2
\$\begingroup\$

The board in question is a four layer board, with a dedicated ground plane as well as a dedicated power plane. Capacitors are MLCC, imperial 0402 (metric 1005). MCU is an STM32.

The question now is how to combine decoupling capacitors and vias. The best way would be to use plugged via-in-pad, but sadly for this project this is out of the question.

So now the question is whether I should a) connect the vias and with the pads of the capacitdor with traces (for superior manufacturability but inferior impedence) or whether I should b) place the vias so close to the pads of the capacitor, that no traces are needed.

Regarding a): When I leave something like 0.2 mm of distance between the annular ring of the via and the pad of the capacitor, there will be solder mask between the annular ring and the pad of the capacitor. This solder mask will prevent solder paste getting sucked away from the pads of the capacitors, down the vias.

See the 2010 article "Five Via-In-Pad Myths" from cadence.com: "That little mask dam will stop solder from flowing into the via and everybody will be happy." The downside: due the relatively long traces, the inductance will be relatively high. The following picture shows a capacitor (imperial 0402/metric 1005), a 0.2 mm wide trace, a 0.6 mm diameter via (hole size: 0.35 mm) and 0.2 mm distance between the annular ring and the pad of the capacitor (the mask dam mentioned above): enter image description here

Regarding b):

The second option could lead to problems during assembly (solder paste getting sucked into the via) but will have superior inductance, as the annular rings of the vias overlap with the pads of the capacitor. No traces are used. See the 1997 article "Bypass Capacitor Layout" by Dr. Howard Johnson: "series inductance is impacted favorably by ... vias jammed up next to the pads (no traces)". See this picture in which the right edges of the via holes line up with the left edges of the pads of the capacitor:

enter image description here

\$\endgroup\$
2
  • 1
    \$\begingroup\$ If this is a volume product, talk to the fabricator and get an answer specific to your situation. If this is a one off design that you are going to hand assemble, the second is absolutely fine. \$\endgroup\$ Commented Feb 26, 2021 at 16:58
  • \$\begingroup\$ The 2010 article seems really good to me. Virtually everything I wanted to say was mentioned in that article. If you feel it is necessary to put the via in the pad, you need to use special via treatment as outlined in that article. The only thing the article doesn't mention is that via in pad will probably increase PCB cost. \$\endgroup\$
    – user57037
    Commented Feb 26, 2021 at 17:41

2 Answers 2

6
\$\begingroup\$

If you don't want the solder to be sucked into the via, then the via has to be a bit away from the pad, and a solder mask sliver between them, but you could:

  • move the vias so they're closer together to reduce via pair inductance, they don't have to be aligned with the pad centers

  • widen the trace from the via to the cap pads to reduce trace inductance, I mean the trace can be as wide as the via and the pad, that doesn't cost extra ;

  • widen the trace from VDDA pad to cap pad (again, lower inductance, although it's very short)

Here's a FingerPaintCAD example, a bit ugly but you get the idea, just apply a wider trace width...

enter image description here

If it is super important that this pin has a very low inductance to ground you can also use a lower inductance cap (LLL, NFM). But I don't see a GND pin next to it, which means the manufacturer of the chip didn't think there was a need to create a low inductance power/ground pair, so most likely it does not need any fanciness.

\$\endgroup\$
0
2
\$\begingroup\$

Unless you have a very specific problem you are trying to solve, I would not butt my vias up against the pads. The potential for manufacturing problems is just too high.

The tiny tiny inductance added by those small traces is (likely) dwarfed by the inductance of the capacitor you are mounting there and will be inconsequential in the vast majority of applications.

\$\endgroup\$
1
  • 1
    \$\begingroup\$ I've seen some companies do that and others don't ever do it. In most cases I think it's to be avoided. If this is a low-volume or prototype, it's not generally a concern. \$\endgroup\$
    – jwh20
    Commented Feb 26, 2021 at 16:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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