Re: [ARTIQ] starter ARTIQ hardware for neutral atoms

[Sébastien Bourdeauducq via ARTIQ]

Hi,

On Wednesday, March 08, 2017 06:11 AM, Neal Pisenti via ARTIQ wrote:

* For ARTIQ core device, we would ideally jump straight to using a
Kasli, but as that isn't likely to be done in the next few months, I was
planning to use a KC705 as the core.


The "EEM" DDS/synth Kasli extensions may not necessarily be ready before 
Kasli. So I don't see how the KC705 helps - is it because you want more 
extensions that one Kasli would support? Supporting this KC705 scheme is 
more gateware development and one more configuration that needs to be 
documented, packaged and maintained. Maintainance means that we need to 
check regularly (preferably automatically) that it keeps working when we 
modify ARTIQ and fix any bugs that pop up. It takes work.



* KC705 has 2x FMC headers, which would drive 1x VHDCI carrier card,
providing 8x IDC for EEMs. We would buy FMC -> VHDCI adapters for this
interconnect.


What adapters in particular? http://www.ohwr.org/projects/fmc-vhdci? We 
didn't check compatibility of any of those.



**Specific questions**:

* what limitations are there (latency/bandwidth/etc) on daisy-chaining
additional Kasili DRTIO modules off of the single KC705 SFP?


While the hardware could do it, daisy-chaining Kaslis is not supported
by the current gateware plans. The plan is to use a Metlino, which has
many available transceiver links (mostly to the microTCA backplane, but 
there are also 3 SFPs), as a central device with a direct link to every 
satellite device. If daisy-chains are implemented, there would

be virtually no impact on bandwidth, and the latency would increase by
roughly ~100-200ns per hop.

Instead of the daisy chain, it is also possible to have one Kasli as 
central device driving directly other Kaslis with its SFPs (note that 
one SFP will be used for Ethernet). There are no current gateware plans 
for this either (so this would need funding), but it is less difficult 
to develop and has less latency.



* Is there an estimate on the timescale for finished Kasli?


It is not funded yet, but I think this should happen in a few months. 
Then there will be another few months before it begins to become usable.


Sébastien

___
ARTIQ mailing list
https://ssl.serverraum.org/lists/listinfo/artiq

Re: [ARTIQ] starter ARTIQ hardware for neutral atoms

[Neal Pisenti via ARTIQ]

>
>
> On Wednesday, March 08, 2017 06:11 AM, Neal Pisenti via ARTIQ wrote:
> > * For ARTIQ core device, we would ideally jump straight to using a
> > Kasli, but as that isn't likely to be done in the next few months, I was
> > planning to use a KC705 as the core.
>
> The "EEM" DDS/synth Kasli extensions may not necessarily be ready before
> Kasli. So I don't see how the KC705 helps - is it because you want more
> extensions that one Kasli would support? Supporting this KC705 scheme is
> more gateware development and one more configuration that needs to be
> documented, packaged and maintained. Maintainance means that we need to
> check regularly (preferably automatically) that it keeps working when we
> modify ARTIQ and fix any bugs that pop up. It takes work.
>

Valid point -- my main thought with this setup is exactly what Tom
articulated; the KC705 -> VHDCI is a stopgap until Metlino/Kasli
development is completed. And it would allow us to start using digital IO
immediately via the PCB_3U_BNC EEM, and we might spin up some in-house
boards for multichannel DAC over SPI. We shouldn't need the DDS/synth
modules for some time still, since all of our DDS currently exist as "slow
peripherals" in the artiq sense.

The other short-term alternative I see is adapting either the nist_clock or
nist_qc2 builds, and making our own FMC -> ___ breakout. But piggybacking
on the EEM architecture seems cleaner and more future-proof. When the Kasli
is done, we would either abandon the KC705 for a Kasli as core, or
(depending on the cost) plan for 1x Kasli as DRTIO connected to the KC705
SFP. That would certainly put a limit on how many things we can control,
but will probably satisfy our needs in the foreseeable future.

Even though the uTCA/Metlino setup will be exceedingly nice, for neutral
atom experiments it is currently overkill. But whether the upfront cost of
a uTCA chassis & associated hardware is worth additional development
headache to exclusively use Kasli baords...? Remains to be seen.


> * KC705 has 2x FMC headers, which would drive 1x VHDCI carrier card,
> > providing 8x IDC for EEMs. We would buy FMC -> VHDCI adapters for this
> > interconnect.
>
> What adapters in particular? http://www.ohwr.org/projects/fmc-vhdci? We
> didn't check compatibility of any of those.
>

When I chatted with Joe, we looked at this one:
http://www.ohwr.org/projects/fmc-dio-32chlvdsa/wiki/Wiki

@Greg -- are the board files available for this? I looked on the Creotech
website but couldn't find it for sale anywhere (which is odd, because I
swear I saw it there before, for ~ $500). Even so, I wonder about pricing
it out at MacroFab or one of these other small-turn PCB companies before
spending $1k on a pair of them.

Cheers,
Neal
___
ARTIQ mailing list
https://ssl.serverraum.org/lists/listinfo/artiq