A Spartan 6 XC6SLX45T 484pin BGA Adapter Board

[plasmo]

Double row 2mm headers are inexpensive, so you can have 300 breakout pins on a 3”x3” pcb. That’s lots of pins!

[barnacle]

Seconded: Samtec. Also, Harwin.

[GARTHWILSON]

I've bought directly from Samtec several times, without going through a distributor.  They allow it because they have way too many combinations of options for a distributor to stock them all and move them while the platings are still fresh and all that.

[ElEctric_EyE]

After some research I've done, these headers/pins with indeed be spaced at .1" (2.54mm) not the .050" as I was dreaming about. The pins are not thick enough at .050" spacing, plus plugging/re-plugging one of these bastards in would be a nightmare. I always lean towards machine tool sockets due to reliability and ability to transfer current. Some .050" rated @1A and .1" rated @3A. Just 1 consideration. This beginning stage proving a challenge already!

So, if the design were to incorporate 25 pins circumference, then 23 pins circumference, 21 pin and 19 pins, there would be 352 I/O pins that could be routed from the FPGA. This would be 4 rows of pins located around the outer perimeter of the board. I think I'm getting close to target, but I have a few other things to consider... 25 pins spaced .1" apart is just above 2.5". Again 3" x 3" board may be the settling point with some breathing room.

Thanks for all your inputs!

[GARTHWILSON]

ElEctric_EyE wrote:

I always lean towards machine tool sockets due to reliability and ability to transfer current.

In my experience (not just hobby but also decades of using them in our products we make and sell at work), the non-machine-tool ones are totally reliable when they have two contact surfaces like the tiny tuning-fork-shaped things, whereas the machine-tool ones get loose after very few insertion/extraction cycles.  For extraction, you just need to keep the male and female connectors parallel as you pry the connectors apart, particularly if you have the really low-profile ones on which the tuning forks will get pried open in the end positions and not hold the pin tightly after that unless you remove the housing and carefully squeeze (bend) the two sides of the fork toward each other again.

[ElEctric_EyE]

I already chose this type, 48x Aries 40-pin SIP (Single Inline Package) and a 4" x 6.5" vector board. These SIPs can be trimmed as needed. From this I should be able to determine the final adapter board size. Of course I am aiming for as small as possible, <3" x 3". The vector board may or may not be for this project but will be used for initial placement of the headers. Just 1 step at a time, tomorrow I measure when these parts come in!

[Dr Jefyll]

ElEctric_EyE wrote:

This would be 4 rows of pins located around the outer perimeter of the board.

Hmm, yeah -- makes sense. But this might be more than a tad awkward if it ever becomes necessary to un-plug the module from its motherboard/whatever. If you haven't already, I suggest you give this issue some thought while you're still in the planning stage.

One option I considered for a similar assembly was to include a simple, homebrew extractor, consisting simply of a captive nut and a bolt at each of the four corners (probably inboard, within the perimeter formed by the connectors). The arrangement would be such that the module and the mobo could be separated by turning each of the four bolt heads a little at a time, prying the assembly apart equally from all four corners. If that seems unclear I can provide more detail.

This will have a cost in terms of consuming a bit of real estate on the module, but the benefit might outweigh that.

- Jeff

[GARTHWILSON]

Dr Jefyll wrote:

this might be more than a tad awkward if it ever becomes necessary to un-plug the module from its motherboard/whatever. If you haven't already, I suggest you give this issue some thought while you're still in the planning stage.

One option I considered for a similar assembly was to include a simple, homebrew extractor, consisting simply of a captive nut and a bolt at each of the four corners (probably inboard, within the perimeter formed by the connectors). The arrangement would be such that the module and the mobo could be separated by turning each of the four bolt heads a little at a time, prying the assembly apart equally from all four corners. If that seems unclear I can provide more detail.

Jeff and I talked about this a few years ago, resulting in this image:

thumbscrew2.gif (23.19 KiB) Viewed 3713 times

[Dr Jefyll]

GARTHWILSON wrote:

Jeff and I talked about this a few years ago

Wow -- I had utterly forgotten!

Looking at that diagram, there are some details I would consider changing. The nuts do need to be restrained from turning, but soldering each one to the board seems suboptimal due to the weak-ish joint that results and especially because of the rather large chunk of board area that's consumed by the nut and the surrounding solder joint. That entire area is prevented from having traces on its surface, so from a routing perspective I'd prefer a scheme that has no excluded area other than the bolt hole.

So, instead of nuts soldered to the board, I would consider using a short length of bicycle spoke or similar material which perhaps by brazing would have nut #1 and nut #2 attached at its ends. This is sufficient to restrain the nuts from turning, and we use another such assembly to link nut #3 to nut #4. Or, similarly, each of the two nut-and-spoke-and-nut assemblies could be replaced by a piece of 1/4" by 1/4" metal bar, drilled and tapped at each end. (Or make the bar out of nylon, or even have it 3D-printed.)

Optionally, you could even attach the halves and create just one rigid assembly that includes all four of the threaded holes. Done properly, this'd mean every bolt is assured of being perpendicular to the surface, which means it doesn't necessarily need to be sharpened to a point nor have a guide hole to receive that point. Also optional is the use of thumb screws, as opposed to a more ordinary screw/bolt.

-- Jeff

[ElEctric_EyE]

Very good idea guys, I will try to utilize this idea if I can!

Here's an initial layout estimate using a spare vector board. The square in the middle is just a placeholder for the BGA, about 24mm x 24mm. This arrangement has 31x4 pins on the outside perimeter, then 29x4 pins and 27x4 pins for a total of 348 pins. Plenty!
I'm thinking now, that the 3x VR's and JTAG interface will have to go outside of the I/O pins, which will enlarge the 6 layer board to no larger (hopefully) than 3.5" x 3.5" (89mm x 89mm)

[barnacle]

I suspect you counted the corners twice - I get only 336... I hope that's enough

Neil

[ElEctric_EyE]

You are correct sir! Good eye, thank you for spotting that Neil.

[ElEctric_EyE]

I don't know what is going on with the IC supply but I cannot find the Synchronous RAMs I was using for sale and the 1 I found from a manufacturer is almost 5x cost and only 3 left in inventory. This is not good news! Can someone tell me what is going on? This supply problem can't still be related to Covid is it? I mean I am checking direct from manufacturers website, Cypress and GSI Technology were the 2 suppliers I've used in the past. I have searched Avnet and Mouser for stock and there is nothing. The speed must be higher than the 1080p pixel clock, preferably 250MHz speed grade although they sell 167MHz?
This project will not fail! If I have to de-solder the 100-pin TQFPs from old projects, I will do so. Today I will also check my stock to see if I have any unused, that should be fun as I have not checked in over 5 years!

1 other note, my progress will be slow for the next couple weeks as I go on vacation next week for 1 week. But my goal is to get prepared for when I return, before I leave.

BTW, anyone notice I chose the 'T' version of the Spartan 6? It has high speed serial and I believe the possibility for HDMI input and output!

[barnacle]

I note that the part is twenty years old - have they simply stopped making it?

[ElEctric_EyE]

I'm sorry I posted a 4mx8 part from an uncompleted project. I'm using a GS8320xxx 2mx8 250MHz part. They are being sold and even better I left myself a present in the form of 4x unused 2Mx8 still wrapped and untouched in a tray.

I will make 4x adapter boards for these ICs also, probably get away with 4 layer boards for those. I left my notes at home so I'll post more details tonight in about 8 hours. Moving right along!