First of three parts: Increasing focus on cost; why it’s so hard to shrink MEMS chips; what’s missing out of the foundries; what it costs to design a MEMS device.
Semiconductor Engineering sat down to discuss the challenges of MEMS with Rakesh Kumar, senior director of the MEMS program at GlobalFoundries; Tak Tanaka, managing director for Applied Global Services at Applied Materials; Paul Lindner, executive technology director at EV Group; and Alissa M. Fitzgerald, founder and managing member at A.M. Fitzgerald & Associates.
SE: What’s happening in the MEMS industry today?
Tanaka: In the conventional semiconductor market, there are lots of new technologies being discussed like 450mm. But in the conventional semiconductor market, my customers—the device makers—are consolidating. The big guys control the industry. On the other hand, in MEMS, we have hundreds of customers. So that’s good for equipment makers like us. In MEMS, I see lots of new applications coming. So we think MEMS is a growing and emerging market. We see double-digit CAGR growth. That will continue for the next few years.
Fitzgerald: If we look at it from a historical perspective, there has been a series of waves in the MEMS industry. If we look at the mid-1990s or so, the driving technology was the accelerometer. And the booming market was automotive. Accelerometers were going into air bag deployment systems and electronic stability control systems. Then, around the early 2000s, it was Texas Instruments’ DLP and ink-jet technology. All of the money and growth was coming from ink-jet and DLPs. Now, in 2013, we are seeing gyros and inertial sensors. There is a concentration of these products from a few manufacturers. It’s the typical story that a few manufacturers hold the power and experience most of the growth. We are at an interesting point right now, because those sensors have been going into consumer electronic products like smartphones and tablets. But we might be getting close to saturation. Smartphone sales have slowed down, at least in North America and Europe. So, the question is how much more will this growth continue for these types of MEMS devices? And what is the next MEMS device coming along that will create this cycle of growth again? We will see perhaps different players come to power for that type of device.
Lindner: We see changes in the field. We used to mainly focus on reliability in the discussions with customers, especially in the automotive area. Cost was always a topic, but it was more of a side discussion. And now, we face the cost question more and more with customers. Customers are under pressure to shrink their devices and produce more cost-effective products. And certainly, that has an impact with equipment offerings. On the other hand, we see a lot of activity among hundreds of potential customers. They are bringing out new devices and sensors, such as environmental sensors. We are quite optimistic about the MEMS market and we expect double-digit growth rates.
Kumar: From a historical perspective, MEMS have been used in the automotive industry for a long time. Now, what we see is an acceptance and adoption of MEMS in other markets. In the last five years, MEMS have grown more than 10% to 15% in terms of CAGR. In the next five years or so, analysts are predicting more than 20% growth. And once people become more aware of MEMS, and what they can do in terms of providing more value, MEMS will change the landscape. Intelligent sensors are just one example of such products. There are application developers right now who are ready to make these things into a reality.
SE: What are the key similarities and differences between MEMS and semiconductors?
Fitzgerald: There are some fundamental differences. In the semi world, your building block is the transistor. MEMS are mechanical structures. When you shrink them, strange things start to happen. The best analogy is the tuning fork. Big tuning forks have a lower pitch than the tiny tuning forks. As you shrink the size, it proportionally might be the same. But as you shrink the size from the big tuning fork to the little tuning fork, you change the resonant frequency. The resonant frequency goes up. This is why you can’t shrink MEMS devices. If you shrink your accelerometer or gyro, you are changing the fundamental mechanical behavior of that device. You’ll change the sensitivity. All of that changes. So, we can’t do dimensional shrinks the way the CMOS industry can. In hindsight, one aspect that has made the semiconductor industry so profitable is that we had a roadmap. Moore’s Law told us that it would be worth it to invest in smaller lithography nodes. It would be worth it to invest in larger wafers, because we can predict the economics of that. It’s very obvious and very linear. And the industry can move in lockstep. But that’s not the case with MEMS. So, we have mechanical structures that can’t be shrunk. We have a diversity of mechanical architectures. There are commonalities between accelerometers and gyros, but they are very different. The processes to make those devices are very different. So, this is where people are struggling with MEMS in terms of how do we get this industry on a roadmap. It also makes investing difficult. You don’t have the predictability of saying: ‘300mm will one day become a good bet.’ We don’t have that in MEMS.
Kumar: MEMS can’t scale like CMOS. But if we look at different technologies like nano-sensors, we can talk about utilizing the properties that can be changed at the nano-scale. Nano-MEMS structures, where you have piezoelectric MEMS with a very small nanowire, can be used for sensing. So there are the non-conventional mechanisms. In this case, MEMS can be scaled down, but they definitely do not follow the scaling law as in the semiconductor industry.
Tanaka: I was involved in the TFT industry for a long time, so I can compare TFT and MEMS. In TFT, you have the pixel size and the substrate size. From the equipment point of view, our challenge is scaling or shrinking. MEMS is a mechanical structure, not just an electrical structure. All of the devices have different mechanical architectures. That requires different processes. The process tools are the same, like CVD, PVD and etch. But every customer has a slightly different usage for each piece of equipment. That’s good and bad for us. The good thing is we have more opportunities. On the other hand, if a hundred different customers require a hundred different processes, then that’s one of our challenges. So in general, for equipment vendors like us, standardization is better. But overall, we need to come up with different strategies to address the diverse MEMS industry.
Lindner: There are some opportunities for 3D integration. I personally believe in the integration of MEMS with CMOS. There is potential to shrink further and combine them even in advanced nodes. We are also starting to hear about MEMS on 300mm wafers. Could MEMS be produced in 300mm? It comes back to the same equipment that has been developed for 3D-ICs. So, this equipment could be utilized for MEMS. In an evolutionary aspect, you could even combine advanced CMOS nodes with the mechanical structures of MEMS at the wafer level with small form factors. The business question is who needs these trillions of sensors?
SE: What are the big challenges for MEMS?
Tanaka: From the equipment standpoint, I will try to separate the business and technical challenges. We have a lot of technology requirements from customers. For MEMS, the wafer size is not so important. We are talking about 200mm and below. So, the total equipment business opportunity for 200mm is not as significant, compared to 300mm in the semiconductor business. Then, you have to come up with very specialized equipment for each customer. So what we are doing is modifying existing equipment to make MEMS or sensors. In conclusion, we need to adapt ourselves to a smaller, but emerging market size.
Fitzgerald: If you are designing a MEMS device based on new sensor technology, you will likely spend five years and $50 million getting from first concept to market. Our company is a design and product development business for MEMS. In terms of the challenge we see as designers, our customers come to us with a specification sheet with a desire to create a certain type of functionality or technology. And then we have to figure out how to execute that. There are two major challenges we face. One is a lack of foundry-specific process information, and that can actually be drilled down to the equipment level. Ultimately, the foundry is using a piece of equipment. If we can know ahead of time how that piece of equipment forms a certain structure, then that would be enormously valuable to us as designers. For example, we will design and simulate a mechanical structure. We will have the expectation that if we put a block on the mask to get a five-micron-wide structure, and then when it is etched, we will get a five-micron-wide structure. All of our simulations are built around that assumption. But we know in reality, there are things like lateral undercuts that might be just a quarter micron on either side. So now my five-micron structure is four-and-a-half-microns in reality. The problem is that this is a mechanical structure and it’s a little bit different in size. And now, the behavior is different. Because we don’t have the information ahead of time, this extends the design-fab-test cycle. So we have to design based on best knowledge. The foundry does their best job. Then we have to measure and dissect and see what we have. And then, we go back and re-design and re-fabricate over and over. This is what consumes years and tens of millions of dollars when you extrapolate into developing a gyro or complex device. So all told, there are two pieces we would love to have. We would love to have more foundry and process-specific information in the early part of the design cycle. The dream would be to have actual physical simulation of some of these processes. For example, in the CMOS world, a lot of the physical processes of implant are very well characterized. The physics are well understood, so you can simulate. For example, if you dose a certain amount of boron at a certain energy into silicon, you can predict exactly where those atoms will end up. We don’t have that kind of simulation for MEMS. We need that kind of simulation for DRIE, as one example. DRIE is incredibly sensitive to silicon loading, pattern geometry and the location on the wafer. But we don’t know what the tool is going to do ahead of time. And neither does anyone else. And so if we had that kind of physical simulation, that would make a world of difference in our ability to shorten the design cycle for MEMS.
To view part two, click here.