Amino and HW_Slices

[dan]

Thanks Al, I think I have a good view now on what amino is about.

I guess the main keys to open source hardware is to have well defined connectors and SW components that map to them.

The rest is horse for courses.

[Folknology]

Well I hope the work we have done over this period will help towards the general Xmos opensource efforts, sharing is important.

Do you have any schematics, design or general info for the slices idea, work that you guys have perhaps already experimented with, more details would be appreciated, I am especially interested in how you intend connecting things and using what ports etc..

regards
Al

[dan]

We're coming round to the idea that the main thing to get nailed down is the types of connectors and which pins of them are used for what. We may for example have a small connector like on amino and a larger one which exports more or less a whole core's worth of IO.

We have some initial connectivity thoughts in a table which I am currently formatting into .rst. So, not a schematic or design yet, just some ideas in a table. I'll stick it in the slices repo later today.

The thing to realise is that this slices project is still very embryonic, but we hope to firm it up quite quickly, not least because we are in need of more general XS1-L based development boards both for our own use and for our customers and community. This means its an ideal opportunity to get ideas from xcore users incorporated.

[Folknology]

We have some initial connectivity thoughts in a table which I am currently formatting into .rst. So, not a schematic or design yet, just some ideas in a table. I'll stick it in the slices repo later today.

Cool I will take a look later.

FYI we intend fixing the Amino slot/blade pinouts this month and there will likely be only minor changes, config and IO power lines may revert back to a slightly older Amino design as the new version has thrown up some issues already, 2 steps forward, 1 step back ;-).

regards
Al

[dan]

Ok, take a look here:

https://github.com/xcore/hw_slicekit_sy ... nector.rst

[Interactive_Matter]

Ok, take a look here:

https://github.com/xcore/hw_slicekit_sy ... nector.rst

Hi, I had a look there and I have some (limited) experience with the XC-1A and XC-2.

Regarding the conectors I am a bit in trouble to understand it right.
Basically there are two different connectors - why?
What I really like about the XC-Kits is that the connectors are more or less interchangeable. So why don't make 5 connectors, that all look the same?

I know a good argument against it is that it would make it hard to combine the connectors (e.g. using two of the 5 to get a full 16 bit port). I still hope that this can be solved with a clever hardware design (e.g. if you use 5 different IDC connectors you block some pins because they are bigger than 0,1" headers - but if they just block some additional GND/VCC ports it is bearable if they are accessible elsewhere.
My experience with FCC connectors are somewhat limited - especiall regarding plugging/unplugging often - since that is what I do with my dev boards often.

What I also really like and (somewhat dislike) is the ability of the XC kits to provide 5V. The XMOS seems to be 5V tolerant (checked that with the G4 datasheet - and that is acc. to my memory) - so it is a great oportunity to interface with some 5V HW - and you always have them - without using level shifters or external power supply. I am just a bit curious how robust the 5V connection of the XC boards is against drawing too much current, connecting the power supply the wrong way. Is something like this planned for Amino/Slice Kit?
I also really like that the 3.3V line got some overhead so that you can power more devices ffrom the XMOS, makes the design often easier.

Now to the somewhat unrelated part: I really dream of an XC-0 kit. Similar to XC-1 and XC-2 but without the LEDs (to free some 1bit ports - they are precious to my understanding) without the Ethernet or USB connection (they eat up ports & resources). They can be easily added by external hardware (slices here we come).
Why XC-0? Since the 4 core devices have much more processing power (RAM/threads) to build real sophisticated solutions.
And perhaps it could be a good idea to have 1-4 core options which are compatible with all Amino/Slice add on boards.

Sorry for this unrelated stuff, but I just had no better place to put it ;)

Marcus

[dan]

"Basically there are two different connectors - why?"

Hi Marcus,

There are a number of reasons for this, some of which you have noted. Firstly, we need to export all the IO from each core of the L2. This is because we wish to target some interfaces that basically need all the IO from one core. Secondly, the IO exported by a core on an L2 is not symmetric, in the sense that for ports 1M to 1P, they overlay an 8 bit port. Also, some connectors will need to carry clock and power and having them all carry these functions may be inefficient.

I think there is plenty of useful discussion to be had about whether we should have only 1, or 2, or more connectors, but I'm not clear from your post what advantage you perceive in having them all identical even if we could achieve that. Given that each connector requires extra space on the board, and also its own cable, is there a compelling reason to take this hit?

Regarding 5V tolerance, the G4 is tolerant but the L series on which these boards are to based is not. You may well get away with it on the XS1-L but it is definitely not a practice we would encourage.

I also really like that the 3.3V line got some overhead so that you can power more devices ffrom the XMOS, makes the design often easier.

That is certainly I feature we should aim to retain with this system I think. However, at the moment the slicekit system we want to define is really just about the connectors and a few other physical considerations. However I am inclined to agree that since we have VDD over the connectors we should define exactly what users of these connectors and designers of peripheral boards might expect from the core power supply.

I really dream of an XC-0 kit. Similar to XC-1 and XC-2 but without the LEDs (to free some 1bit ports - they are precious to my understanding) without the Ethernet or USB connection (they eat up ports & resources). They can be easily added by external hardware (slices here we come).

Sure, we would have a G4 core board on the roadmap, or once the slicekit standard is finalized, anyone could produce their own version of one. We also plan a 2X L2 core board (e.g. an "L4"), where the L4 is constructed from 2 x L2 chips on the board.

Is there any reason you would prefer to develop with a G4 rather than a 2XL2 solution?

Thanks for your interest so far - please keep the comments coming.

[Interactive_Matter]

…

I think there is plenty of useful discussion to be had about whether we should have only 1, or 2, or more connectors, but I'm not clear from your post what advantage you perceive in having them all identical even if we could achieve that. Given that each connector requires extra space on the board, and also its own cable, is there a compelling reason to take this hit?

What I really like about having a number of similar connectors (best 4bit alternatively to 8bit alternatively to 16bit and so on ;) ) is that you can easily add your slices to any of those connectors and mix and mash them with every protoype in a new variety. I am doing this with my XC kits quite often.
By that I hope to get a more modular system. (If modularity means adding more L2 boards - nice too)

Regarding 5V tolerance, the G4 is tolerant but the L series on which these boards are to based is not. You may well get away with it on the XS1-L but it is definitely not a practice we would encourage.
…

Ok, that is the dea of the 5V idea.

…

Sure, we would have a G4 core board on the roadmap, or once the slicekit standard is finalized, anyone could produce their own version of one. We also plan a 2X L2 core board (e.g. an "L4"), where the L4 is constructed from 2 x L2 chips on the board.

Is there any reason you would prefer to develop with a G4 rather than a 2XL2 solution?

No. The 5V is the only reason to have G4 - bu that is a very minor reason

[dan]

What I really like about having a number of similar connectors (best 4bit alternatively to 8bit alternatively to 16bit and so on ;) ) is that you can easily add your slices to any of those connectors and mix and mash them with every protoype in a new variety. I am doing this with my XC kits quite often.
By that I hope to get a more modular system. (If modularity means adding more L2 boards - nice too)

I understand the goal here just not sure how it could be achieved. Perhaps when you have a moment you could outline your ideal connector and how it maps to the various chunks of the L2 IO (plus VDD/GND) and put a short write up with some tables in the slicekit repo?

Costings for the connectors would be great too because then we can compare the dollar costs involved.

Ok, that is the dea of the 5V idea.

yeah sorry about that!

[Interactive_Matter]

Perhaps I can spend an hour or two later in the day. Expect to get something. Hope it will be useful.

The longer I think about the 5V the more important it gets for entry level kits. I think experts can live with 3.3V (experts = people knowing the nowadays nearly everything is 5V).

So if slice kit is meant to replace Arduino (;)) 5V would be important