Project update 41 of 75
This update covers: returning from Zhuhai to Shenzhen, meetings with Nexell that went extremely well, the production plan for the EOMA68-A20 and Micro-Desktop with Mike currently doing PCBA.
After signing the rental contract and registering with the police for a "Foreigner Temporary Accommodation Certificate", as advised by the agency as it apparently makes applying for visas easier, there was nothing more that needed to be done in Zhuhai (except wait for Allwinner to make decisions), so I made my way to Shenzhen.
There are three (sensible) ways to do this: ferry, 200 km/h train, and bus (or, coach). Just for the sheer perversity of it, I opted for the bus. Around 80RMB ($8), this took two and a half hours. No toilets on-board, no stopping. Err… luckily the back was completely empty, there were curtains that could be drawn all the way round and I drink a lot of water so… err… never mind. Anyway moving very very swiftly on: the actual journey was totally uneventful, the entire back seat row completely free (so I could lie down)… the only problem being that there was a slow leak from the engine compartment, pushing a tiny amount of exhaust fumes into the back. Being tired, I tolerated it and adjusted the air conditioning accordingly. So. I’ve thoroughly "explored that space," as the Zen Buddhists like to say. I shan’t be doing it again.
Turns out that Nexell and Samsung have a long history of which I was completely unaware. If you recall the GPL-violating CT-PC89e with the S3C6410 processor, that processor was actually designed by Nexell and dual-marked as a Samsung processor! Whereas I previously believed that it was impossible to gain access to Samsung SoCs due to them setting a MOQ of 50,000 and resisting dealing with China, the actual reason is because of the relationship with Nexell. So if you want access inside China to the same processors that are dual-marked "Nexell" and "Samsung", you go to Zhongwei Semi. (If you’re reading this and would like an introduction, contact me I’ll be happy to do so.)
The day after arriving back in Shenzhen I’d arranged to meet Yong-Soo, Zhongwei Semi’s CEO. I was amazed at what unfolded very quickly, as he is clearly an extremely smart, very laid back, and genuine person. We sorted out an NDA, which I signed the following day, and that afternoon began work on the EOMA68-S5P6818 schematics and PCB layout. Their K-Project Reference Design module is 70mm x 48mm. EOMA68 PCBs are 78.1mm x 47.3mm. Adapting the schematics for a first revision took one day. Adapting the PCB for a first revision took around three days. It’s only been a week and the entire layout is almost completely finished!
Now, here’s the thing: this is an 8-core 1.4ghz 28nm processor where the PMIC does 3 AMPS at 1.5 volts for the DDR3 RAM. The layout is also using four DDR3 x16 ICs, which for the four gigabit variants - 512mbytes RAM each - are currently around \$2.30. Remember the prices for the eight gigabit variants? \$9. NINE dollars for a RAM IC. The design’s also using eMMC so it can go up to 32GB storage. The processor cost I haven’t got yet (and won’t be able to tell you what it is), but it’s not going to be cheap. This is going to be a monster board, basically. It may have to be priced somewhere around \$120 - double that of the EOMA68-A20, with an option for 32GB costing even more. We still have yet to explore the power budget, and it may be necessary to ramp down to 800mhz on DDR3L and to cap the processor speed to 1ghz or to have a special cpufreq setup which limits 4 of the cores to say 300mhz and allows the other four to run at the full 1.4ghz. We just have to see what happens.
The next phase here is to get the schematics signed off: Zhongwei is working closely with Nexell back in Korea to do extremely comprehensive and thorough reviews. Despite how small their company is, I’m deeply impressed with their technical competence and the resources they have access to. Once the schematics are signed off, the PCB review will be next, followed by running it through a Simulation. that’s a first for me! Clearly they have access to some heavy-duty software, as part of the collaboration with Samsung. Manufacturing of 5 samples we’ve agreed to do (normally they do MOQ 20 which is too much: I can’t justify spending $10k), as well as a MOQ of 1100 units (because the RAM happens to be ordered in trays with weird quantities). Basically I could sense that they preferred the manufacturing to be handled by their own factory (for reasons unstated but clearly there), so decided to go with the flow.
It’s worth emphasising at this point: no, if you’d like an EOMA68-S5P6818 instead of an existing pledged EOMA68-A20, you can’t - sorry. The reason is very simple: Mike’s factory has made room in their extremely busy schedule to manufacture the planned and clearly stated production run. I’ve negotiated the components with the suppliers. They’re expecting that money to be handed over. They’ve put a lot of time and effort (eating into their profit margins) into preparing the quotes (which are both time-sensitive and quantity-sensitive). If I tell them, "I’m sorry, we’re actually going to only be ordering 50% of the boards we agreed to… but in six months we really really really promise we’ll order some more boards even more than the previous one I just reneged on our deal with you over, really really, I sooo promise!!!", how well do you think that’s going to go? Do you think they’ll ever speak to me again? So we’re sorry, but we can’t offer a \$120 board which is still in the planning and prototyping phase instead of the \$60 one that’s entering production in the next few weeks.
Importantly, it’s worth pointing out that, thanks to the past five years of checking and re-checking the EOMA68 Standard, the new EOMA68-S5P6818 will work perfectly with the existing micro-desktop and laptop housings. I’ll make absolutely sure of it, as that’s exactly what EOMA68 is all about: making it possible to upgrade without throwing away perfectly good hardware.
Overall I’m just amazed by Zhongwei, especially when compared to my experience with Allwinner. From meeting Zhongwei’s CEO, meeting their team (ten people) I’m pretty much done within eight days; getting the PCB done and assembled should be around 2 weeks, we can begin testing end of November, because the source code is publicly available. Now compare that to four years of reverse-engineering effort required with Allwinner. Allwinner’s early success has made them believe that they can succeed as a software company, which is setting themselves up for some extremely tough lessons now that they’re entering the IoT market with TinaOS. If you thought it was bad with ten million IoT devices with default passwords "admin, admin" being used in a DDOS botnet to attack DynDNS, wait until that number goes up an order of magnitude. The Internet pipes between China and the West are already saturated 80% with botnet traffic as it is.
Nexell, on the other hand, by working in partnership with Samsung and cooperating with a small, dedicated Chinese design company, Zhongwei Semi, can be much more effective because they know they haven’t got time to mess about trying to keep the software a "secret”. Their software is public, their datasheets: public. Cooperation with FriendlyARM: public wiki with schematic PDFs and HOW-TOs on installing Debian.
I’m loving it, here. The typhoon that hit 100 km away was enough to cause the roof to leak (we’re on the top floor of a 30-story apartment building) and the ceiling plaster on the concrete roof to fall down: I just thought it was hilarious, and went and got a mop and some buckets. The people here are great: they’re relaxed, enjoying life. I just wish they didn’t smoke so much, but they’re young so they believe they can get away with it… so they do. I’m slowly training the people in the 10-bed dorm to switch lights out, not bring cigarettes into the room. One guy has really bad sleep apnea: I’m going to have to find someone to help translate to explain to him that he seriously needs to get that sorted.
Food: for several weeks now I’ve got away with coconut milk, rice, noodles, vegetables, eggs, seaweed snacks and the occasional pork sausage. The second day into beginning with the EOMA68-S5P6818, I was craving salmon and even more coconut milk. I found a local Japanese Sushi bar and absolutely stuffed my face for a whopping $12, then went down to the supermarket and bought two litre cartons of coconut milk. I’ve been back every day since. Apparently when I do intensive PCB design work for twelve hours a day seven days straight I must be needing a hell of a lot of Omega 3 and 6, as well as MCT oil (MCT is what your brain is mostly made of).
So, the basic plan is: do the Micro-Desktop and EOMA68-A20 production as soon as possible (planned production for November if it’s convenient for Mike), put the EOMA68-S5P6818 into prototyping and testing (talking to Crowd Supply to see if they’re happy to put it into another crowdfunding campaign). The Laptop PCB prototyping can then take place in an overlapping fashion in parallel with those pipelines, and I can also get the breakout boards and the passthrough card designed. In talking with Crowd Supply’s CEO Josh Lifton, I’ve learned that when a project starts shipping, they see a lot of new orders, which makes sense. However in this particular case, I am reluctant to take additional orders for the EOMA68-A20 (unless they’re a client who would like MOQ 2,000), and would prefer to go ahead with the S5P6818 instead. But, we have to see how this all works out. With Zhongwei / Nexell’s help I think it will go well. And Josh and his team need to be happy with the plan, as well. Lots more going on, behind the scenes - for another time.