EOMA68 Computing Devices

An Earth-friendly way to easily upgrade and fix your own computer

Feb 17, 2017

Project update 44 of 75

Progress, Physics, Taiwan, and More

by Luke Kenneth Casson Leighton

In this update, I bring you up to speed on… everything!


Visiting Taiwan

I like it here: it reminds me of Costa Rica. Huge swathes of mountainous jungle where humans are trying to keep nature at bay with concrete, tarmac, wires and metal. For the most part they succeed. We’re staying 30 minutes East of Taipei City by bus, which is a shorter ride in to, for example NTNU, than many of its students and professors have to take from living inside Taipei City. The location happens to be a "famous tourist trap" known both nationally and internationally for its incredibly smelly "stinky tofu." There’s also a really famous cafe that we encountered accidentally. It’s very strange to be in such amazing art decor quiet and respectable surroundings, eating waffles and the best panini sandwich I’ve ever tasted, overlooking a combination of jungle, mountains, river gorge, concrete, 5-metre-high industrial tanks next to somebody’s vegetable garden, and a building constructed from single-wall bricks and corrugated iron where there’s evidence of at least two people sleeping in a 2.5 x 2 m shack right next to the highway, making a living selling drinks to passing motorists. This really stands out in contrast given the context that Taiwan produces some of the most innovative industrial machinery and electronics in the world.

We’re in a 4.5 x 4 m room that has been extremely conscientiously designed, which was the best that could be found on short notice, for short-term rent, particularly given that we’re foreigners. It’s one of seven such rooms on this ground floor, clearly designed for students. We have "room service:" tiny ants kindly come in from all corners (and some edges) and tell us what we need to clean up. There’s a small fridge with an actual CRT TV sitting on it, a desk with a pull-out tray, above are some cupboards, underneath that some hangers. The bathroom is 1.5 x 1.25 m in the corner of the room and has floor-to-ceiling tiles and a drain-hole in the middle to let water out (a feature I really don’t understand why it’s not available in Western culture). Properly cleaning the bathroom is real simple: use the shower, hose everything down. Done. And to think of some of the fluffy carpets I’ve seen in some people’s bathrooms, where one small "accident" by a young child doesn’t bear thinking about…. Anyway: the room’s compact, functional, and "somewhere to sleep," but the lack of sunlight is exacting a heavy price. We have to eat out, locally every day. That’s okay for a while but it’s getting a bit much.

Setting up a Lab

My friend has a family house where he’s been setting up small-scale prototyping and manufacturing equipment. He justified it very, very easily: if you are making two or three products a year and each one costs \$1500 USD to \$2,000 for the complex ones and \$500 for the simpler ones, and you have to go through at least 2 or 3 revisions per product, that’s a hell of a lot of money and you might as well get your own pick-and-place machine, SMD IR oven, CNC milling machine, 3D printer and everything else needed to prototype and even manufacture your own products.

So he bought a $5,000 commercial PNP machine a few months ago, and already he is fed up with it. It’s fine if you have it set up (which takes almost a day including checking errors made in the PCB design) and all fine if you are making 1,000 to 10,000 units on it, but the bugs in the proprietary software which cannot be corrected are really annoying him. So I am helping him to assemble an OpenPNP design, with a view to modifying and adapting it to cater to the circumstances that he is operating under (not the circumstances that the proprietary PNP machine is forcing him to operate under).

I’ve already thought of a design for a "tape feeder" that can take either a full reel or a strip of as few as 20 components and only wastes about four or five of them if they’re 0402 sized. We’re looking forward to implementing it and, instead of having component reel feeders fixed to only one side, it has a modular design where individual reel-feeders can be removed (and stored, with their feeders still attached), with space to slot reel feeders on all three sides of the machine. This is really important in the context of manufacturing a PCB, because if you have more than 22 different types of 0402 components using this proprietary machine you have to do all your boards with one run (stacking them up carefully so that the components don’t fall off - one knock and you have to chuck it), then change all the reels which takes at least two hours, do the next set, and so on. So the more components you can get round the sides of the machine, the better.

He also bought a \$200 "AliExpress-sourced" CNC milling machine, made from really thick and heavy 10 mm PVC plate and bits you’d find from any 3D printer store. It’s proven incredibly useful and I am amazed it functions so well for such low-cost parts. No end-stops in sight. The downside is the state of software libre for CNC machining has a lot to be desired. BlenderCAM apparently is great, but anyone who’s used Blender knows that Blender was designed to be as user-hostile as possible so that the company who made it could easily justify the training and tutorial courses that they sold. That didn’t work out so well, the company went bust, and the investor agreed to sell the source code for \$100k if the Management, developers and customers could raise that: they did… in just six weeks. But that hasn’t made Blender any less user-hostile. So we are running some proprietary software that has been custom-written to deal with CNC.

Ultimately, my friend would like to be totally independent and return to India, setting up a lab there and contacting universities to invite them to learn and train to make electronics products locally, as he can see the writing on the wall clear as day: India is set to become the place where China outsources its manufacturing. They have a vast untapped labour force, and the cost of living hasn’t been disrupted by idiotic Western journalists blundering about where they’re not welcome, unaware that a $0.50 a day wages for a single bed in a dorm factory where there’s light, warmth, warm water, hot food from vendors nearby and a door that closes and locks is infinitely better than the circumstances that the average Foxconn China worker’s grandparents were living in as little as fifty years ago: \$0.01 to \$0.05 wage a day for working in a corner of a field, sleeping under a corrugated iron roofed open shack with cows for warmth and company, a cold water pipe in the corner with not enough pressure to provide guaranteed water every day let alone every hour, and you starved for three months of the year during winter.

Anyway, I am seeing how amazingly fulfilling it is to be independent, being able to assemble your own PCBs. I already know what it’s like to make your own casework, but to be able to manufacture it… so we are looking up how to do injection-molding, metal CNC machining and much more. I’ve already pointed him at Julia Longtin’s innovative aluminium casting techniques (using 1000 W 2nd-hand industrial microwaves for two hours instead of burning two days worth of propane), I’m investigating OpenSLS because I feel that the level of detail in the laptop casework parts would make it more appropriate to laser-sinter injection-mold parts as opposed to CNC milling them…

… and ultimately this comes down to cost. To get someone else even to 3D print 150 sets of casework @ \$USD 70 each is around \$USD 10,000. If I can set up some low-budget open equipment for even remotely close to that kind of money, it’s a straightforward business decision. The fact that it’s "open" doesn’t even really come into the equation (it does, but you know what I mean).

Intel’s "World first Credit-card-sized Modular Computer".

Thanks to an alert from the arm-netbooks mailing list members, we had a bit of a laugh reviewing the BBC’s "over-excited coverage" of the "Innovative! World First!" Intel "Compute Card." Reviewing Sharp and other manufacturers’ engineering teams’ comments and the various press releases and the coverage on Ars Technica and other places, the similarity to EOMA68 - even some of the wording from the eco-computing whitepaper I wrote two years ago - is absolutely amazing.

There was a bit of a scary moment that they had done something as stupid as re-using PCMCIA without thinking through the consequences of doing so, so the first and most important urgent thing to do was to identify the connector and the interface. It looks like, from about 40 seconds in to the BBC’s CES video, that it’s simply Mini-PCIe, with a shroud around it that anyone familiar with the 2600 and NES Games Consoles will instantly recognize. Mini-PCIe would be a very good choice (if that’s what they’ve used), because it has PCIe (one lane), USB2, I2C and a few (dedicated) GPIOs. There’s one good thing about this design choice: Intel will need to start making dedicated one-lane low-power PCIe 3D GPUs to put into the Base units. If you try to find such ICs, there’s only one: the Volari X10 last seen in the OpenRD Ultimate. It’s 2D only, it’s VGA only, and it’s an insane $12 USD in volume.

The upside of Intel being involved in this modular concept is that by the time it hits the Linux community they’ll realise and deal with the severe shortcomings of the current "it’s monolithic, it’s all IBM PC architecture, we don’t need to take account of any other design strategies, we don’t need to split out device drivers" design approach, and perhaps speed up the devicetree "overlay" patches and subsequent userspace changes as well. Here’s more information on devicetree overlays which is very very important to have in order to support EOMA68 Housings properly. I appreciate I’m being a little unfair with the "It’s All IBM PC Architecture" comment, but that’s just history/reality: the pain being undergone by the Linux kernel development teams at the moment is down to the fact that in software you can’t predict or code for future hardware (because you don’t have any to test with… yet), and you’re always struggling to catch up with the past. Things change, priorities change.

On the mailing list we did a comprehensive analysis of whether or not Intel’s approach would succeed or fail. By the way: to any Intel employees, management, or especially directors reading this: I’m keenly aware that you’ve been subsidising (paying) OEMs to design products based around your current smartphone / tablet SoCs, then subsidising (incentivising) the wholesalers to stock them, then subsidising (with partial refunds) end-users as well to actually buy the products. I can only wonder if the announcement last year that you’re not going to be in the smartphone / tablet space any more probably coincided with a "quiet word" from someone in the FTC. I have told you at some length that the x86 architecture’s compact instruction set, from which you made your money 40 years ago when RAM was so insanely costly it was necessary to make the instructions as compact as possible, that the price you now pay is a huge overhead in the instruction decode engine to achieve a clock-for-clock equivalence with RISC designs. With power being a square law based on clock rate, and having to operate at 30% to 40% faster base clock to achieve the same equivalent clock speed as RISC, you are NEVER going to win this one as it is simply too costly for you to remain at least an entire generation of geometry ahead of the competition.

You NEED to move to a pure RISC architecture and instruction set, and deploy the strategy utilised in the Loongson 3G architecture: hardware-emulate the top 200 most popular instructions and leave the rest to software if you are to be a player in this game at all. However, you already sold off the team that created the DEC Alpha because they produced the PXA architecture that pissed all over the Intel Atom, your pride and joy, and rather than accept that and promote the hell out of it, you artificially limited the clockrate to 600 MHz so that it wouldn’t compete with the Intel Atom, and sold off the entire PXA division as quickly as you could. Marvell bought it, instantly ramped the clockrate to 1200 MHz without making a single design change and made a fortune. Except they didn’t understand that the team who designed it were crucial to its success, so have squandered the opportunity, just as you did.

Anyway, with that as perspective, I don’t expect you to listen, which means the space is clear for people like me to not be concerned about your move into the modular area. Your "Compute Cards" will be Google’s "Project ARA," which, as they didn’t understand either, has to have independent and crucially truly open architecture and standards unencumbered by patents for it to work. Don’t get me wrong, everybody benefits from what you’re doing, but soon the companies that you’re (almost certainly paying and) working with will realise that they’ve locked themselves into an Intel-only strategy, Intel-only chipsets, Intel-only marketing, and Intel-only ecosystem, where, because of the lack of other SoCs which you’ll be actively discouraging because you’re a "competitive business," they’ll soon work out that you could, at any time, pull the plug on both them and the entire Compute Card Concept just as you pulled the plug on the Smartphone / Tablet business only last year, and they’re left high and dry, with the customers going "I bought this over-expensive hunk-o-junk thinking I would be able to upgrade it indefinitely for at least the next fifteen years, WHERE’S MY REFUND or would you prefer a class-action lawsuit based on your deceptive marketing, instead?"

Independent open standards are IMPORTANT.

Anyway. Intel: I look forward to you finally producing an SoC that has the unbelievably-poor and fatal design and security flaws fixed, through the removal or publication of the private keys for the on-board ME, that also is affordable and has a guaranteed lifetime of at least 5 years (unlike all the current offerings which are only around for about 8-9 months and are only available to the "partners" you’ve established) and I particularly look forward to you sponsoring me to design an Open EOMA68 Card based around it.

Call me. We need to talk. Before it’s too late for you.

The Passthrough Card

Using my friend’s equipment, I assembled the first prototype Passthrough Cards, a few days ago. It had taken only a week to design, the PCBs arrived a couple weeks ago. One of them we even managed to put the TFP401a, STM32F and the new JAE mid-mount HDMI connector on, successfully. It was an interesting exercise and hugely exciting to be assembling my own PCB. Using the proprietary PNP equipment - getting it set up - was an unbelievable test of my patience. After checking things thoroughly I applied power and learned rapidly that I’d managed to reverse the order of the USB-OTG pins (5-1 instead of 1-5) and also I’d made a mistake on the SOT-89 3.3 V regulator’s layout. To Be Corrected for revision 2. With a bit of messing about I now have one PCB which is set to be a good candidate for "fully working."

Over the next couple of days, I aim to test it by setting up an EOMA68-A20 to output 800x600 from its HDMI port as a fixed configuration, plug its HDMI out into the Passthrough Card’s HDMI INPUT, and with an oscilloscope see what the TFP401a does. If it’s producing signals then I’ll put a PCMCIA header on the PCB and plug it into a Micro Desktop Housing and hook up a VGA monitor to it. If it doesn’t work then I will have to start writing some quick software, to get the STM32F072 to emulate an I2C EEPROM.

Just as with Adafruit’s board, the TFP401a is normally connected directly to an I2C EEPROM that is programmed with the EDID data that you anticipate using with the timings set up for the fixed LCD that you aim to use in the monolithically-designed product that the TFP401a is in. However, that strategy doesn’t work here, as this is a Passthrough Card, the LCD that you’re going to be using cannot possibly be so specified as a hard-coded entry in an EEPROM, doing so entirely defeats the object of the exercise. Therefore, what you have to do instead is provide a dynamic means by which the EDID data is fed to the TFP401a, which becomes one of the jobs of the STM32F072. In the case of the Micro Desktop, it’s going to have to bit-bang I2C on a pair of GPIO lines that I connected up to the VGA EDID I2C port (deliberately keeping them away from the EOMA68 I2C port), then, having read that data, feed it to the TFP401a on request as an I2C client. When plugged into the laptop housing, it will be able to provide a fixed EDID set of data for the specific LCD that’s in the laptop.

Basically the entire STM32F072 will need special "per-housing" firmware to be uploaded. Luckily this is a software libre project, so the chances of that firmware ending up in debian and other distro’s repositories are very high. It’s going to be a bit like the usb-modeswitch stuff. In fact, it could probably use usb-modeswitch. Hmmm…

Visiting NTNU

I just wanted to say thank you to Jacky and Professor Shen for inviting me to give a talk to NTNU’s Robotics students on such short notice. I really appreciated meeting you. Thank you also for such a wonderful lunch, and also for the gift of the locally-made tea. To you, the backers, as you can see, I was invited to speak at NTNU. This has opened my mind to some possibilities, in particular, the idea of doing a Ph.D. in HEP (high-energy particle physics) and just being around to make use of the huge innovation opportunities and eco-system here in Taiwan. It would be irresponsible and much too forward of me to assume that I can talk publicly about the ideas that are occurring to me, in case it is inferred that I am overstepping boundaries of propriety, here, so let me be indirect, mention that Taiwan is where TSMC is based, and not discuss it further at this point. When and if there is appropriate news with permission and consent to speak, you will be the first to hear about it.

The Extended Rishon Model: 31 Years and Counting

Some people reading this may be hugely surprised by the mention of particle physics on a crowdfunding campaign’s update. It’s been a long-standing interest of mine (31 years and counting), dating back to my time at Stonyhurst College around 1986 when I independently derived what I later learned to be Haim Harari’s Rishon Model. You would not believe the hostility that the mainstream scientific community has towards people who work outside of the peer-reviewed process. Mind you, if you have ever read some of the discussions on sci.physics.particle it is somewhat understandable. With particle physics being such a long-standing interest, during this New Year Break I encountered some papers which caught my attention, and, given that it was holiday time and the China Factory was very busy as was my host as well, I began investigating a lead that I had been looking for, for a long, long time.

Turns out that it was a legitimate one: as a reverse-engineer I had been expecting to find mathematical "signposts" for some time, but, honestly, the maths that is being used in the mainstream community (known as the "Standard Model") is so fantastically complex I can barely understand even a single word let alone an entire paragraph. As an aside, I am telling you, the reader, that I really do know what it feels like to read some of the stuff that I write, with all the buzzwords and techie jargon. One word or acronym takes an entire paragraph - or a day - to understand. In mathematics, those paragraphs and explanations are often given in terms of (if you follow the rabbit-hole) an infinite and recursive loop of self-referring mind-melting, insanity-producing web pages and documentation. So I have been searching for alternative clues that are easier to follow, and I found one, in the field of optics… not particle physics.

Now, if you read up on the Standard Model, looking for an actual basis in reality, there isn’t one. Absolutely nowhere does anyone actually say or describe the actual fundamental "hardware" that our universe is running on. As a software engineer and reverse-engineer, that’s what we call a "Bad Sign." With leads in the High-Energy Physics community being not just a dead-end but with their own peers acting in a flat-out hostile manner, I investigated "alternative" theories instead. This was, putting it diplomatically, an interesting experience in its own right. However, one theme kept coming up: those people who investigated light as being the basis of particles seemed to actually get somewhere.

So that’s the lead I began following (just over five years ago). It led me through the field of optics: some experimental research in "Braided Light" from 2015, an experiment to create "mobius light circles" again from 2015. A paper by a mathematician named Castillo who worked out a way to map Jones Calculus onto Pauli Matrices and Poincare Spheres. I found that in the field of optics, they are making huge advances, making hugely important discoveries, but that if you speak to anyone in this field, and mention "High Energy Physics" or "Particle Physics Theories" then instead of being interested and courteous, they become very very nervous and quickly and politely find a way to shut down an ongoing conversation.

The breakthrough is: with sheer bloody-minded persistence substituting for actual mathematical knowledge, the application of the reverse-engineering tool of searching exhaustively for repeated occurrences of "keywords" (and certain mathematical symbols), I have worked out the mathematical conditions under which it’s possible to "superimpose" circles of light in ways that will not destabilise. I knew it was possible, because the reverse-engineered evidence overwhelmingly said so: I just didn’t know how, up until a few days ago. It’s a hugely important mathematical discovery in its own right, just in the field of optics alone… but getting the word out to the right scientific communities is a task I cannot do without the framework of an Academic setting. The outright hostility of reviewers, even leading them to make comments that clearly indicate they haven’t even read the peer-reviewed references where I simply restate a premise that was already well-known and been accepted for publication decades ago, is phenomenal and a real indictment of the peer-review Journal and publication process as a whole. Not to mention the fact that the mental resources required to understand the current approach are just… fantastically complex, it’s no wonder they’re not making progress. Science has got itself into a pathological state. I’m aware of this. I’d like to help break them out of that so that the resources of these amazing people can be put to better use elsewhere. This is not a new argument: it’s one that’s been voiced in respect to String Theory very often. All I’m saying is: it applies to the Standard Model as well. As a community, they’ve made some erroneous empirical (reverse-engineering) deductions, many decades ago. The theory’s become too complex for anyone to have noticed the mis-judgements, and outsiders are NOT WELCOME to point that out. That’s pretty much the definition of a pathological state.

So, coming back on track: the reason I am mentioning this in the context of the crowd-funding campaign is that I need a context - a reason - to stay in Taiwan. One that has sponsorship or funding, does not involve "employment" (another word for "slavery" in the context of the "intellectual property" ownership contracts so prevalent and accepted without thought for the consequences pretty much everywhere in the world) or other means which would interfere with the role of being "Guardian of the EOMA68 Standard," does not interfere with but instead actively supports my responsibility towards fulfilling the promises that I have made to you, and for which it would be not just relatively easy but also is actively endorsed by its Government’s encouragement of foreigners to come to this country, for the purposes of study.

All of that is a prosaic way of reminding you (and myself) that the shoe-string budget we’re on to develop and roll out an entire computing platform requires us to be resourceful at every turn, even when it comes to where and how we live. From the start, I’ve known I’m going to have to fulfill the promises during this project (not just this campaign) in phases (the A20 Card and Micro Desktop being part of that first phase), introduce new campaigns with upgraded Cards (such as the RK3288 one) in an overlapping manner, and find a suitable country and way in which I can stay financially independent, all at the same time. It’s a huge complex set of inter-dependent conditions, which I love finding myself in the middle of: it’s where I thrive and where I do best, tackling simultaneous avenues and approaches all at once, planning ahead, adapting accordingly so that the chances of success are greatly increased. If one approach fails there are multiple others already lined up to take its place. Taiwan (at least in Shenken) has a much lower cost of living, paradoxically has a hugely advanced innovative industrial base, is right next door to China where they speak Mandarin, too, and has extremely prestigious universities. China was, honestly, too weird. Visas are issued 30 days at a time and may be revoked at any time. Hong Kong is just far too expensive. Taiwan is just the right balance where I can get things done.

The EOMA68-A20 Cards

The Micro Desktop was done, pre-production prototypes are good. The EOMA68-A20 Card however is being a pain. Just that one connector, the mid-mount Micro HDMI, has cost about \$4,000 USD in PCB prototyping to find and confirm, alone. Last time the formerly-found Amphenol Micro-HDMI connector which went obsolete on us, it was around \$10,000 to \$15,000 thrown away. It’s annoying, but part of what needs to be done. The JAE replacement however is perfect. It’s on the Passthrough Card, and on the RK3288 PCB. It’s also much better than the Amphenol designed part which required "solder reflow" (placing it by hand then applying a secondary rather expensive process). The JAE DC3 part has its pins sticking out like a QFP IC, so you can use automated pick-and-place then put it straight into an IR oven to bake it, and it will go on, perfectly. I confirmed that with my friend’s equipment.

What needs to be done is the casework. A new fascia plate from 0.1 mm stainless steel needs to be designed, a jig made to cut away the centre of the plastic PCMCIA casework (that would cost several thousands to redesign, so we went with the preexisting one from the factory) and the plate put in its place. I have a design in mind, what I don’t know at this phase is what techniques to use to get it made up. Whether to pay someone to CNC machine the tools, whether to make a CNC mill and then make the tool myself, whether to pay someone to just make QTY 1000 of the fascia plates, whether to get an OpenSLS built so that I can make the tooling myself, whether not to use a stainless steel plate but to get a DLP 3D printer and make replacement PCMCIA plastic casework from scratch, and so on.

Bear in mind that the fascia plate also has to be designed for the Passthrough Card (there are only 100 of those pledged for), but there is the upcoming RK3288 Card which has to be assembled and tested soon before returning to Zhuhai…. my friend’s approach of buying the equipment and doing things himself is really, really making a lot of sense. The issue I have is, firstly I need somewhere to live and keep all the equipment (I can’t be moving around month after month), and secondly, I need equipment that can do quite a high level of detail in a short amount of time. And at a reasonable cost within the available budget. CNC milling has me concerned that it will simply take too long, and the software is mostly proprietary (and so cannot be adapted).


It’s a long update, because it’s been a long time, and there is a lot going on in many different areas, where this is not a "normal" crowdfunding campaign. Over ten separate PCBs, and four separate sets of casework. I have to concentrate to write these updates in one go (over 120 minutes straight, just for this one), and, knowing that it would take quite a lot more energy and effort than I have capacity for, most days, I’ve been delaying writing this one…. so of course it gets longer as a result. Go figure.

There’s a lot to consider and take into account. I have to think not just how to fulfill this campaign but also to look after myself and my family, as well as plan ahead with a strategy that will ensure the success over the next decade and beyond of EOMA68 as a whole, not just deliver on my immediate promises to you, the backers. That has to be done as well as plan ahead for the success of the larger goal of reaching mass-volume with a long-term, cost-saving, eco-conscious, ethical range of products. Anyone who has some ideas, input or ways in which that can be made easier to achieve, and still be true to the ethical basis on which the project has to be completed, do contact the mailing list or reach out on the usual IRC channel #arm-netbooks on freenode.

Sign up to receive future updates for EOMA68 Computing Devices.

Subscribe to the Crowd Supply newsletter, highlighting the latest creators and projects