Issue



Emulation-based virtual manufacturing techniques for integrated MEMS


12/01/2008







Maurice J.A. Delafosse, Dalsa Corp.
Gerold Schröpfer, Coventor

Integration of MEMS devices and CMOS electronics on one die can bring powerful capabilities, including reduced system power consumption and improved volume-to-area ratio to automotive, biomedical, RF, photonics, information technology, and other applications. Because MEMS devices require interaction with their environments beyond just electrical signals, the development of fabrication and packaging processes is difficult. Mainstream CMOS IC processes are often insufficient for integrated MEMS/CMOS systems.

To surmount this challenge, developers of integrated MEMS systems often use simulation software to model the process. Simulation is extremely useful at certain points, such as once the full process has been tested and verified, or when a particular step requires improvement, or when it is the first step on a virgin wafer. In general, however, precise simulation of each individual process step is less important as a good, close-approximation overview of the complete integrated process.

To gain that overview, emulation software combined with silicon-accurate 3D visualization capabilities and backed up by experimental calibration is a more productive approach. This is because emulation software enables developers to use a “virtual manufacturing” concept to make process development for integrated MEMS systems more efficient. By taking into account all of the specific information for each process step, the technique efficiently flows that information through the entire process, and also from department to department within the organization. Engineers can know all pertinent information about what has been done previously so they can anticipate what needs to be done for the present fabrication step and for subsequent steps, and so they can react quickly to uncertainties and issues that crop up.

Dalsa Corp. and Coventor Inc. have teamed on several projects that have employed virtual manufacturing methods to develop, produce, and package versatile integrated MEMS devices. Dalsa’s core competencies are in specialized integrated circuit and electronics technology, software, and highly engineered semiconductor wafer processing. Products and services include image sensor components (CCD and CMOS); electronic digital cameras; vision processors; image processing software; and semiconductor wafer foundry services for use in MEMS, high-voltage semiconductors, image sensors, and mixed-signal CMOS chips. Coventor, meanwhile, is the world leader in design automation for micro- and nanoscale devices and systems. Coventor’s MEMulator software uses volumetric pixels technology (“voxels”) to build highly realistic 3D models that show the precise impacts of process changes. These are realistic virtual prototypes, not simply idealized geometries, as can be seen in Figure 1.


Figure 1. MEMulator builds highly realistic 3D models. (Design by U. of Waterloo, Ontario, Canada)
Click here to enlarge image

The three following examples show how a virtual manufacturing approach can be of benefit.

Support for process integration/process flow development

Process engineers often know what to expect from a specific process step, such as deposition or etching, and thus can predict its geometrical effects. But this gets tricky when brand-new process flows are created from novel combinations of existing process steps, or when a specific physical design layout needs to be rendered. One example is when multiple etches are performed sequentially, in several mask layers, where each etch strongly depends on the outcome of the previous one. In these cases, even for experienced process engineers, it can be difficult to estimate the outcome without a proper representation of the starting point.

In this case, virtual manufacturing comes in handy–it allows engineers to readily develop and test a large number of process step combinations and novel process flows using multiple physical design patterns, which wouldn’t be possible in a real manufacturing environment or would be costly to perform. The model for each process step can be calibrated through experimental measurements, resulting in very reliable result-models that can be reused in multiple combinations. Emulation software takes these verified models and creates a complete process flow, and variations of it with various physical design layers, in a few hours or less. All of the software code is saved and is used to speed up process development.

An example of the processes Dalsa emulated before building silicon was a fabrication process for an integrated microphone. Dalsa already had emulation-scripts samples for the desired process steps, while the preliminary product’s layout was supplied by the customer. The product required a big cavity in the sensing area. Dalsa used Coventor’s MEMulator emulation software not only to evaluate the impact of various etch methods on the final die size, but also to show the impact of each choice on the previous and following process steps. This was important because there were material stacks needing to be exposed to chemicals in different ways (i.e., dry or liquid etch) and at different etch-rates. Each option could have been modeled by hand by a couple of engineers at a table, but the emulation software provided a formal, reusable way to do the job. In addition, the emulations could have been conducted by a non-engineer since the emulation-scripts samples already existed.

Validate and prepare a design before fabrication

Emulation also gives engineers the ability to do virtual test runs to verify that a device design is compatible with the manufacturing process, and that the 3D result is as expected. Moreover, design mistakes and shortcomings can be identified, even if they are compatible with 2D layout rules. Other pre-production advantages include:

  • Ability to efficiently model any design changes,
  • Test-run multiple designs and test structures,
  • Improve and optimize device design for given process constraints,
  • Prepare a mask layout for fabrication,
  • Bridge the gap between lithography mask creation and process engineering, and
  • Optimize a 2D mask layout, taking into consideration 3D manufacturing aspects.

For example, one project required Dalsa to fabricate the mechanical portion of a one-die tire pressure monitoring system. The processing was complicated because the die was an assembly of three wafers, each wafer processed on both top and bottom, and all of them sequentially bonded one to the other as the overall process reached completion. There were a number of important process recipes of various types to queue: photolithography, oxidation, deposition, etching, implantation, bonding, and grinding. Although the result of each recipe was well-understood and well-controlled by the responsible engineers, they still needed to know what to expect as inputs in order to predict how the device would look when leaving their processing stage. Emulation software enabled that information to flow from one department to the other in a visual format. Corrections and adjustments were done in minutes. Screen captures for PowerPoint presentations enabled the sharing of information via different media with people who had different levels of technical understanding or different domain expertise.

The software enabled Dalsa to present the customer with a visual image of how its product would look after processing, with all the associated impacts of different processing choices clear to see. In fact, upon seeing such presentations, it happens from time to time that a customer will ask Dalsa to modify the proposed process or to modify the product design. Since the process hasn’t been built and launched at this point, both resource and financial cost of those moves is nil compared to what it would be during fabrication–or even worse, after initial devices have been built.

Inspect and control wafers during and after fabrication

An emulation-based virtual manufacturing approach also gives fab personnel a precise degree of control during and after the fabrication process. For example, it allows:

  • 3D visualization of each process step,
  • The ability to follow wafer processing step-by-step and to anticipate the next step,
  • Technical staff/equipment operators to use 3D virtual prototypes (or printouts) for quality control after each critical fabrication step, and
  • Comparison of manufacturing output with specified output, for final quality control and sign-off.

Click here to enlarge image

Emulation software also can be used for failure analysis in conjunction with analytical techniques (FIB, SEM, TEM, etc.), and for documentation and training of fab personnel.

Click here to enlarge image

As an example of this control, a key concern at Dalsa and elsewhere is potential contamination by “flying” structures–pieces of products that break off a previous wafer and “fly” around inside the production equipment–created by mismatches between the physical design layout and the fabrication process. Catching these potentially dangerous structures can be done by automated layout verification (i.e., design ruler check) but only after a complete 3D comprehension of the process and product has been achieved.

For example, on one Dalsa project to manufacture a resonator, a deliberate partial release of material was necessary. It was important to identify realistic release and residue patterns in order to calibrate the amount of released material, so as a first approximation Dalsa engineers performed a data-shrinking step using standard layout tools. However, they quickly resorted to emulation software because it gave them realistic release and residue patterns and enabled the creation of a model that took into account the impact of the preceding fabrication steps, all in about the same time as the data-shrinking. Once certain “forbidden” shapes of the released material were identified, a PowerPoint presentation with 2D and corresponding 3D images was created to transfer that information to the process engineering and inspection staff. Being in 3D, it was as if they had actual pieces of silicon to compare with the product. All of this was done in just a couple of hours using the base functions of MEMulator.

Dalsa has used emulation software to do much more than just create models that some might consider “more than average yet still simple.” For example, during fabrication of one device, metal residues were becoming trapped in the corners of the product, and the mechanism by which that happened wasn’t understood. Using emulation software, Dalsa engineers modeled the ripples created by the cycles of a DRIE etch on the edges of a polysilicon layer. With that information, new device shapes and different design rules could be evaluated without having to produce experimental silicon. The models were so realistic that when the MEMulator results in grayscale imagery was compared with SEM images, the differences weren’t obvious (see Figure 2).

Conclusions

Virtual manufacturing–process emulation backed up by experimental calibration–is a more productive way to build integrated MEMS/CMOS systems than process simulation. It provides a link between fab and design, is a lower-cost and faster technique, and provides a unique method to understand and improve design and process interaction, and wafer control/inspection. It leverages the enormous knowledge and tooling that has been developed in CMOS semiconductor fabrication, and transfers that knowledge to the MEMS world.


Figure 2 : Actual SEM images (top) vs. Coventor MEMulator/SEMulator 3D rendering (bottom) of DRIE etched polysilicon edges
Click here to enlarge image

null

Click here to enlarge image

Integrated MEMS/CMOS development and prototyping can be done at less risk using emulation tools such as MEMulator because new process flows based on existing semiconductor CMOS know-how can be created. Design and process information can be readily shared between fabless design houses and foundries, and the ultimate goal of faster time-to-product can be more easily achieved. Another part of the return-on-investment is the decrease of both development cost and time with less iteration, plus the increase of overall quality, reliability, and efficiency comforting current customers and maximizing the appeal for new customers.


Maurice J.A. Delafosse is MEMS product engineer, DPFS department, at Dalsa Corp. in Bromont, Canada.
Gerold Schröpfer
is director of European operations and foundry partner program for Coventor in Sarl Paris, France.