FEDEVEL
Platform forum

AviusAquila , 07-23-2025, 09:41 PM

I'm designing a PCB on KiCAD for an iMX6, based generally on the SABRE schematics. I've seen the OpenRex designs - thanks @Robert Feranec very much for those. Knowing that you've got the chip fanned out with just through hole vias is one of the reasons I've worked up the courage to try this.

I was considering using Blind vias to make my life a little easier. The following 10-layer stackup is what I thought I could use:
- L1 : Top Component
- L2 : Signal
- L3 : Ground
- L4 : Split Power
- L5 : Ground
- L6 : Signal
- L7 : Split Power
- L8 : Ground
- L9 : Signal
- L10: Non-critical Signal / Components

The Blind vias I intend to use are mechanically drilled vias with sequential lamination, because that is what my manufacturer can handle. Specifically, I intend to use the following vias:
- Through L1 - L10
- Blind from L1 - L2
- Blind from L10 - L9
- (maybe) Buried from L3 - L8

This is my thinking :
- L2, L6, and L9 are layers with good ground references next to them.
- L1 really is mostly going to be components and dogbones. The impedance control on it may be less critical. The high speed traces which could have been on L1 / L10 otherwise will be forced onto inner layers.
- The L1-L2 blind vias will keep the ground plane on L3 a lot more solid than it otherwise would be. It'll also leave a lot of room on the other signal layers, to not have the high via count slicing through the entire IC region on all layers.
- L3-L8 buried vias can stitch the ground planes together, while not interfering with the outer layers.
- Since the non-through-all vias don't overlap, these can be regular drills.

Obviously, this is a highly unusual stackup. I would think it would have been a lot more common to do this, but from what I can see, it isn't. What would break if I used this stackup?
As an aside, I've seen that the OpenRex has 4 nearly full planes worth of power. I am really hoping that I can get away with just 2. 😦

AviusAquila , 07-24-2025, 09:02 AM

Additional Notes :

AviusAquila , 07-24-2025, 09:02 AM

1. Diminishing Returns from L1 Impedance Control

Busses like DRAM need all the traces to be identically routed, or in specific groups. Since there are generally atleast a few pins on the inner rows, they do need to dogbone and drop to inner layers. Because of this, pins toward the edge of the IC also need to quickly go to the same inner layer. With this reasoning, I reach the conclusion that long impedance controlled traces on the component layer are unusual.

This is what I saw on a previous HDI design, where the top and bottom layers get pretty filled up with components and capacitors respectively, and most of the critical signals dropped to inner layers immediately with via-in-pad. I actually ended up requesting impedance control from the manufacturer only on internal layers.

My thinking is that if I route only non critical signals on these outer layers, and the critical traces only spend the fanout / dogbone duration on them, then having ground right next to it becomes less important.

AviusAquila , 07-24-2025, 09:02 AM

2. Stipulations / Assumptions

I make the following assumptions. The question may be read as "Assuming these are true, what other issues might arise?". If any of these assumptions raise a red flag, please feel free to beat me over the head with it

Materials and thicknesses will be viable
Impedance calculations will be done and traces will be correctly sized for the layer they run on for L2, L6, L9.
L1 and L10 will have no meaningful impedance control
I intend to ask for L6-L7 to have a thicker separation to prevent the split plane from causing too much issue.
I'm also making the potentially problematic assumption that the 3 layers will be enough for all the impedance controlled traces. I suspect that removing the top and bottom layers from the space available for controlled signal routing will be the thing that answers the question of why this is a bad idea.

QDrives , 07-24-2025, 02:58 PM

So a reverse build-up. With mechanical drill, there is little sequential about it, unless you do the buries from L3 to L8.
Even if your traces on L1 and L10 are short, the noise will still couple to L2 and L9.

AviusAquila , 07-24-2025, 05:10 PM

Yes,
- drills on (3-8) which become buried at the end,
- two panels pressed on either side of it, ie, (1-2) and (9-10) with drills that end up becoming blind,
- and finally through drills from (1-10).

This is what I understand from their capabilites. Yes, I think sequential lamination is the incorrect term. The places talking about it were mostly referring to 2+N+2 stackups with laser drills.