Tiny Difference

The snow is falling thick and fast this winter. Our mile-and-a-half of steep gravel road has become so clogged with snow that our snowplow operator, Charlie, can no longer push the snow off to the side. There's no place left to put it! Charlie's no slouch. He usually drives an F-350 truck with an 8-foot snow blade. His rig weighs about 9000 pounds (loaded with rocks) and even with all that he just doesn't have enough traction to get the job done. So, today we ordered him some heavy machinery. On Friday he'll be moving snow with a full-sized road grader. This vehicle weighs 50,000 pounds and has tires that stand almost five feet tall. I call it the "ultimate off-road vehicle". With that sort of horsepower at his command Charlie should, as Larry the Cable Guy says, "Git 'er Done".

Regarding Anjaly's message below, we never satisfactorily determined the problem with his layout (or if he even had a problem), but I did come up with some tips that might help yours! Read on.


Tiny Difference

Anjaly Chandran writes:

I am puzzled by the skew difference on a differential pair and would like your inputs (see Figure 1). Each purple trace is 24mil wide and 2.4 mil thick with a 13-mil dielectric thickness. The board is made from FR408 material.

Each trace is routed as a single ended pair with almost 75mil clearance between the two traces of each differential pair to reduce inter-pair coupling. Layer 2 provides a solid ground underneath the traces. The trace lengths are matched to within 2 mils, but when I measure the skew I find almost 5 ps difference in delay between the two traces. One thing to point out is that these traces are tapered as they approach the package pin, but the tapered length is almost the same, differing only by 3 mils or so. Also, there is an AC coupling cap (2.2nF) on each trace. To cause a 5 ps delay on a micro strip requires a 30-35mil trace length difference (assuming ~140ps/inch).

What could cause the skew between the traces? Any help is highly appreciated.

Wow! Measuring a tiny time difference like 5 ps can be quite challenging. Anjaly will need well-matched, skew-calibrated probes and perfectly symmetric attachments to the board. Before I discuss the attachments, let's look at his layout.

My rendering of Figure 1 does not show the SMA launches. They begin just off the left side of the drawing, connecting through blocking capacitors to the two purple traces that are the focus of this newsletter. The SMA layout for the top trace is made just a tad longer than the one on the bottom. I call that "pre-skewing" the connections, and it was done for a reason that will become clear in a moment. From point a, working to the right, segment a-f comprises four 45-degree turns, two going left and then two going right. Any skew accumulated in the two left turns is precisely compensated by the equal and opposite skew accumulated in the two right turns.

After those turns the signal encounters a straight segment f-g, also balanced, and then a weird tapered part g-h. The taper connects the widely spaced, loosely coupled trace pair to the fine-pitch geometry of the IC soldering pads.

In segment g-h the lower trace appears clearly longer than the upper trace. Anjaly knew this, so he pre-skewed the SMA connections to compensate for the skew he knew would accumulate in the tapered area. Overall, from SMA to IC, the total lengths are matched to within 0.002 in.

So why does Anjaly measure a 5-ps difference in delay? Here are some of my guesses.

  1. The ground terminals of the surface-mounted SMA connectors sit on large layer-1 ground pads that may not be well connected to the actual ground layer that the traces use, layer 2. Near the SMA connectors I see only two ground vias in the layout. Something about the shapes of these layer-1 ground regions and the inadequacy of their grounding to layer 2 looks suspicious. Each SMA connector conveys signal current and also returning signal current (ground current) to your board. Small differences in the returning signal current paths (i.e., ground pathways) on the two connectors could easily cause the 5-ps difference he observes. Recommendation: add more ground vias to the SMA layout, and make the SMA layouts the same.
  2. Anjaly's test board was likely hand-soldered. The blocking capacitors may not have been held firmly against the board during soldering. If one capacitor stands higher above the board than the other, that would cause skew. A difference of 1/64th of an inch in height, going up and back down, would contribute the 1/32nd difference in length that bothers him. Recommendation: test at least three boards to see if the delay differences are repeatable. [Note -- Anjaly subsequently did test a second board, from another supplier, and came up with only 2 ps difference].
  3. When measuring tiny time differences, always reverse the probes to see if the skew stays with the layout or moves with the probes. Five ps of skew can easily be caused by probe mis-calibration. Test your probes all by themselves, on the simplest layout you can manage, to confirm that you can repeatedly observe skew differences as small as 5 ps. To make the test, you will be attaching four probes (two in and two out). A nominal positioning error of, say, 0.01 in. would be more than enough, when multiplied by four attachment points, to completely swamp the measurement you are trying to make. Anjaly must get his probe placement accuracy down to something like 0.003 in. to obtain a good measurement. That amount represents one-tenth of a turn on a 36 tpi threaded SMA connector. Therefore, always use a calibrated torque wrench on SMA connections. At the IC attachment end of the connection, provide a guide hole on the board (accurate to 0.003 in.) to help align the probes. You can press a hardened steel pin into the hole to help hold the microprobes in precise alignment. Recommendation: make a simple test layout that is obviously symmetric and practice using the probes on the test layout.
  4. The layer-1 grounded metal regions may influence the signal traces. I realize Anjaly kept the layer 1 ground regions at least 75 mil. away from the signal traces at all points. That would be far enough on an ordinary board, but at his exaggerated trace height of 13 mils that spacing provides a spacing-to-height ratio of only about 6:1. That ratio suggests to me perhaps as much as 3oupling from each trace to its peculiarly-shaped layer-1 ground region. Not much coupling, but it could be having an effect. And it doesn't contribute anything to the system performance. Why have these regions? Recommendation: move the layer 1 ground regions even further away from the signal traces, or eliminate them altogether except under the SMA connectors where they are required for SMA soldering.

If you have other ideas for Anjaly, please let me know and I'll pass them along. Or, if you join my next public seminar Feb. 5-8, 2007 in San Jose, CA, we can talk about them together.

Best Regards,
Dr. Howard Johnson

I'm talking to the folks at National Semi. about filming another Bob Pease "Analog by Design" show this Spring . If you haven't checked out Bob's programs, I can highly recommend them. You can find all his shows, including the ones in which I appear, archived on YouTube. Search for, "Bob Pease Show"