Thanks for the discussion,
I'm using modules as attached on photo (GY-521). It's less convenient to me to manufacture those modules myself, and source the chips form known source. QFN packed is not so friendly to solder. Since I only need like 10pcs a month average.
Actually, these chips are quite solderable using just hot air.
I haven't even used solder paste on mine, just tinned the pads. It helps to not solder the central pad - it must not be connected anyway and is there only for mechanical/thermal relief.
Alternatively, if you are doing 10 or so a month, it might make sense to buy a stencil and use solder paste and a hot plate reflow already.
The thing is that its hard to know modules for whats suppliers are using geniue chips. I doubt that IC manufactures are selling chips form bad batches.
I don't think the chips are fake, these MEMS chips are really difficult to duplicate. It is not a FET or a voltage regulator where you could replace the original with a cheaper relabeled knock-off and nobody will notice the difference. There aren't that many companies in the world capable of making these MEMS sensors. Either you are getting manufacturing rejects, as mentioned by others, or the boards are crap - Google will find you plenty of stories about these Chinese breakouts being unreliable either because of mistakes on the boards, poor soldering or both.
Its also possible to buy a module with 9DOF chip MPU-9250. But that might be the same story.
It is. If you want a reliable breakout for MPU-9250, get the Drotek board:
http://www.drotek.com/shop/en/home/421-mpu9250-gyro-accelerometer-magnetometer.html?search_query=Breakout&results=34But beware, it doesn't fit into a breadboard because the pins are offset in a weird way
Only not convectional thing I do with chips is that I drive the modules on VCC pin form 3.3 (so it goes through 3.3 LDO on the module), and I have pull ups on my PCB also for SDA ans SCL (of course connected to 3.3V). I'm using it with LPC1343 from NXP. Standard circuit using hardware I2C.
Um, that could be actually your problem - who knows what that LDO does in such out-of-spec situation. It may not be simply passing the current through when the Vin is below or at the Vout. If it oscillates, that could cause trouble. Or you might be getting some sort of voltage spike into the MPU that kills it over time. However, without seeing the schematics of what you have done it is difficult to say what could be wrong.
Re pull-ups - do you have pull-ups on your own PCB too? If the breakout has its own pull-ups, perhaps the I2C drivers in the IMU are getting overloaded? But that would require a really very small value pull-up resistors.
When speaking about pull-ups - you do have a pull-up on the AD0 pin, right? I was banging my head over a problem for quite a while when my MPU-9250 was randomly hanging or suddenly producing completely BS data. That pin is the LSB of the I2C address and if you leave it floating, the IMU will randomly switch I2C addresses, breaking the communication.
Maybe somebody have some different module/chip (3DOF or more), that were tested on larger quantities even bought from china? Or some more friendly package with leads (LQFP etc, just not QFN or BGA).
Sadly, no. These sensors are all made for the smartphone market, so they all come in these horrible lead-less packages. However, the Drotek module above is French made.