On Packaging of MEMS. Simulation of Transfer Moulding and Packaging Stress and their Effect on a Family of piezo-resistive Pressure Sensors
Abstract
Micro Electro Mechanical Systems (MEMS) produced to date include IR detectors, accelerometers, pressure sensors, micro lenses, actuators, chemical sensors, gear drives, RF devices, optical processor chips, micro robots and devices for biomedical analysis. The track for tomorrow has already been set and products like 3D TV, physician on a chip, lab on a chip, micro aircraft and food safety sensors will be developed when the technology matures and the market is ready.
Todays MEMS fabrication is typically based around a silicon substrate and borrow batch fabrication processes from the IC industry. Many of the developed MEMS products have never left a laboratory environment because they are fragile in the macro environment. The way to deal with this is to provide proper packaging so that they can be handled. This poses one of the major challenges in the MEMS industry. Not many packaging techniques have been commercially developed for MEMS and companies that have overcome the packaging problems very seldom reveal their packaging techniques. Functional problems that could be associated with a MEMS structure are often amplified by the package. The reason for this is often associated with packaging stress. Packaging stress related problems is what has kept many promising products from emerging on the market. Even the commercially available pressure sensors and accelerometers have packaging stress problems, but most of them have been overcome. A first step towards solving these challenges is to localise, quantify and understand the critical packaging stresses that act in a packaged MEMS device.
The goal of this work was to understand how packaging stresses act in a plastic moulded MEMS chip. The work has been threefold; simulation of transfer moulding, static stress analysis of the plastic capsule after moulding and modelling of the piezo-resistive behaviour of a MEMS pressure sensor.
This dissertation is divided into 9 chapters. Chapter 1 introduces the concept of level-0 and level-1 packaging and looks at different techniques of obtaining the different packaging levels. It introduces the Small Outline Package (SOP) which is the package that has been simulated in this dissertation.
Chapter 2 gives the background in the theory that has been used to complete this work. It starts by discussing the chemistry and mechanics of thermosetting polymers. Then the rheological behaviour of Epoxy Moulding Compounds (EMC) in a transfer moulding process is discussed.
The experimental results from the thermomechanical material characterisation of the EMC are presented in Chapter 3. The material was found to have a Tg of 130oC and coefficient of linear expansion of /oC and /oC below and above Tg respectively. It was further found that the material showed linear viscoelastic behaviour. Stress relaxation tests were run to obtain the relaxation coefficients needed for accurate modelling. The material was found to behave in a thermo rheologically simple manner and the WLF shift function was used to describe the time-temperature superposition principle.
Chapter 4 addresses the applicability of the plastic processing simulation code, C-Mold, for simulations of MEMS packaging in a SOP. It was found that the 2.5D simulation technique used by the software was inadequate for simulating the polymer filling of the SOP in question. This conclusion was drawn because 3D flow effect were observed in the moulding cavities. The cause for the 3D flow effect was the height of the SOP which was relatively large in order to accommodate for the MEMS device. However, the software proved to be very useful for balancing the runner system.
Chapter 5 starts with the development of a novel method for calculating the accurate piezoresistance for implanted silicon piezo-resistors. The method let each finite element in a piezoresistor region represent one resistor in a resistor network. The total resistance was then found by simple resistor summation. This method was then utilized on a silicon diaphragm pressure sensor, which had four piezo-resistors implanted into the top surface. The resistors on the diaphragm formed a Wheatstone bridge and the change in piezo-resistance, as a result of applied pressure and hence change in the stress field, was transformed into an electrical signal by proper post processing. The model was built from the design specifications of a commercially manufactured die. The results were compared to the production measurements and matched the data within one standard deviation. It was found that the level-0 package had an effect on the output signal. This work is believed to be the first to report an estimation of the distortion effect that a level-0 package has on a sensor signal with temperature.
Chapter 6 presents the model of the complete MEMS pressure sensor component encapsulated by EMC in a SOP. The EMC was treated as being elastic and temperature dependent. The method that was developed and calibrated in Chapter 5 was used as an indirect measure of the accuracy of the FEM model. It was evident that the package had a profound effect on the sensor signal. This was consistent with the actual measured data. The match of the signal data was not satisfactory. The signal values for two of the four service temperatures lay outside 3 standard deviations of the experimentally measured results. The estimated sensitivity of the die also fell outside 3 standard deviations for three of the four service temperatures.
A special vector plot was developed to understand how the pressure, or packaging stress, from the EMC effected the signal and sensitivity of the sensor die. The numerical simulations were done assuming a stress free temperature of 175oC, the moulding temperature. The packaging stress was found to increase with decreasing temperature. This was the effect of the subsequent increase in ΔT as the service temperature decreased.
The signal at zero pressure was found to shift as a function of temperature. This was caused by the packaging stress and a corresponding stress-field-shift on the diaphragm. The origin for this shift was an uneven packaging stress between the front and the back side of the sensor die. At -7oC, the pressure on the front and the back was 30 and 20MPa respectively. This caused an uneven bending moment on the membrane long sides and resulted in a shift in the stress field.
Chapter 7 elaborated the model one step further by treating the EMC as a viscoelastic material. The result of using the viscoelastic material model showed a reduction in the packaging stress due to stress relaxation. Viscoelastic materials are temperature and strain-history dependent. It was therefore necessary to run the model through the same processes posed by the manufacturing of the MEMS and SOPs. These included a set of thermocycles between -40oC and 125oC before the signals as a function of temperature and pressure were taken. The thermocycles were found to have a positive effect on signal shifting. Less signal distortion was seen with more cycles. The estimated and measured signal- vs. temperature-values matched within two standard deviations. The estimated sensitivities did not match the experimental measurements any better than those obtained for the elastic case. It was also found that sensitivity was nearly independent on packaging stress, but significantly dependent on pressure loading conditions.
The use of the viscoelastic model gave an improvement in simulated signal accuracy over the elastic model. It became clear that the EMC had to be treated as a viscoelastic material.
Chapter 8 concerned the change in material properties of the EMC and the impact this had on the FEM results. It was found that the behaviour of the MEMS pressure sensor was greatly affected by such changes.
Chapter 9 present the concluding remarks of this study.