Why MEMS Need to Be More Like ASICs

They are undeniably pervasive—each of us uses equipment that employs them on a daily basis. I'm referring to those marvels of micro-miniaturization, microelectromechanical systems, or MEMS. You find them in everything from automobiles (accelerometers for air bags) to inkjet printers (micro-droplet ink dispensers). However, the promise of MEMS in several other applications continues to be just that, a promise. The "hot" markets where MEMS can make a significant performance or price impact, biomedical and communications, are still waiting for microelectromechanical technology to become feasible when placed in a production environment. Why, you may ask, is this not happening?

The answer is cost! MEMS development has progressed to a point where the underlying technologies offer several performance advantages over strictly electronic components in many applications. However, the cost of manufacturing and testing MEMS chips is holding back their use in all but the very highest volume applications. MEMS devices require specialized processing, which adds cost beyond "standard", more mature, and less complex CMOS processes. Comprising mechanical structures, MEMS also need a means of protecting individual die on a wafer during die separation and assembly—so-called zero-level packaging. Further adding to the cost budget is the need for specialized test equipment to verify both electrical and physical structures within individual MEMS devices. The bottom line is that MEMS, even more so than ASICs, are expensive to develop and bring to manufacturability. This is why MEMS for low-to-medium volume applications have been slow to reach production. However, there is a potential way to reduce this prohibitive cost, similar to what developers currently do with many ASIC devices—platform-based design.

The advantage of platform-based design for ASIC developers and vendors is that a single ASIC platform, a device with functional blocks that can be added or deleted from a specific design, can be used in several applications. Instead of fully developing a different ASIC for, say, four variants of a device, designers just have to "personalize" the functionality of the ASIC platform for each application, significant reducing design and verification time and, of course, cost. In theory, a platform-based design methodology could also be used for MEMS designs.

Opponents of a MEMS platform methodology can argue that MEMS devices are too complex for platform-based design. These people may point to the need for individual design, verification, and testing of each different MEMS-based chip. However, the same argument was also used, initially, against platform-based ASIC design. MEMS chips are like ASIC chips with an additional dimension—physical structures are added to electrical structures, making MEMS development a multidimensional problem. Nevertheless, the complexity of an SoC—a completely electronic-domain structure—is extremely high, when you consider that a typical chip may contain millions of logic gates plus analog/mixed-signal, exotic memory, and sometimes RF blocks. Adding one or more MEMS structures to the chip is just adding another level of complexity.

I don't mean to belittle the effort designers would need to develop production-viable MEMS platforms. There is a huge amount of research and development effort needed to make MEMS-platform-based design a reality. And, like ASICs, a MEMS platform will probably only be useful if considered for a single type of MEMS structure—for example, separate platforms for optical, microfluidic, RF, and motion-enabled MEMS. However, the huge (and growing) costs associated with leading-edge microelectronic devices—NRE, processing, and test—require us to think "out of the box" for less expensive ways to develop the next generation of MEMS-based devices. Platform-based design just might be the way to go.