Making things can be hard. Sometimes making things is extra hard. The past couple weeks has been one of those times.
We got our EmotiBit Beta boards back from our manufacturer and everything was wrong. Although it’s a board manufacturer I’d worked with at a previous startup, on this occasion, of the 84 boards we ordered, only 25 arrived by the fourth day after the scheduled arrival. We tested all 25 of the boards and none of them worked. None. Digging a little deeper, we found 6 different failure modes.
Having just dropped a big chunk of change on this manufacturing order, we would have been pooping our pants at that point, but we’d done our due diligence. In addition to employing a careful review process for every part, schematic, layout, and manufacturing file aspect of the board, we’d also done a bare-board prototype run that we hand assembled and those boards worked 100% perfectly. So we had a high-degree of confidence that we could figure out what was going on.
The first thing we noticed was that the snaps completely popped out of the board as soon as we tried to plug in the EDA electrodes. Digging a little deeper, we recognized that was because the manufacturer had clearly removed the solder paste around the snaps from our design files. It’s a bummer because it’s not super fun hand soldering those big metal snaps that sink a lot of heat, but it was definitely a fixable problem, so we continued looking for what was causing the other failures on the boards.
As we continued looking over the boards we found that one board was fully missing at least one part. Continuing on our failure root-cause journey we realized that many of the failures seemed to be occurring at or near the magnetometer, so we focussed our microscope on that part to see what could be going on. To our chagrin, on many boards the solder job on the BMM150 was bad. Really bad. The part was often shifted off the pads and often exhibited what obviously appeared to be cold solder joints.
Although it was clear that these solder joints could potentially cause some of the problems we were seeing, to prove that it wasn’t a design issue, we had to show to ourselves (and our manufacturer) that we could fix the board by fixing the soldering. So we picked a board where only the magnetometer was exhibiting failure and set about resoldering the part. That’s not a totally trivial matter because the part has 12 solder pads in a 1.5mm x 1.5mm space and 4 of the pads are fully under the chip in a so-called ball grid array (BGA) form factor. It tooks some time huddled over a microscope removing the chip, cleaning up the 0.2mm diameter pads, delicately painting solder paste onto to each, and using a hot-air rework gun to flow each solder ball to its mated pad without any cross-talk to the nearby pads spaced only 0.2mm away. It took some work to get it just right, but my days as a neuroscientist doing micro brain surgery definitely pulled through, and after reworking the BMM150 we managed to get one board in the batch that was 100% working.
We breathed a sigh of relief knowing for certain that our design was good, but unfortunately that was just the beginning of our manufacturing saga. It was super frustrating that we’d paid a premium (~double) for quick-turn manufacturing to hit our deadlines, but sometimes these kind of mistakes happen, so we reached out to our manufacturer detailing the problems we’d found to ship the 25 boards back and get them (as well as the remaining >⅔ of our order) fixed. Our biggest concern was that we get boards that are 100% working shipped to our Beta customers. Because it’s always an act of faith to send product into the world, we asked our manufacturer for a number of data points to assure that the boards we got back were working and would continue to work in the hands of our Beta partners.
Pages and pages of emails and phone calls later, the reworked boards were scheduled to arrive later that week, but Friday came and went with no boards. Finally, on Monday 36 of our 84 ordered boards arrived on our doorstep. We eagerly opened the box like kids on Christmas unwrapped the bubble bag and “POP” one of the EDA snaps immediately fell out. If the board was working, it might well have measured my disappointment. We then set about the dogged business of testing boards. Board 1… fail. Board 2… fail… 3, 4, 5, 6, 7, 8, 9, 10, 11… after 15 failed boards in a row, we decided we’d seen enough. On one board the magnetometer initially wasn’t working, but when we breathed on it to test the temperature and humidity sensors, the magnetometer started working! Eee gad!
This wasn’t going to work. There’s no way we could ever ship these boards to customers with confidence. Although we were now 11 days past the original scheduled board arrival and also past the date we promised we would ship to customers, we had no choice. We had to lot-reject the boards and start over with a new manufacturer. Although it’s definitely not how we wanted to spend our time over the last few weeks and it interposes a few weeks delay to get EmotiBits into our Beta partners hands, we’re absolutely committed to creating a truly outstanding bio-wearable sensor that enables our partners realize their wildest imagination. Even with this manufacturing setback, our EmotiBit Alpha boards are still measuring a high level of excitement and confidence!