I've got a design that worked great on the two boards I hand assembled, but more than half of the boards from the local assembly shop are bad.

I've traced the most common failure mode to an unstable reference clock from my processor to the ethernet PHY. I guess the PLL is not locking properly in some cases.

About the only thing I've found (and it's possibly a big thing) is that in an effort to squash the area down I somehow ended up with the 24MHz crystal for the system clock (that is fed to a PLL for the Ethernet reference clock) -very- close to the DC/DC converter's shielded inductor. The shielded inductor is at 45 degrees orientation to the crystal, but one corner is within 20 mils of the side of the crystal! Oops.

I've been able to move this crystal to about 160 mils away now, which is the best I can do without some serious rework. I've seen a layout example in the processor's layout notes that appear to show the crystal about 100 mils away from the inductor (the DC/DC is integrated in this processor package), so I'm thinking that's okay. The evaluation board has them about 250 mils apart, but it doesn't look like that distance was a significant factor in that design (although it could have been). It just looks like a convenient spot for both components.

My big concern at the moment is... did I fix the problem? How likely is it that a crystal 20 mils from a shielded inductor could cause problems? The odd thing is I have 6 boards that have so far behaved perfectly, and about 5 that have this reference clock PLL problem. I'm not sure why it's not all of the boards, unless it's just how individual tolerances add up.

I could have bigger signal integrity issues here... but then much more demanding parts of the processor layout (DDR2 memory) seem to be behaving well. No boards show any hint of problems there or anywhere else.

The most likely cause of my woes has been the local assembly shop. I have a very low confidence level in the boards I got from them. I've found a large number of mistakes. One board's been working since I replaced the crystal... I didn't see any oscillation on the scope, but under a microscope it definitely appeared to have connectivity. However, replacing crystals didn't help any of the other boards.

I just wish I had a concrete, fixed problem for this next board revision instead of a bunch of "It'll probably work now"...

Here's a picture of before and after (the crystal is slightly larger in Y than it's footprint): enter image description here


1 Answer 1


My guess is that the inductor is probably not a major contributing factor to the failures. I say this because it's a inductor, and as such produces a magnetic field. The quartz in the crystal is not particularly magnetic. Second, you note that the inductor is shielded, so the external leakage field should be small. Still, the crystal output is pretty high impedance and induction could be adding a small voltage in series with it. This should be small and not much issue if the crystal is being driven properly with the right load caps.

I would look closely at the crystal circuit. This sounds like maybe you are using too low of a drive level setting or the load caps aren't right. What does the datasheet say the load capacitance of the crystal needs to be? What size cap do you have on each side? The chip driving the crystal may have different drive level settings if it's intended to run with a wide range of crystals. At 24 MHz you almost certainly should be using the highest drive level choice.

Look at the crystal driver output pin (input to crystal) with a high impedance probe. That should be a good strong signal. It could be sortof square wave looking with a bit rounded corners or nearly a sine wave. It should be a few Volts p-p, usually half the supply voltage or more. Check the frequency carefully to make sure it's not running on a harmonic. If it is, it's a indication one or both load caps aren't connected or are too low. Then look at the crystal output. That should be a nice 24 MHz sine wave at least a Volt p-p, preferably a little more. Does the unit suddenly work right when you put the scope probe on a crystal lead? If so, it's again missing, bad, or incorrect load caps.

  • \$\begingroup\$ I'm actually confused by the load caps here. There's no info in datasheets concerning crystal specs of any kind except for "24MHz". I went with what the eval board used, which was an 18pF crystal with 10pF load caps. I experimented with different load caps (20pF, 33pF) with no effect. I recall the INPUT half being about 1V p-p. I'll measure the output from the processor this evening. The odd thing is this drives the whole processor, and everything else appears to work. It's just the Ethernet reference clock PLL that's messed up. (unless everything else can just deal with the variability) \$\endgroup\$
    – darron
    Oct 19, 2011 at 17:56
  • 1
    \$\begingroup\$ @Darron: The spec for the load caps comes from the crystal, not the microcontroller. 10pF sounds quite low, especially if the crystal wants 18pF. Note that is what the crystal wants to effectively accross its leads. The load caps are in series from that perspective, but there is also stray capacitance accross them and the output impedance of the crystal driver. The crystal may be unstable by getting close to overtone mode. Try two 22pF caps and see what happens. \$\endgroup\$ Oct 19, 2011 at 19:20
  • 2
    \$\begingroup\$ @Darron: You say the processor works, but have you checked the frequency and jitter? Run a short loop that just toggles a output pin and see how solid and accurate that is. \$\endgroup\$ Oct 19, 2011 at 19:21
  • \$\begingroup\$ I'll do a pin test tonight. I did try low-20s pF load caps (I forget exactly what right now, away from my lab), and a pair of 33pF caps. Both had no discernible effect on the blurring waveform of the reference clock. I thought it seemed low too, but trying other values didn't seem to do anything. \$\endgroup\$
    – darron
    Oct 19, 2011 at 19:45
  • 1
    \$\begingroup\$ Are you really really sure you are running the PLL in the micro within its parameters? Some newer micros can have quite complicated clock chains, and it's possible to set them up so that one or more sections are out of spec. Usually there is a min/max frequency for each node in the clock chain. Make really sure these are all adhered to. \$\endgroup\$ Oct 20, 2011 at 14:05

Your Answer

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

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