Annular rings... are they really needed?

Thread Starter

alex_d

Joined Jul 13, 2008
5
Like this poster, I am trying to use PCBexpress to make a PCB for a 1mm pitch BGA. Their steep annular ring requirement (smallest via is 8mil hole + 2x8.5mil ring) is making things impossible for me. I'd like to understand better why we need annular rings or why I can't make do with a much smaller one.

Specifically, I read two things. The first, repeated in a few places on the web, is that without a ring the connection between the trace and the in-hole plating is physically weak and liable to crack. The second, that I saw in the above-linked thread, is that a too-small or non-existant ring will cause the copper inside the plated hole to get eaten out and the via will stop being a via (I guess the etching fluid gets inside).

The first scenario doesn't sound too scary. If I just use a thick trace (eg a 10mil trace to my 8-12mil hole) then there'll still be plenty of area to bond if the hole is misaligned to begin with. I'm not concerned about even a 10% failure rate, for example. The second scenario sounds a lot scarier. Also I would like to not use an annular at all on the bottom of vias that aren't connected, but according to scenario #2 that would mean certain death.
 

Thread Starter

alex_d

Joined Jul 13, 2008
5
Follow up:

Has anyone tried via-in-pad? From reading about it, it seems if I use the small 8mil holes AND tent my vias on the bottom (AND maybe apply some sort of filler, like extra solder paste), I could pull it off. Thoughts?
 

nanovate

Joined May 7, 2007
666
You get what you pay for generally. If you need tighter tolerances then you'll need to pay more. Get a custom quote by contacting the mfg directly.
 

Thread Starter

alex_d

Joined Jul 13, 2008
5
You get what you pay for generally. If you need tighter tolerances then you'll need to pay more. Get a custom quote by contacting the mfg directly.
Not exactly. The interneticity (I'm just going to make up my own buzz words) of the pcb websites gives you a lot for a lot less. I've tried getting a few custom quotes, and it's generally twice as much even for the same specs. Wouldn't be so bad if it was a $30 vs $60, but I need 50 sq in of 6 layers. The price is ~$500 at pcbexpress and $1000+ for a custom quote (even from Sunstone Circuits themselves).

Of course if I don't have a choice... but I think I do. The specs aren't set in stone: they're a measure of probability and a very conservative one. Now I'm not going to second-guess them without understanding the details (hence this thread), but I'm not afraid to either.


Anyway, I realized that the first PCB I did with them (which was a pretty simple passive thing) wasn't in hindsight exactly to their specs. It had some vias with 5mil annular rings and also a couple mechanical holes that had no annulars. Unfortunately I don't have a microscope (yet) but the holes look all drilled with very high precision (there was only one via that I could see was a couple mils off) and the holes without annular rings were still perfectly plated through. I think that answers my questions for the time being, but I'm still very interested in the theory of how plating and etching gets done so that you can have an etched signal layer and a plated hole touching each other.
 

nanovate

Joined May 7, 2007
666
The price is ~$500 at pcbexpress and $1000+ for a custom quote (even from Sunstone Circuits themselves).
Like I said you get what you pay for. These internet deals are geared toward "standard" easy-to-meet specs. They've streamlined it for quickturn. When you start pushing the limits of the PCB mfg they either won't do it or they will charge you for the extra work (time) they have to spend. In any case they will rarely guarantee anything when you go outside their capabilites. You also need them to do electrical testing if you are going to do 4 layers or more-- unless you have the equipment to inspect and test yourself.

Call one of these PCB companies and ask to speak to their engineering dept. maybe someone there can explain their specific processes and limits.
 
Last edited:

Thread Starter

alex_d

Joined Jul 13, 2008
5
I was a little pissed at first by your non-constructive answer, but actually talking to a PCB company on the phone turned out a good idea.

They confirmed that you don't need an annular ring if you're not connecting a trace. Annular rings are a mechanical joint between the layer's copper and the through-hole's plating for when the board is soldered and the expansion coefficients are slightly different.

I didn't ask them if it's ok to have no annular if you make the trace wide enough (since it's just mechanics I'm going to trust my speculation and say yes), but they said something weird to me that I think adds up to the same thing. They said they can do a 1 mil annular on an 8 mil post-plated hole. Yet 1 mil is less than the thickness of the plating itself. Ie the drill is 12 mils, the pad is 10 mils, and the post-plated hole is 8 mils. The pad is the plating. I should call them back about that.

In any case, I'm pretty confindent I can push the specs and pull off the deed with an internet pcb company. I just have to choose one that does electrical testing. PCBExpress doesn't, but PCB-Pool does. They're cheaper too. Even if it doesn't work, giving it a shot will cost 4x less than "doing it right."
 

nomurphy

Joined Aug 8, 2005
567
Never tent the vias, it is very unreliable and some PCB vendors will purposely avoid tenting vias (which is really bad, if you wanted them).

When a layout design requires it, always indicate the vias are to be "plugged" not tented.
 

Thread Starter

alex_d

Joined Jul 13, 2008
5
I'm actually doing PCB-Pool, but their specs are identical to PCBExpress.

I've just submitted a test board with my poor man's BGA breakout. It's got two traces between each pin on the inner layers that I routed as 5mil width/5mil separation/7mil antipad annulars (I've been careful to budget the vias themselves as 12mil since "8mil" figure is post-plating and thus a lie. The total is 5*2+5+7*2+12=41mil=1mm). I used a cool technique where the vias connecting to inner traces don't have any pads per se (there was just no room) but instead have "half-pads" consisting of 12mil traces that teardrop down to the 5mil ones. The traces I connected to testpoints. There's capacitor pads on the bottom to get a feel for soldering 0402s and even 0201s and I've ordered a cheap z80 microcontroller to practice soldering the BGA.

If all this succeeds, I'll be able to route my 484-pin fpga on a 6-layer cheapo internet board. Which will be quite an achievement.
 
Top