What Happens When Routing Over Splits In Power Planes — Transcript
Full transcript
- 0:00Hello everybody, welcome back to Altium
- 0:02Academy. I'm your host Zach Peterson and
- 0:04today we're going to be looking at
- 0:07routing over splits in power planes.
- 0:10Should you do it? We're going to answer
- 0:11that question in this video. Now, Burt
- 0:14Simonovich recently put out a paper in
- 0:16Signal Integrity Journal that examined
- 0:18this and I wanted to examine the same
- 0:20thing in Altium Designer and Simbeor.
- 0:23What I'm going to do in this video is
- 0:24show you how to set up a test board
- 0:25properly in order to do a simulation of
- 0:28this and then we'll look at what happens
- 0:30when we actually route over a split
- 0:32between two large power rails with
- 0:34stripline routing. Make sure to hop into
- 0:36Altium Designer and follow along and
- 0:38let's get started.
- 0:50Now, the situation that we want to look
- 0:51at in this video involves stripline
- 0:53routing and it typically arises when you
- 0:55have six-layer boards. Typically, when
- 0:57you have a six-layer board that's
- 0:59designed for high speed, you'll have
- 1:01signal on the top layer and then you'll
- 1:03have
- 1:04symmetric signal on the bottom layer.
- 1:07Then on layers two and layer five
- 1:12you will typically have ground
- 1:14and then another layer with ground.
- 1:16Now, one of the reasons that you might
- 1:17use a six-layer board is because one of
- 1:20your internal layers can also be sig.
- 1:22So, we have some signal routing in this
- 1:24layer, but then you can typically have a
- 1:27power layer here as well. Now, one thing
- 1:30that often happens in these situations
- 1:32with power layers is you may have two
- 1:34different rails. One routed with V1 and
- 1:38then the other routed with voltage V2.
- 1:40Now, as you're routing your signal layer
- 1:42around in this internal part of the
- 1:43stackup, you may end up routing over one
- 1:46of these splits in this power layer.
- 1:49Does this split actually matter? Well,
- 1:51that's the situation that we want to
- 1:52look at in this video and we're going to
- 1:54run some simulations to determine
- 1:56whether or not it does actually matter.
- 1:58I think the intuition here is that
- 2:00because you have a ground plane above
- 2:03this signal layer, then the signal layer
- 2:05is still going to be strongly coupled to
- 2:07the ground plane, and you may expect
- 2:10that the impedance deviation here in
- 2:12this region where you have the split
- 2:14between your power rails might be
- 2:16negligible. And as a result, you can
- 2:19probably ignore this. Well, I think this
- 2:21is one of those things that require some
- 2:23context. At certain frequencies, it may
- 2:25not matter, and at other frequencies, it
- 2:27might actually matter. So, this is one
- 2:29of those things that you really should
- 2:30simulate. The reason is that because
- 2:33here in this region where you have the
- 2:35solid power plane, you have the
- 2:37impedance determined by layer two and
- 2:40layer four. And so, you would then put
- 2:42layer two and layer four as your
- 2:43references into the impedance
- 2:44calculator, such as if you're using the
- 2:47layer stack manager in Altium Designer.
- 2:48However, in this region, your reference
- 2:50is no longer layer four. Your reference
- 2:53is actually layer two in this region,
- 2:55and then layer five down here. So, you
- 2:58have slightly more distance from the
- 3:00signal to the bottom reference layer,
- 3:03which could then create your impedance
- 3:04deviation. Will this actually matter?
- 3:06Well, it depends on the input impedance
- 3:08looking over this direction through the
- 3:11signal layer. Now, this is something
- 3:13that we can actually simulate in an
- 3:15electromagnetic simulator. And what we
- 3:17want to do is look at the S parameters.
- 3:19So, since we're dealing with an input
- 3:21impedance deviation right here in this
- 3:23region, we would then want to calculate
- 3:25our S11 value for the traces that are
- 3:28being routed over this region. Now,
- 3:30another question that you might ask is,
- 3:32is there going to be a difference for
- 3:34single-ended signals versus differential
- 3:36signals? So, for single-ended signals,
- 3:39the only reference you have are these
- 3:41plane layers above and below the signal
- 3:43layer. But, if you have a differential
- 3:45signal, the differential signal can
- 3:47provide the return path, and it
- 3:48functions as the reference for one of
- 3:50the other traces. So, is that going to
- 3:52produce a difference? We're going to
- 3:54look at that situation as well. We'll
- 3:55look at single-ended S parameters and
- 3:59we'll look at
- 4:01differential S parameters when we do our
- 4:03simulation. Now, let's jump into Altium
- 4:05Designer. We'll take a look at how to
- 4:07set up a test board for this type of
- 4:09situation and then we'll throw that into
- 4:11a simulator and we'll get our S
- 4:12parameters and we'll see what the
- 4:14difference is between routing over solid
- 4:16power versus routing over split power.
- 4:18So, now I'm inside of Altium Designer
- 4:20and what I want to do is show you how to
- 4:22set up a test board in order to
- 4:24accurately perform simulations for this
- 4:26power plane split. So, here inside of
- 4:27Altium Designer, what I've done is just
- 4:29created kind of a simple model where I
- 4:31have two different sizes of gaps for our
- 4:33power plane split and then I've created
- 4:35an identical set of interconnects over a
- 4:37solid ground and power plane. Now, if we
- 4:40look at the stack up, you can see here
- 4:42where our power ground and signal planes
- 4:44are located. Here on the top and bottom
- 4:46layers, we have uniform ground. Here,
- 4:49layer two is our power layer, that's
- 4:51where our split is going to be and then
- 4:53layer three is our signal layer, so
- 4:54that's where our signals are being
- 4:56routed. In this example simulation model
- 4:58that I've created, I've placed two
- 5:00different size gaps. So, here I have a
- 5:02400 mil gap and here I have a 200 mil
- 5:04gap. And then if you look at the
- 5:06schematics here, I've defined
- 5:08single-ended and differential traces
- 5:10that we can use to simulate what happens
- 5:12for our different types of signals when
- 5:13they route over these gaps. Now, on each
- 5:15end of these nets, what I've done is
- 5:17I've created these dummy components and
- 5:18these dummy components are basically
- 5:20just a pad. That's going to make it very
- 5:22easy to assign a port on each end of
- 5:24these interconnects when we then import
- 5:26this into the simulation tool. Now, one
- 5:28thing I'd like to do before we jump into
- 5:30simulation is to just discuss these via
- 5:32transitions. Now, you can see here I've
- 5:34placed some stitching vias around these
- 5:36signal vias and then you can see that
- 5:38there's kind of a default anti-pad
- 5:40around these signal vias. Depending on
- 5:42the frequency range that you're working
- 5:43in, you would actually want to design
- 5:46these vias so that they have a 50 ohm
- 5:48impedance specifically in the frequency
- 5:51range that you care about. For this
- 5:53demo, we're not going to worry about
- 5:54that because that's a whole 'nother
- 5:56level of complexity, but in general, if
- 5:58you were going to create a simulation
- 5:59model and whether you were going to use
- 6:01ANSYS or use Symbeor or some other
- 6:03simulator, you would want to make sure
- 6:05that that via transition provides
- 6:07correct matching. If you can't do that,
- 6:09you would then need to relocate the pad
- 6:12for this port to an internal layer and
- 6:14eliminate those vias. Now, the other
- 6:16thing that you might want to do, again,
- 6:17depending on the frequency range you're
- 6:19working in, is move the location of this
- 6:22split closer to your input port. So, for
- 6:24lower frequencies, you can have this
- 6:26somewhat farther away, whereas at higher
- 6:28frequencies, you would want to move this
- 6:31gap closer to the input. So, to do that
- 6:34is pretty simple. You're, of course,
- 6:35just dragging the gap here from these
- 6:37polygons closer to your input side.
- 6:40Then, you would just size it so that you
- 6:41have the correct gap size. Here, I'm
- 6:42going to leave it here and just in the
- 6:44middle, and we're going to see what
- 6:45happens when we simulate the S
- 6:47parameters. We're going to look at how
- 6:48the different size gaps influence our S
- 6:51parameters in different frequency
- 6:53ranges. For this set of simulations,
- 6:55we're going to use Symbeor. To get your
- 6:56board into Symbeor, you would just do an
- 6:59ODB++ export, and then you can re-import
- 7:01the board back into Symbeor and start
- 7:03selecting the nets that you want to
- 7:05simulate. So, now I've brought the board
- 7:06up in Symbeor, and now I can start
- 7:08selecting the ports that I want to
- 7:10simulate. Here, when I'm in Symbeor,
- 7:12what I can immediately do is just run an
- 7:14ERC, and I can see what the expected
- 7:16impedance deviation is along this
- 7:19section of interconnect. So, you can see
- 7:20here for the differential pair with the
- 7:22200 mil gap that we have a 6 ohm
- 7:25deviation in the differential impedance.
- 7:28Here, if we go back to the single-ended
- 7:30port and I run the same, you can see
- 7:33that I get an 8 ohm deviation. So, it's
- 7:35actually much more significant when you
- 7:37have the single-ended signal versus the
- 7:40differential signal. For the
- 7:41single-ended signal, it's a almost 20%
- 7:44deviation in the impedance, whereas it's
- 7:46only a 6% impedance for the differential
- 7:48pair. So, that should underscore the
- 7:50importance of having differential pairs
- 7:53and sometimes having them spaced closer
- 7:55together. When they're closer together
- 7:57and you route over this kind of split,
- 7:59you're going to have a smaller deviation
- 8:01in the differential impedance. That's
- 8:03because one trace always references the
- 8:05other trace. Next, let's start selecting
- 8:07our ports here and then we can set up a
- 8:08quick S-parameter simulation. So, here
- 8:10I'm going to select our 200-mil gap, our
- 8:12400-mil gap, and then our uniform plane.
- 8:15And I'm going to run a fast SI
- 8:17simulation to get the S-parameters. So,
- 8:19now we have our return loss data pulled
- 8:21up and the return loss data pretty
- 8:23clearly shows what happens when you have
- 8:25that split in the plane, especially when
- 8:27it's a reasonably large split. Here,
- 8:29when we look at really low frequencies,
- 8:31we see that all the S-parameters pretty
- 8:33much overlap with each other. And this
- 8:35is when we're sub 100 MHz or right
- 8:39around 100 MHz. Once we start to get
- 8:41into the gigahertz range, we really
- 8:43start to see the deviation in the
- 8:44S-parameters really quickly. Here, the
- 8:46blue curve is for S33 and that's for
- 8:50this middle trace right here. So, that's
- 8:52where we have the uniform plane. Then
- 8:53here, the red is S11, so that is for our
- 8:57top left corner here. That's where we
- 8:59have our smaller gap. And then here, the
- 9:01brown or the orange curve, that is our
- 9:04trace down here in the lower right,
- 9:06which is our 400-mil gap. So, this is
- 9:08where we start to see a pretty big
- 9:09deviation in the S-parameters in the 1
- 9:12to about 7 or 8 GHz range. And you can
- 9:16see here that we do get pretty high
- 9:18return loss compared to the trace over
- 9:21uniform planes. However, is this too
- 9:24much reflection? Well, as we can pretty
- 9:26clearly see in this case, the return
- 9:28loss does get pretty high, but it
- 9:30doesn't jump up above -10 dB. In a lot
- 9:33of cases, this would probably still be
- 9:35acceptable even if we're routing over
- 9:38these large gaps. Now, once you get
- 9:40around 10 GHz and higher frequencies, we
- 9:43see that the S parameters all start to
- 9:44overlap each other a little bit. So,
- 9:46this is because the dominant return loss
- 9:49mechanism switches from the gap in the
- 9:52power plane to the loss or the impedance
- 9:54deviation that's created by the loss in
- 9:57these interconnects. Now, remember, in
- 9:59Altium Designer, it's determining a
- 10:01lossless impedance, but the real
- 10:02interconnects have dielectric loss
- 10:05tangent as well as skin effect that
- 10:07create deviations in the impedance, and
- 10:09that's what's creating these reflections
- 10:10at much higher frequencies. So, our
- 10:12conclusion is this, at really low
- 10:14frequencies, sub 1 GHz, you're really
- 10:16not going to notice a difference. Once
- 10:18you get into the low GHz range, you do
- 10:20start to see a large difference.
- 10:22However, with the small spacing that we
- 10:24have between ground and signal, it
- 10:26doesn't create so much loss that the
- 10:28interconnect may not still be usable.
- 10:30Once we get to really high frequencies,
- 10:32the split in the plane becomes less
- 10:35important, and you have other loss
- 10:36mechanisms that are creating the
- 10:37impedance deviation that you see in
- 10:40these S parameter plots. Now, let's take
- 10:41a look at the differential S parameters
- 10:44for all of these differential pairs in
- 10:45this test board. We have all of the
- 10:47differential S parameters prepared. Now,
- 10:49what you can see here is that all of the
- 10:52differential S parameters are pretty
- 10:53similar. We have the same ordering here
- 10:56for these curves by color. The gapless
- 10:58and the small gap and the large gap are
- 11:01all really similar all the way up to
- 11:03really high frequencies. This can be
- 11:05explained in a couple of ways. First of
- 11:07all, as we saw earlier, the impedance
- 11:09deviation when routing over that split
- 11:11in the power plane was smaller for the
- 11:13differential pairs. Remember, it was
- 11:14only 6%, whereas for the single-ended
- 11:16traces, it was almost 20%. Next, we have
- 11:19the same change in the dominance of the
- 11:22return loss from low to high frequencies
- 11:25as we had in the single-ended case. So,
- 11:27just because we're in differential
- 11:28pairs, doesn't mean that the loss
- 11:30tangent and the skin effect go away.
- 11:33They also will dominate and create an
- 11:35impedance deviation seen looking into
- 11:37the interconnect. And so that's what's
- 11:39creating this big deviation at higher
- 11:41frequencies and then creating this
- 11:42higher return loss rising just above -10
- 11:45dB. So this channel would still be
- 11:47appropriate all the way up to about 20
- 11:49GHz, but to really perfect this we would
- 11:52want to do what I stated earlier both
- 11:54for the single-ended transition and the
- 11:56differential transition. And that would
- 11:58be to optimize these vias so that we get
- 12:01a really flat near 50 ohm single-ended
- 12:04impedance or 100 ohm differential
- 12:06impedance within the bandwidth that we
- 12:08care about. That's going to help
- 12:09eliminate some of this discrepancy and
- 12:11then it's going to give us a little bit
- 12:12more accurate view into what's happening
- 12:15with our system. Now as I've shown in
- 12:16some other videos, we really don't care
- 12:18about the via impedance until we get
- 12:20above a couple of GHz. So if we were
- 12:23only looking from for example
- 12:25uh up to 2.5 GHz here, we could get a
- 12:28really clear view of what's going on in
- 12:30both sets of curves. Here if we just
- 12:32zoom in to the 2.5 GHz range for our
- 12:36single-ended uh traces, we see here the
- 12:38really big impedance deviation that we
- 12:40get when there's a gap versus no gap.
- 12:42And it's a it's a little over 20 dB. For
- 12:45the differential pairs, we have a much
- 12:46smaller deviation between the return
- 12:48loss curves, so it's a little bit more
- 12:50difficult to conclude that the split in
- 12:53the power layer is the sole cause of all
- 12:55of this behavior. To really nail that
- 12:57down, we would definitely want to
- 12:59optimize those vias or we could bypass
- 13:01the vias completely and just move those
- 13:03pads into the inner layer and then
- 13:06assign our port directly to those pads
- 13:08on the inner layer. Make sure to take a
- 13:09look at the blog that's linked in the
- 13:11description to learn more about this and
- 13:12I'll discuss this further in that blog.
- 13:14Thanks for watching everybody. Make sure
- 13:15to hit that subscribe button, hit that
- 13:17like button, leave your comments and
- 13:18questions in the comment section, and of
- 13:20course don't forget to call your
- 13:21fabricator, folks. We'll see you next
- 13:23time.
- 13:31Mhm.
About this transcript
This page contains the full transcript of What Happens When Routing Over Splits In Power Planes by Altium Academy, generated from the public captions YouTube serves with the video. The transcript has 2,619 words across 382 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.