Contractor: Stanford University
Agreement Number: F30602-96-2-0308
Project Number: HJ1500-3221-0599
DARPA Order Number: N-7-0504/AO E117/12
Contract Period: September 13, 1996 - August 12, 2000
Date: June 1, 1999
Sponsor: Department of the Air Force
Technical POC: Robert W. Dutton
Report Prepared by: Robert W. Dutton
Task 2
Task 3
Develop and demonstrate fully-integrated, FEM-based prototype solver capabilities to model behavior and fabrication process-dependency of MEMS devices. This will include capabilities to consider multi-physics and materials dependencies as well as other process induced factors--geometry effects due to deposition/etching. The overall tool integration strategy will be to develop and test key components that overcome limits currently seen to be "bottlenecks" in commercial systems. Application-specific lumped modeling will be developed and used to guide the overall direction of the CAD efforts, including improved parameter extraction schemes that cross-link device performance with layout parameters that can in turn support design (and hopefully optimization) of practical MEMS devices (i.e., RF switch).
Approach:
Progress:
Most of the work in the past quarter focused on consolidating, confirming and correcting observations and conclusions of the past two years. In particular, measurements were made on dies from the MUMPs 29 run. These dies do not have connections to gold to avoid the influence of gold on material properties. The calibration methodology was shown to be extremely good, with the simulated behavior of the canonical benchmark being within 3% of the measured behavior.
The Zygo interferometer was outfitted with micromanipulators to allow interferometric measurements under electrical probing. Measured deflection-voltage characteristics compared very well to simulations. Careful measurements of buckling amplitude reveal that probe pressure from the micromanipulators affect the stress state of the beams. The amount of pressure can be controlled and minimized by observing the interference fringes under the Zygo. Care must be exercised when probing under a regular optical microscope -- high probe pressures can change buckling amplitudes by about 0.1um.
The web site describing the canonical test problem, along with figures, data, input decks and publications is now up and running at http://www-tcad.stanford.edu/~chan/mems/canonical/index.htm . The latest measurements and calibrations will be included shortly.
As discussed in detail in the previous quarterly report, the most significant obstacles in realizing MEMS geometry from process simulation using the domain decomposition technique [1] is the need to generate more "solid modeling friendly" deposited surfaces and the reduction of unnecessarily refined surfaces produced by the pseudo-3d technique [2]. Progress has been made on multiple fronts in efforts to reduce the data complexity in the 2-D and 3-D regions of the domain decomposition.
Specifically, in the 2-D regions a new non-uniform sampling technique has shown great promise in reducing the number of faces in the resulting solid model. The algorithm is straight forward and attempts to remove as many points as possible in the resulting deposited profile while maintaining a user specified volume (e.g. 99% of the original volume). Using the volume of the resulting solid seems to be a reasonable metric for data reduction since the resulting mass is significant in mechanical and modal analysis.
In the pseudo 3-D regions, the code has been enhanced to allow both non-uniformly spaced cuts and the non-uniformly sampled 2-D simulated deposition profiles as described above. This has led to a significant reduction in computational time and a reduction in the number of faces in the resulting surface for the 3-D region.
Also in the area of geometry, initial results from a 2-D selective etch are promising. The effort leverages on a levelset code under developed for another research project. The geometry server developed so far has relied on geometric etches, but fundamentally the same concepts used to integrate the results of deposition simulations can be used to incorporate the results of etching simulations.
Difficulties/Problems:
Develop extraction schemes for materials properties using simplified MEMS test structures. In combination with the CAD tools efforts, the physical models (including microstructure) will be evaluated and modified as needed. The scope of material characterization efforts include both standard metal layers (possibly graded as well) and reliability issues such as plastic yield and fatigue. The extraction and validation steps for physical models is a key linking between the three sub-tasks.
Approach:
Progress: 1. Fatigue Study
Peak stress relaxation to a plateau stress has been observed during cyclic testing of Al microbeams. The amount of relaxation has been shown to be dependent on the strain rate during cycling and is, in all cases, smaller than the relaxation observed during monotonic testing. An experiment of cyclic loading followed by stress relaxation at constant strain has yielded additional relaxation. The plateau stress after cyclic loading has been identified as a metastable stress state.
A two-mechanism model with a fast and a slow relaxation time - corresponding to recoverable grain boundary sliding and dislocation motion - has been proposed to explain the observed relaxation behavior in Al thin film samples during monotonic and cyclic loading. Anelasticity has been identified as the dominant deformation behavior in such samples as small strains.
Iridium thin film samples have been tested under monotonic and cyclic testing conditions. Lenear-elastic behavior has been observed up to fracture. Unloading of iridium samples has shown no hysteresis of loading and unloading curves. Fracture stresses of up to 3.5 GPa and fracture strains of up to 0.75% have been measured. A Young's modulus of 490 GPa as compared to the literature value of 528 GPa (7.2% deviation) has been extracted. Cyclic loading experiments of iridium microbeam samples showed small relaxation of the peak stress at the initial strain (around 10%). This initial relaxation cannot be explained by anelasticity since reloading is not observed at the end of the test. The relaxation is more likely due to plastic deformation. No relaxation of the peak stress was observed for further cyclic testing on the same sample at increased strains.
2. In-situ Stress Measurement
At EG&G IC Sensors, we have explored the wafer-to-wafer bonding of the membrane-wafer to the tip-wafer, as well as the final dicing of the bonded set to form the stress measurement sensors. One set of the devices have been successfully bonded and diced.
Before bonding, we thoroughly cleaned the wafers at IC Sensors. Although we were allowed to align the wafers from bonding, a process engineer at IC Sensors bonded and diced them. To protect membranes from being damaged or contaminated by dicing fluid, we used blue-tape (commonly used for dicing single-stack wafers) on the back and UV tape on the front. The UV tape is very rigid and does not sag onto the membranes that are 0.5mm below the top surface. After dicing, we exposed the tape to UV light to release it from the top of the membrane-wafer. It is later peeled off from individual devices prior to packaging.
The first set of wafers did not bond well. The wafers delaminated soon after bonding. We attributed the cause to the fact that the bond-rings on the tip-wafers were too narrow. This was confirmed with optical microscopy. The tip-wafer bond-rings were originally designed to be narrower than those of the membrane-wafer to ensure minimum effective bonding area. However, during the patterning of the tip-wafer bond-rings, which comprise 100A/100A/8000A of Cr/Pd/Au, over-exposure and over-developing of the photoresist (mask for etching Au) caused narrowing of Au lines. Moreover, undercutting during the etching of Cr caused the Cr under the Au lines too narrow to promote adhesion. To address this problem, we stripped off the metals on the four remaining Tip-Wafers and re-patterned the bond-rings. We developed an improved lithography recipe to minimize over-exposure and over-developing of the masking photoresist, so that the bond-rings could remain as wide as possible at the end of the process. We also minimized the undercutting of Cr by strictly controlling its etching time. After the re-patterning, both the bonding and the subsequent dicing succeeded on the second set of wafers. However, many of the membranes ruptured after the final dicing. So far, we have not found the cause of the rupturing. The fabrication yield is about 79%.
Instead of bonding the remaining wafer sets right away, we decided to package and test some of the devices in this batch first. Transistor outline headers (12-lead) were chosen to package each individual die. The new packages have several advantages over the hybrid size-braze packages used last year. They are more reliable electronically and would be more suitable for mounting and wiring in the evaporation chamber that we plan to use.
Difficulties/Problems:
Design and implementation of MEMS structures, including materials parameter extraction, will be used to test the CAD and physical models based on MEMS devices of interest to DARPA. The experimental measurements made with these structures will facilitate the critical evaluation of models, physical parameters and overall simulation accuracy of CAD for MEMS devices. The test vehicles from this work also support canonical benchmarking of both new materials for MEMS and accuracy of Composite CAD in MEMS applications, specifically the RF switch.
Approach:
Progress:
Attempts over the last half year to develop single film membranes with advanced materials processing using Ir,W and Mo have demonstrated at least one very exciting result based on Ir. However, useful into a useful RF MEMS process integration has been frustrated by a range of film stress gradient and other process control issues during deposition. In situ, real-time stress measurement (see Task #2) is of great potential value to remedy these limitations and emphasis in this subtask has shifted towards support of process development needed to create this new diagnostic tool.
Given the process development problems in depositing layers of Ir, we were only able to achieve a useful films on a few runs. The results of which were described by Guido in his fatigue studies. In his work he showed that Al films undergo anelastic fatigue while the Ir films don't. This further indicates that Ir is indeed a stronger candidate for Rf MEMS applications than Al.
In contrast to the thwarted efforts to develop a single layer metal RF MEMS switch, processes that use multi-layered (or composite) films and sputtered silicon have shown ongoing progress and promise.
The materials used are thin encapsulating conductive films of Al or TiW layered both above and below mechanically stiff layers of PECVD silicon nitride or sputtered silicon, respectively. Using this approach we successfully demonstrated a MEMS device (continuously variable capacitor) that is identical in fabrication sequence to the RF MEMS Switch. This composite film approach was initially put aside as it was determined to be a much more difficult structure to model. Now, given the process difficulties outlined above and our success with the variable capacitor MEMS device we have returned to this approach to complete an RF MEMS switch. In fact two mask sets are currently in the works with processing beginning later this summer. These projects discussed below in the technology transfer section will be using composite layered film approaches for the RF MEMS switches.
The sputtered silicon process was demonstrated using both wet-related oxide sacrificial layers and dry-released polyimide sacrificial layers. Without cladding, the conductivity of the sputtered silicon was insufficient for high frequency work but is suitable for most physical sensors based on electrostatic measurement of a mechanical deformation. Cladding the sputtered silicon in TiW increased the conductivity by several orders of magnitude. The higher melting temperature and fracture strength of silicon relative to aluminum, as well as it's lower thermal expansion coefficient, make it less susceptible to warping during the plasma release. The Zygo interferometer was used to characterize the released curvatures. As a demonstration of CMOS compatibility, released structures were fabricated atop pre-fabricated CMOS wafers.
In the area of fatigue characterization and modeling, there is now even greater interest within Subgroup #3 and from the perspective of application drive in the need from such modeling. Namely, the use of Aluminum hinge layers is of critical importance in these new composite layered devices. Hence the detailed characterization and modeling efforts related to fatigue in Al films is directly relevant to ongoing developments of a composite material RF MEMS switch.
Al can still be used in a composite layered structure of an RF switch as long as it was strengthened by another stronger material. Or if it doesn't act as a mechanical layer at all, such as would be the case if one laid down a strong hinge of Silicon nitride or Ti/W and deposited a thick plate layer over this hinge and patterned it so that the deformable hinges weren't covered with Al.
The single layer approach (for the mechanical layer of an RF MEMS device) is still viable for a conductive film like iridium, due to it's relatively high conductivity, stiffness and apparent inability towards mechanical fatigue. Two process problems continue to be a challenge for Ir based MEMS devices. First the ability to minimize stress gradients in these layer are still going to need further development. Presumably through the use of insitu stress gradient measurement tool that allows for realtime operator control. Second improved lift-off patterning approaches that don't fail during the deposition process step or destroy the underlying circuit layers during the removal of excess metal.
During the dry Plasma release of Al over organic release layers (i.e. Polyimide) we have found a correlation between the amount of curling (stress gradients) and the surface temperature at the released metal film (actually the correlation is with more applied RF power which equates to higher surface temperatures at the released Al or metal layer). Due to Al very low melting point and stiffness compared to Ti or Ti/W, we expect to see little or no effect on the amount of curling with increasing surface temperatures. To test this theory we are going to perform a set of experiments where we release Ti and Ti/W layers at increasing RF power densities (thus increasing surface temperatures). We will then measure the results on the Zygo interferometer and compare to our results with Al. If this hypothesis proves correct then we will quickly have a working single layer process for RF MEMS using these two sputtered films.
Composite films with conductive layers (such as Al or Ti/W) encapsulating either PECVD SiNx or sputtered silicon support layer remains our most viable option for the near term, and in fact our current projects to fabricate RF MEMS switches will us these approaches until the single film processes mature.
Difficulties/Problems:
Technology Transfer:
An RF MEMS switch, designed by the Kovacs group collaboratively with AFRL, Hanscomb AFB, now includes a similar effort with Hughes Electronics/Raytheon research labs in Malibu CA. All layouts also include a subset of the Testarosa mechanical test structures which continue to be a boon for process control and characterization. These switches will be built onto Sapphire substrates and eventually integrated into our successful Silicon On Sapphire/CMOS (SOS-CMOS) process, which we have developed for high voltage circuits (up to 300V on chip from 5V power supply) and photovoltaic arrays that could potentially be used to drive the MEMS switches with only light supplied as power. This indicates that a completely integrated device using only light for input power could be used to operate these RF MEMS switches. We are currently working on further collaborations with them.
II. Financial Information: AF F30602-96-2-0308
| Total Amount Funded | $2,378,700 | |
| Salaries and Wages | $640,836 | |
| Staff Benefits | 100,303 | |
| Travel | 33,887 | |
| Expendable Mat./Services | 197,271 | |
| Repairs and Maintenance | 19,843 | |
| -------- | ||
| Total Direct Costs | $992,140 | |
| Capital Equipment | 128,593 | |
| Student Aid (Tuition) | 110,687 | |
| Indirect Costs | 541,994 | |
| Total Expenses to date | $1,773,414 | |
| ---------- | ||
| Funds to date, 3/31/99 | $ 605,286 | |