#milkymist IRC log for Saturday, 2012-09-08

--- Sat Sep 8 201200:00
sb0got some kintex 7 prices... XC7K160T-1FBG676C is 210E07:05
sb0that's still expensive, but not as bad as the $1200+ from online shops07:05
sb0wpwrak, so, how do you feel about kicad for m3 design?07:15
lekernelhmm... where can I find a 10GbE PHY with datasheet not requiring a NDA?12:54
wpwraksb0: people have made multilayer designs of comparable complexity with kicad, so i think it should be able to handle the task. maybe it would benefit from a few additions, though, e.g., a trace length calculator16:35
lekernelwpwrak: there will be DDR3 (at 1.8Gbps/pin, which will require matched track lengths) and differential traces carrying 5.4Gbps of data17:19
lekernelaccording to http://dangerousprototypes.com/forum/viewtopic.php?f=2&t=2822&start=30 there's no serious length matching ...17:25
wpwraksomething easier would already do the trick: have a means to measure your tracks. since you route manually anyway, you can then make corrections as needed. so you wouldn't even need to specify this to the DRC.17:30
wpwrak(of course, having DRC watch such issues would be an additional benefit in the future)17:31
lekerneldo you have links to complex kicad boards?17:35
wpwrakwe have one even in the qi-hw universe: http://projects.qi-hardware.com/index.php/p/xue/17:37
wpwrak(only routed, never built, though)17:37
wpwrak(wp.pdf) i see a few good items :) e.g., selection is indeed a crying shame at the time17:39
lekernel(xue) hmm, it's not fully routed17:42
wpwrakyes, they stopped somewhat in the middle of the project. but what they did looks pretty decent.17:43
lekernelthere isn't any track length matching too17:43
wpwraktrack matching may be messy. three choices: 1) figure out a workflow that lets you do it without adding/changing code. 2) add a trace length calculator to kicad (you should have connectivity information inside pcbnew). 3) write some processor for the .brd file or maybe even gerber.17:46
wpwrak3) may be easier but a bit less convenient17:46
wpwraknot sure if 1) is possible without going insane ;-)17:46
wpwraksince i hate C++ and even more the way it's used in kicad, i tend to choose 3) over 2) when i need some features. a bit of perl can work miracles ;-)17:47
wpwrakan "offline" trace length checker could also be easily extended to do DRC on trace length17:49
wpwrakmy usual kicad setup is to have things on all my three screens. so a routing process with external tools could have eeschema on the left (for reference and for swapping pins and such), pcbnew in the middle, and a few xterms with tools on the right17:50
lekerneland same things for differential pairs...?17:56
wpwrakthey're just a special case of length-matched traces, no ?17:57
lekernelyou have to respect the spacing between the two signals too17:57
wpwrakdoes this have to be precise ? or is just "about the same path" combined with the common sense "avoid parallel traces" enough ?17:58
lekernelI'm thinking of something 40 differential pairs (80 individual signals) + around 200 signals that need to be length matched (in 8 groups)17:59
lekernelat 5.4Gbps I guess it has to be quite precise, yes18:00
lekerneleg http://www.nxp.com/documents/application_note/AN10798.pdf18:00
wpwrakah, lots of interesting rules :)18:03
wpwraksounds like 3) would be preferable over 2) for getting started. first find a nice workflow, and only then worry about integration18:03
lekernelI'm a bit worried about disaster scenarios like a couple of expensive respins followed by the conclusion that kicad is not suitable and then switching everything to altium/pads and trying again...18:06
lekernelespecially since few layout people would work with kicad18:06
wpwrakdo you think altium/pads have secret knowledge ? like the alchemists did ? :)18:07
lekernelno, but there can be practical issues, like kicad DRC not spotting some errors that we painfully find after the board is fabricated and tested18:13
wpwrakah well, that's always a possibility18:14
GitHub161[migen] sbourdeauducq pushed 1 new commit to master: https://github.com/milkymist/migen/commit/51f9a2a963c52cff1174954ca9f48fc9e3a20ab623:18
GitHub161[migen/master] fhdl/namer: simplify + more relevant names - Sebastien Bourdeauducq23:18
--- Sun Sep 9 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!