August 26, 2013
Two “Universal” MEMS Processes
Teledyne DALSA and Leti Proffer Platforms
There is no standard MEMS process. Not even close. We’ve talked about this before, along with incipient efforts to rid MEMS of its “one device, one process” reputation.
If you have your own fab and you’ve figured out how to make stuff, then this is probably just fine for you. But if you’re a foundry looking to leverage your technology and expand your market, then making the technology less custom would seem like a good thing. Of course, even though there are benefits to “standards,” there are those who voice concerns that they lock out good new ideas. What we’re going to look at today is probably a good illustration of both the good and the bad about standards.
Recently, two companies have come forward with very different visions of platform processes for sensors. They both aim at being “universal” in the sense that they can accommodate different sensors. Their current incarnations are rather limited in that regard, but their roadmaps move them further in the direction of something you could consider “more universal” (which I put in quotes, since “universal,” like “unique,” isn’t relative: if something is 99% universal, then really, it’s simply not universal).
They both start by trying to find a better way to put accelerometers and gyroscopes on the same die. Why is this hard? If nothing else, simply because accelerometers like to be damped and gyroscopes don’t. According to Teledyne DALSA, air (or gas in general) is the main contributor to damping (although there are other effects), which is good for an accelerometer and not good for a gyro.
So, ideally, you want a carefully-controlled gas pressure in your accelerometer – one that you’ve included in your modeling – and a perfect vacuum in your gyroscope. Teledyne DALSA says that efforts to combine both in the past have compromised by putting some in-between pressure in both the gyro and the accelerometer, meaning that both were suboptimal.
Teledyne DALSA’s solution to the problem is their MIDIS (“mee-DEES”) process. It’s a complex beast (although that seems to be par for the course with MEMS), requiring a stack of four wafers to enable a sensor-plus-ASIC solution. And key to making this work is an “antechamber” – space around the sense elements for the accelerometers and gyroscopes. For the accelerometer, it’s pressurized; for the gyroscope, it’s evacuated.
“How do you do this,” you may wonder. There are a couple of keys to it: allowing different pressures in the two chambers and keeping CMOS circuitry away from the MEMS.
Image courtesy Teledyne DALSA
The two-pressure thing isn’t too hard. You start with a “handle” wafer that has the bottom of the antechambers; you bond the “device” wafer – the one that has the actual sense element – above it, and then there’s one more wafer that has upper antechamber recesses and TSVs that gets bonded on top. In addition to housing the bottom of the antechamber, the handle wafer also has – hmmm… what to call it… a port? – drilled under the chamber leading towards the back of the wafer.
This sandwich is assembled in a vacuum, so both the gyro and the accelerometer start with a vacuum. After it’s all sealed, the back of the handle wafer can be ground down, exposing that port thing. Now you can selectively expose the port to the specific pressure of the specific gas that you want, and then either plug it or put on a cap wafer to seal it. Now, at least for the moment, you have the vacuum you want and the pressure you want where you want them.
But that’s not good enough: vacuums don’t last forever. Why? Because nature abhors them, of course. Atoms and molecules tend to diffuse and break free and invade the once quiet, empty space, and your vacuum dissipates.
The MIDIS process was described via a webinar, and this issue attracted a lot of questions: the usual solution to this weakening vacuum problem is to use a “getter” – a material that absorbs the errant particles, keeping them out of the mechanical works. But MIDIS doesn’t use any getters.
Here’s the issue with getters, according to Teledyne DALSA: the kinds of contaminants most problematic are ones that come from CMOS – low-Κ dielectrics in particular. They have semi-organic materials as well as porosity that can absorb and release moisture and other contaminants. Other invaders come from the atmosphere in a fab, including argon, helium, methane, and other volatile organics.
And, they say, the standard getters simply aren’t good at sucking up these materials. Not to mention the fact that, at some point, the getters become saturated and stop working. So the MIDIS solution relies on keeping the ASIC circuitry on a completely different wafer, and then they take advantage of a particular silicon characteristic: at around 1000 °C, they say it acts like a getter. So you can do a one-time sweep of all of the cavities by heating the wafer assembly to a temperature that wouldn’t be possible with a CMOS wafer, and then you’re done, because there’s no CMOS stuff around the cavities to outgas.
While a separate CMOS die is critical to this working, they’re not proposing simply putting the two dice next to each other and wire-bonding them. Instead, the upper layer of the MEMS stack – with the TSVs – provides a platform where a micro-bumped ASIC die can mate. Such TSVs will provide a far superior (i.e., lower-inductance) electrical connection than wire-bonds would.
Of course, this brings up the old question about whether to bond wafers or use known-good ASIC dice to minimize loss. They’re designing both dice – MEMS and CMOS – to yield at around 99%, so the combined yield due to the occasional working die being mounted on a flawed MEMS element (or vice versa) will be around 98%. That small 2% loss is far cheaper than the cost of pre-dicing and then assembling one die at a time.
While the focus so far is only on the accelerometer and gyroscope, they’re also looking to include a pressure sensor as well, venting through the handle wafer to gauge absolute air pressure.
Meanwhile, Leti is proposing a completely different integration platform, one they call M&NEMS because it integrates some micro- and nano-scale elements. Their idea is to pair a thick proof mass (micro-scale) with a thin piezoelectric wire (nano-scale) that they call the “gauge.” Because of the difference in thicknesses, the stresses are magnified in the gauge, making it much more sensitive.
Image courtesy Leti
They show arrangements for in-plane and out-of-plane sensing, and for accelerometers and gyroscopes. Their gyros need a vacuum, but only a rough one, so they also don’t need getters. The reason they can get away with this is simply due to the stress magnification that the thin gauge provides. This makes the setup sensitive enough that they say they can tolerate a low-Q environment.
Of course, they also need to solve the problem of providing a vacuum for the gyro and a damped environment for the accelerometer; they say that they are patenting a solution to this, so I don’t have any more detail at this point.
In addition to those two sensors, they also have one for a magnetic sensor: they put a magnetic material on a movable proof mass. As the magnetic layer tries to orient itself with an external magnetic field, it moves the proof mass, which stresses the gauge, which causes measurement.
They’re also working on additional sensors to add to the mix: pressure and microphone. It bears noting that the history of pressure sensors and microphones is replete with problems involving, not the MEMS die itself, but the packaging. Some folks have even thrown up their hands in disgust* and given up.
So this is where it becomes useful to review the purpose of a universal platform like this. On the one hand, yes, you can integrate multiple devices on a single monolithic platform. But do you want to do that? Or is the benefit simply that you can make a number of different separate sensors using a common process?
You could argue that much smartphone space could be saved if you have a single die that acts as IMU and pressure sensor and microphone. Then again, the microphone has to go in a specific spot; the others less so. Some have argued against the integration of a magnetic sensor with other sensors, since optimal placement of the magnetic sensor may depend on local magnetic anomalies on the board. Couple that with the packaging challenges of getting the cavities and ports right for pressure and microphone, and it’s enough to make you question whether there really is a need for a single “11-axis” sensor.
On the other hand, harder problems than this have been solved, so I’m not ruling it out. Just sayin’… And I’m assuming that Teledyne DALSA and Leti are taking the non-die stuff into account as they expand their integration.
Coming back to our opening premise, we have here two different ways of trying to address multiple sensors with a single standard process. But the two approaches are also completely different from each other, highlighting the fact that, in the absence of an enforced standard, innovative ideas have a better chance of seeing daylight.
*OK, OK… you got me. I don’t really know if they expressed disgust. Heck, I don’t even know if they threw their hands up. Author’s license…
[Edited to read 98% yield, not loss]
Posted on August 26, 2013 at 11:23 AMWhat do you think about these two approaches to universal MEMS processes?
Posted on August 30, 2013 at 10:15 AMBryon has done a nice job, as usual, of explaining these two standardized MEMS process offerings in a technically coherent way. At this point, the Teledyne–DALSA approach sounds closer to meeting volume production needs whereas the Leti approach is more of an R&D platform. One point that should be made is that the Leti approach may enable a game changing reduction in the die footprint of the MEMS sensors, from hundreds of microns on a side to tens of microns. That benefit was my takeaway from hearing a Leti presentation on this process. An observation: Piezo resistance was a popular sensing mechanism in the early days of MEMS, but capacitive sensing has dominated in recent years. The Leti approach could auger a comeback for piezo resistance. What goes around comes around.
While standardized MEMS processes like these two will no doubt gain traction, there will be multiple "standardized" MEMS processes, each with variations and customizations for different MEMS designs. Both of these processes appear well suited for inertial sensors (accelerometers and gyros), but I would hazard an educated guess that either process will require substantial modifications for microphone designs, or other types of MEMS. The MEMS industry needs better software tools to cope with the continuing reality of multiple processes. Otherwise, the traditional approach of building and testing until it works will continue to exact a heavy toll in terms of development expense and time to market.