9
\$\begingroup\$

With the way things are headed, more and more functionality moves into a single chip each year. However, one thing that seems to remain completely untouched by this is MEMS devices like accelerometers and gyros.

Despite many device classes practically requiring accelerometers, integrating MEMS into chips seems astonishingly rare, except for a few expensive (and weak) outliers by ST and Bosch. I assume the reason is technical.

In particular I'm interested in the following questions:

  1. What makes them so rare?
  2. Do process differences have an impact on this?
  3. How do the components that do exist circumvent these problem(s)?
\$\endgroup\$
  • 2
    \$\begingroup\$ In addition to the process/cost reasons mentioned below, I'd add that you probably don't want a SoC to integrate this. For the same reason that manufacturers have multiple variant of accelerometers in their catalog: they have different specs/accuracy/price/etc... and you want to be able to select the one that fit your needs. Having the wrong one integrated in the chip doesn't serve any purpose except making it more expensive. \$\endgroup\$ – dim Apr 17 '18 at 15:24
  • 2
    \$\begingroup\$ They don't even integrate MEMS chips into other MEMS chips sometimes. Some of the types of sensors you mention have a number of interconnected dice inside one package. \$\endgroup\$ – Spehro Pefhany Apr 17 '18 at 17:10
  • \$\begingroup\$ You mean IMU's, the chips are integrated via die bonding. Not everyone has that capability. Expensive is a relative term, I think they are cheap at only 7$. If you think that you could integrate an acceleratometer, magnetometer and gyro on on a PCB (which would be larger size) let me know. Keep in mind a few years ago IMU's were larger and much more expensive \$\endgroup\$ – laptop2d Apr 20 '18 at 18:45
9
\$\begingroup\$

If you meant your question as "why do we not integrate them into a full SOC", then I am afraid I do not really answer your question below. Else:

In addition to the reasons already given here is that they not just require extra steps, but compromises on steps. In other words, your MEMS part will not be as good (or as cheap) when it is integrated with CMOS than it would be if it was done with a separate process. The CMOS would not be as good as a dedicated CMOS process either (for example, heating steps of your MEMS part would impact the doping profiles of your CMOS devices. Many cutting steps like plasma etching, DRIE, etc, use large fields and can cause charge to damage devices). However, it is done: take this example from the Melexis MLX90807/MLX90808 pressure sensors (source).

Die graphic

Because of this, it is often cheaper to just use different processes and connect in-package. Here is an example fro mCube (source). You can see two die in the top-left picture. According to the source, the top die is connected to the bottom one with through-silicon-vias.

Die connected with bondwires

An example where bond wires are used to interconnect multiple die (source):

enter image description here

\$\endgroup\$
8
\$\begingroup\$

The same reason why you do not find DRAM memory inside controllers: the process steps to make these are too different from the standard CMOS process.

Even adding something like OTP to a device can mean an extra 4 or 5 process steps which makes the chips more expensive.

I don't know if EEPROM is just about cost effective or if they add these because otherwise they can't compete with micro-controllers which do have them.

\$\endgroup\$
2
\$\begingroup\$
  1. There's no demand for them.

  2. Yes, processes differ. MEMS use process steps like DRIE and wet etching that aren't needed for normal ICs. Including these steps is (extremely) costly.

  3. The existing components don't circumvent this problem, they focus on being components where there's enough demand to (hopefully) offset the increase in price caused by having the extra process steps.

\$\endgroup\$
  • \$\begingroup\$ I think the first point is a contradiction with the second two. The problem seems to be cost process cost, not demand. \$\endgroup\$ – Azsgy Apr 17 '18 at 7:27
  • 1
    \$\begingroup\$ Well, they're inherently linked through supply and demand. Including the processes is expensive. In very specific cases you can win back the investment of having these expensive steps. These cases are determined by how much demand there is. In a consumer driven market, the demand will determine which products are put onto that market. ICs are a predominantly consumer driven market. \$\endgroup\$ – DonFusili Apr 17 '18 at 7:33
0
\$\begingroup\$

Because the accelerometers and gyros need a inertial mass, higher the mass higher the sensitivity of the sensor. Higher the mass, higher the dimensions. Including into a chip increases the cost more that any other peripheral.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.