Thank you to absolutely everyone who came to FOSDEM. RISC-V gets interesting. A reminder that EOMA68 is a Certification Mark. EOMA68-A20 card supply chain progress.
It was hectic but absolutely wonderful to meet so many people at FOSDEM. I actually had to keep a public wiki page to track where I was supposed to be and whom I was supposed to meet. Fortunately there were some wonderful volunteers who came along and helped run the stand: I am so grateful for your help and for the opportunity to meet you.
For people who weren't able to attend or who had other talks to go to, the wiki page above contains links to YouTube videos, including playing Bodhrain, fiddle, and spoons with Howard Chu and his friend, Martin. The one thing that I didn't have time to check was the actual contents of the talk that I gave on Sunday! It was so hectic, with so many people to talk to, with a virus of some kind making it harder and harder to actually speak. So I apologise, the intended published talk about EOMA68 doesn't match up with what I actually said.
One thing: I got a chance to donate the GNU Guix team with an EOMA68-A20 Computer Card and Microdesktop. The reason why: they are an important strategic Libre Software Team that intend to help the project by making sure that the OS that they are developing is available for the EOMA68-A20 Card. To be clear, as I already have been on the mailing list, there are a LIMITED number of Cards available for allocation for SPECIFIC strategic purposes that are clearly, demonstrably, and directly beneficial to the EOMA68 Project, and that the pre-production Cards are NOT available for "general purchase for end-users". If you would like one of the (small number of) EOMA68-A20 2.7.4 Cards, you will need to make it clear - publicly
Friday before FOSDEM 2018 I got to attend a birds of a feather session for RISC-V, and got to privately see the SiFive 64-bit developer board that had literally just been assembled. They actually managed to use it to run the presentation that they later gave. Deeply impressive. I met Richard Jones (fedora-riscv maintainer) and Karsten (rebootstrap and debian-riscv maintainer). I also had the opportunity to meet with some people who suggested that doing a secure RFID / NFC embedded 32-bit RISC-V processor for actual proper secure auditable crypto would not be a dumb idea. I quickly pointed them in the direction of the rhombus tech pinout page, made them aware of the librecores project and the Madras RISE Group's 180 nm foundry, and apart from that forgot to get their contact details. Oops. So much to do in such a short span of time.
Then, in the past few days, which have been extremely hectic with a lot of travelling with very little actual opportunity to fire up the laptop properly, I met with some old business associates whom it turns out were, when working for VLSI many decades ago, the people responsible for ensuring that ARM's early designs were actually manufacturable (translation: before they were presented to VLSI, they weren't manufacturable. And didn't work. At all). I am extremely lucky to know them, and in an interesting twist one of them, Norman, was the original person who was prepared to help with the 2012 effort to create an FSF Endorseable processor.
Now, the thing is that gaining the trust of Foundries is itself extremely difficult. Minimising risk and the cost of mistakes even more so. With Norman on board, he could not only help advise and ensure that the pitfalls are avoided but also he is experienced enough to be extremely well-respected throughout the entire ASIC Industry. The thing is: he's semi-retired, and is well off enough not to be persuaded by offers of money. By actually showing an interest in the stories that he was willing to share, and also by demonstrating that I did actually genuinely appreciate them and also understood their significance, and also by sharing some of the missing pieces of the puzzle about the link between DEC, StrongARM, Intel's PXA Division and a few more things besides, I believe I may have actually managed to convince him to, well, if not actually help then at least maybe participate from the sidelines.
One important piece of advice he gave was: for goodness sake do a test chip in something nice and big like 180 nm (or its derivatives). He informed me 180 nm is not only by far the world's most popular and lowest-cost process (an 8" 180 nm wafer only costs about $600 USD) but also several improvements to the 180 nm process have allowed the same equipment to go down as far as 110 nm in some cases. And it may be the case that you can do a DDR3 PHY in 110 nm.
Oh, I also had an opportunity to speak to Madhu's team lead for the RISC-V main core: Neel. They have only just finished the TSMC 20 nm tape-out, and are taking a bit of a break. I had a brief opportunity to get across to them that it is critical that this entire project be run off of libre-hosted services, particularly as anyone who wishes to contribute from the libre community to the software will take one look at a proprietary-hosted repository, snort or barf or fart loudly, and never enquire about the project ever again.
There is much to communicate and coordinate here! I haven't even had a chance to bring Madhu's team - or the anonymous backer - up to date properly on the various efforts and input from the other teams and sources!
A discussion came up on the mailing list, and it was a bit of a surprise. EOMA68 is a Certification Mark. I will do a more in-depth exploration and explanation of this another time, but basically it overrides the GPL (and any other software or hardware license) and any expectation on the part of hobbyists and businesses that they may do what they wish with it is false. The most critical reason is for user safety, where batteries are involved and a mistake by someone not correctly implementing the standard could cause a lithium battery fire.
Lastly, for the EOMA68-A20 Card, Mike is following up on pricing for Samsung DDR3 RAM, which appears to be available. I want one other guaranteed supply of a completely different RAM IC and I will tell him to go ahead with the 10 samples... but not until we have an absolute guarantee of availability. Then once the samples are done we have a clear and extremely simple and straightforward production schedule to follow.