Hello Fomu backers! You may have noticed that the shipping date on the Crowd Supply website has been slowly slipping. This is because of a manufacturing delay, and today I would like to share with you the reason behind this delay and what we’re doing to fix it. It all has to do with the manufacturing of the actual printed circuit board, which is a complex beast. We have a solution to this problem now, so read on to find out more!

Tiny and Complex

Fomu is a complex board, and that complexity is designed to bring you reliability. This includes things such as TVM (i.e. ESD) protection and a ton of decoupling capacitors. These all mean that Fomu will continue to work even if you zap it, and even if it’s running a very complicated bitstream. In order to fit all of this functionality onto a tiny circuit board, we used very small components.

This sort of thing is normal for products such as mobile phones, and surprisingly wasn’t the source of any of the manufacturing issues. We haven’t had any assembly issues with Fomu, once we got the process established. There were a few hiccups when bringing up the production line, but that’s normal. If we can provide the assembly factory with good raw circuit boards, they can do a good job producing Fomus.

All That Glitters is Hard Gold

Circuit boards can perform multiple roles. Usually, they exist to connect various components together and hold them in place. Sometimes they are used as connectors, as is the case for things like PCI Express cards. Fomu also uses the PCB material as a USB connector, rather than using an off-the-shelf USB connector. It’s Fomu’s signature feature.

With circuit boards, the copper finish will oxidize and form a non-conductive coating when exposed to air. In order to prevent that from happening, some sort of coating is applied. Electroless Nickel Immersion Gold (ENIG) is relatively common, where a very thin layer of gold and nickel is placed on the circuit board. This is an inexpensive process, but it is only guaranteed to be reliable for about 200 insertion/removal cycles. ENIG plating will continue to work after that number of cycles, possibly for a very long time, but it’s still a low number.

With Fomu, we didn’t feel that 200 cycles was enough, so we opted for a different method on the actual connector surface called "hard gold". This electroplates a larger amount of gold onto the connector surfaces, which should be good for at least 5,000 insertion cycles.

The electroplating process involves zapping the circuit board with very high voltage, and this is where the issues happened. The electroplating process wasn’t precise enough, and they ended up electroplating too much. Unfortunately, because the Fomu PCB is so dense, the electroplating process created short circuits between areas of the circuit board that shouldn’t have been connected. When this happens, there’s nothing you can do but start all over again.

Before a factory will even attempt to produce a circuit board, they usually do what’s called a "Design For Manufacture" review. This DFM review usually contains lots of helpful critiques of the circuit board, including areas where they have concerns. When we did the first run of the PCB back in March, they did flag out some issues which we addressed. But they never raised any concerns about the density of the PCB.

The factory has attempted this process seven times now, and they’ve been having about a 20% success rate. This low rate is why we have had to push back the delivery dates — if you have no circuit boards, you have nothing to put components on!

Changes to Improve Yield

Since the issue happens during hard gold electroplating, one option is to get rid of the electroplating entirely. This would certainly solve the issue, but it means that the resulting Fomu is less reliable and will fail sooner. We promised hard gold plating, and the factory has proven that they are technically possible to produce, so that wasn’t a viable solution.

It may be possible to change the PCB slightly to improve yield even more, by spacing out components a little bit more. Changing component position will require us to reprogram all the pick-and-place machines, as well as to recreate the case. This also isn’t a viable solution.

At this point, we’re planning on moving PCB manufacture to a different factory entirely. This is a dramatic step for a small hardware project, because changing PCB facilities requires that you start from scratch and pay any Non-Recurring Engineering (NRE) costs again. It also takes about three weeks to get the new factory set up, so we wanted to avoid switching unless it’s absolutely necessary.

We’ll know whether we’ll need to switch PCB factories by sometime next week, and we’ll have production back on schedule as soon as that’s resolved.