When a MEMS fab doesn’t deliver: assessing a new manufacturing source
In some cases the business climate was simply unfavourable – fabs pursued customers in the slumping markets, such as telecom, or they went after customers in promising markets before the technology was proven. Existing RF switch designs, for example, required complex process technology and considerable non-recurring expense (NRE) resources, not to mention long qualification and design-in times before volume production could actually commence.
In other cases, fabs had strong R&D capability for prototyping but they and their customers had little or no experience in designing product and developing processes for volume production. Management and engineering staff were not grounded in commercial reality, and the fabs tried to be all things to all people, despite wide variations in design and process technology for every new device.
The result was that foundries and their customers were forced to focus on product development as well as process flow and standardisation. Since all the steps were inextricably intertwined, the result was an explosive multiplication of challenges rather than simple addition of a few incremental hurdles. It was never clear whether the developed process was the right process, and the fabs were never able to hit a stride in continuous wafer flow.
While the rash of closures is unfortunate for the companies and for their employees, and doesn’t help the reputation of the MEMS industry, it’s the customers who are left without a manufacturing source in which they had invested considerable time and money. These fabless organisations, if they survived, need to look for new manufacturers and in most cases will need to hand over more NRE dollars to recreate a process. They will also need to compensate for the delay in time-to-market.
Once burned, twice shy
At Silicon Microstructures (SMI), a fully owned subsidiary of Dortmund-based Elmos Semiconductor AG, we’ve had a good deal of experience with issues involved in transferring MEMS device manufacturing and processes from fab to fab. SMI has been on both sides of the relationship – originally as a fabless supplier of MEMS pressure sensors and now as an experienced MEMS foundry. The recent acquisition and integration of a new facility taught us even more about moving technology processes. At the same time we continue to work with customers on the rebound from negative experiences with failed or inexperienced fabs. These experiences have enabled us to develop a list of key considerations for fabless companies seeking to assess new manufacturers, as well as some technical tips to guard against future problems.
These considerations should enable device developers to maximise manufacturing investment by focusing on return in the areas of time-to-market, lowering costs, increasing performance and protecting intellectual property (IP). Obviously, an overarching goal is the formation of a win-win business arrangement, including a short-term revenue plan with a forward path. While a fabless developer may not have the leverage of a Qualcomm in seeking favourable terms from a fab, it should certainly be entitled to as much information as it needs to make a decision that affects the future of the company.
We’ve already pointed out that R&D organisations are not candidates for volume manufacture. Of course, there is a stage - usually at the rough conceptual period - where a relationship with an R&D-based fab can be very helpful. However, once product development nears commercialisation and volume, the team should focus on finding a fab that is running product now instead of soliciting business based on promises for the future – it’s more likely to be there down the road.
For fabless developers, time-to-market is paramount. As Jeffrey Moore emphasises in “Living on the Fault Line”, market share in emerging technology will go to the organisation that launches the product first and defines the space. Similarly, keeping costs low is highly desirable in manufacturing if a fabless developer is intent on large volumes where a premium can be charged for new or improved applications. As the product becomes commoditised, margins will shrink, so the economies of scale offered by a high-volume fab become critical.
Finding a MEMS supplier that has broad-based experience in volume manufacturing, with fixed processes and a loaded foundry, will usually mean quick and effective development of process variations on a base technology in time to meet a product introduction window – as long as the product development team provides fixed specs. A well-loaded fab (with the expertise to produce both its own products and contract products) also offers customers a better cost position than a fab working on R&D revenue, and provides the required experience in protecting design IP.
Control as a key factor
The primary key to selecting a fab that meets time, cost and performance requirements is to find out how well controlled it is. It will be important to know the exact processes used, and the statistical tightness of the process parameters. Do the fab have detailed control charts? A good high-volume facility will be able tell you its cycle times, yields, the number of reworks its undertakes, test methodology, and even the scrap percentage, which can provide insight into the effectiveness of its processes.
Uniformity of process goes a long way towards improving part performance. Process uniformity can be determined by a number of factors.
* Recipe control
* The list of standardised processes: doping, furnace procedures, etch rates, deep reactive ion etch (DRIE) rates, lithography line width control, metal thickness and composition
* Statistical process control (SPC/AEC)
* Consolidation of processes to a minimal number to get statistically significant numbers of wafers of each process type and to permit process centring and monitoring.
The downside of process uniformity is that product designs need to fit into the existing process, but reputable commercial fabs also have more tools available to help adapt the design to fit the process, including newer methods of film deposition and the latest DRIE techniques.
Measuring capability through infrastructure
Second, it’s crucial to know what infrastructure a prospective MEMS fab has - both procedural and physical - and whether it is state-of-the-art enough to allow flexibility in approach. This will also have a major bearing on cost and time-to-market.
From a procedural perspective, advance planning is the key to a streamlined infrastructure, enabling a smooth flow in developing the process and qualifying the prototypes for time and cost savings. A good MEMS fab will have implemented specific quality procedures such as advanced product qualification planning (APQP), during the design phase. APQP procedures also cover test and measurement, including accelerated lifetime test, if required. These tests will need to be device-specific, but a volume foundry will be able to save time by using similar parts or processes as points of departure. Following advance planning guidelines provides the ability to have zero defects in production.
Having a developed infrastructure to monitor processes is often where typical start-up fabs come up short. Process monitoring is essential to verify whether processes are really accomplishing what the fab thinks they’re accomplishing. During this production part approval process (PPAP) phase recipes and variables are frozen, accelerated lifetime test is completed, and full production can commence with no change. While the fab may eventually adapt the product and process as a result of a long-term continuous improvement programme, such modifications require going back through the APQP process with more capability analysis and control planning.
A state-of-the-art physical infrastructure obviously offers significant benefits, especially with well-trained operators. The most recent tools can be used in a variety of ways, and expert staff can get the most out of their equipment, from steppers to DRIE machines. In some cases operators can take the equipment beyond what the equipment suppliers thought possible. Less experienced teams, who may be more technicians than true operators, are unlikely to be able to squeeze as much from their tools.
Infrastructural factors also include fab maintenance and environmental aspects. Is the facility clean and smoothly maintained? On-time deliveries and consolidation in materials buying all have an effect on hitting a market window, so it’s critical to be aware of a foundry’s track record in these areas. Does the facility have adequate process piping, air handling and temperature control? Do they have any records of maintenance problems?
One quick way to screen candidates is to make sure they are compliant with either semiconductor industry standards (ISO) or applicable market standards such as TS6949 (for automotive). Certification is not a sole gauge of competence, but it offers a general assurance that the entity knows about quality control.
Communications is key
The organisation of different functions within the communications channel, the implementation of the chain of command and the experience of the people chosen for customer – supplier interface, all will have a significant impact on timing and cost of product delivery. Having knowledgeable teams on both sides, as well as set procedures for jointly resolving technical issues, facilitates smoother development and minimises misunderstandings.
Team members on each side should communicate by function, enabling smaller issues to be resolved quickly and more important challenges to be elevated to a team-wide level, as needed. A typical team might have a member designated for each of the following: business/general technology questions, product, process, qualification, and production issues. The reliance on established standard processes, distinct programme phases with sign-off at each stage, and a good reporting structure, will make the communications between teams easier and faster.
Working out payments
For a fabless company faced with closure of or non-delivery by its first-choice or existing manufacturer, it’s worth working together with the prospective new fab to come to a favourable arrangement. Often the second foundry will take on the incremental costs associated with running the new part. However they will need to charge NRE costs for adapting their standard process to the design and for building prototypes. In the cases of some of the failed fabs, the failure was in part attributable to insufficient NRE for development of some particularly complex devices.
Protecting IP
The question of IP ownership and exclusivity tends to be a hot issue, especially with start-ups, but in fact, the solution is quite straight-forward, revolving around both parties’ market priorities. In a typical arrangement the fabless customer brings the general concept of layout and process sensitivities, and the fab offers the process technology to make the layout happen as closely as possible to the customer’s specs. The advantages are that the customer owns the design and gets it to market faster with the fab’s (modified) processes. A reputable fab works with many customers but doesn’t ever disclose confidential information, since providing manufacturing for multiple customers is at the heart of its business.
In some unique cases, for example with special coatings such as vapour-deposited protection or actuator materials, or with a special sequence to create three-dimensional structures, the fabless developer may have processes and as well as actual recipe that they own and have patented as part of their IP portfolio. This IP they may license to the foundry. While having an exact recipe is a helpful starting point for a foundry, most contract fabs don’t operate this way, preferring to leverage their own investments in a consolidated group of controlled processes and recipes where they have expertise and can offer repeatability.
Technical considerations
If a fab change is necessary, what can the customer do to help streamline its own transition from one fab to another?
The fabless customer is faced with a choice of how to proceed. The first option is for the customer to offer the prospective foundry complete documentation of everything done to date, with detailed process information and manufacturing tolerances. In this approach, the most important step is to organise everything into a complete design binder, showing modelling data, process sensitivity and constraints on the design due to process variation. This avenue enables the fab to quickly and directly assess whether it can meet the process requirements. However, in this case the foundry is not responsible for meeting performance specifications.
The second option is to provide the fab with functional performance requirements for the end product, which might include angle of deflection of a micromirror, high or low sensitivity for a pressure sensor, or desired throughput of a microfluidic nozzle. In this scenario the fab has more latitude in process adaptation and in meeting necessary yields.
If prototypes and masks were produced by the previously used foundry, this helps validate the process and documentation. A customer can also make the prototypes available for reverse engineering, if required. The customer’s masks must still be made compatible with procedures in the new fab -- typically alignment targets, castles and in-process testing and process control monitors still need to be dropped in by the foundry engineering team. Without prototypes the prospective fab will be forced to provide its best estimate, and will need to clarify that this is a higher-risk situation.
Layout files are the typical starting point. Engineers will need to go through the structure and verify that it meets the design rules of the foundry, working in the alignment targets and process control monitors. Remaining steps include having the reticles and photoplates shot so they are compatible with procedures in the fab, and then program them into the stepper according to the layout. The process engineering team will need to evaluate process sensitivities as well. In the case of a less developed product, many of these requirements must be resolved before the process work is started.
One factor that is often troublesome in a fab-to-fab transfer of a completed design is a lack of information on background development and on unusual characteristics of previous processes. Many times, even undesired characteristics and parasitics may make a device work - thus better control can actually result in a problem.
A classic example is variations in actual values or profiles of stress in thin films and dissimilar materials such as nitrides, oxides or polysilicon. Similarly, parasitic leaching of charge in process steps, dissimilar grain structure in deposited metals, oxide density or variations in raw wafers can cause problems. Process control and test monitors may vary between fabs, and alignment structures can vary.
Exact fabrication tools and chemicals may also vary. As examples, differing standard silicon etch recipes can create corner compensation problems, conversion from contact lithography to stepper can require careful assessment, and exact recipes for resist coating may give different results depending upon wafer topography.
Looking to the future
Finally, the foundry should leave a fabless provider well set up for the future. The chosen fab wants the customer’s repeat business and will help ensure a straight migration path to manufacture of the customer’s next generation of device or of a related device in the same product family.
Author:
Jim Knutti is CEO of Silicon Microstructures (SMI), a high-volume manufacturer of MEMS-based pressure sensors, inertial and other MEMs devices for a range of industries. He can be reached at jim.knutti@si-micro.com
