posted by Bryon Moyer
Changing wafer size is a big deal. You can kiss all your old equipment good-bye and usher in a whole new suite. So what happens when you’re planning to use that wafer size for a new technology node? You really don’t want to have to have two sets of production equipment, one for each side of the wafer-size shift. But it would also be rough to develop a new wafer size at the same time as developing a new technology node. That’s risk upon risk.
I talked with Ludo Deferm at Semicon West, where 450-mm was all the rage. But this excitement is clearly about things yet to come: there’s not much equipment available yet; only one item – KLA-Tencor’s blank wafer metrology unit – has been announced. And that just ensures that you’re starting with a good blank wafer. The rest is yet to come.
And imec sees 14 nm being the starting node for 450 mm. But the 450-mm R&D facility that imec just got government help for isn’t going to be started until 2014 – you can do the math on when it’s likely to be up and running. So if we had to wait for that before we could develop 14-nm technology, we’d be a long ways away.
As it is, imec is doing 14-nm development work on 300-mm wafers – it’s just that that equipment won’t be used for production. It’s just to get the process itself up. Clearly it will take some freshening up on the new 450-mm equipment when it’s ready. By that time, they’ll already be developing the 10-nm node.
As a curious side fact, he noted that a 200-mm cleanroom is actually more expensive to build than the 450-mm facility. That’s because, back then, the whole room had to be clean. Now everything is sealed in FOUPs, so, while it’s probably not a good idea to be tracking mud into the room or smoking, the level of cleanliness in the room is actually less than it used to be. Inside the equipment, however, there’s little forgiveness for the slightest intruder.
More on the Flemish investment can be found here…
posted by Bryon Moyer
We’ve spent a lot of time talking about MEMS in these pages, but not all sensors are MEMS. At Semicon West, I got to talk to Becky Oh of PNI Sensors. They make geomagnetic sensors with noise characteristics around 30 times better than Hall-effect sensors, as she tells it. But they’re not MEMS sensors.
They use a material whose permeability changes with the magnetic field. They wrap this material in a coil and then put the resulting variable inductor in an RLC tank and use the oscillation frequency as an indicator of the magnetic field.
Of course, like all sensors, they need software. They include low-level code that corrects for local magnetic anomalies, although, on its own, the sensor needs some movement to calibrate itself. Combined with a gyro and accelerometer, they can distinguish between an external field change and actual movement of the sensor.
Given the discussions we’ve had on sensors being different from each other, making universal fusion harder, her perspective was that the data returned by sensors are generally quite similar: it’s the APIs that tend to vary. And, of course, software uses APIs, not data directly, so those calls end up masking the similarities between sensors.
Given the size of their sensors (not huge, but not MEMS), they’re not looking to sell into phones. They have background in military, navigation, and virtual reality applications; they’re looking to grow further into games and TV controllers. These are somewhat more forgiving than some of their earlier markets in that they are, more or less, pointing applications, and absolute heading isn’t critical for those apps in the way it is for navigation apps.
They’re building a new fab in Santa Rosa, CA. It seems to be part of a trend to bring manufacturing back to the US – as long as the manufacturing line itself doesn’t require hiring any real people. The key to these fabs is automation: everything is done by computer. There is a need for a hundred or so well-trained people that can work (and repair) the equipment, which is less than would have been used in the past or overseas, but more than would be in the US if they remained overseas.
Even though they aren’t MEMS, they’re joining the MEMS Industry Group, since there’s a lot of commonality with the ecosystem and other players there. And it appears that the Group will let them in…
posted by Bryon Moyer
I recently got a chance to talk with Hillcrest Labs, another big mover in the motion market. In fact, their pedigree sounds remarkably like that of Movea, whom we’ve looked at in the past: starting with interactive TV and transitioning to broader motion.
They’ve gone on to develop a gesture library (released, but, to date, un-officially-announced) that includes more than 50 gestures, including the typical control gestures, numerals, letters, cardinal directions, and rotations.
We discussed the usual sensor fusion stuff, which, of course, is now bread and butter for them. They do this in a sensor-agnostic way, writing drivers from the data sheets and using an internal lab to characterize the sensors to obtain data not available in the datasheets.
The drivers exist at the lowest level of the software stack, protecting the upper layers from the specifics of different sensors. Sensor fusion is above that. But there is actually something they point to between the driver and the fusion that they feel themselves to be particularly strong in: calibration.
And this isn’t just individual sensor calibration; they do it across sensors (so, for instance, a compass and gyroscope can cross-calibrate each other). This might sound like fusion, but the goal is different. The goal of fusion is to use multi-sensor data to get some higher-order information. The goal of calibration – even if done across sensors – is to make sure the data itself is accurate.
A few months ago they announced their official entry into the mobile market; you can see the details of that in their release.