6
\$\begingroup\$

To avoid to use very small and expensive crystals I'd like to use a bigger one and place it to opposide side of PCB to save space.

My question is: Is it possible to place an SMD oscillator on opposite side and connect it through vias?

In my case I've a 4 layer PCB. I'd like to know if this will harm or affect crystal signal quality and if it's used/unused practice.

Here's an example image: enter image description here

EDIT: In this example and in my question I refer to a crystal (no oscillator). I've not placed capacitors to have a more clear example. EPSON TSX-3225 38.4000MF10Z-AS3 is a device that I'd like to use.

\$\endgroup\$
  • \$\begingroup\$ I do not think you would want to put the crystal exactly there on the other side of the PCB. Depending on the number of layers, it could create some type of signal interference between traces. \$\endgroup\$ – 12Lappie Nov 8 '16 at 18:47
  • 2
    \$\begingroup\$ Don't worry, it will work. But where are the capacitors \$\endgroup\$ – Gregory Kornblum Nov 8 '16 at 18:47
  • \$\begingroup\$ As far as I know, it is recommended to place the oscillator on the same side and as close as possible. See this post \$\endgroup\$ – Nazar Nov 8 '16 at 18:51
  • 1
    \$\begingroup\$ @GregoryKornblum I think this is one of those oscillator ICs that require only a decoupling capacitor on the power rail. \$\endgroup\$ – user110971 Nov 8 '16 at 18:58
  • 1
    \$\begingroup\$ If it's an oscillator, why there are two traces connecting to the IC? What is the PN? \$\endgroup\$ – Gregory Kornblum Nov 8 '16 at 19:01
6
\$\begingroup\$

You can do that, as long the frequency is not too high. A few 10's of MHz should be fine. As always, if the frequency is high, you should consider the signal integrity. The vias introduce impedance discontinuities, which can make the clock non-monotonic from the reflections. This is a problem, if you do not have a Schmidt trigger inputs.

Another thing to consider is that it is usually more cost effective to have the boards pass through a one side pick and place process. It depends on your volume and manufacturing strategy. I've seen board where it is cheaper to go for blind and buried vias, than to put components on both sides.

|improve this answer|||||
\$\endgroup\$
  • \$\begingroup\$ "With a few 10's of MHz" do you think 50Mhz it's enough slow? \$\endgroup\$ – Singee Nov 9 '16 at 10:45
  • \$\begingroup\$ @Singee depends I guess. Is it a square wave oscillator? 50Mhz has a period of 20ns. Probably a rise time of 1ns. This gives \$0.35 / f_{\mathrm{rise}} = 350 \mathrm{MHz}\$ bandwidth. I would be a bit concerned at those frequencies. You can try putting it in spice by calculating the transmission lines. You have trace -> via (PCB thickness) -> trace. \$\endgroup\$ – user110971 Nov 9 '16 at 10:54
  • \$\begingroup\$ Unfortunately rise time is not specified on datasheet of p.n. that I've added to my question. Maybe on oscillator/mems it's specified but on crystals I've never found this information (suppose they've a sinusoidal wave) \$\endgroup\$ – Singee Nov 9 '16 at 11:04
  • \$\begingroup\$ @Singee it depends on the type of oscillator for the IC. At this point I would build a prototype to see if it works. Or you could do single integrity simulations, if you have the tools. \$\endgroup\$ – user110971 Nov 9 '16 at 11:09

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.