As the MEMS market forces continue to apply enormous pressures on device makers to reduce costs, companies are continually striving to make their development and manufacturing cycles more efficient. One approach has been to design around well-defined processes, or at least process modules, to accelerate time to market. We recently spoke with Alissa Fitzgerald of A.M. Fitzgerald & Associates and Peter Himes of Silex about their joint efforts in this area through a recently announced partnership. In this detailed interview, Alissa and Peter discuss the emerging trends with fabless MEMS manufacturing, provide an overview of existing commercial MEMS processes, address intellectual property and cost challenges, and explain their approach.
MEMS Journal: What are the main bottlenecks in the MEMS device development cycle? Please group these by category and briefly describe them.
Alissa Fitzgerald: My comments here are limited to fabless MEMS development. In contrast, designers at IDMs have the advantages of a captive fab and years of characterized process data.
Access to process information: very few foundries provide process-specific data to fabless designers. In some cases, the foundries have the data, or know it "in their fingertips", but can't or won't give it to the designers. In other cases, it's simply too expensive to do process measurements and materials characterizations. These types of studies are very expensive and not every customer would be willing to pay for that work. The data is also proprietary and so it is understandable that foundries may not be willing to release it.
Without process-specific data, when creating a new MEMS design, designers will guess (within a tolerance band) of what the foundry or a process recipe can produce. The designer's guesses do not always match reality, so when the processed wafers come back not matching the models, a lot of detective and iterative design work is needed to figure out what changed and then to converge the design and process. This leads to multiple design-fab-test cycles which are very time consuming and expensive.
Process qualification: every new process flow requires qualification by the foundry. The main problem is that many process steps, particularly etch and DRIE, are *very* sensitive to pattern geometry, pattern density and silicon loading (amount of bare silicon interacting with the etch plasma).
Unexpected process interactions happen all the time, and not all of them can be predicted. So the foundry must run many wafers through a new process to verify the process steps, set process windows, and establish quality control guidelines. It takes a lot of time and money -- at least a year and usually several hundred wafers.
MEMS Journal: Which MEMS foundries and IDMs are emerging with unique processes or process modules around which designers can develop their MEMS devices?
Alissa Fitzgerald: Several MEMS foundries today offer a multi-project wafer (MPW) run based on a fixed process flow. The process flow is typically generic (meaning, not optimized for a specific MEMS device) and reflects the process strengths of the individual foundry. These MPWs are generally offered to designers as a platform for exploration, not volume production (with the exception of the NF Shuttle from InvenSense). In fact, some of the MPWs (such as those offered by Europractice IC and ePIXfab, who are brokers of foundry processes, similar to MOSIS) specifically restrict commercial production and only allow research and development work. As a result, most MPWs are utilized by academic groups or research groups doing early stage explorations. Very few designers today set out to develop a commercial device on a MEMS MPW platform. We think this should change.
Here are the foundry and IDM processes that we are aware of: MEMSCAP MUMPS family, Sandia Summit V, Tronics' HARM, imec SiGe, Leti photonics, LioniX photonics and microfluidics, X-FAB inertial sensors, InvenSense NF Shuttle. Europractice IC and ePIXfab organize MPWs runs on some of these processes.
Silex takes a unique approach, which is to standardize process modules, and then to let the designers build the process flow from those modules. Silex calls this their SmartBlock® approach. It allows the designer much more flexibility while still maintaining the advantages of a standardized process.
Table 1. An overview of commercial MEMS processes.
MEMS Journal: How are you working with Silex and what is the RocketMEMS program all about?
Alissa Fitzgerald: All of the processes listed above are MPWs where the customer must supply the device design. AMFitzgerald's RocketMEMS program takes a different approach -- we ask the customer to supply a spec sheet, and then AMFitzgerald provides the design. Many OEMs don't know how to design MEMS, and they don't want to start learning. We will do the design and layout for them.
MEMS Journal: How are you able to accelerate the MEMS device development cycle?
Alissa Fitzgerald: We are taking a new approach to MEMS development -- letting the process drive design. AMFitzgerald has been working with Silex for many years and we are quite familiar with their processes and rules. The SmartBlock approach provides us with established, production-proven modules, for which process data is available.
We developed a pressure sensor architecture designed specifically for the Silex processes, leveraging its features and capabilities. The result is a device with a low-risk, short development cycle. The two main bottlenecks to MEMS development, described above, are removed.
AMFitzgerald's half of the partnership is design and customer relations, and Silex will take care of wafer production. We think it makes good business sense for developers and foundries to partner in this way. It is a non-exclusive arrangement that we each hope to replicate with other partners, over time.
MEMS Journal: Could you please expand on this and list 3-4 SmartBlock examples?
Peter Himes: The SmartBlock approach has evolved over the many years of supporting custom process development and integration for our customers. Starting from the assumption that every MEMS program is different and will require a custom process flow, Silex created a framework for process integration that breaks flows into discrete elements that can be reconstituted for new processes. These “process elements” are the SmartBlock elements which form the constituents of new flows. This approach is critical to supporting a wide number of programs (Silex typically has 50 engagements ongoing at any time) on common fab lines, and recognizes the need for custom process development or tuning on a case by case basis.
There are actually two aspects of the SmartBlock approach: the first being how the process flows are defined and constructed inside of Silex. This is Silex proprietary, and it is not like a “library” that we can give to customers to use. This is why working with AMFitzgerald up front has been so crucial: it is only by close collaboration that we can define and establish a process flow tailored for specific product needs in advance of the actual design work. With this framework established, AMFitzgerald can engage with their end customers for custom product development with confidence of manufacturability on a production line. The inherent risks and unknowns of typical MEMS developments are minimized or eliminated.
The second aspect of SmartBlock elements are the customer integrated features we offer, best known of which is our Sil-Via® TSV technology. This gives high value-add to customers and, once again, de-risks the use of a technology like TSV. In fact, today over half of Silex customers use Sil-Via in their products.
We do not offer any product platforms directly to customers. The SmartBlock approach is a methodology (and internal framework) for process integration to realize customer’s process platforms with high confidence.
MEMS Journal: What about costs? Does the customer need to achieve a certain volume to be able to afford producing at Silex?
Alissa Fitzgerald: No, there are no volume minimums. The benefit of RocketMEMS is that AMFitzgerald aggregates customers into a MPW run, so customers with small volume needs can get cost-effective production as part of a larger group of wafers.
Peter Himes: Note that from Silex perspective (and AMF I am sure), we see this as an opportunity to expand the served market for MEMS foundries to a much wider range of customers who would consider customized MEMS but where traditional hurdles of cost, time, risk and engineering expertise make it impossible. In our view, the “flood of sensors” requires mass customization, which is only possible through programs such as RocketMEMS.
MEMS Journal: Are there certain MEMS device types for which the SmartBlock approach is especially well suited?
Alissa Fitzgerald: SmartBlock elements can create any MEMS process flow, and as such will work for any MEMS sensor type. It is an interchangeable set of process steps used every day at Silex for custom process integration.
MEMS Journal: In your partnership with Silex, are you focusing of specific types of MEMS devices first and why?
Alissa Fitzgerald: We are focusing on MEMS devices that leverage the processes that have been in long production at Silex. We are starting with pressure sensors because that is where AMFitzgerald has seen the most customer demand recently and Silex has a decade of manufacturing experience with them. Other MEMS device types will be announced later.
MEMS Journal: What are the IP issues and how are you dealing with them? In other words, who owns the MEMS device design IP? Is it AMFitzgerald, Silex or the customer?
Alissa Fitzgerald: The IP is straightforward. Silex owns the process, and some aspects of it are protected by Silex patents. The sensor design is based on public domain knowledge, AMFitzgerald's design know-how and proprietary process data. Pressure sensors have been developed over the past 40 years, so there is a lot of expired art and public domain information.
The designs will not be portable, they can only be fabricated at Silex because they contain Silex's process IP and also because they have been optimized for the Silex process modules.
We are targeting customers who value getting the silicon chips they need quickly, not owning IP.
MEMS Journal: Who are these types of customers? Can you give a few specific examples?
Alissa Fitzgerald: These are customers who have an urgent market need for their system-level product. Many OEMs regard MEMS as a commodity component technology, similar to a resistor or capacitor. The MEMS sensor is just one of a thousand parts in their bill of materials. Design and process IP ownership is not a concern because they are (correctly) focused on adding value elsewhere in their system, such as in packaging, board integration, or firmware. We think they will be quite comfortable with this arrangement.
MEMS Journal: So, does the customer have any IP at the end? If not, what is the value that they are bringing into the supply chain? Without any IP, why don’t they just buy off the shelf components?
Alissa Fitzgerald: This is a shift in thinking for MEMS people. We like to think our technology is so special and important. But to some customers, it’s not. To them, MEMS is just a sensor in a control loop, and the customer places much higher value on their control circuit design, or their signal processing algorithms. So for some customers, the real value is in getting the chip they need at the right time and the right price, not its IP. The market is ready for their system product and they’re solving a sensor supply problem. Despite the large number of off-the-shelf MEMS readily available, we still find customers who need customized sensors. Keep in mind that most off-the-shelf MEMS have been engineered to serve the consumer electronics market. If one’s sensor needs are outside the operating spec for a typical mobile phone, such as for medical or aerospace applications, a custom sensor will probably be needed.
MEMS Journal: Do you already have success stories or illustrative case studies that you can point to?
Alissa Fitzgerald: We have just launched the RocketMEMS program and are taking in customers in Q1 2013. We have been fielding inquiries from medical device companies and aerospace companies. We are planning our first run in Q2 2013. Please check back with us in 3-6 months for an update!
Peter Himes: From a foundry perspective, the challenge with MEMS is still that each product presents unique processing needs, driving the "one product, one process" approach. This can sometimes result in a radical redesign of a product to make it manufacturable on a given foundry, adding significant cost and time to the MEMS time-to-market challenge. It also severely restricts fab portability, as processes are generally foundry-specific.
MEMS Journal: Will this SmartBlock approach somehow enhance fab portability? If so, how? And is it in Silex’ interest to do so?
Peter Himes: A SmartBlock approach can provide the framework for fab portability, for sure, but it is only a starting point. Internally at Silex, as you know, we have separate 6-inch and 8-inch lines, so “fab transfers” occur internally on a regular basis. As the market expands and we need to bring on a Fab 3 (wherever and whenever that will be), a SmartBlock approach will once again be critical to building in portability to our process methodology.
At Silex, our core competency is based on this process integration and development for manufacturability of complex MEMS devices, and we believe this will still be our main customer engagement path for the foreseeable future. However, we see the opportunity by working with AMFitzgerald to turn this paradigm on its head by integrating the process design directly into the product design architecture, and create a new approach which can not only circumvent this product-process dance, but make MEMS accessible to a much broader audience.
Our approach is the basis for how our customer integration engineers work with customers to create a process runcard for their project, however it is not quite a "library" that we can hand over to a MEMS designer to construct a custom flow. In MEMS, much of the processing knowledge is contextual -- based on what came before in terms of geometries, structures, films, etc. So while we have frameworks within which we work, and definitely do have standardized blocks which are common across process flows, a well-engineered process still depends on close collaboration between design and foundry from the start.
This is where we think that working in advance with AMFitzgerald can make a difference -- by working with them ahead of time to define the process windows, AMFitzgerald can know a priori what class of products can be designed, with very high degree of first pass success. Smart design of the AMFitzgerald process allows them to offer this in a MPW format, further lowering the barriers to customer's "real product" design needs.
MEMS Journal: Is there something significantly different about what AMF is doing as opposed to the usual activities of Silex’ own customer integration engineers?
Alissa Fitzgerald: AMFitzgerald is supplying sensor design and process integration expertise to the customer. Silex’ engineers are providing only process integration expertise in response to a customer’s existing design. Silex’ engineers are not creating new designs for the customer.
Peter Himes: I agree -- Silex is adamantly *not* offering any product platforms, as this violates our “pure play” philosophy. MEMS is still an innovation game and our success is on being a trusted partner of the innovative MEMS companies. We have seen again and again in the foundry space that, where foundries fall to the temptation of offering their own product capabilities, they lose the ability to engage with the market leaders in the same product space.
*********************************************
This article is a part of MEMS Journal's ongoing market research project in the area of MEMS manufacturing. If you would like to receive our comprehensive market research report on this topic, please contact Dr. Mike Pinelis at mike@memsjournal.com for more information about rates and report contents.
Copyright 2013 MEMS Journal, Inc.
Comments