0.5mm pitch BGA too small to bother with?

[FrankBuss]

You can use more than one external SRAM in parallel. For 16 bit 2 MB you need 21 address lines, 16 data lines, and 3 control lines (2, if CE is always on), so 40 lines per 2 MB chip. There are 240 QFP pin FPGAs like an old Cyclone II, which is sufficient for your project, with 142 usable IOs. 3 SRAM chips would be no problem for it and you would still have 22 IOs for parallel video output (VGA would be the simplest way, 4 bit R2R per color component) and for buttons, LEDs etc. This would simplify the design, because you could use one SRAM for the Z80 and z8002s, and another one or two for the graphics RAMs, which might need to be faster so that you can't share it with the CPU RAMs. And it can be done with a 2 layer board and soldering is no problem.

A general purpose FPGA board with 3 x 2 MB SRAMs, 16 bit data bus, could be useful for other projects as well.

[T3sl4co1l]

Yeah, guessing you'll spend more on the tight spec PCBs.  Go with the easier FPGA.

Tim

[ajcrm125]

I came across this video of Dave himself pulling this off with a 0.4mm pitch BGA on a 2-layer board.
I guess it's not that crazy??

[exmadscientist]

It seems to me that you would probably be best served by moving up to the FTG256 package. Its ball pitch is 1.0mm, which is a lot more forgiving. Particularly if you only need a modest number of external I/Os, you should be able to get away with a 4-layer board and 5/5 design rules + 18/10 vias (as you get with, for example, OSH Park's 4-layer service).

Dave is away to get away with 2 layers because the device in that video is tiny (only 36 pins!) and he uses 4/4 design rules. Trying to do a 2-layer 24MHz design with an Artix-7 is just asking for pain. 4-layer boards are relatively affordable these days and you'll save more in development time than the boards end up costing you.

You also may run into trouble with running out of memory on the XC7A35T. 193kB is getting pretty close to its maximum capacity of about 200kB, and you can easily lose several BRAMs due to the FPGA's restrictions. You might have to step up to the XC7A50T. Fortunately it isn't too much pricier and comes in the same packages, so it's a drop-in replacement.

[ajcrm125]

It seems to me that you would probably be best served by moving up to the FTG256 package. Its ball pitch is 1.0mm, which is a lot more forgiving. Particularly if you only need a modest number of external I/Os, you should be able to get away with a 4-layer board and 5/5 design rules + 18/10 vias (as you get with, for example, OSH Park's 4-layer service).

Dave is away to get away with 2 layers because the device in that video is tiny (only 36 pins!) and he uses 4/4 design rules. Trying to do a 2-layer 24MHz design with an Artix-7 is just asking for pain. 4-layer boards are relatively affordable these days and you'll save more in development time than the boards end up costing you.

You also may run into trouble with running out of memory on the XC7A35T. 193kB is getting pretty close to its maximum capacity of about 200kB, and you can easily lose several BRAMs due to the FPGA's restrictions. You might have to step up to the XC7A50T. Fortunately it isn't too much pricier and comes in the same packages, so it's a drop-in replacement.

All very good points.  I guess I was just afraid of having to go the 4-layer route due to pricing.  I was just eye-balling some boardhouse prices and it seems going from 2-layer to 4-layer will double or triple the board price.  But if I order say 100 PCBs or so (which will be the case) perhaps the delta is smaller?

[Korken]

A quick search on pcbshopper.com with 5x5 cm, ENIG and 100 boards gave 90 USD total for 2 layers and 140 USD total for 4 layers.

This gives some feel for it, about 50% more for a 4 layer board.

[MagicSmoker]

...I guess I was just afraid of having to go the 4-layer route due to pricing.  I was just eye-balling some boardhouse prices and it seems going from 2-layer to 4-layer will double or triple the board price.

The only time it makes sense to (attempt to) go with 2 layer on a design like this is when the product is both cost sensitive and will be made in relatively high volumes (greater than 1k, at least). It will be *much* harder to lay out a board with a 0.5mm pitch BGA if constrained to 2 layers (indeed, it might be impossible) and the time you will spend ripping up and retrying routes and the time you will spend debugging noise problems because you can't have a proper ground plane will absolutely and totally swallow the extra cost of going with a 4 layer board from the beginning.

Exceptions: you are doing this as a hobby and you are a masochist.