1
\$\begingroup\$

I'm working on a project that involves multiple (many) BGA packages, namely interface converters such as Ethernet PHYs connected to a central FPGA.

I've already worked out a (tentative) layer stackup in terms of what signals will run over which layers, but I need a good strategy for assigning pins. Obviously, a PHY will fan-out in a pretty distinct order, but I'd like to have a decent idea of how to assign the pins in an informed manner so that way I'm not having to get shifty with vias to rearrange the trace order during the layout.

To illustrate, the interface I'm working with is standard GMII, so a 16-bit bus with control signals. What are some good methodologies for assigning pins from an FPGA that correlate into the most logical fan-out order for traces leaving the FPGA?

\$\endgroup\$
3
\$\begingroup\$

If the design isn't fixed yet (i.e. pins are free to be changed), then you have some flexibility in what goes where.

The first step is to determine what pins are absolutely fixed. For example the FPGA may only have certain resources on some pins - dedicated clock inputs spring to mind, as do high speed transceivers. Depending on the clock network some clock inputs can only feed some PLLs for example, so you need to decide early one where these clocks are going.

The second step, decide where interfaces can go - you want parallel interfaces to use pins close to each other, the last thing you want to do is route some pins only to realise that one bit of an interface is on the opposite side of the FPGA! This step also involves looking at voltages. FPGAs tend to have different 'banks' which each have different IO voltage references. You can't(*) have an interface running at 2.5V sharing with one running at 1.8V. You also ideally want all pins of a parallel interface to be on the same, or adjacent banks, preferably ones in the same corner of the FPGA (there are four corners, Top-Left, Top-Right, Bottom-Left, Bottom-Right, and each tend to have resources dedicated to that corner).

Once you have a map of where things can go, so you have decided your 16bit interface can use pins n-m or whatever, the exact order doesn't actually matter. You can rearrange in that group quite readily - because if picked correctly, the FPGA can simply remap the pins as needed as long as you were careful in your groupings.

Third step is to decide what layers interfaces are going on - try to minimise the amount of times high speed signals need to change layers as vias are nasty at high frequency.

Fourth step, pick any order in your grouping and start routing from the various devices towards the FPGA, preferably doing the most important ones first. Once you get to the FPGA, you can see what pins could do with being remapped - say you end up getting back to the FPGA with the data pins in an obscure order, simply reassign the pins to untangle them (easier to remap than to start adding vias to cross over each other).

Once you have a good idea of the mapping - maybe you've routed a fair few of the traces and are now confident that your choice of pins is routable. The fifth step is do a test compile - compile the FPGA and make sure it can use that pin mapping! I've had it happen to me once before that I've chosen assignments and accidentally used a dedicated programming pin for one bit in a data bus - I sent off the design without test compiling only to find out when the board got back that the one pin can't be used. The test compile will ensure that your pin mapping is compatible with the FPGA.


Note (*) Well you can sometimes, depends on the FPGA.

\$\endgroup\$
  • \$\begingroup\$ Great answer - and that final paragraph (about testing the configuration in the FPGA) is important; I too have burned myself on a couple low-end boards by not double-checking pin compatibility. Due to signal integrity, I plan on only having two vias in the signal path of each data line - one at the pad-end of each BGA's terminus - on FPGA and PHY side. I'm just attempting to streamline by making informed judgment on what goes where so that the number of iterations of Schematic->Layout->Schematic can be kept to a minimum. \$\endgroup\$ – ecfedele Nov 28 '15 at 1:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.