An accelerated thermal cycling test for RF-MEMS switches
- First Online:
- Cite this article as:
- Mulloni, V., Sordo, G. & Margesin, B. Microsyst Technol (2016) 22: 1585. doi:10.1007/s00542-015-2780-4
Thermal cycling tests are an important part of the standard space qualification procedure for RF-MEMS devices. Standardized tests are rather demanding in terms of equipment, sample preparation and experimental time. In this paper, we present a fast thermal cycling test aimed at obtaining a rapid selection among different switch geometries in terms of thermal cycling resistance. Seven different switch typologies are examined and tested, evidencing that the most important source of deterioration is the mechanical deformation of the movable membrane due to changes in internal stresses. For this reason, the clamping typology is the most important design parameter to take into account in order to achieve a good thermal cycling resistance. Contrarily to expectations, the cantilever structure is not the most promising switch typology, because while the main part of the movable membrane is stress free, the anchoring part of the structure is on the contrary heavily stressed and modifies with the temperature change. The most promising structure is a four-anchored switch with a flexural clamping structure that allows the thermal expansions in the horizontal direction, while the data for all the clamped–clamped structures investigated are similar and their differences can be interpreted in terms of different distribution of the thermal strain over the whole membrane.
Because of their outstanding RF performances such as high isolation, low insertion loss, low power consumption and high linearity, RF-MEMS switches have the potentiality for a very broad exploitation. In satellite networks application these switches have additional advantages, being also lightweight, of low volume, and very resistant to high acceleration because of their very small mass. On the other hand, reliability issues are the principal barrier to the wide utilization RF-MEMS components and, in view of possible space applications, switch reliability performances are rather demanding because they include additional mechanical, thermal and long-term functioning performances with respect to standard applications.
In the last few years, several switch configurations have been modeled (Rebeiz and Muldavin 2001; Senturia 2001), fabricated and tested in order to realize mechanically stable, low-loss, negligibly charged devices potentially suitable for satellite applications (Daneshmand and Mansour 2007; O’Mahony et al. 2014). The vast literature on the topic of MEMS switch reliability reports several studies on switch failure or performance degradation, including important factors such as maximum number of cycles, charging (Van Spengen 2012) ambient contamination (Blondy et al. 2007), proper packaging, contact degradation and long-term-stress functionality (Mulloni et al. 2014a). Only recently, mechanical issues have been considered in greater detail (Mulloni et al. 2014a; Ryan et al. 2014), being recognized as one of the most important source of failure in RF-MEMS. For space application purposes, the mechanical stability to thermal cycling is a very important reliability parameter, because in the typical space conditions thermal changes are quite relevant and rapid. For this reason thermal cycling testing is one of the characteristic normally requested in space qualification procedures (Lucibello et al. 2015).
Thermal cycling degradation depends on both switch design and structural material properties. The aim of this work is to address the thermal cycling performances of RF-MEMS switches as a parameter depending on the specific design of the switch. To this respect, the qualification procedures (Lucibello et al. 2015) employed for thermal cycling testing are rather complex and extremely time consuming. It is then very useful to employ a preliminary and fast test to discriminate among different design strategies and to recognize the switch features that are more influenced by the thermal cycling effects on the specific switch. In this way it is possible to select the most resistant typologies, and also outline the criteria to design new and more thermal cycling resisting devices.
2 Switch design and fabrication
In order to test the mechanical stability to rapid temperature variations, different mechanical structures were studied, including one cantilever switch, four clamped–clamped structures (two shunt capacitive and two ohmic series) and two circular, four-anchored switches. This selection covers all the most common switch geometries.
3 Thermal cycling test
The standard space qualification procedure includes, as already mentioned, thermal cycling tests. In particular, MIL-STD-883H standard requires the samples to undergo to 500 thermal cycles of 10 min each, with the temperature varying between −55 °C and +125 °C. This is a rather long procedure, which necessitates of a fast and dedicate climatic chamber, with a large temperature interval and a total duration time of the test of several days (Lucibello et al. 2015). Sometimes it is also useful to measure the device functionality after a fraction of the total number of cycles, and this increases the total time required for the experiment. Finally, the test must be performed on packaged devices, and the only package typology already qualified for space is a dedicated ceramic LTCC package, which is extremely costly. An important disadvantage of the measurement in package is also the impossibility to operate a post-failure analysis of the device, in order to detect the reason for failure or performances deviations.
It is clear that the implementation of this test is not easy in common research laboratories, but requires a highly dedicated equipment, sample preparation and a relevant number of packaged samples in order to obtain some meaningful statistic. It is then very useful to have a faster and simpler thermal cycling test, not to ensure qualification standards but to highlight the main factors influencing thermal cycling degradation and to allow a selection of the most promising switch designs. In this way the qualification effort can be limited to the geometries that have more chances to pass the test and hopefully the preliminary results may provide some guidelines to design more thermal cycling resistant switch typologies.
The criteria employed for the definition of the rapid thermal cycling are the following. First of all the number of cycles must be strongly reduced, being the most time consuming aspect of the standard test. We tentatively set the number of cycles at ten, because this is a number sufficiently manageable in a normal research laboratory. The temperature interval was then restricted to the range above 25 °C. The reason for this choice is the assumption that the damage induced on the device by the thermal cycling is due to the mechanical deformations occurring in the movable membrane because of the variation of its internal stress. These deformations happen mainly because the internal stress changes from tensile to compressive when raising the temperature due to the thermal expansion mismatch between the gold membrane and the substrate (silicon) (Mulloni et al. 2010). When the stress becomes compressive enough, the membrane buckles in order to minimize its stress, deforming its shape. For our devices, this is known to happen around 85 °C (Mulloni et al. 2013a), but a further increase in temperature makes the deformation more pronounced. If the deformation is large enough, the original shape is then not perfectly restored during the cooling phase and the repetition of this process results in a distorted shape even at room temperature. The extent of this deformation is however heavily dependent on the design and especially on the clamping geometry which sets the constraints for the thermal mismatch.
Limiting the temperature range to 25–155 °C reduces the oven specification, since no external cooling is required during cycling, but is not expected to change the result of the test. It should however be noted that in principle at very low temperature the stress of the movable membrane could become extremely tensile, causing cracks in the membrane. This has however been verified not to happen for the type of structures investigated in this paper (Lucibello et al. 2015).
The time duration of the single cycle has however be kept the same as in the standard test, that is 10 min for cycle, because it is well known that the evolution of internal stresses in metals depends of the speed of the temperature change. This posed the problem of the oven choice, because the typical laboratory oven is too slow for this test. A good alternative choice was found in a hot plater set at the maximum temperature close to another one at ambient temperature, with the samples moved from one to the other every 5 min.
Optical profiler analysis performed before and after the thermal cycles allows a precise evaluation of the mechanical deformations evidencing important quantitative differences among the different switch structures studied, as will be shown in the following. As a tentative thermal cycling test we measured some specimen of the switches investigated before and after a thermal treatment of ten cycles of 10 min each in the range 25–155 °C.
This preliminary test evidenced that the strong reduction in the number of cycles needed to be compensated in some way, because the mechanical modifications before and after cycling were barely detectable. For this reason we decided to repeat the test changing the temperature range to 25–200 °C, exploiting the higher temperature as an accelerating factor for mechanical changes, and maintaining also a temperature interval and increasing/decreasing rate close to that of the standard test. In this way it has been possible to study in great detail the mechanical variation due to thermal cycling for the single switch geometries presented above.
4 Finite element simulation
Out-of-plane deviations for the simulated switch membranes corresponding to the switches in Fig. 1
Max. def. 125 °C
Max. def. 175 °C
Max. def. 200 °C
Shape change II
4-anchored I (e)a
Twisting of anchors 0.50
Simulations results are of course dependent on the mechanical parameters of the membrane. In particular, residual stress at 25 °C and stress gradient are strongly process dependent and cannot be extracted from literature data. They must be the same for all the simulated structures, being all made with the same material and process. Cantilever structures are especially sensitive to stress gradient, and quite insensitive to residual stress (Rebeiz and Muldavin 2001). This allowed us to determine the value of stress gradient in the gold membrane with some precision simply reproducing the correct tilting upwards of the cantilever switch at 25 °C. Stress gradient was approximated simulating the thin gold layer as a bilayer, with the two layers having different values of residual stress. The stress difference was determined to be 24 MPa, with the upper layer more tensile than the bottom one. The average residual stress value was then determined setting this parameter in order to have a buckling temperature close to the experimental value of 85 °C (Mulloni et al. 2013a) for all structures. This turned out to be compatible with measured values for very similar structures (Mulloni et al. 2013b).
It can be noted from Table 1 that in general the temperature increase from 125 to 175 °C does not change the general shape but simply enhances the out-of-plane distortion, effectively acting as accelerating factor. However, this does not hold for all the switches. In particular, the clamped–clamped switch (b) changes its shape between 125 and 175 °C and again between 175 and 200 °C, revealing some general instability in the membrane shape, while the distortion of the four-anchored switch (f) is entirely on the thin anchors, which are twisted at 125 °C but bend markedly upwards at 175 and 200 °C.
Especially noteworthy is the deformation of the four-anchored switch (e), because is practically negligible at every temperature. This is an important result because this type of structure is expected to be extremely resistant to thermal cycling for all investigated maximum temperatures.
The cantilever structure is also supposed to be very resistant to thermal cycling, being its average stress virtually zero. In the simulation the vertical deviations for this structure are small but not totally negligible, with the upward tilting progressively reduced with increasing temperature. While the stress in the membrane is really zero, the same cannot be said of the anchor, which is on the contrary extremely stressed. Is the progressive stress change in this region with raising temperature that causes the membrane deflection.
5 Optical profiler analysis
Three-dimensional (3D) pictures of the investigated switches were taken before and after the thermal cycling test for all switch typologies by means of an optical profiler (Zygo New View 6300). In order to better detect the mechanical changes caused by the thermal treatment, the most powerful tool is not the direct comparison of the 3D-pictures before and after treatment, but the construction of a new image obtained by their point-to-point difference in the vertical direction. This method evidences the differences due to the thermal cycling and allows for a clear determination of the amount of vertical mechanical change and its exact localization on the membrane. The pixel-to-pixel subtraction of the two images is feasible using the profiler software, and the only operation that needs to be carried out with care is the perfect alignment of the two images before subtraction.
Summary of experimental maximum vertical deformation for the different switch typologies
Maximum deformation (µm)
Clamped–clamped Ohmic I
Clamped–clamped Ohmic II
The first differential images are those of the clamped–clamped capacitive switches (a) and (b), reported in Fig. 3 together with their profile difference along the horizontal symmetry line. The two switches have a very similar behavior, as predicted in the simulations. In both cases the deformation is more pronounced in the central and narrower part of the switch, where it is around 0.4 µm. This means that after thermal cycling the membrane is slightly deformed upwards, which again agrees with the simulation. It seems that in this case the thick gold reinforcement does not have much effect, probably because it is compensated by slightly longer anchors. This type of deformation is not expected to be fatal for the switch functioning, but it simply means that the actuation voltage could by a few volts higher. It may be a problem, however, if the increase of the actuation voltage is not compatible with the application requirements or if the switch itself tends to stick at high voltages (Mulloni et al. 2014b). In spite of the high value of out-of-plane deformation resulting from the simulations, the fraction of non-reversible deformation for these structures is quite low.
When we move to analyze the two ohmic clamped–clamped switches, we note, instead, a clear difference between the two typologies investigated. Switch (c) is clearly more deformed than the switch (d), which has a deformation in the central part very similar to the capacitive switches already discussed. The two ohmic structures are very similar and this result is somehow surprising. They differ, basically, only for the distribution of the reinforcing gold layer on the movable membrane (Mulloni et al. 2014b). The simulation results of the previous section offer a possible explanation, given by the double change of membrane shape of the switch (d) with increasing temperature. These shape changes contribute to reduce the thermal strain in the membrane, which is then lower and more distributed. The reinforcements on the membrane of this switch are only transversal, allowing the switch to bend in its central part, while in the switch (c) the reinforcement makes the structure very rigid in the central part of the membrane. This means that almost all the thermal strain is concentrated on the thinner parts close to the anchors, which experience a large distortion and are consequently more irreversibly deformed. It should however be noted that the actuation voltage of the structure (d) is a few volts higher than that of (c) at 25 °C (Mulloni et al. 2014b) This could imply that the second structure is more rigid and robust, partially explaining the result found.
The third comparison is between the two four-anchored switch structures, reported in Fig. 3. The difference in this case is enormous, because while the first one is deeply deformed after thermal cycling, the other is almost unaffected by the treatment. This is especially evident when looking at the profile comparison of the two switches, and is in total agreement with the simulated data. Since the switches differ only in the anchor form, which is straight in one case and more complex in the second, it is clear that the explanation for the different behavior must be found on the anchors (O’Mahony et al. 2014). It should be noted that in general, the larger the deformation during heating, the less reversible is the process during cooling. This implies that structures where the thermal strain is very large and localized are likely to be the more deformed during thermal cycling experiments. In this particular case, the vertical deformation is so reduced in the second switch because the thermal expansions of the gold is partially distributed laterally, while in the first case all the deformation is concentrated only in the vertical direction and on a very thin part (anchor) which cannot freely expand. The fraction of non-reversible deformation is in this case especially high, as can be seen comparing the data of Tables 1 and 2.
Finally, the cantilever switch variation is reported in Fig. 4. In first approximation, as already mentioned, the cantilever structure should be unaffected by the residual stress variation due to the change in temperature, because is free to expand during heating. For this reason the change after cycling is expected to be minimal. However, when we look at the experimental data, we found that the cantilever shape has remarkably changed. It should be noted that the cantilever structure is not flat at the end of the fabrication process, because of the development of stress gradient during the release phase (Mulloni et al. 2010).
The cycling lowered the bending of about 1 µm as shown in Fig. 4. This means that also the cantilever structure is sensitive to thermal cycling, at least when the upper temperature is increased up to 200 °C. In this case, however, the measured deflection is larger than the simulated value at 200 °C, making clear that for this structure more complex phenomena, not included in the static simulation, are taking place during thermal cycling. What it is likely to happen is a relaxation of the local stress in the anchoring zone induced by the rapid thermal changes.
This explanation is supported by the switch profile in Fig. 4, which shows that the change is localized in the anchor zone, the only non-reinforced part of the movable membrane. Other test performed on the same structure, with longer cycling times and very few cycles evidenced that the change in tilting angle is strongly dependent on the specific maximum temperature and cycling time, limiting the quantitative value of this accelerating test for the cantilever switch.
Several switch geometries have been investigated to test their resistance to thermal cycling, and a fast semi-quantitative test was defined in order to select the more promising structures and to understand the most important factors governing thermal cycling resistance. As a general rule, the mechanical modifications due to thermal cycling can be understood as related to the thermal induced stress variation in the switch mobile membrane. The thinner and narrower parts in general and more specifically the anchors, turned out to be the membrane parts where the irreversible thermal strain is most concentrated. From these considerations it follows that a thermal cycling resistant design implies a uniform distribution of the thermal strain in order to maximize the reversibility of the deformation phenomena, and, quite obviously, a mechanically robust structure. The robustness of the design can however be partially an inconvenience for other switch desired specifications, such as low actuation voltage and minimal charging.
Finally, a specific structure very resistant to thermal cycling has been selected. The specific design of the anchors of this structure allows for the thermal strain to be partially distributed laterally, where the anchors are free to expand, strongly reducing the deformation caused by the clamping constraints. In fact, this structure does not buckle, and this opens possibilities to exploit this design strategy also in other applications where thermal stability or wide temperature operating range are required.