There were those who said history would not repeat itself. But evidence continues to accumulate that the MEMS sector will follow a similar path to CMOS logic.
That path is one in which integrated device manufacturers (IDMs) that do everything thing under one roof gradually give way to those choosing one side or other of a dual fabless-foundry business model, where there are those that specialize in manufacturing in volume and those that specialize in design.
The difference between MEMS and logic -- and the reason previously given for the continued supremacy of the IDM in the MEMS domain -- is that in the case of MEMS the manufacturing process is fundamentally linked to the design possibilities. In contrast, over 30 years of digital CMOS development has led to a remarkable consensus on the best ways to implement integrated circuits and a degree of separation of design from process implementation. The fabless avoid manufacturing infrastructure costs, while those that incur them spend big for economies of scale and then manufacture for many customers and do not compete with them.
In CMOS logic this has led to the rise of such foundries as the Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) and fabless chip companies such as Qualcomm, Broadcom, and Mediatek. It is also notable the way a traditional IDM, Advanced Micro Devices, evolved to become fabless and threw off its manufacturing operations, which became the seed crystal for the creation of GlobalFoundries.
Meanwhile the latest MEMS status report from Yole Développement indicates the start of similar process. (See MEMS Leaders Under Pressure, Says Yole.) It remains true that there almost as many detailed MEMS processes and packages as there are MEMS products, but the volume is met by a much-reduced set for such things as inertial sensors, pressure sensors, and microphones.
Leaders under pressure
The Yole status report shows that some of the former leaders in MEMS -- the likes of Texas Instruments (digital light processing) and Hewlett Packard (microfluidics for inkjets) -- are seeing sales decline. STMicroelectronics, the MEMS market leader in 2012, which operates a split strategy where it does MEMS manufacturing for some customers and also produces under its own name, has also come under pressure. Meanwhile, Yole is tipping InvenSense and mCube as rising stars of the MEMS firmament.
It is a financial formula that works. The fabless get a faster, lower-cost route to market that helps them compete against the IDMs, while the foundries acquire process know-how, which they may one day be able to offer to other MEMS startups. Although the true standardization of manufacturing processes has yet to happen in the MEMS sector, it appears the economic benefits of disaggregation into a fabless and foundry duality will not be denied.
MEMS transmission, rack, and polysilicon mirror: The transmission and linear rack elevate the mirror located in the lower-right of the frame.
(Image courtesy of Sandia National Laboratories, SUMMiT™ Technologies, www.mems.sandia.gov)
Of course, these things are never simply black and white. In logic production IDMs have coexisted with foundries for many decades. In last few years the interdependence of design and process has increased, as have the variety of processes (planar CMOS, FinFET CMOS, FDSOI, and other variants). Similarly in the MEMS domain ST's replacement at the top of the MEMS vendor ranking is Robert Bosch, which is a classical IDM with its own dedicated MEMS wafer fab.
Nonetheless, expect future MEMS startups to be predominantly fabless. They will be looking for foundries to work with them on process development and carry the cost of expensive wafer processing machines, and the likes of TSMC and GlobalFoundries appear ready to step forward to do that. IDMs such as Bosch and STMicroelectronics will also be tempted to keep wafer fabs full by running third-party MEMS wafers, which brings them into the part-time foundry category.
This financial slope led to the foundry business solution in digital logic, and it will do the same for MEMS -- and it probably won't take decades to do it.
— Peter Clarke is a freelance writer for EE Times Europe covering analog and sensors.
This article was originally published on EE Times Europe.