Pre-production prototypes of the Microdesktop and EOMA68-A20 boards were initiated last month, samples were completed last week and tested. Short version: another pre-production run is needed.
It’s just one of those things, but the decision has to be made to do another pre-production set of prototypes for both the Microdesktop PCB and the EOMA68-A20 PCB, both for different reasons. The sequence of events went something like this:
So a combination of classic mistakes, as well as finding out that the two key components (Micro-HDMI mid-mount and Micro-USB mid-mount), which took eighteen months to negotiate and find suppliers for from outside of China, have much better, much easier-to-get alternatives.
The good news is that the 2 GB RAM upgrade worked perfectly at 384mhz. This was a high-risk modification to the DDR3 RAM Layout, learning from the Cubietruck’s schematics how to support four DDR3 8-bit RAM ICs. It’s something that’s not documented by Allwinner! Some undocumented pins had to be used, in particular an extra address line (A15) had to be added, moving aside other tracks in what is an extremely complex layout with timing-critical track lengths, to make room for the extra line. I’ll be absolutely honest: I really really don’t like doing DDR3 RAM layout. It’s risky, it’s fantastically complex, and it requires some extremely comprehensive and thorough analysis. Luckily, the risk paid off, and we can go to production with 2 GB of RAM.
The other good news: the previously-selected mid-mount Micro-HDMI connector was a compromise as it was a "reverse" type. The "reverse" type is at a different height in the middle of the PCB from the other two components (Micro-USB and Micro-SD), meaning that aesthetically it was less than ideal. By being forced to use Runde’s non-reverse-type of mid-mount Micro-HDMI, the three connectors are all inline, resulting in better aesthetics as well as a stronger face-plate.
The mistake made on the Microdesktop v1.5 PCB was just one of those things. The initial schematics were done nearly three years ago: it’s literally taken this long to spot the mistake! It was only discovered through comprehensive testing of the SD card interface, switching off the use of the "Detect" GPIO and resorting to "polling" mode (a mode which the sunxi 3.4 kernel is capable of supporting). In "polling" mode, the 2nd SD card was detected perfectly at 1-bit, 2-bit and 4-bit modes. However when using the "Detect" GPIO, it failed to be detected. Only then was it noticed that the GPIO selected was not an EINT (external interrupt).
I do find it hilarious that as the designer of EOMA68 I can make these kinds of basic mistakes. It’s good that I do so, so that I can emphasise to other people - before they spend their money on PCB design - how critical it is to double-check these things!
The other thing is that we’re taking the opportunity to switch to Runde’s commonly-available USB-A connectors for the Microdesktop PCB. It’s almost identical to the TE part, but the "peg" holes are in different locations. I’ve simply added extra ones so that either connector could be used. They don’t have the VGA D-Sub themselves but have agreed to source one for us, given that we’re using them for supply of literally every single connector on every single PCB designed so far, they’re happy to do this. It’s quite funny that they have absolutely no English at all, and I can’t speak Mandarin. Mike’s kindly acting as translator.
I’m basically done here in Shenzen for this round: Mike will sort out the new revisions of both PCBs, and through ZhongweiSemi, I’ll be waiting for Nexell to do the PCB review (including Simulation) for the EOMA68-S5P6818. Whilst that’s all happening, I’m returning to Hong Kong tomorrow, and will be applying for another visa (business visa this time) - actually three visas, one for each of us in my family. The remaining luggage and family will then travel to Zhuhai, where we’ll get set up and settled in. By that time, Mike and ZhongweiSemi should respectively be ready, and it will be time to return to Shenzhen - a place I am really beginning to enjoy, although it is extremely weird and I will never be able to call it home. Knowing this makes me appreciate my time here even more.
Other bits and pieces:
EOMA68 is too big for smartphones. EOMA50 was designed in parallel with EOMA68, but was shelved pending finding suitable processors. Now that one’s been found, the plan is to do a hybrid phone.
However, none of these ideas will interfere with the immediate tasks of the next few weeks and months, unless someone happens to step forward with sufficient sponsorship funding in order for me to find and pay someone else to do the PCB and Casework designs, while I focus on sorting out this current crowd-funding campaign. We’ll see what happens.