RELIABILITY ISSUES IN RF-MEMS SWITCHES SUBMITTED TO CYCLING AND ESD TEST

Transcription

1 RELIABILITY ISSUES IN RF-MEMS SWITCHES SUBMITTED TO CYCLING AND ESD TEST A. Tazzoli, V. Peretti, R. Gaddi, A. Gnudi, E. Zanoni, G. Meneghesso DEI, University of Padova, Via Gradenigo 6/B, 5 Padova, Italy, tel , fax ARCES-DEIS, University of Bologna, Viale Risorgimento, 6 Bologna, Italy, tel ABSTRACT RF-MEMS switches have potential prerogatives better than traditional solid state devices, but the presence of mechanical movement introduces new classes of reliability issues that are not found in traditional devices. In this work we have carried out an extensive electrical characterization in order to identify the dynamic response of RF-MEMS switches driven in different conditions of voltage and actuation time. Furthermore, probably due to their recent introduction to the market, the robustness of MEMS submitted to ESD stresses has been also poorly investigated. We have studied, for the first time to our knowledge, the effects of TLP-ESD events on RF-MEMS switches identifying a very critical ESD sensitivity. INTRODUCTION Micro Electro Mechanical Systems (MEMS) devices for radiofrequency and microwave applications, already acclaimed in the past decade as one of the most promising emerging technologies, have recently received further attention for their ability to implement reconfigurable passive networks for future generation multiple standards and multiple frequency wireless terminals []. These devices have potential performances that can surpass the limits of their current equivalent implementations using more traditional solidstate technologies. In fact, all RF-MEMS devices (not only RF switches) maintain good miniaturization and they can be integrated with solid state circuits, either above IC or in the same package. These kinds of devices exhibit almost zero power consumption, extremely good linearity, both in terms of IIP and IIP (both > 7dBm), and very low losses (high Q), making them very suitable for tuning. Concerning RF-MEMS switches peculiarities, they achieve very low insertion loss (<.db up to GHz) while maintaining high isolation (>db). From the technology standpoint, they tend to require low-cost fabrication processes compared to RF and microwave solid-state integrated circuits. Despite these positive aspects, all the above and further benefits go along with a series of shortcomings, mostly related to the poor maturity of still evolving design methodologies and fabrication processes. The paradigm shift that these new technologies have introduced is also accompanied by a lack of standardization in the fabrication process, with a big impact on the design flow, limited reliability data compared to microelectronics standards, and poor knowledge of ageing mechanisms and reliable design practices. In terms of future applicability of these technologies to high volume, low cost-per-unit applications, such as the wireless telecommunications market, there are still many applications to be realized. Conventional solutions are currently reaching levels of cost reductions through batch processing that could hardly be tackled by emerging technologies. The RF-MEMS technology does not only need to prove as performing and reliable as its established competitors, but it must provide for completely new functionalities and unexplored circuits and systems solutions, to gradually find its way into the commercial RF microwave market. The presence of mechanical contact introduces a whole new class of reliability issues related to both mechanical and electrical phenomena [,]. Cycled mechanical deformations and steady-state vibrations introduce new stress mechanisms on the structural parts of these devices. Mechanical relaxation of residual material stress, plastic deformations under large signal regime, creep formations and fatigue can all impair the stability of electro-mechanical device behavior and eventually cause device mechanical failure. Finally, other surface effects such as oxidation or absorption can also result in changes of effective mass or stress of a moving or vibrating structure, causing stability issues and device failures. Probably the most important effects impairing device functionality is the stiction of the mechanical parts that reached contact, which is the inability to restore the device to its resting position after the actuation stimulus has been removed. Several factors have been known to cause stiction: capillary effects due to changed environment conditions, electrostatic charge accumulation or redistribution within dielectric layers, and micro-welding of metals due to DC or RF power. Electrical ohmic contacts occurring between two metallic surfaces can also suffer from stability problems due to cycling, resulting in changes in ohmic contact resistance. The causes can be diverse, such as surface contaminations, material transfer and erosion, and surface changes due to absorption or oxidation. This yet non-exhaustive collection of possible reliability and failure mechanisms occurring in RF-MEMS devices depicts a very complex scenario for lifetime testing. For this reason this new class of devices, needs the definition of new methodologies, device degradation models, and accelerated tests criteria, all covering different and cross-coupled physical domains. In this work we have carried out an extensive electrical characterization in order to identify the dynamic response of RF- MEMS switches [] driven in different conditions of bias and actuation time. We have found that an optimum actuation voltage must be chosen as a trade-off between good switch transmission and isolation properties and the need to avoid bouncing phenomena when the actuation voltage has been applied. Moreover, the degradation of shunt RF-MEMS strongly depends on the duty cycle of the actuation voltage waveform and so on the time that the switch spends in the actuated state. Another important aspect concerning the reliability of MEMS devices in general that has been poorly investigated is the sensitivity to the Electro Static Discharge (ESD) or Electrical Over Stress (EOS). Using a Transmission Line Pulser - Time Domain Reflectometer (TLP-TDR) on wafer system we have stressed RF- MEMS switches in different conditions with interesting results. We have found that TLP-ESD pulses around 5V that corresponds to a very low HBM event applied to the actuation pins produce potentially dangerous high currents exposing issues related to the device electrode layout. Moreover, the TLP-ESD pulses of around 5V can easily lead to device destruction indicating the necessity of a detailed study of the ESD sensitivity of these RF-MEMS switches /6/$. 6 IEEE IEEE 6CH778 th Annual International Reliability Physics Symposium, San Jose, 6

2 6 Actuation Delay [ms] Actuation Voltage [V] FIGURE. LAYOUT AND CROSS-SECTION OF AN INTERDIGITATED RF-MEMS SHUNT SWITCH. TECHNOLOGY AND DEVICE DESCRIPTION The technology utilized for the fabrication of ohmic RF-MEMS switches uses a surface micromachining process based on electrodeposited suspended gold for the membrane layer. The same gold layer is also available as low-loss RF signal path, while a high resistivity poly layer is used for actuation electrodes and a further Al- Ti-TiN multilayer for RF-signal underpasses. A detailed process description is given in [5]. A shunt switch configuration has been the focus of this work, as described by the layout and cross-section of Figure. An interdigitated topology is adopted for signal underpass and actuation electrodes []. Input and output RF ports are physically connected by the signal fingers and the switch is therefore normally close. A thin evaporated gold layer is deposited on top of the signal underpass electrodes, to achieve a gold-gold ohmic contact upon actuation. Direct contact is allowed only between the plate and the signal electrodes, by raising the signal underpass metal above the level of the poly electrodes, through the placement of poly dummy rectangular bricks. Therefore the actuated grounded plate gets in contact with the signal underpass, creating a low resistance path to FIGURE. ACTUATION DELAY AS FUNCTION OF THE ACTUATION VOLTAGE. ground that blocks the RF signal. A 5µm thick electroplated gold layer is used for the plate to improve its rigidity, while a thinner (.5µm) gold membrane implements the four suspending beam springs. In order to achieve lower actuation voltages, the straight beam structures have also been substituted with meander-shaped springs. The electrostatic pull-in voltage is typically around V for the straight beams design, reaching below V for the meanderbased devices. The maximum width of the suspended structures is µm, due to requirements for the release etch step. This leads to the perforated plate structure with x µm holes with µm separation. This characteristic, typical for surface micro machined devices, is a key factor in defining viscous damping phenomena for the dynamic behavior of the device operating in non-vacuum conditions. ELECTRICAL CHARACTERIZATION The stability and reliability of direct metal-metal ohmic contacts during operation can be hampered by several degrading mechanisms, such as micro-welding, electro-migration, metal softening, erosion and material transfer, and surface contaminations. Both the number S [db] V Actuation -5 V Actuation Time [s] FIGURE. SWITCH TIME RESPONSE OF A RESISTIVE SHUNT MEMS FOR TWO DIFFERENT ACTUATION VOLTAGES. FIGURE. S SCATTERING PARAMETER OF A RESISTIVE SHUNT MEMS FOR TWO DIFFERENT ACTUATION VOLTAGES: 5V AND V.

3 5. Voltage Actuation [V] Actuation signal Out signal -.E-.E- 8.E-.E-.8E V Output [V] S [db] Duty cycle 5% (5us) Duty cycle 5% (5us) Duty cycle 75% (75us) Dc = 75% Damaged Device Dc = 5% Dc = 5% E+ E+ E+ E+ E+ E+ E+6 Time [sec] Number of cycles FIGURE 5. UNEXPECTED ACTUATION BOUNCING OBSERVED WHEN A V ACTUATION VOLTAGE IS APPLIED. of cycles and the total time the switch spends in the actuated state are critical for these degrading mechanisms. In order to characterize the electrical behavior of MEMS switches we have set-up an ad-hoc measurement system. We used the internal RF signal generator of a 6 GHz Vector Network Analyzer, a GHz oscilloscope, a solid state pulser, a self made voltage amplifier, and an envelope detector. Figures and show how the actuation delay time of resistive shunt MEMS decreases on increasing the actuation voltage. Furthermore larger actuation voltages improve the S (not shown) and S (see Fig. ) scattering parameter. However, large actuation voltages produces unexpected bouncing as shown in Fig. 5. This clearly indicates a trade-off procedure that need to be carried out in order to have the optimum actuation voltage required for adequate stability and reliability. In order to characterize the changes of the scattering parameters of these kind of devices subjected to many thousand of actuation cycles a semi-automatic measurement system has been developed. The system consists of a Vector Network Analyzer, an arbitrary waveform generator and a self-made voltage amplifier in order to bias the actuation pad of the switch with a user-selectable shape, period, duty cycle, and voltage waveform. All the instrumentation is remotely controlled by a program written in LabView. The flexibility FIGURE 7. S SCATTERING PARAMETER EVOLUTION IN A RESISTIVE SHUNT MEMS DURING CYCLING STRESS WITH AN ACTUATION VOLTAGE DUTY CYCLES OF 5% - 5% AND 75%. of the system makes it easy to program the number of actuation cycles before a successive measurement of the scattering parameters. In order to correctly monitor the degradation of the switch under test and to reduce the measurement time we choose a logarithmic-step stress. During the measurement of the scattering parameters the switch is actuated with a square pulse for ms and biased with a GHz RF signal. As previously stated, the time that a switch spends actuated plays a big role in the degradation rate of the device. Figures 6 and 7 show the evolution of the isolation (S ) and transmission (S ) parameters of ohmic shunt switches stressed with pulses of different duty cycles (5%, 5% and 75% with a period of ms). All examined devices exhibit a decrease in performance after some thousand cycles, accompanied by a heavy degrade of the S and S parameters after million of cycles. The device stressed with 5% duty cycle fails to actuate several times already after some thousand cycles, but never fails completely. On the other hand, the device stressed with a duty cycle of 75% completely fails after just cycles, a very poor result if these switches have the ambition of substitute traditional solid state switches. Dc = 5% 6 5 Failure S [db] Duty cycle 5% (5 us) Duty cycle 5% (5 us) Duty cycle 75% (75 us) Dc = 75% Damaged Device Dc = 5% Idut [A] ESD-TLP RF-IN RF-OUT - E+ E+ E+ E+ E+ E+ E+6 Number of Cycles Not Actuated FIGURE 6. S SCATTERING PARAMETER EVOLUTION IN A RESISTIVE SHUNT MEMS DURING CYCLING STRESS WITH ACTUATION VOLTAGE DUTY CYCLES OF 5% - 5% AND 75% Vdut [V] FIGURE 8. TLP I-V CHARACTERISTICS A RESISTIVE SHUNT MEMS TESTED BETWEEN THE (ELECTRICALLY CONNECTED) IN- AND OUT-RF PAD. AFTER THE FAILURE POINT AN OPEN CIRCUIT IS OBSERVED.

4 5 5 Dielectric Breakdown 5 Ideal Open Circuit Idut [A] RF GND RF-IN RF-OUT ESD-TLP ACTUATION Voltage pulse [V] 5 Measured Ideal Short Circuit Vdut [V] FIGURE 9. TLP I-V CHARACTERISTIC OF A RESISTIVE SHUNT MEMS TESTED BETWEEN THE (ELECTRICALLY NOT CONNECTED) ACTUATION AND RF GROUND PLATE PADS. DIELECTRIC BREAKDOWN IS OBSERVED AT RELATIVELY LOW ( V) DISCHARGE. ELECTROSTATIC DISCHARGE EOS/ESD phenomena are universally recognized as a big reliability issue for all kinds of devices and, especially in the silicon IC world, ad hoc protection structures have been developed in order to prevent any failure or degradation of the electrical characteristics of the stressed devices. EOS/ESD stress can cause structural damage also to MEMS switches, impairing either the mechanical functionality of the devices or electrical characteristics such as scattering parameters. Some works in literature have shown that the ESD sensitivity of many MEMS components is a real issue and usually only V of HBM test is sufficient to produce failure [6,7]. On the contrary of traditional microelectronic circuits, stand-alone MEMS structures do not have protection mechanism to avoid damage caused by electrical overstresses. We have tested RF-MEMS switches by means of a Transmission Current pulse [A] Ideal Short Circuit Measured Ideal Open Circuit Transition to low impedance Time [ns] FIGURE. TLP CURRENT EVOLUTION IN A RESISTIVE SHUNT MEMS DURING THE TDR-TLP PULSE AT THE DIELECTRIC BREAKDOWN POINT SHOWN IN FIG. 9. DASHED AND DOTTED LINES REPRESENT THE EXPECTED BEHAVIOR IN AN IDEAL OPEN AND SHORT CIRCUIT ELEMENT Transition to low impedance Time [ns] FIGURE. TLP VOLTAGE EVOLUTION IN A RESISTIVE SHUNT MEMS DURING THE TDR-TLP PULSE AT THE DIELECTRIC BREAKDOWN POINT SHOWN IN FIG. 9. DASHED AND DOTTED LINES REPRESENT THE EXPECTED BEHAVIOR IN AN IDEAL OPEN AND SHORT CIRCUIT ELEMENT. Line Pulse (TLP) technique that is a widely used testing method which allows not only a device standard characterization, but also a detailed investigation of the device in Electrical-Over-Stress / Electro-Static-Discharge (EOS/ESD) regime [8]. The TLP adopted in this study works on the constant 5 Ohm impedance Time Domain Reflectometer (TDR) method [8] for wafer level testing, and it is able to produce a rectangular shaped pulse with a duration of ns and a rise time of some hundred of picoseconds. To study the susceptibility of RF-MEMS to ESD we have submitted shut switches to TLP stress in three different configurations. In the first case a TLP stress was applied between IN and OUT RF pads. In this configuration a not-actuated resistive shunt switch is like a short circuit between the IN and OUT pads. As it is possible to see in Figure 8 an equivalent resistance of around Ohms can be obtained up to a current level of about 5 A. This failure current level, obtained with a TLP system with a pulse of ns can be correlated to a good Human-Body-Model (HBM) robustness of around 7.5 kv [8]. Beyond 5 A, the equivalent resistance largely increases becoming an open circuit possibly due to high current induced metal opening. In the second configuration the TLP stress was applied between the actuation pad and the RF-ground, where no electrical connection is present. The actuation structure can be modeled as a capacitor so, as depicted in Fig., at low applied voltages the I-V plot is similar to an open circuit. When the applied voltage reaches around V, the device does not behave anymore as an open circuit, and a large current is measured through the device reaching the Dielectric Breakdown point, see Fig. 9. The voltage and the current waveform pulses during the TLP test at the Dielectric-Breakdown point of Fig. 9 is shown in Fig. and Fig.. In the highlighted regions the measured current and voltage waveforms clearly show a transition between an open-circuit regime to an almost short circuit regime (see, for comparison, the ideal open and short circuit curves in dashed line in Figures and ). During this TLP stress, a spark is observed (with an optical microscope, see Fig. ) outside the intrinsic MEMS. This suggests that the high current observed in this test is due to a dielectric discharge between metal lines and not due to the intrinsic MEMS. The dielectric breakdown does not correspond always to device destruction, however, if multiple TLP ESD stress are applied (after dielectric breakdown point) device destruction is always reached (after 5- pulses), possible due to metal fusion. We believe that the dielectric breakdown point leads to

5 5 RF GND RF-IN RF-OUT Idut [A] ESD-TLP ACTUATION NO FAILURE Vdut [V] FIGURE. TLP I-V CHARACTERISTIC OF A RESISTIVE SHUNT MEMS TESTED BETWEEN THE (ELECTRICALLY NOT CONNECTED) ACTUATION AND RF-IN PADS. DIELECTRIC BREAKDOWN IS NOT OBSERVED IN THIS CONFIGURATION UP TO 8 V TLP DISCHARGE. a partial degrade of metal lines. This can be confirmed by the third TLP stress. In this case the TLP stress was applied between the actuation pad and the RF-IN/OUT pad (open circuit) and no Dielectric-Breakdown is observed up to V of applied TLP stress, see Fig.. The difference between these two last cases is probably due to a different metal-line configuration of these electrodes. A further confirmation that the discharge does not interest the intrinsic MEMS is given by the cycling test shown in Fig. where more than one million cycles applied and monitored in a virgin and in an ESD-TLP stressed RF-MEMS are shown. In order to make a comparison between virgins and ESD stressed switches we stopped the TLP system after that the device reached for the first time the Dielectric Breakdown point. Then the MEMS was electrically stressed for one million of cycles in order to see if the ESD stress caused some changes in the scattering parameters. Clearly, as depicted in Fig., in the RF-MEMS subjected to a NON destructive TLP test the isolation scattering parameter (S ) not only does not degrade, but on the contrary, for some unknown reasons, shows an improvement in the switch insertion loss (S decreases) and present a lower degradation rate (lower increase of the insertion loss S parameter during cycling). This aspect requires S [db] ESD-TLP stressed Fresh Device E+ E+ E+ E+ E+ E+ E+6 Number of cycles FIGURE. S SCATTERING PARAMETER EVOLUTION DURING CYCLING STRESS IN RESISTIVE SHUNT MEMS. CLOSE SYMBOLS: FRESH DEVICE. OPEN SYMBOLS: DEVICE AFTER ESD-TLP STRESS UP TO THE DIELECTRIC BREAKDOWN POINT (SOFT-FAILURE) SHOWN IN FIG. 9. NO BOUNCING AND LOWER DEGRADATION RATE IS OBSERVED IN THE AGED DEVICE. FIGURE. TYPICAL DAMAGE OF AN OHMIC SWITCH SUBMITTED TO TLP STRESS BETWEEN THE ACTUATION PAD AND THE GROUND. THE ACTUATION LINE IN THE HIGHLIGHTED REGION APPEARS BURNT (WHERE A SPARK CAN BE SEEN DURING THE TLP STRESS). THE REITERATION OF THE TLP STRESS BRINGS TO THE DESTRUCTION OF THE DEVICE. further investigation. However, if several TLP stresses beyond V are applied from actuation pad and the RF-ground (TLP stress type ii), a catastrophic degradation is observed due to an open circuit happening in the spark region suggesting that better layout design rules are required in order to reach reasonable ESD robustness. Similar results have been found on ohmic resistive switches and on capacitive shunt switches. CONCLUSIONS The reliability of RF-MEMS switches has been deeply investigated obtaining important information on the optimized biasing conditions (actuation voltages and duty cycle) and on the necessity of adequate design rules in order to reach reasonable robustness of these devices against Electrical Overstress and Electrostatic Discharge. We have found that large actuation voltages improve RF performances, but, at the same time, can produce unexpected bouncing. Furthermore, devices stressed with a duty cycle of 75% completely fail after just cycles. This is clearly a very poor result. Concerning the ESD sensitivity, this kind of devices are highly exposed to failures, at very low HBM event too. Summarizing the stresses carried out in three different configurations, we have found the following results: TLP stresses applied between IN and OUT RF pads (i) make devices fail at a current level of about 5 A, equivalent to a HBM robustness of around 7.5 kv. If TLP stresses are applied between the actuation pad and the RF-ground (ii), devices do not behave anymore as an open circuit when the applied voltage reaches around V. In this case a dielectric discharge occurs between metal lines but it does not interest the intrinsic MEMS. In the last case, where stresses are applied between the actuation pad and the RF-IN/OUT pad (iii), no failure is observed up to V, probably because the different layout design of involved metal lines. The electrical cycling characterization of devices submitted to ESD stresses has taken to strange results. We have found in fact that scattering parameters not only do not degrade, but usually they show an improvement in the switch insertion loss and present a lower degradation rate. To conclude, a detailed study of the ESD sensitivity of RF-MEMS switches is needed.

6 ACKNOWLEDGMENTS The authors would like to thank Karl Süss for providing a PM8 probe station in order to build the TLP-TDR on wafer system. This work was partially supported by the PRIN 5 entitled Reliability of RF MEMS for high frequency applications. REFERENCES [] G. Rebeiz in 5 MTT-S Int. Microwave Symp. 5. [] J. Maciel, in Proc. of the 5 IEEE MTT-S Int. Microwave Symp. 5. [] J.L. Zunino III, et Al.,Proc. 5 NSTI Nanotechnology Conf. and Trade Show (Nanotech), Anaheim, CA, May 8-, Vol., 5, pp [] R. Gaddi et. Al, NSTI Nanotech, Boston, March, vol., pp.7-. [5] F. Giacomozzi, et Al., Electromechanical Aspects in the Optimization of the Transmission Characteristics of Series Ohmic RF-Switches, in Proceedings of the 5 th MEMSWAVE Workshop, Uppsala, Sweden, June - July, pp. C5- C8. [6] J. A. Walraven et al, Human Body Model, Machine Model and Charged Device Model ESD Testing of Surface Micromachined Microelectromechanical Systems MEMS, EOS/ESD Symp., pp [7] J. A. Walraven, Failure Mechanism in MEMS, ITC, pp [8] A. Amerasekera, C. Duvvury, nd Ed., John WILEY and Sons,, pp