ICM-20948 Breakout for SlimeVR

VR body trackers are expensive, so I figured what the hell, why not make my own? SlimeVR has a cool offering, but with supply chain issues they’ve had trouble getting things shipped out and have gone through a few board redesigns trying to use what parts they can source. They’ve open-sourced their software (with plans to open-source the hardware, but no date on that yet), and the idea of not completely reinventing the wheel is pretty appealing to me. I started reverse-engineering their schematics based on the pictures I could find of their trackers, but in doing some research stumbled on a very cool project- TheButlah’s ferrous_slime.

To summarize what’s already in the documentation, it’s an open-hardware implementation of a SlimeVR compatible board, based on the newer ESP32 C3 which is faster, more energy efficient, and based on RISC-V rather than xtensa, which means compiling code is much easier, and supported by more languages (such as Rust!). It’s got on-board jumpers to support a number of different IMUs, and a 2.54mm header for a primary IMU daughterboard. After some comparison the supported IMUs by the SlimeVR firmware, the most-accurate, low-power IMUs I was able to source were a handful of ICM-20948s from Alibaba. Support was still marked as “experimental” in the firmware, but I was ready to iron out some bugs if it meant building my own trackers on the cheap. I placed an order, and got to work designing a breakout.

Building the Breakout

Most IMUs currently supported by SlimeVR’s firmware operate on 3.3v- the same as the ESP at the heart of the tracker. (Un)fortunately, the ICM-20948 operates at 1.8v- that’s great for battery life, but it means that a breakout is a bit more complicated as we need to do some logic-level translation to enable communication between the ICM and the ESP. I never made it past the second year of my EE degree, so my circuit design skills are rusty at best. In particular, I had no idea how to choose a logic level converter (open drain? tri-state? directionality?) that was suitable for I2C, the protocol the ICM and ESP use to communicate. Thankfully, GitHub user dtornqvist had already designed a breakout for the ICM-20948, a DIP format breakout with on-board 3.3v logic level conversion performed by an NXP NTS0104. This converter was available in abundance from the usual suppliers and so, armed with that information, I fired up KiCad and got to work.

Revision a

My first attempt was certainly an attempt. It wasn’t until after I had submitted the order that I realized I had used a 1.27mm pitch header, rather than a 2.54mm one (no wonder the IMU looked so big). Additionally, I vastly underestimated how small the BGA package I chose for the voltage regulator would be (0.5mm is, in fact, very small).

The NTS0104 comes in what I believe is a proprietary package from NXP. I had to build the footprint myself, but thankfully the spec was publically available, and Mouser even had a .step file for the 3D model.

The first batch showed up and I attempted to assemble one so I could at least test the rest of the board and debug the I2C communication to make sure the IMUs were good (you never know with Alibaba stock sometimes).

This quickly became more work than it was worth, and after an hour of frustration I went back to the drawing board to fix my mistakes and order a new batch. No more BGA, the correct header, and a slightly thinner profile so that there wasn’t any overhang beyond the ferrous_slime board. A little branding wouldn’t hurt, either…

Revision b

The second attempt was much better. Routing on the smaller space proved to be a bit of a challenge (though nothing compared to another project I have in the works), but ultimately it came together without too much headache.

The silkscreen resolution on a board this small clearly loses some definition, but it’s still something I’m very happy with! The manufacturer that did the PCBs somehow ran out of both the voltage regulators and logic level converters in the time between orders, so I had to source them myself and assemble them via hotplate reflow. Thankfully, even with my less-than-surgical fine motor skills, the new components were all large enough I was able to place them by hand and get them attached to the board.

There are still some issues I need to look into, such as the charging LEDs not working. Electrically, they seem fine, and the LEDs on my boards aren’t burned out, so I think perhaps my BOM didn’t specify the correct voltage for the LEDs and the charger IC isn’t capable of driving them.

I grabbed some cheap LiPo batteries, the typical 600mAh form factor is perfect for these boards. There were a few failures (I ordered enough to make 10 boards) and lost components as well as failed assemblies meant I could only finish 7. Conveniently, I only needed 7 for complete full-body tracking- so there wasn’t any functionality loss. After tweaking the firmware slightly and programming the ESPs via the built-in USB port, everything fired up seamlessly and worked without a hitch, much to my surprise! I did have to fix a couple of components (this was my first hot plate reflow project), but nothing too painstaking. There was only one thing left: making them wearable!

Designing a case

Somehow, it feels like 90% of the CAD I do for making parts ends up being fabricating cases for bare PCBs. I have zero formal instruction on CAD in general, so bear with me. I recently moved from a CSG workflow to a Parametric one, and while it’s definitely been the right choice, I still have a lot to learn.

I started the design process with simple goals- building a small case that would hold everything, not obstruct any ports or LEDs, and be ergonomic enough that wearing it for hours at a time isn’t painful. Trusty calipers in hand, I grabbed some measurements, added some tolerances, and started building.

At this point in the process, I sent it to a few friends to review, and got one consistent piece of feedback- ditch the strap anchors, and add a 1/4"-20 thread. See, initially, I had planned on just buying some velcro strapping and running it through the anchors to strap it to myself. As is always the case though, I never seem to know when a problem has already been solved. Commercially available straps with an embedded 1/4"-20 bolt are widely and cheaply available, and certainly much more comfortable to wear (they’re used for the Vive trackers, for example).

Putting threads directly on a 3D printed part is generally a bad idea. In my experience, they’re never as smooth as I want them to be, and require substantial cleanup to be usable. They’re also not very durable. Instead, it’s common practice to embed a hex nut into the part. So, I simply cut a hexagonal pocket in the bottom of the case, added an oversized through-hole for the bolt in the bottom, and thickened it up a bit so there was clearance for the battery (would be a shame if over-tightening the bolt caused a fire). I cut a slot in the top and modeled a slide-in lid with hexagonal cutouts so you can see the LEDs (they’re not placed on the sides of the board, so side-view holes aren’t an option without some optics wizardry).

And off to the printer it went.

You can clearly see that my printer needs some adjustments, but I am very happy with the results. I ended up printing everything at 102% scale, and with -0.1mm of horizontal expansion to get everything to fit nicely. Sadly, halfway through printing the cases my printer broke, so I’m not quite finished yet. The nut works great, and I’m able to secure it squarely to some old camera tripods I have lying around, so I’m confident it’ll work nicely with the aforementioned straps.

Conclusion

That’s about it! TheButlah was kind enough to review my design, and after ensuring there weren’t any license violations (thanks for the conformation, dtornqvist!), the breakout design was upstreamed to the same project that the ferrous_slime board is in! You can check it out here. I’ve exported a CPL and BOM, so you can have these ordered yourself alongside some ferrous_slime boards to build your own trackers- just be wary sourcing the ICMs, as they’re in the NRND (not recommended for new designs) lifecycle phase. I may end up doing some research on more available IMUs and making another breakout, as I have some friends who want trackers like this.

The case is available on Thingiverse if you get this far, and want to make some yourself!