MEMS: Learning the lessons from the fabless chip industry

Paul Boughton
As product developers look to MEMS devices to enable significant new features, the technology and business of MEMS is following paths already familiar to the semiconductor industry. Tony McKie reports.

MEMS devices such as motion detectors, miniature silicon microphones and wafer-level actuators are increasingly found in consumer products, as designers seek to create new and differentiating features and miniaturise products without trading performance. And as product designers delight their customers with new features only possible with the use of microelectromechanical systems (MEMS) technology - such as high-end photographic performance from a camera-phone handset - end-users' expectations will continue to increase. As demand for MEMS-dependent functionality grows, despite the effects of the current financial crisis, the MEMS industry must innovate new functions, improve processes, and establish a lower cost base for MEMS production.

Advances in production processes for MEMS devices hold the keys to meeting these goals. The processes used for etching, for example, have a major influence on the design techniques that can be applied, the materials that can be used, and the size of the features that can be created. Aspects such as wafer-to-wafer repeatability and the total cycle time per wafer are also dependent on the type of etch process used.

Etching is the process of releasing the micro-mechanical structure from the sacrificial material on which it is formed. The structure may be the cantilever beam of an accelerometer or gyroscope, or the silicon membrane of a MEMS microphone. Established etching processes are based on wet chemistry, such as liquid hydrofluoric acid (HF). Although the HF acid is effective in removing the sacrificial material, the behaviour of the liquid tends to restrict the shapes and dimensions of features that can be achieved. In the production of a MEMS silicon microphone, for example, the removal of sacrificial material from beneath the membrane - which is necessary to enable the membrane to vibrate - requires apertures to be created in the membrane itself to allow the HF acid into contact with the sacrificial material. From the point of view of microphone performance, the aperture size should be as small as possible. However, a certain minimum size is required to permit adequate flow of HF acid.

There are also limitations on the types of materials that can be used. Since liquid HF attacks silicon dioxide, it is not compatible with MEMS structures that use oxide as a mechanical material. HF acid is also incompatible with metals. Hence liquid HF places major restrictions on MEMS technology going forward.

To overcome such restrictions, and to achieve greater repeatability, reliability and selectivity in single-wafer or batch processing, next-generation etch processes exploiting Sacrificial Vapour Release (SVR) permit the use of a broader variety of structural materials, including metals, are being adopted. SVR chemistries such as vapour-phase hydrofluoric (vHF) for oxide-release etch, or Xenon difluoride (XeF2) for silicon-release processing, will allow new types of MEMS devices to emerge. Vapour-release processes also provide greater freedom for designers to optimise structures and materials to achieve the desired device properties within the smallest possible dimensions.

The introduction of the RF-MEMS variable capacitor, which is produced using SVR with XeF2 chemistry to etch sacrificial silicon without attacking the device's aluminium support layer, highlights the extra freedoms now available to device designers. As processes evolve in this way, MEMS production will become compatible with CMOS processes and CMOS fabs, leading exciting future generations of components combining MEMS and CMOS elements within the same package. Processes compatible with metals are necessary to achieve this class of device, not least to enable attachment of the MEMS device using aluminium bond pads. With investment in suitable MEMS processes, it is technically achievable to converge MEMS and CMOS fabrication on a single line.

History shows an earlier migration from relatively imprecise and restrictive wet chemical processing to vapour-phase processing. As the semiconductor industry progressed towards sub-micron and subsequently smaller design rules, these successive node shifts forced the pace of advances in wafer-level processes, including vapour etching. With process improvements such as these, the semiconductor industry has been able to pursue Moore's Law, to deliver more highly integrated, better performing devices; improvements in process capability, leading to more good components per wafer, have also contributed to significant savings in the cost of each unit produced. The MEMS industry can expect to achieve a similar 'coming of age' as it transitions to future generations of manufacturing processes.

Another significant trend set by the semiconductor industry, which has led to tremendous advances in device performance, functionality and cost-down, has been the emergence of the fabless chip company. By focusing on developing device IP, and purchasing foundry services, this model has successfully reduced start-up costs for design-led chip companies. Who knows how many devices currently on the market may never have come to fruition if their creators had also had to find the necessary investment to build a manufacturing line to deliver their bright ideas to market?

The MEMS sector is now displaying a similar trend. Around 30 per cent of the companies memsstar is currently working with are fabless IP developers. Non-captive MEMS foundry capacity is also increasing; semiconductor foundries can convert existing CMOS lines for MEMS production relatively easily, and new service providers are also entering the marketplace. As has been seen in the semiconductor industry, this should increase the pace of MEMS IP development, and should also lead to rapid increases in process capabilities and yield. The time to market for new MEMS device designs should also become shorter, as foundries seek to standardise MEMS fabrication processes to deliver faster turnaround and lower costs for customers.

The MEMS industry can benefit from experiences acquired in the semiconductor sector. Much existing knowledge surrounding vapour processing is directly applicable, and will serve to accelerate manufacturing improvements. Also, the fabless model has been adopted more quickly. This will help to increase the pace of device development as well as process improvement, driving MEMS-enabled features into diverse types of consumer, commercial, industrial, medical, scientific and automotive products.

Tony McKie is with Point 35 Microstructures in Livingston, UK.