Investigating control strategies for the Phicom 3 wirebonder

Transcription

1 Investigating control strategies for the Phicom 3 wirebonder T. Kok DCT Traineeship report Coach(es): Supervisor: H.M.J. van de Groes M. Steinbuch Technische Universiteit Eindhoven Department Mechanical Engineering Dynamics and Control Technology Group Eindhoven, August, 2006

2 2

3 Contents Summary 9 Nomenclature 11 1 Introduction Background Problem Statement Outline of Thesis Introduction to the Phicom The bonding process The Phicom System identification Measurement setup Unconstrained phase Transfer Modeling the system Constrained phase Transfer Modeling the system Conclusions Control strategy Problems DAC output limit Measurements Switching control Stability Simulations Conclusions and Recommendations Conclusions Recommendations

4 A Measurement Setup 33 A.1 Experiment Setup A.1.1 List of Used Equipment A.1.2 List of Used Software A.2 Description of Experiments A.2.1 FRF measurements A.2.2 Impact measurements B Simulation model 39 B.1 List of Used Software B.2 Description of Simulink model

5 List of Tables 3.1 Parameters of the unconstrained situation Parameters of the constrained situation A.1 Equipment used A.2 Software used B.1 Software used B.2 Component description of the simulink model

6 6

7 List of Figures 1.1 Endproduct of the Phicom Bonding proces Schematics of the Phicom z-axis Sensitivity measurement FRF measurement of the unconstrained situation th order model of the unconstrained situation Measured and modeled plant FRF measurement of the constrained situation th order model of the constrained situation Measured and modeled plant FRF measurement of both situation Responses with DAC output limit applied for different impact speeds Simulink model of the switching control strategy Simulation results without switching Simulation results with switching A.1 Sensitivity measurement A.2 Schematic representation of the PMAC A.3 Coherence of the measurements of both situation A.4 Measurement setup A.5 Trajectory used for impact measurements

8 8

9 Summary At Philips ITEC a complete semiconductor line is developed and constantly improved. One of the machines in line is the Phicom wire bonder. This machine, and especially its z-axe, will be the subject of this report. The problem is to speed up the total trajectory which consists of two separate phases. The first phase without contact with the surrounding environment and the second phase with contact. In the first phase a very fast trajectory should be followed with as little tracking error as possible. For the second phase a constant force should be exerted to the environment. The project goal is to investigate possible control strategies which might be able to control the total trajectory with high performance. For a better understanding of the system, a system identification by means of a FRF measurement was done. Because of the very different dynamical properties of the separate phases this was done with and without contact with the environment. Models were made for simulation purposes and these models were fitted onto the obtained FRF data. Three different control strategies were investigated. The first idea was to use a normal linear controller to control both stages. Another idea was to use a DAC output limit in order to limit the motor force. This would mean that the first stage could be controlled by a high performance controller while the second stage is only controlled by a constant force. The third option is to control the system with a switching controller. Separate controllers could then be designed for the separate phases. From the obtained FRF data and earlier investigations it became clear that a normal linear controller would not be able to control the system with a high performance. The DAC output limit will not be able to control both phases because of a dominant resonance at 500Hz of the system with contact. A switching controller should, although not yet tested in practice, be able to control both stages with high performance. Further research should incorporate the implementation of a switching control strategy. Also small changes in the dynamics of the system should be researched to allow higher bandwidths of the separate phases. 9

10 10

11 Nomenclature Symbols A, B, C, D State space matrices x(t) State u(t) Input signal y(t) Output signal r(t) Reference signal n(t) Noise signal e(t) Error signal P (s) Physical plant C(s) Feedback compensator S(s) Sensitivity Constrained plant m b m t k c k bt d c d bt k 0 Bondhead mass Transducer mass Cross spring stiffness Stiffness between bondhead and transducer Cross spring damping Damping between bondhead and transducer Total internal gain 11

12 Unconstrained plant m bt k c k env d c k 0 Total mass Cross spring stiffness Environmental stiffness Cross spring damping Total internal gain 12

13 Chapter 1 Introduction 1.1 Background Philips ITEC is developing and constantly improving a state of the art semiconductors production line. This line starts with wafers, which are cut in dies, and ends with ready to distribute semiconductors. The first machine in line, the ADAT, picks up the die from the sawn wafer and attaches it to the lead frame. Extra connections, which are usually needed, are made by means of wire bonding. This is done with the Phicom. The lead frame, connected to the semiconductor chip, is then molded into a package by use of the Mutiplunger. Finally the end product is tested with the µparset. Figure 1.1 Endproduct of the Phicom. The subject of this report will be the second machine in line, the Phicom. This machine is used to make gold wire connections between the die and the lead frame as illustrated in figure 1.1. Ultrasonic welding is used to make these connections. The welding unit of the Phicom has four axis, two (x and y) for the positioning of the weld in the xy-plane, one (z2) for the wire transport and one (z) for the bonding. This report will focus on problems, which are encountered when higher output rates are needed. Here only the z-axe, which is used for the actual bonding, will be discussed. 13

14 1.2 Problem Statement The main problem is to bring down the time per weld while maintaining or improving the product quality. The bonding process exists of two phases, the unconstrained phase before touchdown and the constrained phase after touchdown. The problem of the first phase is to control the bondhead to track a very fast trajectory with as little tracking error as possible. In the second phase the force exerted on the weld should be controlled to get and stay within some limits as fast as possible. This leads to the following problem statement. The goal of this project is to design a controller that is able to control both stages of the welding process. 1.3 Outline of Thesis In Chapter2 an introduction is given to the Phicom3 split-control, the most important parts will be explained. Chapter3 gives an identification of the dynamics of the system which can be used for simulation purposes. Different control strategies are explicated and simulations and experiments are done in Chapter4. Finally, Chapter5 give the conclusions and recommendations. 14

15 Chapter 2 Introduction to the Phicom 2.1 The bonding process Wire-bonding is the process that makes connections between the bare semiconductor and the lead frame. This is done with very thin (23µm) gold wires, using a so-called ball-wedge bonding strategy. The ball is made by a spark coming from a spark-unit (not shown in figure 2.1). This ball is then pressed on the die and the connection is made by means of ultrasonic welding. The bondhead is then transported to the leadframe where a wedgeweld is made. The force applied to the weld should be high enough to get a good weld but when it is too high the semiconductor could be damaged. The process is illustrated by figure 2.1. Figure 2.1 Bonding proces. 2.2 The Phicom As stated before only the z-axis will be discussed in the report. A schematic of the Phicom z-axis is shown in figure 2.2. The most important parts will be explained. 15

16 Figure 2.2 Schematics of the Phicom z-axis. A voice-coil motor serves as the actuator of the system. Small upward translations at the actuator side will be translated into small downward translations at the product side and visa versa. A cross-spring bearing acts as a hinge. The position is measured at the actuator side with an encoder (not visible in figure 2.2). A transducer transfers the ultrasonic movements from a piëzo to the tip (capillary). The gold wire runs trough this tip. 16

17 Chapter 3 System identication In chapter 2 an introduction to the phicom was given. This chapter will focus on the dynamics of the phicom z-axis. The wire bonding proces consists of two phases, the unconstrained phase with the capillary free in the air and the constrained phase with the capillary pressed on the product. These phases have very different dynamical properties. Therefore two different identifications will be done. 3.1 Measurement setup A frequency response measurement for both situations is done using siglab. Figure 3.1 shows the setup for the measurement. SigLab n r + e C(s) + + u P (s) y Figure 3.1 Sensitivity measurement The measurements are done in closed-loop for obvious reasons. SigLab is used to create noise and measure this input and an output signal. The reference will be kept zero (r = 0). Measurement setup is further explained in appendix A. The sensitivity is calculated from the measured signals. When written as a function of C(s) (the controller) and P (s) (the 17

18 plant) the sensitivity reads: S(s) = P (s)c(s) (3.1) Of which C(s) is known and S(s) is measured. The plant is derived using this formula: P (s) = 1 S(s) C(s)S(s) (3.2) 3.2 Unconstrained phase Transfer The bode diagram of the calculated transfer of the unconstrained plant is depicted in figure 3.2. To proof the validity of the measurement a coherence plot is shown in figure A.3 of appendix A. Figure 3.2 FRF measurement of the unconstrained situation Modeling the system A model of the system is needed for simulation purposes. Figure 3.3 shows a simplified 18

19 Figure th order model of the unconstrained situation. model of the system in the unconstrained phase. This model is used to derive the statespace equations of the system. The standard form for state space equations is: ẋ(t) = A(t) x(t) + B(t) u(t) y(t) = C(t) x(t) + D(t) u(t) (3.3) Within this situation the system is and thus the system-matrices are time invariant. The following equations are derived: ẋ(t) = k bt k c m b d bt d c m b k bt m b d bt m b k bt m t d bt m t k bt m t d bt m t ( ) y(t) = k x(t) + (0) u(t) x(t) m b 0 0 u(t) (3.4) This model is fit onto the plant derived from the measurement. Table 3.1 shows the values of the parameters used. 19

20 Parameter Value Description m b 0.017[kg] Bondhead mass at tip m t [kg] Transducer mass at tip k c 0.5[ N m ] Cross spring stiness at tip k bt 150[ N m ] Stiness between bondhead and transducer at tip d c 0.35[ N s m ] Cross spring damping at tip d bt [ N s m ] Damping between bondhead and transducer at tip k Total internal gain Table 3.1 Parameters of the unconstrained situation. Figure 3.4 shows the frequency response of the modeled and the measured plant. Figure 3.4 Measured and modeled plant 3.3 Constrained phase Transfer Another frequency response measurement is done using siglab. The bode diagram of the calculated transfer of the constrained plant is depicted in figure

21 Figure 3.5 FRF measurement of the constrained situation Modeling the system Again a model of the system is needed for simulation purposes. Figure 3.6 shows a sim- Figure th order model of the constrained situation. plified model of the system in the constrained phase. This model is used to derive the 21

22 state-space equations of the system. The following equations are derived: ( ) ( ẋ(t) = k c k env d c x(t) + 1 ( m bt ) m bt m b y(t) = k 0 0 x(t) + (0) u(t) ) u(t) (3.5) This model is fit onto the plant derived from the measurement. Table 3.2 shows the values of the parameters used. Parameter Value Description m bt 0.017[kg] Bondhead mass at tip k c 150[ N m ] Cross spring stiness at tip k env [ N m ] Stiness between bondhead and transducer at tip d c 2.81[ N s m ] Cross spring damping at tip k Total internal gain Table 3.2 Parameters of the constrained situation. Figure 3.7 shows the frequency response of the modeled and the measured plant. Figure 3.7 Measured and modeled plant 22

23 3.4 Conclusions Two different models where made of the unconstrained respectively the constrained situation. These models where fitted onto the measured data. The obtained models should be able to describe the behavior of the two phases of the bonding process. 23

24 24

25 Chapter 4 Control strategy In chapter 3 the mechanics of the Phicom where analyzed. A FRF measurement of both situations was done and simplified models where fitted on this data. This chapter will focus on controlling both phases of the bonding process. First the problems that occur in the controller design for this system will be explained. In Paragraph 4.2 the idea of limiting the output of the Digital to Analog Converter (DAC) is explained and tested. Paragraph 4.3 will focus on the use of a switching controller. 4.1 Problems The Phicom uses standard PMAC controller cards (schematic representation is available in figure A.2 of appendix A). Therefore it is only possible to implement control strategies that are allowed by this kind of controller. The simplest control strategy is to find a controller that suits both phases of the bonding process. This section will show why it is not possible to get a high performing system using a single controller to control this system. Figure 4.1 shows the bode diagrams of both situations. In order to get a high bandwidth for the unconstrained situation a phase lead filter should by applied. The system needs extra phase up to the crossover frequency. After this frequency roll-off is needed. The constrained situation has one large resonance at around 500Hz. The controller should suppress this resonance with for example a notch filter. This notch filter would add phase lag. The problem is that the bandwidth of the unconstrained situation should be around 250Hz, which is very close to the resonance in the constrained situation. This means that phase lead is needed for the unconstrained situation, where the notch should be applied for the constrained situation. This is not possible with a single controller. 25

26 Figure 4.1 FRF measurement of both situation. 4.2 DAC output limit One way to control the both situations is to apply a DAC output limit. The idea is that when impact occurs the force delivered to the product should be kept constant. A limit applied to the DAC output means that the force delivered by the motor is also limited. The limit is reached very fast after impact because of the fast growing position error. This way the unconstrained situation is controlled by a high performance controller and the constrained situation by a constant force. When a high performance system is wanted using this kind of control strategy the DAC limited plant, here the constrained situation, should have a good open loop step-response. The response should be fast while suppressing struck resonances Measurements In order to test this idea a measurement setup was made with a piezo force sensor. The setup is shown in figure A.4 of appendix A. A DAC output limit is applied, which ensures that the motor output force will stay below 0.5N. Measurements are done with different constant impact speeds using a trapezoid shaped trajectory. The trajectory with a 20mm/s impact speed is shown in figure A.5 of appendix A. Figure 4.2 shows the force response of the system for different impact speeds. The welding process, and thus the time the capillary is pressed onto the product, takes about 15ms. 26

27 (a) 10mm/s (b) 15mm/s (c) 20mm/s (d) 25mm/s Figure 4.2 Responses with DAC output limit applied for dierent impact speeds. Therefore the force response should be much faster than 15ms. It becomes clear that the resonance at 500Hz in the constrained situation is not damped enough to get a satisfactory response. When a 10mm/s trajectory is used the resonance is not struck but the response takes about 30ms to settle which is also to slow. 4.3 Switching control Another possible solution to the problem mentioned above is the use of a switching controller. A switching controller can switch its parameter set when needed. For the Phicom this would mean that parameters are switched when impact occurs, and back again when the capillary looses contact with the product. Although the theory behind a switching controller seems very simple care should be taken with respect to stability. Also it should be possible to implement this kind of controller in practice Stability Stability analysis of a switching control scheme is divided into three parts ([1] and [2]). The first two parts should prove asymptotic stability for the unconstrained and constrained 27

28 situations respectively. The third part should prove stability for the transition phase. Figure 4.1 show the bode diagrams of both situations. With an appropriate controller it is possible to get stable high performance systems for the separate situations. So the only part that should be analyzed is the transition phase. In [1] it is stated that the transition phase could only be instable when an infinite number of switches occurs. Otherwise the system ends in one of the two situations mentioned above and so the system stabilizes. Although mathematical proof is very difficult only one assumption has to be done to ensure a finite number of switches. If bouncing is not allowed only one switch is made. It is possible to do this assumption because of the nature of the system, e.g. the position of the environment and the impact speed are accurately known Simulations In chapter 3 models were made for both situations. A simulation of the switching control strategy using these models will be the subject of this section. Matlab/Simulink model A Matlab/Simulink model was developed in order to test the theoretical feasibility of the switching control strategy. Figure 4.3 shows the simulink model used. The main components of this model are explained in tabel B.2 of appendix B. Figure 4.3 Simulink model of the switching control strategy. 28

29 Results Figure 4.4 shows the result of a simulation when switching is not applied. In figure 4.5 the result of a simulation with switching applied is shown. The switching of parameters is done after 2 milliseconds. These plots show that in theory a switching controller could improve performance. Also it is shown that the performance improvement is directly linked to the switching delay witch is mainly caused by the time needed for touchdown detection. Figure 4.4 Simulation results without switching. 29

30 Figure 4.5 Simulation results with switching. 30

31 Chapter 5 Conclusions and Recommendations The Phicom3 split-control is the next generation wirebonder. Next generation indicates that a jump should be made with respect to speed and performance. Advanced control strategies might be able to achieve this. 5.1 Conclusions The Phicom z-axis was analyzed and models were made for it. Three different control strategies were explicated: single controller Due to the resonance in the constrained situation it is not possible to obtain a high bandwidth controller for both phases. The resonance frequency is around 500Hz which is approximately the same as the wanted bandwidth. This resonance should be suppressed by the controller which accounts for extra phase loss while the constrained situation needs extra phase around the crossover frequency. It is not possible to achieve this by applying a single controller. DAC output limit DAC output limit control is a good alternative when the dynamics of the uncontrolled situation have a good step response. The constrained situation does not have a good step response. The resonance of 500Hz is clearly visible and is damped to slow. This control strategy is therefore not applicable for this system. Switching control Although not yet tested in practice switching control could in theory overcome the problems that occurred when other control strategies where applied. Simulation results show big improvements with respect to resonance suppression. Care should be taken with respect to stability when this control scheme is implemented in practice. 31

32 5.2 Recommendations Due to practical limitations of the PMAC controller card it was not possible to implement a switching control strategy. The next generation controller cards (the FlexDMC) should be able to switch parameter fast enough to allow a switching controller. This lead to the first point of further investigation: The switching control scheme should be implemented en tested. This should at first be done with controllers which are stable in both situations to test the possibilities of the FlexDMC. After these first tests higher bandwidth controllers should be implemented which are probably not stable in both situations. Robustness with respect to stability of the controlled system should be a point of focus. Another point of investigation is the mechanics of the system. Both situations suffer from parasitic dynamics witch introduce unwanted resonances. The unconstrained situation suffers mostly from resonances which are introduced by the cross spring hinge ([3] and [4]). Apart from these unwanted dynamics the constrained situation also suffers from the lack of stiffness of the transducer and yoke. To allow higher bandwidths of the separate situations when a switching controller is used the mechanics of the system should be improved. 32

33 Appendix A Measurement Setup A.1 Experiment Setup A.1.1 List of Used Equipment Hardware FRF measurements Phicom3 split-control PMAC controller card SigLab Notebook Impact measurements Scope Piezo Piezo amplier Description Machine to be analyzed Controller installed on the phicom Dynamic Signal and System Analyzer With SCSI-card and SigLab installed Fluke PM3394A Autoranging combiscope Used as a force sensor. Output : 550mV/N Table A.1 Equipment used. 33

34 A.1.2 List of Used Software Software FRF measurements Matlab 6.5 SigLab 3.28 Impact measurements Fluke view Matlab 6.5 Description Data processing, plotting Matlab package for communicating with the SigLab hardware Software package for communicating with the scope Data processing, plotting Table A.2 Software used. A.2 Description of Experiments A.2.1 FRF measurements Sensitivity measurements are done using the setup shown in figure A.1. SigLab is used to create noise and measure this input and the output signal. The reference will be kept zero (r = 0). SigLab n r + e C(s) + + u P (s) y Figure A.1 Sensitivity measurement A schematic representation of the PMAC controller card (C(s)) is given in figure A.2. The settings for this controller are: Ix08 Ix09 = 12 (Internal gain) Ix30 = 4000 (Proportional gain) Ix31 Ix32 = 2000 (Derivative gains) Ix33 Ix39 = 0 (Integral gain, feedforward and biquad parameters) As stated before the measurement is done for the unconstrained and the constrained plant. Extra care should be taken for the constrained situation to keep the capillary pressed on 34

35 the environment at all time. To accomplish this the capillary should be pressed onto the environment with a high enough force before the measurement starts. 35

36 z Ix35 Ix z 1 + Ix32 Ix z 1 Ix z 1 + Ix33 Ix Ix Ix36z 1 +Ix37z Ix38z 1 +Ix39z 2 2 DAC 15 z 1 + Ix31 Ix z 1 Figure A.2 Schematic representation of the PMAC 36

37 The coherence of the measurements is shown in figure A.3. The plot shows that the coherence is above for all frequencies higher than 40Hz. Therefore the measurement can be used to reconstruct the plant s transfer function good enough. Figure A.3 Coherence of the measurements of both situation. A.2.2 Impact measurements Impact measurements where done using the setup shown in figure A.4. The piezo amplifier converts a force of 1N to an output of 0.55V. Measurements are done with different constant impact speeds using a trapezoid shaped trajectory. The trajectory with a 20mm/s impact speed is shown in figure A.5. 37

38 Figure A.4 Measurement setup Figure A.5 Trajectory used for impact measurements 38

39 Appendix B Simulation model B.1 List of Used Software Software Matlab 7.1 Simulink 6 Description Data processing, plotting Modeling and simulating of the control strategy Table B.1 Software used. B.2 Description of Simulink model Component Unconstrained system Constrained system System switch Impact disturbance Description A model of the plant and the controller are used to simulate the unconstrained situation. The constrained system contains the unconstrained as well as the constrained controller. The unconstrained controller is used to simulate the fact that the controller is switched milliseconds after the impact is made. Used for the switching from the unconstrained system to the constrained system. Produces an impact disturbance at the moment of impact. Table B.2 Component description of the simulink model. 39

40

41 Bibliography [1] Tarn, T. J., Y. Wu, N. Xi, A. Isidori, Force Regulation and Contact Transition Control, IEEE Control Systems, Vol. 16, No. 1, pp , Feb [2] Moorehead, Stewart John, Position and Force Control of Flexible Manipulators, Waterloo 1996 [3] Ahuis J., Modal analysis Phicom 3+, Jun [4] Kok T., FRF-measurements with different cross-springs, , ITEC Nijmegen. [5] Rodermond S., Impedance control of a PHICOM 3+, May [6] Verkooijen R.A.A., Impact Modeling and Control of a PHICOM 3+ wire bonder, December [7] Delta Tau Data Systems, Inc. User Manual Turbo PMAC, 6 may. [8] Groes, H.M.J. van de, Modeling Z-yoke, a, ITEC Nijmegen. [9] Groes, H.M.J. van de, PMAC controler calculations in scope, c, ITEC Nijmegen. 41