The ADXL345 is a MEMS accelerometer made by Analog Devices. It’s a popular device among hobbyists because of its low cost, easy availability and rich feature set. But apparently you should be careful about where you buy them: one of my readers ended up with a bunch of ADXL345s that had significant offsets, measurement axes that didn’t work at all, and an inoperative freefall detection mode. After spending lots of time trying to get them to work, he decided to send them to me instead and hopefully find out what was wrong with them.
The front and back look pretty normal at first glance. But if we slightly change the illumination angle, we can already spot something unusual.
Behind the printed markings we find another set of laser markings. The top and bottom lines say the same thing (“945B” and “PHIL”), but the two lines in the middle are slightly different: “#934” instead of “#935”, and “2029” in place of “9697”. These numbers indicate things like the manufacturing date, a batch number or a factory indentifier and don’t affect the functionality or performance of the chip. However, the fact that there was a first set of numbers which was apparently deleted and replaced by another set is not a good sign. It could mean that this chip was part of a faulty batch meant to be destroyed, but ended up in the hands of an unscrupulous dealer who re-marked the top side and sold them as good parts.
If we etch away the top plastic, we find this two-chip module on the bottom layer. There’s a MEMS chip in the middle and a readout chip on the right, connected to each other by nine bond wires. There’s another bond wire from the top of the MEMS chip to a pin that’s also used by the readout chip, which is most likely the ground pin.
The readout chip is a typical mixed-signal IC. At the bottom there’s a row of bond pads that connect to the MEMS chip, with a three-channel analog-to-digital convert above it. The rest of the chip is a digital engine that drives the ADC, stores the results and sends them out through an SPI or I2C bus.
In the bottom-left corner we find the chip ID, confirming that this is an ADXL345.
In the other corner are the 2008 copyright date, ADI’s logo and a set of revision letters (all “A”s).
If we take a closer look at the MEMS chip, we find that it consists of two layers of silicon glued together. The top layer (which is in focus on this picture) acts as a protective seal for the MEMS layer below.
For now, we can only see the bond pads of the bottom layer. It also shows a 2007 copyright date and an identifier “ADXL335”. That last bit is interesting: it shows that this chip uses the same MEMS die as the ADXL335, which has an analog output. If that’s the case, then only the readout chip is different between these two products. But then you’d also expect the specifications to be similar, which isn’t the case at all. The ‘335 has a maximum range of +/- 3 g in each direction, while the ‘345 can go all the way up to +/- 16 g. It could be that the limiting factor for the ADXL335’s performance is actually the readout chip, or perhaps ADI found out that the MEMS die could perform with much higher accelerations than initially thought.
The outside is a bit boring however, so we should find a way to separate the two halves. If we look at the stack from the side, we see that the bottom layer is a thick slab of silicon, which probably carries the MEMS structure, while the top layer is a much thinner piece of silicon. The two are glued together through some sort of bonding layer in the middle.
... continue reading