-off between the design optimized for best performance at larger gaps and best performance at short gaps. The traces for this geometry can be easily designed by choosing the minimum feature size and the maximum sensor area. The capacitor width can be calculated as: w 3u 5 n 1 (6.1) where u is the unit or minimum feature size (the traces’ width) and n the number of fractal levels. The number of levels should be chosen so the capacitor fits within the maximum sensor area. The following figure shows how the layout would look like the following figure for three levels. Figure 6.2: proposed central fractal geometry design. In this case the fractal has three levels. 95 The following table gathers several suggested designs that could be used and estimates the alignment that would achieve assuming that it reaches a particular minimum gap Design number Min feature size w strip-group 1 w strip-group 2 w strip-group 3 w strip-group 4 w strip-group 5 w strip-group 6 Min gap Max gap #groups n Xmin scaled Xmin calculated 1 1 um 1-3 um 1-15 um 5-75 um 25-375 um 2 1 um 1-3 um 1-15 um 5-75 um 25-375 um 0.125-1.875 mm 20 nm 25 um 5 240 nm 4.8 nm 20 nm 0.1 mm 7 34 nm 0.68 nm 3 0.1 um 100-300 nm 0.1-1.5 um 0.5-7.5 um 2.5-37.5 um 12.5-187.5 um 0.0625-0.9375 mm 10 nm 62 um 9 26 nm 0.53 nm 4 1 um 1-3 um 1- 291 um 0.1 – 1.5 mm 20 nm 0.1 mm 49+2 5.8 nm 0.11 nm Table 6.1: Suggestions for sensor design. The three first designs (1, 2, 3) are based on the central fractal geometry, while design 4 is composed of 2 wide electrode groups and 49 thin equal electrode groups. Designs 1 and 2 have a minimum feature size of 1μm, so the fabrication can be done by optical lithography. Design 1 has a size of 0.375x0.375 mm2, while design 2 is 1.875x1.875 mm2. Design 3 has a feature size of 0.1 μm and and area of 0.9375x0.9375 mm2 Design 1 tries to fit the sensor within 0.5 mm long, and therefore it has 4 levels. This limits the number of electrode groups to 5, which worsens the sensor accuracy. As well, due to the size limitation, the widest electrode is only 25 μm width. This makes the initial alignment for large gaps more complex. Design 2 is an extension of design 1 with one more group, so the area is 1.875x1.875 mm2. This improves the alignment accuracy and makes the initial alignment for larger gaps less demanding. Design 3 assumes that the minimum feature size is 100 nm. This allows the sensor to have more electrodes groups in the same area, and therefore it increases the alignment accuracy. 96 Design four does not follow exactly central fractal geometry. Instead, it is composed of 1 wide group with 0.1 mm wide electrodes to increase the accuracy at large gaps. The rest of the areas is used for as many 1 µm wide electrode groups as could be fit, which is 49 groups of 2 electrodes each. This approach increases the number of electrode groups n, and therefore the alignment accuracy. As well, its wide group of 0.1 mm electrodes allows a more accurate alignment for large gaps. The disadvantage of this approach is that the large group has to do the alignment for all the gap range from 0.1 mm to 1 µm, where the 1 µm wide groups take over. This could potentially lead to instability in the alignment. 6.3 Future work New geometries should be explored to improve capacitance sensitivity. Differential capacitors, which are mentioned in Section 2.1, Figure 2.2-2.3, may improve capacitance sensitivity. Furthermore, assuming that the geometry is perfectly symmetrical, they result in zero capacitance rather than a maximum capacitance at central aligned position. To overcome the problem of connections to both plates described in Figure 2.2, one could employ a geometry such as that illustrated in Figure 6.3. Figure 6.3: Differential capacitor with connections on only one side. The voltage supplies are of equal magnitude but are 180o out of phase with each other. The arrow illustrates the output to a readout circuit. The electrodes with the same number are connected. 97 Simplifying the capacitances in Figure 6.3 to C14, C24, and C34, the circuit illustrated in Figure 6.4 can be extracted from Figure 6.3. Figure 6.4: Circuit from scheme Figure 6.3. The capacitors in the circuit in Figure 6.4 are arranged in start configuration. Figure 6.5 illustrates the circuit transformed to the equivalent star-triangle configuration [6.1] Figure 6.5: Equivalent circuit of Figure 6.4 with star-triangle transformation. The output current of the circuit in Figure 6.5 can be calculated analytically by neglecting fringing fields. The capacitances of the circuit in Figure 6.4 are calculated in Figure 6.6 using the parallel infinite plate capacitance equation. 98 Differential capacitor 1.4 Capacitance [fF] 1.2 1 C14 C24 0.8 0.6 0.4 0.2 0 -0.5 0 lateral displacement [um] 0.5 Figure 6.6: Capacitance circuit Figure 6.4. The geometry of capacitors, as in Figure 6.3, assumes a gap of 0.1 μm, bottom electrode width 1μm, and top electrode width 4 μm. The impedances in Figure 6.5 are calculated using the capacitance of Figure 6.6; Figure 6.7 shows the results. 15 16 x 10 Differential capacitor Za Zb Impedance [Ohms] 14 12 10 8 6 4 2 0 -0.5 0 lateral displacement [um] 0.5 Figure 6.7: Impedances of circuit Figure 6.5. Figure 6.8 shows the output current, assuming opposite voltages in the voltage supplies in the Figure 6.5 circuit. 99 -16 6 x 10 Differential capacitor Current [Amps] 4 2 0 -2 -4 -6 -0.5 Current through Za Current through Zb Output current 0 lateral displacement [um] 0.5 Figure 6.8: Currents of circuit from Figure 6.5. The output current is zero for zero displacement. Non-differential capacitors have the disadvantage that when close to central alignment, the output becomes flat due to fringing field effects, as shown in Figure 5.8, which decreases the sensor resolution. Differential capacitors have the advantage that not only do they not loose sensitivity when near central position, but they also allow a simple means of detecting alignment by giving no output current. 100 Appendix Appendix A: Calculation of equivalent output capacitance MATLAB code: syms ig v c12 %a=23, b=24, c=34, d=51, e=41, f=25, g=12, h=13, j=35, k=45, %capacitances [ia,ib,ic,id,iee,iff]=solve(‘1/c23*ia + 1/c13*(ia+ib+iff-id-iee)-v’ , ‘1/c24*ib + 1/c45*(ib+ic-iee) — 1/c35*(id+iee-ib-ic-iff) + 1/c13*(ia+ib+iff-id-iee)-v’,’1/c34*ic + 1/c45*(ib+ic-iee) — 1/c35*(id+iee-ib-iciff)’ , ‘1/c15*id — 1/c13*(ia+ib+iff-id-iee) + 1/c35*(id+iee-ib-ic-iff)’,’1/c14*iee — 1/c13*(ia+ib+iff-idiee) + 1/c35*(id+iee-ib-ic-iff) — 1/c45*(ib+ic-iee)’ , ‘1/c25*iff — 1/c35*(id+iee-ib-ic-iff) + 1/c13*(ia+ib+iff-id-iee)-v’,’ia,ib,ic,id,iee,iff’); % Ra *ia + Rh *ih -v , Rb *ib + Rk *ik -Rj *ij +Rh * ih -v , Rc *ic + Rk * ik - Rj * ij , rd *id - rh *ih + Rj * ij , Re *iee - Rh * ih + Rj *ij - Rk*ik , Rf * if — Rj *ij + Rh *ih — v ig=simplify(ia+ib+iff); cout=simplify(ig/v)+c12 Equivalent capacitance: cout=(c45.*c13.*c15.*c23+c45.*c15.*c35.*c23+c24.*c13.*c15.*c23+c24.*c15.*c35.*c23+c24.*c45.*c35. *c13+c14.*c35.*c45.*c23+c25.*c23.*c45.*c15+c45.*c25.*c34.*c14+c14.*c13.*c45.*c23+c14.*c13.*c35. *c23+c23.*c13.*c35.*c45+c34.*c13.*c15.*c23+c34.*c15.*c35.*c23+c35.*c14.*c13.*c25+c13.*c25.*c34. *c24+c34.*c25.*c45.*c13+c34.*c35.*c13.*c25+c13.*c25.*c35.*c45+c34.*c25.*c15.*c23+c34.*c25.*c13. *c15+c34.*c25.*c15.*c35+c25.*c45.*c15.*c34+c34.*c25.*c24.*c15+c34.*c45.*c24.*c15+c34.*c35.*c24. *c15+c34.*c13.*c25.*c23+c34.*c35.*c24.*c13+c15.*c24.*c45.*c23+c15.*c14.*c13.*c25+c15.*c45.*c25. *c35+c15.*c24.*c25.*c23+c14.*c24.*c35.*c23+c14.*c24.*c45.*c35+c14.*c24.*c45.*c23+c14.*c24.*c35. *c13+c14.*c24.*c45.*c13+c14.*c15.*c34.*c24+c14.*c25.*c34.*c24+c14.*c24.*c13.*c15+c14.*c24.*c15. *c35+c14.*c24.*c15.*c23+c14.*c45.*c13.*c25+c14.*c24.*c13.*c25+c14.*c24.*c25.*c23+c14.*c24.*c25. *c35+c14.*c35.*c24.*c34+c14.*c24.*c45.*c34+c35.*c24.*c13.*c25+c45.*c23.*c25.*c14+c25.*c45.*c13. *c15+c25.*c24.*c13.*c15+c25.*c24.*c15.*c35+c25.*c14.*c15.*c35+c15.*c24.*c45.*c13+c14.*c25.*c35. *c45+c14.*c13.*c25.*c23+c25.*c23.*c15.*c14+c34.*c24.*c45.*c13+c34.*c24.*c13.*c15+c25.*c15.*c34. *c14+c35.*c25.*c34.*c14+c24.*c15.*c35.*c45+c45.*c13.*c25.*c23+c24.*c13.*c25.*c23+c45.*c15.*c23. *c34+c24.*c45.*c13.*c23+c24.*c35.*c13.*c23+c14.*c45.*c23.*c34+c14.*c23.*c34.*c25+c14.*c23.*c34. *c15+c35.*c14.*c23.*c34+c34.*c35.*c23.*c13+c34.*c45.*c23.*c13+c14.*c15.*c35.*c23+c14.*c13.*c15. *c23)./(c24.*c35.*c23+c35.*c14.*c34+c24.*c45.*c35+c24.*c45.*c23+c14.*c35.*c23+c14.*c45.*c23+c35 .*c23.*c45+c24.*c35.*c13+c24.*c45.*c13+c45.*c25.*c34+c45.*c15.*c34+c34.*c25.*c23+c34.*c13.*c25 +c34.*c25.*c35+c14.*c13.*c35+c13.*c35.*c45+c15.*c34.*c24+c14.*c13.*c45+c25.*c34.*c24+c14.*c35. *c45+c25.*c34.*c14+c15.*c34.*c14+c45.*c15.*c35+c45.*c13.*c15+c45.*c15.*c23+c14.*c13.*c25+c14.* c25.*c23+c14.*c25.*c35+c24.*c13.*c15+c24.*c15.*c35+c34.*c13.*c15+c34.*c15.*c35+c34.*c15.*c23+c 34.*c35.*c13+c34.*c35.*c23+c34.*c45.*c13+c34.*c45.*c23+c24.*c15.*c23+c45.*c13.*c25+c45.*c25.*c 23+c45.*c25.*c35+c24.*c13.*c25+c24.*c25.*c23+c24.*c25.*c35+c14.*c13.*c15+c14.*c15.*c35+c14.*c1 5.*c23+c35.*c24.*c34+c24.*c45.*c34+c14.*c45.*c34)+c12; 101 Appendix B: APDL macro for capacitance extraction simplified model table 3.2 !simplified model ! -----------------4--! -----3---! --1-- --2— ! -----------------5--finish /clear /prep7 /title, multielectrodes capacitor chip simulations !-----------geometry parameters (um)-----------------g=.1 !gap t=.1 !thickness w=1 !width wtop=2 !width top electrode tox=.5 !oxide thickness tsub=381 !Si substrate wchip=500 !width chip !there should be 27 groups of electrodes instead of 1, so the capacitances !are around 27 times smaller. wchip=15mm/27= .5mm so capacitances are equivalent displfinal=5 !displacement (final position) displinitial=0 resolution=.1 points=(displfinal-displinitial)/resolution !points step=(displfinal-displinitial)/points s=1 s1=3 s2=15 s3=75 s4=0 !separation between two bottom electrodes !separation units within first group !separation subgroups in 2nd group !separation subgroups in 3rd group st= 2*w+2*s-wtop !separation n1=3 n2=1 n3=1 n4=1 !number units first group !number subgroups 2nd group !number subgroups 3rd group !number subgroups 4rd group nelect=3*n1*n2*n3 wunit=2*w+s1 !width of unit,i.e. two bottom and 1 top electrode w1=wunit*n1+2*s1 w2=w1*n2+(n2-1)*s2 !width 2nd group 102 w3=w2*n3+(n3-1)*s3 !width 3nd group w4=w3*n4+(n4-1)*s4 !width 4th group wt=w4+displfinal-displinitial !total capacitor width *dim,c12,array,points !to store capacitance and displacement *dim,c13,array,points *dim,c14,array,points *dim,c15,array,points *dim,c23,array,points *dim,c24,array,points *dim,c25,array,points *dim,c34,array,points *dim,c35,array,points *dim,c45,array,points *dim,displacement,array,points !----------material and element properties-----------et,1,121 !for air, 2-D, 8-node charge-based electric element et,2,110,1,1 !to model open of 2D unbonded problem !1: voltage, 1: 8 node quadrilater et,3,121 !for si dioxide mp,perx,1,1 !relative permittivity 1 for air mp,perx,3,3.7 !relative permittivity 3.7 of SiO2 emunit,epzro,8.854e-6 !to obtain pF/um !--------------model---------------------------------displ=displinitial *do,id,1,points,1 !------displacement loop---------/prep7 !removing model *if,id,gt,1,then cmdele,1 cmdele,2 cmdele,3 cmdele,4 !delete assembly cmdele,5 asel,all aclear,all !clear mesh adele,all !delete areas lsel,all ldele,all !delete lines *endif !drawing electrodes spos=0 stb=wunit-wtop align=(w4+displ)/2 *do,i4,1,n4,1 *do,i3,1,n3,1 103 *do,i2,1,n2,1 *do,i1,1,n1,1 rectng,spos-align-wtop/2+stb/2+displ/2,wtop/2+spos-align+stb/2+displ/2, g/2,g/2+t !top electrode 3 rectng,spos-align-displ/2,w+spos-align-displ/2,-g/2,-g/2-t !bottom electr 1 rectng,spos-align+w+s-displ/2,spos-align+2*w+s-displ/2,-g/2,-g/2-t !bottom elec 2 spos=spos+wunit+s1 *enddo spos=spos-s1+s2 *enddo spos=spos-s2+s3 *enddo spos=spos-s3+s4 *enddo !oxide rectng, wchip/2, -wchip/2, g/2+t, g/2+t+tox rectng,-wchip/2,wchip/2,-g/2-t,-g/2-t-tox !Si substrate rectng,-wchip/2,wchip/2, g/2+t+tox, g/2+t+tox+tsub rectng,-wchip/2,wchip/2,-g/2-t-tox,-g/2-t-tox-tsub !gap rectng,-wchip/2,wchip/2,g/2+t,-g/2-t !air diag=wchip*1.7 cyl4,0,0,diag,180 cyl4,0,0,diag,180,0,360 !infinite element cyl4,0,0,diag*1.1,180 cyl4,0,0,diag*1.1,180,0,360 aovlap,all numcmp,area numcmp,line /pnum,area,1 aadd,nelect+9,nelect+10 aadd,nelect+6,nelect+7 numcmp,area !-------------meshing---------------------------------!10,11: substrate !12,13: ring !14: oxide top !15: oxide bottom !16: air !17: gap !----specifying divisions per line !electrodes lsel,s,line,,1 104 *do,j,3,nelect*4,2 lsel,a,line,,j *enddo lesize,all,,,40 !oxide lsel,s,line,,nelect*4+2 !line oxide-substrate lsel,a,line,,nelect*4+4 lesize,all,,,1000 lsel,s,line,,nelect*6+4 !line gap-oxide right lsel,a,line,,nelect*7+8 lsel,a,line,,nelect*6+2 !line gap-oxide left lsel,a,line,,nelect*7+2 lesize,all,,,1000/2 !—meshing mshape,1 !triangles mshkey,0 !use free meshing type,1 amesh,nelect+8 !gap type,3 amesh,nelect+5 !oxide amesh,nelect+6 type,1 smrtsize,1 !way way faster with smart mesh amesh,nelect+7 !air !—meshing infinite elements lsel,s,line,,nelect*5+6,nelect*5+7 lesize,all,,,1 smrtsize,off type,2 mshape,0 mshkey,1 amesh,nelect+3 amesh,nelect+4 !-----------setting up flag and electrodes--------------csys,1 !cylindrical nsel,s,loc,x,diag*1.1 sf,all,inf !flags !top electrodes *do,i,1,nelect,3 *if,i,eq,1,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo 105 cm,cond3,node !bottom electrode left *do,i,2,nelect,3 *if,i,eq,2,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond1,node !bottom electrode right *do,i,3,nelect,3 *if,i,eq,3,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond2,node !top substrate electrode asel,s,area,,nelect+1 lsla,s nsll,s,1 cm,cond4,node !bottom substrate electrode asel,s,area,,nelect+2 lsla,s nsll,s,1 cm,cond5,node allsel,all finish !--------------solutions----------/solu solve cmatrix,1,’cond’,5,1 displacement(id)=displ c12(id)=cmatrix(1,2,2) 106 c13(id)=cmatrix(1,3,2) c14(id)=cmatrix(1,4,2) c15(id)=cmatrix(1,5,2) c23(id)=cmatrix(2,3,2) c24(id)=cmatrix(2,4,2) c25(id)=cmatrix(2,5,2) c34(id)=cmatrix(3,4,2) c35(id)=cmatrix(3,5,2) c45(id)=cmatrix(4,5,2) !saving results *cfopen,c12,txt *vwrite,c12(1) (F20.16) *cfopen,c13,txt *vwrite,c13(1) (F20.16) *cfopen,c14,txt *vwrite,c14(1) (F20.16) *cfopen,c15,txt *vwrite,c15(1) (F20.16) *cfopen,c23,txt *vwrite,c23(1) (F20.16) *cfopen,c24,txt *vwrite,c24(1) (F20.16) *cfopen,c25,txt *vwrite,c25(1) (F20.16) *cfopen,c34,txt *vwrite,c34(1) (F20.16) *cfopen,c35,txt *vwrite,c35(1) (F20.16) *cfopen,c45,txt *vwrite,c45(1) (F20.16) *cfopen,displ,txt *vwrite,displacement(1) (F20.16) *status,c12 !to get something on screen displ=displ+step finish *enddo !---finishes displacement loop *status finish 107 Appendix C: APDL macro model for large gap and displacement range table !Jose Medina, ECE Northeastern, October 2006 !Capacitive displacement sensor design !code models metal electrode, oxide and Si chip ! electrodes: ! -----------------4--! -----3---! --1-- --2— ! -----------------5--finish /clear /prep7 /title, multielectrodes capacitor chip simulations /nerr,-1 !limits warning written to file.err to 1 per command !-----------geometry parameters (um)-----------------g=20 !gap t=.3 !thickness w=1 !width wtop=2 !width top electrode tox=.5 !oxide thickness tsub=381 !Si substrate displfinal=10 !displacement (final position) displinitial=0.5 resolution=.5 points=(displfinal-displinitial)/resolution !points step=(displfinal-displinitial)/points s=1 s1=2 s2=13 s3=65 s4=0 !separation between two bottom electrodes !separation units within first group !separation subgroups in 2nd group !separation subgroups in 3rd group st= 2*w+2*s-wtop !separation n1=3 n2=3 n3=3 n4=1 !number units first group !number subgroups 2nd group !number subgroups 3rd group !number subgroups 4rd group nelect=3*n1*n2*n3 108 wunit=2*w+s1 !width of unit,i.e. two bottom and 1 top electrode w1=wunit*n1+2*s1 w2=w1*n2+(n2-1)*s2 !width 2nd group w3=w2*n3+(n3-1)*s3 !width 3nd group w4=w3*n4+(n4-1)*s4 !width 4th group wt=w4+displfinal-displinitial !total capacitor width cx=0.01 !finer mesh box x cair=2.5 cxchip=1.5 !times wider than capacitor width wt *dim,c12,array,points !to store capacitance and displacement *dim,c13,array,points *dim,c14,array,points *dim,c15,array,points *dim,c23,array,points *dim,c24,array,points *dim,c25,array,points *dim,c34,array,points *dim,c35,array,points *dim,c45,array,points *dim,displacement,array,points !----------material and element properties-----------et,1,121 !for air, 2-D, 8-node charge-based electric element et,2,110,1,1 !to model open of 2D unbonded problem !1: voltage, 1: 8 node quadrilater et,3,121 !for si dioxide mp,perx,1,1 !relative permittivity 1 for air mp,perx,3,3.7 !relative permittivity 3.7 of SiO2 emunit,epzro,8.854e-6 !to obtain pF/um !--------------model---------------------------------displ=displinitial *do,id,1,points,1 !------displacement loop---------- /prep7 !removing model *if,id,gt,1,then cmdele,1 cmdele,2 cmdele,3 cmdele,4 !delete assembly cmdele,5 asel,all aclear,all !clear mesh adele,all !delete areas lsel,all ldele,all !delete lines *endif 109 !drawing electrodes spos=0 stb=wunit-wtop align=(w4+displ)/2 *do,i4,1,n4,1 *do,i3,1,n3,1 *do,i2,1,n2,1 *do,i1,1,n1,1 rectng,spos-align-wtop/2+stb/2+displ/2,wtop/2+spos-align+stb/2+displ/2, g/2,g/2+t !top electrode 3 rectng,spos-align-displ/2,w+spos-align-displ/2,-g/2,-g/2-t !bottom electr 1 rectng,spos-align+w+s-displ/2,spos-align+2*w+s-displ/2,-g/2,-g/2-t !bottom elec 2 spos=spos+wunit+s1 *enddo spos=spos-s1+s2 *enddo spos=spos-s2+s3 *enddo spos=spos-s3+s4 *enddo !oxide rectng,-spos*cxchip,spos*cxchip, g/2+t, g/2+t+tox rectng,-spos*cxchip,spos*cxchip,-g/2-t,-g/2-t-tox !Si substrate rectng,-spos*cxchip,spos*cxchip, g/2+t+tox, g/2+t+tox+tsub rectng,-spos*cxchip,spos*cxchip,-g/2-t-tox,-g/2-t-tox-tsub !gap rectng,-spos*cxchip,spos*cxchip,g/2+t,-g/2-t !air diag=((g/2+t+tox+tsub)**2+(spos*cxchip)**2)**.5 cyl4,0,0,diag*cair,180 cyl4,0,0,diag*cair,180,0,360 !infinite element r_infe=diag*cair*1.1 cyl4,0,0,r_infe,180 cyl4,0,0,r_infe,180,0,360 aovlap,all numcmp,area numcmp,line /pnum,area,1 aadd,nelect+5,nelect+6 !air: nelect+11 aadd,nelect+9,nelect+10 !gap: nelect+5 numcmp,area !-------------meshing---------------------------------!substrate bottom,top: 82,83 !inf elem rings bottom,top: 84,85 110 !gap: 86 !oxide top,bottom:87,88 !air: 89 !----defining divisions per line !oxide-substrate lsel,s,line,,326 !top lsel,a,line,,328 !bottom lesize,all,,,1000 !oxide-gap lsel,s,line,,350 !top left lsel,a,line,,377 !bottom left lsel,a,line,,376 !top right lsel,a,line,,431 !bottom right lesize,all,,,500 !infinite elements lsel,s,line,,340,341 lesize,all,,,1 !----meshing !gap smrtsize,1 !way faster and less errors with smart mesh mshape,1 !triangles mshkey,0 !use free meshing type,1 amesh,86 !oxide type,3 smrtsize,off amesh,87 amesh,88 !air type,1 smrtsize,1 amesh,89 !infinite elements smrtsize,off type,2 mshape,0 mshkey,1 amesh,84 amesh,85 !-----------setting up flag and electrodes--------------csys,1 !cylindrical nsel,s,loc,x,r_infe !nplot sf,all,inf !flags !top electrodes 111 *do,i,1,nelect,3 *if,i,eq,1,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond3,node !bottom electrode left *do,i,2,nelect,3 *if,i,eq,2,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond1,node !bottom electrode right *do,i,3,nelect,3 *if,i,eq,3,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond2,node !top substrate electrode asel,s,area,,nelect+2 lsla,s nsll,s,1 cm,cond5,node !bottom substrate electrode asel,s,area,,nelect+1 lsla,s nsll,s,1 cm,cond4,node 112 allsel,all finish !--------------solutions----------/solu solve cmatrix,1,’cond’,5,1 displacement(id)=displ c12(id)=cmatrix(1,2,2) c13(id)=cmatrix(1,3,2) c14(id)=cmatrix(1,4,2) c15(id)=cmatrix(1,5,2) c23(id)=cmatrix(2,3,2) c24(id)=cmatrix(2,4,2) c25(id)=cmatrix(2,5,2) c34(id)=cmatrix(3,4,2) c35(id)=cmatrix(3,5,2) c45(id)=cmatrix(4,5,2) !saving results *cfopen,c12chip,txt *vwrite,c12(1) (F20.16) *cfopen,c13chip,txt *vwrite,c13(1) (F20.16) *cfopen,c14chip,txt *vwrite,c14(1) (F20.16) *cfopen,c15chip,txt *vwrite,c15(1) (F20.16) *cfopen,c23chip,txt *vwrite,c23(1) (F20.16) *cfopen,c24chip,txt *vwrite,c24(1) (F20.16) *cfopen,c25chip,txt *vwrite,c25(1) (F20.16) *cfopen,c34chip,txt *vwrite,c34(1) (F20.16) *cfopen,c35chip,txt *vwrite,c35(1) (F20.16) *cfopen,c45chip,txt *vwrite,c45(1) (F20.16) *cfopen,displchip,txt *vwrite,displacement(1) (F20.16) 113 *status,c12 displ=displ+step finish *enddo !finishes displacement loop *status,displacement finish Appendix D: MATLAB filter for ANSYS capacitance results %’c12, c13, c14, c15, c23, c24, c25, c34, c35, c45, displ’ are input data in this file cout=(c45.*c13.*c15.*c23+c45.*c15.*c35.*c23+c24.*c13.*c15.*c23+c24.*c15.*c35.*c23+c24.*c45.*c35. *c13+c14.*c35.*c45.*c23+c25.*c23.*c45.*c15+c45.*c25.*c34.*c14+c14.*c13.*c45.*c23+c14.*c13.*c35. *c23+c23.*c13.*c35.*c45+c34.*c13.*c15.*c23+c34.*c15.*c35.*c23+c35.*c14.*c13.*c25+c13.*c25.*c34. *c24+c34.*c25.*c45.*c13+c34.*c35.*c13.*c25+c13.*c25.*c35.*c45+c34.*c25.*c15.*c23+c34.*c25.*c13. *c15+c34.*c25.*c15.*c35+c25.*c45.*c15.*c34+c34.*c25.*c24.*c15+c34.*c45.*c24.*c15+c34.*c35.*c24. *c15+c34.*c13.*c25.*c23+c34.*c35.*c24.*c13+c15.*c24.*c45.*c23+c15.*c14.*c13.*c25+c15.*c45.*c25. *c35+c15.*c24.*c25.*c23+c14.*c24.*c35.*c23+c14.*c24.*c45.*c35+c14.*c24.*c45.*c23+c14.*c24.*c35. *c13+c14.*c24.*c45.*c13+c14.*c15.*c34.*c24+c14.*c25.*c34.*c24+c14.*c24.*c13.*c15+c14.*c24.*c15. *c35+c14.*c24.*c15.*c23+c14.*c45.*c13.*c25+c14.*c24.*c13.*c25+c14.*c24.*c25.*c23+c14.*c24.*c25. *c35+c14.*c35.*c24.*c34+c14.*c24.*c45.*c34+c35.*c24.*c13.*c25+c45.*c23.*c25.*c14+c25.*c45.*c13. *c15+c25.*c24.*c13.*c15+c25.*c24.*c15.*c35+c25.*c14.*c15.*c35+c15.*c24.*c45.*c13+c14.*c25.*c35. *c45+c14.*c13.*c25.*c23+c25.*c23.*c15.*c14+c34.*c24.*c45.*c13+c34.*c24.*c13.*c15+c25.*c15.*c34. *c14+c35.*c25.*c34.*c14+c24.*c15.*c35.*c45+c45.*c13.*c25.*c23+c24.*c13.*c25.*c23+c45.*c15.*c23. *c34+c24.*c45.*c13.*c23+c24.*c35.*c13.*c23+c14.*c45.*c23.*c34+c14.*c23.*c34.*c25+c14.*c23.*c34. *c15+c35.*c14.*c23.*c34+c34.*c35.*c23.*c13+c34.*c45.*c23.*c13+c14.*c15.*c35.*c23+c14.*c13.*c15. *c23)./(c24.*c35.*c23+c35.*c14.*c34+c24.*c45.*c35+c24.*c45.*c23+c14.*c35.*c23+c14.*c45.*c23+c35 .*c23.*c45+c24.*c35.*c13+c24.*c45.*c13+c45.*c25.*c34+c45.*c15.*c34+c34.*c25.*c23+c34.*c13.*c25 +c34.*c25.*c35+c14.*c13.*c35+c13.*c35.*c45+c15.*c34.*c24+c14.*c13.*c45+c25.*c34.*c24+c14.*c35. *c45+c25.*c34.*c14+c15.*c34.*c14+c45.*c15.*c35+c45.*c13.*c15+c45.*c15.*c23+c14.*c13.*c25+c14.* c25.*c23+c14.*c25.*c35+c24.*c13.*c15+c24.*c15.*c35+c34.*c13.*c15+c34.*c15.*c35+c34.*c15.*c23+c 34.*c35.*c13+c34.*c35.*c23+c34.*c45.*c13+c34.*c45.*c23+c24.*c15.*c23+c45.*c13.*c25+c45.*c25.*c 23+c45.*c25.*c35+c24.*c13.*c25+c24.*c25.*c23+c24.*c25.*c35+c14.*c13.*c15+c14.*c15.*c35+c14.*c1 5.*c23+c35.*c24.*c34+c24.*c45.*c34+c14.*c45.*c34)+c12; n=length(displ) displsym=[-displ(n:-1:1),displ]; coutsym=[cout(n:-1:1),cout]; windowSize = 20; coutfilteredsym=filter(ones(1,windowSize)/windowSize,1,coutsym); figure(1) plot(displsym,coutsym,’b.’) xlabel(‘displacement [um]’) ylabel(‘capacitance [pF/um]’) title(‘output capacitance, 3x3x3x3 model, g=20 um’) figure(2) plot(displsym,coutsym,’b’,’linewidth’,2) xlabel(‘displacement [um]’) ylabel(‘capacitance [pF/um]’) 114 title(‘output capacitance, 3x3x3x3 model, g=20 um’) figure(3) plot(displsym,coutfilteredsym,’b.’,’linewidth’,2) xlabel(‘displacement [um]’) ylabel(‘capacitance [pF/um]’) title(‘filtered output capacitance, 3x3x3x3 model, g=20 um’) Appendix E: APDL macro to calculate lumped matrix capacitance versus number of nodes per electrode’s horizontal surface !analysis convergency ! -----------------4--! -----3---! --1-- --2— ! -----------------5--finish /clear /prep7 /title, multielectrodes capacitor chip simulations !-----------geometry parameters (um)-----------------g=.1 !gap t=.1 !thickness w=1 !width wtop=2 !width top electrode tox=.5 !oxide thickness tsub=381 !Si substrate wchip=500 !width chip !there should be 27 groups of electrodes instead of 1, so the capacitances !are around 27 times smaller. wchip=15mm/27= .5mm so capacitances are equivalent displ=0 divinitial=5 divfinal=100 step=5 points=(displfinal-displinitial)/step+1 !points s=1 s1=3 s2=15 s3=75 s4=0 !separation between two bottom electrodes !separation units within first group !separation subgroups in 2nd group !separation subgroups in 3rd group st= 2*w+2*s-wtop !separation n1=3 !number units first group 115 n2=1 n3=1 n4=1 !number subgroups 2nd group !number subgroups 3rd group !number subgroups 4rd group nelect=3*n1*n2*n3 wunit=2*w+s1 !width of unit,i.e. two bottom and 1 top electrode w1=wunit*n1+2*s1 w2=w1*n2+(n2-1)*s2 !width 2nd group w3=w2*n3+(n3-1)*s3 !width 3nd group w4=w3*n4+(n4-1)*s4 !width 4th group wt=w4+displfinal-displinitial !total capacitor width *dim,c12,array,points !to store capacitance and displacement *dim,c13,array,points *dim,c14,array,points *dim,c15,array,points *dim,c23,array,points *dim,c24,array,points *dim,c25,array,points *dim,c34,array,points *dim,c35,array,points *dim,c45,array,points *dim,divisions,array,points !----------material and element properties-----------et,1,121 !for air, 2-D, 8-node charge-based electric element et,2,110,1,1 !to model open of 2D unbonded problem !1: voltage, 1: 8 node quadrilater et,3,121 !for si dioxide mp,perx,1,1 !relative permittivity 1 for air mp,perx,3,3.7 !relative permittivity 3.7 of SiO2 emunit,epzro,8.854e-6 !to obtain pF/um !--------------model---------------------------------div=divinitial *do,id,1,points,1 !------displacement loop---------- /prep7 !removing model *if,id,gt,1,then cmdele,1 cmdele,2 cmdele,3 cmdele,4 !delete assembly cmdele,5 asel,all aclear,all !clear mesh adele,all !delete areas lsel,all 116 ldele,all !delete lines *endif !drawing electrodes spos=0 stb=wunit-wtop align=(w4+displ)/2 *do,i4,1,n4,1 *do,i3,1,n3,1 *do,i2,1,n2,1 *do,i1,1,n1,1 rectng,spos-align-wtop/2+stb/2+displ/2,wtop/2+spos-align+stb/2+displ/2, g/2,g/2+t !top electrode 3 rectng,spos-align-displ/2,w+spos-align-displ/2,-g/2,-g/2-t !bottom electr 1 rectng,spos-align+w+s-displ/2,spos-align+2*w+s-displ/2,-g/2,-g/2-t !bottom elec 2 spos=spos+wunit+s1 *enddo spos=spos-s1+s2 *enddo spos=spos-s2+s3 *enddo spos=spos-s3+s4 *enddo !oxide rectng, wchip/2, -wchip/2, g/2+t, g/2+t+tox rectng,-wchip/2,wchip/2,-g/2-t,-g/2-t-tox !Si substrate rectng,-wchip/2,wchip/2, g/2+t+tox, g/2+t+tox+tsub rectng,-wchip/2,wchip/2,-g/2-t-tox,-g/2-t-tox-tsub !gap rectng,-wchip/2,wchip/2,g/2+t,-g/2-t !air diag=wchip*1.7 cyl4,0,0,diag,180 cyl4,0,0,diag,180,0,360 !infinite element cyl4,0,0,diag*1.1,180 cyl4,0,0,diag*1.1,180,0,360 aovlap,all numcmp,area numcmp,line /pnum,area,1 aadd,nelect+9,nelect+10 aadd,nelect+6,nelect+7 numcmp,area !-------------meshing---------------------------------!10,11: substrate 117 !12,13: ring !14: oxide top !15: oxide bottom !16: air !17: gap !----specifying divisions per line !electrodes lsel,s,line,,1 *do,j,3,nelect*4,2 lsel,a,line,,j *enddo lesize,all,,,div !oxide lsel,s,line,,nelect*4+2 !line oxide-substrate lsel,a,line,,nelect*4+4 lesize,all,,,1000 lsel,s,line,,nelect*6+4 !line gap-oxide right lsel,a,line,,nelect*7+8 lsel,a,line,,nelect*6+2 !line gap-oxide left lsel,a,line,,nelect*7+2 lesize,all,,,1000/2 !—meshing mshape,1 !triangles mshkey,0 !use free meshing type,1 amesh,nelect+8 !gap type,3 amesh,nelect+5 !oxide amesh,nelect+6 type,1 smrtsize,1 !way way faster with smart mesh amesh,nelect+7 !air !—meshing infinite elements lsel,s,line,,nelect*5+6,nelect*5+7 lesize,all,,,1 smrtsize,off type,2 mshape,0 mshkey,1 amesh,nelect+3 amesh,nelect+4 !-----------setting up flag and electrodes--------------csys,1 !cylindrical nsel,s,loc,x,diag*1.1 sf,all,inf !flags !top electrodes *do,i,1,nelect,3 118 *if,i,eq,1,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond3,node !bottom electrode left *do,i,2,nelect,3 *if,i,eq,2,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond1,node !bottom electrode right *do,i,3,nelect,3 *if,i,eq,3,then asel,s,area,,i lsla,s !selects lines from selected area nsll,s,1 !selects nodes from selected lines *else asel,a,area,,i lsla,a nsll,a,1 *endif *enddo cm,cond2,node !top substrate electrode asel,s,area,,nelect+1 lsla,s nsll,s,1 cm,cond4,node !bottom substrate electrode asel,s,area,,nelect+2 lsla,s nsll,s,1 cm,cond5,node allsel,all finish 119 !--------------solutions----------/solu solve cmatrix,1,’cond’,5,1 divisions(id)=div c12(id)=cmatrix(1,2,2) c13(id)=cmatrix(1,3,2) c14(id)=cmatrix(1,4,2) c15(id)=cmatrix(1,5,2) c23(id)=cmatrix(2,3,2) c24(id)=cmatrix(2,4,2) c25(id)=cmatrix(2,5,2) c34(id)=cmatrix(3,4,2) c35(id)=cmatrix(3,5,2) c45(id)=cmatrix(4,5,2) !saving results *cfopen,c12convg01,txt *vwrite,c12(1) (F20.16) *cfopen,c13convg01,txt *vwrite,c13(1) (F20.16) *cfopen,c14convg01,txt *vwrite,c14(1) (F20.16) *cfopen,c15convg01,txt *vwrite,c15(1) (F20.16) *cfopen,c23convg01,txt *vwrite,c23(1) (F20.16) *cfopen,c24convg01,txt *vwrite,c24(1) (F20.16) *cfopen,c25convg01,txt *vwrite,c25(1) (F20.16) *cfopen,c34convg01,txt *vwrite,c34(1) (F20.16) *cfopen,c35convg01,txt *vwrite,c35(1) (F20.16) *cfopen,c45convg01,txt *vwrite,c45(1) (F20.16) *cfopen,divconvg01,txt *vwrite,divisions(1) (F20.16) div=div+step finish *enddo !---finishes displacement loop finish 120 Appendix F: input equivalent capacitance for connected plates. The electrodes are numbered as shown in the figure below. Figure. F.1: scheme sensor with plates connected The following figure shows the selected tree and the branches. Figure F.2: Circuit graph with selected tree and branch current references The fundamental loops, branches, and equations are as follows. Loop 2-3 2-4 1-4 Branch 2-3, 1-2, 1-3 2-4, 2-3, 3-4 1-4, 1-3, 3-4 Equation 0 = i23 Z23 – V + i13Z13 0 = i24 Z24 – i23Z23 + i34Z34 0 = i14 Z14 + i13 Z13 + i34 Z34 The cut set equations are as follows i12 = i23 + i24 (F.1) i34 = i24 + i14 (F.2) i13 = i23 + i34 = i14 + i23 + i24 The following code implements the equations in MATLAB syms v c12 [i14,i24,i23]=solve('i23*1/c23 - v + (i14+i23+i24)*1/c13','i24*1/c24 - i23*1/c23 + (i24+i14)*1/c34','i14*1/c14 + (i14+i23+i24)*1/c13 + (i24+i14)*1/c34','i14,i24,i23'); i12=i23+i24; cout=simplify(i12/v) +c12 121 (F.3) The result is cout = (c34*c24*c13+c24*c14*c13+c24*c14*c34+c24*c14*c23+c13*c34*c23+c14*c13*c23+c14*c34*c23+c24 *c23*c13) / (c13*c34+c14*c13+c14*c34+c23*c34+c14*c23+c24*c34+c24*c13+c24*c23)+c12 Appendix G: hspice code readout circuit Figure 4.24 and RC LPF *gregorian SCA .lib 'models2.lib' devicemodels .option post .tran 10us 300ms sweep data=filter .option dcstep=1e-5 .op .data filter rfilter 2meg 20meg 200meg .enddata Vinp in 0 1 M1 in phi1 1 0 cmosn M2 1 phi2 0 0 cmosn Cs 1 2 'V(var1)*1e-9' ic=0 $sensor capacitance E1 out 0 3 2 1e6 Vos 3 0 .01 Cdg 1 out '5e-10' $anti-spike capacitor Cf 2 4 '1e-9' ic=0 $feedback capacitor M5 4 phi1 0 0 cmosn M4 4 phi2 out 0 cmosn M3 2 phi1 out 0 cmosn *RC low pass filter R7 out outf rfilter C7 outf 0 1n *nonoverlapping clocks .param f=1ms Vphi1 phi1 0 pulse(0 5 '0*f' '.1*f' '.1*f' '2.3*f' '5*f') Vphi2 phi2 0 pulse(0 5 '2.5*f' '.1*f' '.1*f' '2.3*f' '5*f') *control capacitance Vc1 var1 0 pwl(0,0.1 100ms,1 200ms,1 300ms,0.1) Rc1 var1 0 10k .end 122