There is a general expectation that 10-DoF solutions will be a high point of sensor fusion implementation in the next couple of years. So, let us focus on the 10-DoF solution and discuss the role of MEMS sensors in it. Three out of four sensors to be used in the 10-DoF solution (accelerometers, gyroscopes, and pressure sensors) are manufactured by using MEMS technology; thus it can be stated that MEMS technology is critical to sensor fusion. Magnetic field sensors typically do not use MEMS (although they could); therefore they will not be discussed here.
MEMS technology has matured in the last decade. It can now deliver reliable, and more importantly, cost effective sensors that are used in mobile devices without having a significant impact on the total cost of BOM. The same is true for other consumer products. There are two key features pertinent to MEMS technology that stand out. The first key feature of many MEMS devices is the creation of movable structures (dynamic MEMS structures). Often for a MEMS device to function something has to move (up and down, or left and right, or back and forth, or tilt) and then that movement is translated into a corresponding signal. The movable structures are diaphragms, cantilevers, beams, proof mass, or some type of combination of these basic micromechanical structures. The second key feature is hermetical sealing or capping. The movable MEMS structures need to function reliably; therefore they need to be protected. Typically, they are hermetically sealed (or capped), creating a controlled environment. Capping is usually done at the wafer level.
Naturally, there are variations in the MEMS process for making movable structures, as well as variations in the capping process, and often the process chosen for a specific device depends not only on the application but also on the vendor’s technology.
Motion and inertial sensing
Accelerometers and gyros are at the core of motion and inertial sensing. Most accelerometers and gyroscopes, when made by the same manufacturer, use the identical MEMS process. The only difference between these two sensors is basically in the geometry of MEMS structures. Movable MEMS structures for accelerometer and gyro are different (one is optimized for linear motion sensing and the other for rotational sensing) but the MEMS process itself, including the capping process, as well as the CMOS process for ASIC chips are the same for both. Today, the majority of 3D-accelerometers and 3D-gyroscopes are manufactured as separate MEMS chips although several companies have already introduced the integration of accelerometer and gyro on a single MEMS chip which leads to a 6-DoF combo solution.
The MEMS dynamic structures used in 3D-accelerometers and 3D-gyros are designed as high-aspect ratio structures. They are made from polysilicon or silicon using a deep reactive ion etching (DRIE) process. The high-aspect ratio approach prevails today in the industry because it provides shock resilient MEMS structures and it also makes use of a capacitive sensing possible. As already mentioned, there are variations in the MEMS process and they are typically associated with a vendor and proprietary technology. Here, we will discuss several examples of accelerometer/gyro products and point out some variations.
One common version of the accelerometer/gyro solution that is based on a combination of MEMS structure with a passive cap and a separate signal processing CMOS IC chip is shown in Figure 2. The MEMS element is made of a thick polysilicon (up to 30 um) by use of DRIE and sacrificial etching. The sealing of MEMS is done by low-temperature frit-glass bonding. The CMOS IC chip in this case seats on the top of the cap. This approach is currently used by STM and Bosch. There are further variations of the passive capping approach whereby the CMOS IC chip is packaged side-by-side to the capped MEMS structure (ADI and Kionix/Rohm use this approach), or the capped MEMS device seats on the top of the CMOS IC chip (Freescale solution).
Figure 2: Accelerometer/Gyro solution with passive capping
There is another version of the accelerometer/gyro solution that involves active capping, as shown in Figure 3. In this case, the MEMS structure is made of silicon crystal by using fusion bonding of two wafers, DRIE, and sacrificial etching. The capping wafer is the CMOS IC wafer; thus the passive cap has been eliminated completely. The sealing is done by low-temperature eutectic metal (different combinations of metals can be used including an AlGe, Au and Au/Sn combination). As is always the case, there are advantages and disadvantages in every approach. For example, a frit-glass sealing is a relatively simple process but it takes up more space on a die to make a good hermetic seal which leads to a larger die size. On the other hand, the metal seal approach requires additional photo steps for metal patterning but it enables the shrinking of a die. InvenSense is the leader in the active capping approach. Their accelerometers and gyros have been accepted enthusiastically by the smartphone OEMs (they are the dominant supplier to the Android OS camp while STM is a dominant supplier to Apple).
Figure 3: Accelerometer/Gyro Solution with Active Capping by InvenSense
The CMOS IC chip is an inevitable part of every accelerometer and gyroscope solution. It plays the crucial role and provides basic functions such as biasing and reference voltages, offset compensation, trimming, self-test, signal amplification, temperature compensation, driving MEMS structures to oscillation (as in gyros), filtering, and analog to digital conversion. The CMOS IC may also contain a DMP (Digital Motion Processor) as in the case of the integrated 6-DoF solution introduced by InvenSense (their series MPU-6000/6500). It should be stressed that the 3D-accelerometer and 3D-gyro, as well as the 6-DoF integrated solutions provide digital output that comply with either the I2C, or the PSI protocol, or both. The I2C and PSI are communications protocols commonly used in the industry.
The examples shown here are representatives of the current inertial sensing products on the market. If one starts looking into the next generation products things look even brighter. As already mentioned, a further integration is already taking place and the form factor is shifting from discrete accelerometer and gyro devices to 6-DoF combo products on a single MEMS chip. This is possible because of the advancements in MEMS technology and packaging. One excellent example of this is STM’s announcement of the use of TSV (through silicon via) technology in MEMS packaging. Using TSV through the CMOS IC chip enables active capping and elimination of the bond wires at the same time, as shown in Figure 4. That way the new 6-DoF combo solution (accelerometer + gyro) is scaled down in all three dimensions (true 3D-scaling down), reducing the cost and improving performance. Very impressive indeed!
Hello Mr. Ristic,
I would like to appreciate the succinct description of sensor fusion provided in the aforesaid article. But, while working on sensor fusion for the the estimation of velocity of a body frame, I came across some real world challenges, for which I seek advice.
*When the mobile device is mounted on a moving body frame, there are 3 coordinate systems, the geomagnetic earth frame, the frame of device mount and the frame of motion of the body itself. In this case, how can we project, the acceleration vector on to the body frame only?
*Furthermore, from inertial navigation, we can use gyroscopes to lock in inertial vectors in the reference frame. But, how do we account for centripetal forces in this case?
It would be really helpful if the doubts enumerated above could be clarified.
The author has to be congratulated for an excellent article on sensor fusion and its future directions.
I also would like to call his attention (since he is local to Silicon Valley) to the event on a closely related topic being held on a Saturday this month (29th):
Next Generation Circuit & Systems, Communication and Sensor Technologies in Mobile Devices
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.