A new inertial measurement unit (IMU), the FIS1100, was announced by Fairchild, and it gives us a couple things to talk about.
First, well… Fairchild. I remember when I started in the industry [kids roll their eyes, “Oh, there goes grampa on one of his stories again”], you’d have these genealogy charts showed how various companies evolved from prior companies. And one of the very few root companies was Fairchild. Anyone who had any time in the industry had, at one point or another, worked at Fairchild.
And then, well, forgive my saying it, but they kinda just disappeared from view.
Well, they’ve now announced their first MEMS product. It’s a six-axis IMU with inputs for an external magnetometer to give nine-axis results. This is more than just attaching another sensor to a bus, since the IMU internals have their own sensor fusion (the AttitudeEngine) for generating quaternion results. They also have higher-level sensor fusion libraries (like body tracking) for execution on a host (those libraries sold separately).
(Image courtesy Fairchild)
And that leads to one of their claimed differentiators. Most IMU vendors specify a level of accuracy for their basic acceleration (linear and rotational) results. Fairchild is specifying the accuracy of their full orientation output, after the calculations. They claim to be the only ones to do that. They specify ±3° for pitch and roll and ±5° for yaw.
They also claim that their algorithms are self-correcting with respect to offset changes (but not drift). And the algorithms leverage human body motion models from Xsens, whom they acquired (and from whom also come many of the body tracking algorithms), allowing enough accuracy to run navigation without GPS input for 60-90 seconds before it gets too far out of whack. (I know, that doesn’t sound that long, but it’s a lifetime in the gyro world, where a few seconds can sometimes be all it takes…)
Other than the fact that we like high accuracy, a primary beneficiary of this approach is power. When calculations are done externally, the sensor data must be sampled frequently for accuracy. With this part, because the calculation is done externally, the output data rate can be slower – saving power. The calculation itself is in the mA-to-10s-of-mA range on a non-aggressive silicon process node with low leakage.
(Image courtesy Fairchild)
The other trick they’ve managed is dual vacuum. As we discussed when covering Teledyne-DALSA’s MIDIS process, accelerometers like some damping – meaning they need some air in the cavity. Gyroscopes, meanwhile, like a high vacuum for best quality. So the accelerometer and gyro chambers have different vacuum levels, with the gyro chamber including a getter to maintain the low vacuum.
They’re also touting through-silicon vias (TSVs) for a smaller footprint, but they’re not using that yet; they’ve put in place a pathway to TSVs. For now, they’re still using wire bonding.
You can find more info in their announcement.