Comments by "Siana Gearz" (@SianaGearz) on "MEMS: The Second Silicon Revolution?" video.

I'm lead to believe that a number of MEMS devices are indeed MCMs - multi chip modules, such as 9-axis IMUs. But i'm not super certain. And yeah putting an existing MEMS onto an MCM leadframe makes sense i suppose. But also it may be unnecessary. Have you seen WLCSP packages? Like damn.
3
You probably already have dozens or hundreds of MEMS IC. The most common is a MEMS clock oscillator, instead of large cans it's now tiny barely die sized chips that have saved an insane amount of space, and space is everything in so many use cases. Then a handset typically also has those but also a MEMS microphone or two and of course inertial and magnetic sensors. A number of life saving medical devices couldn't have existed without getting the electronics down to size or the use of MEMS sensors. And bendable parts (flextures) have already changed the world centuries ago, they enabled book binding. So just not really new.
2
They can be i2s and have an internal DAC, and they can use acoustic lowpass as a reconstruction filter, so they can be fundamentally be executed effectively fully digitally, without ever coming up with an explicit voltage or current representing the amplitude of the signal, but driving the MEMS assembly with a delta-sigma signal directly. I haven't looked at MEMS speakers, but i have seen MEMS microphones with i2s digital interface and analogue both, though i don't know that i've seen a manufacturer spill the beans on how they do it specifically.
2
Electrical Engineering has a number of arcane units, and for better or worse, they're going to stay like that. Some are explicitly confusing, like 1 Mil = 1/1000th of an Inch. And you have a wild mixture of component footprints some of which are designed in millimetres and other in Mils, so good luck choosing a layout grid that you don't hate. We don't love it but it is what it is.
1
@punditgi yeah it's down to parts and when you design something in Europe, you can't just be designing in European parts you know, you're going to have a mixture of American, French-Italian, Japanese, Taiwanese and Chinese semiconductors, and in turn they use standards borrowed from different company standards from somewhere else in the world entirely with a different design culture, the footprints are pretty universal across the world and they are universally a mess across the world. Just some company did something and then the rest did something footprint-compatible and it became a standard eventually. Soviet Union tried to put things on a millimetre grid but that was even more infuriating, since you might have functionally compatible and similar looking chips with both 2.54mm spacing (international) and 2.5mm spacing (domestic) and domestic chips were occasionally produced in export leadframe so you never know what you get and the subtle and prone to bad contact misalignment on a 40 pin socket oh my God, and I can't even bear the thought of a 64-pin chip going through that sort of torture... Oh yeah that kinda still a thing in the rest of the world, with some connectors having same diameter pins but pitch of 2.5 or 2.54mm, so best not confuse the footprints or think of a last minute substitution... But at least that's not chip sockets. Newer footprints tend to be based on metric but we couldn't get rid of occasional Mil footprints in 35 years... So yeah that's how it stays. You can tell I built my first computer in Soviet Union - by soldering it together chip by chip. And I had a curse of finding only an imported Z80. But only domestic DRAM. It was the worst combination.
1
Almost all MEMS devices are i2c if they're low data rate digital and i2s if they're audio specific and digital, but analogue devices exist as well. Their packages and pinouts tend to be engineering driven rather than standardised, especially since the industry is pushing for smaller packages, you can have packages the size of the actual die pretty much, but that leaves no room for standardisation. A neat thing about i2c is that you can have a single bus of 2 data wires running throughout your whole device and all the sensors can be on there. Indeed there's an i2c bus running throughout your PC called SMBus which contains a great number of sensors and also your SDRAM SPD chips and it's also exposed onto PCIe connections. However most sensor ICs are not SMBus compliant, they cannot hotplug themselves an address, you can't auto detect them, you just hardwire them an address from the available ones via pin pullups/pulldowns and then do a register read, about a few seconds of looking in the datasheet and hardwired into your firmware according to your hardware, more complexity really isn't generally desired by the hardware designers. What use-case are you thinking of? USB HID is suitable and Bluetooth (LE) GATT is intended for sensor use.
1
@vputilov As long as there are no address collisions, and every address corresponds also to a particular type of device, you can treat i2c devices as polled hotplug. The addresses aren't software determined, they are determined in hardware at board layout stage, or the device can include address set jumpers. You can also get SMBus to i2c chips which handle just the dynamic address assignment (the adapter chip talks to a sensor on a fixed address, and to host it just gets a valid address from the arbiter) and device type query.
1