This is part 3-1, the beginning of my personal favourite session of the entire series, where we’ll talk about PCB design for the best possible electromagnetic compatibility or EMC.
Section 3-1 Agenda
• Electron “psychology”
• Inductors – friend and/or enemy
• What ground is and isn’t
• Signals as currents, not voltages
• An impedance experiment
• The four ways that noise couples
• Electric field and magnetic field shielding
To begin, I’d like to say that you can read this part of the series without reading any other section first. It stands on its own. The concepts I present here apply not only to all the switching regulator topologies, but in fact, to pretty much all electronic circuits, with the possible exception of radio frequency things. Whether your design is made of analog amplifiers or digital microprocessors or both, the PCB is a critical component. In this session, we’ll begin by looking at inductors and how they can either work for us or against us. Then move on to the ground connection as a practical concept. From there, signals as currents will be explored followed by an experiment designed to explain the differences between resistance and impedance. The final parts of section 3-1 will be exploring how electrical noise gets from its victim to the source, and then a look at shielding from electric fields and magnetic fields.
Electron Psychology 101
When you finish with all three sections of this PCB webinar, you will hopefully have gained some insight into why electrons do what they do. In effect, I hope to convert you into an electron psychologist. Please excuse my poor excuse for a divan, where the electron is sitting. That black bar represents the electrons negative charge. No psychology discussion will be complete without Dr. Freud, naturally. In all seriousness, the best PCB layout engineers I ever had the pleasure of working with were also the best analog electronics engineer. One thing they all had in common was enough experience to know almost intuitively where electrons were going to go. The honest truth is that it’s not always obvious which path an electron will take through a PCB or any other network, but the more you work with them, the more you’ll be able to predict where they go, and ever more importantly, force them to go where you want them to go.
We’ve put our electron on the divan, and now we’re going to ask it, why do you do what you do? If one person ever truly knew exactly where an electron was going, it was this man, Dr. Erwin Schrödinger. Our electron here says as much. Then, of course, Dr. Schrödinger couldn’t tell you exactly where that electron was, but that’s enough quantum humor for one slide. Schrödinger and his wave equations are an excellent way to describe electron behavior mathematically, but we’re going to look at electron behavior from a standpoint that’s nearly all practical. Now, if I ever get to be as respected in scientific circles as he is and was, maybe I can get away with that electrified hairdo as well.
If you watch part two of this seminar, which was all about the buck regulator, you might have gotten the impression that the input capacitors were the most important part of a switcher circuit, but it’s really to power inductor. A good subtitle for this presentation would be the inductor, your best friend when used properly, and your worst enemy when it shows up in paces it shouldn’t. I had the fortune of watching two great engineers present about PCB layout for switchers. One of them, my old boss, Craig, repeated the following phrase like a mantra, “Always minimize stray inductance in the power path.”
Now, stray inductance and parasitic inductance are synonyms. Both refer to the fact that whenever a current flows through a conductor, any conductor, it generates a magnetic field. The characteristics of that magnetic field, which stores energy depend upon many factors. One extremely important factor is the geometry of the conductor. Another is the frequency of that current. PCB design is all about compromise. In this presentation, I plan to explain which portions of the circuit can handle added stray inductance without problems, and which portions of those circuits should be laid out on your PCB for the lowest possible inductance.
Here, another of my terrible drawings, you have a PCB layout engineer sleeping under his desk, but he’s not avoiding work. He’s dreaming about the ideal concept of ground. What we all wish ground was is an ideal conductor with zero resistance and even more importantly, zero impedance. Remember that impedance takes frequency into account, so this magical ground connection will be completely free of inductance and capacitance as well as having zero resistance. Even a true superconductor still has inductance and capacitance, so those dream of ground is unfortunately not a realistic one.
When our engineer wakes up, I want him to avoid a costly mistake, assuming that the ground plane in his PCB is ideal. No matter how solid, thick, or continuous the ground connection is, it’s nothing more than a piece of copper, with non-zero resistance, non-zero inductance, and non-zero capacitance. One of the biggest errors I see in PCB design are connections to a ground plane made wherever it’s convenient, with the assumption that electrical noise will somehow disappear into the ground plane like magic.
Signal Return is NOT Ground
The ground plane or ground connection in your PCB is not a perfect sink that eats electrical noise. Its two job is to act as a reference, representing zero volts for all the different circuits of varying types that you have. Since it has resistance and impedance, as soon as a current flows in the ground connection, a voltage differential is created. That means that two circuits in two different places in a PCB don’t see the same zero volt reference level, even if the ground is a nice solid copper layer.
In theory, the best thing to do is to ensure that every current that goes out has a dedicated return path that is separate from the ground plane. In practice, you often won’t have enough layers to do this. Now, when I say layers, I really mean budget, mostly money, but also time. In a perfect with all the PCB layers you could possibly need, all your return currents will flow locally, with a single point connection to a ground plane that has no current flowing in it, and then everything would truly have an equal reference of zero volts. You probably already guessed that in practice, currents will flow in the ground plane, so the trick is to control where they flow and where they don’t.
Signals As Voltages
We’re all accustomed to thinking of signals as voltages. Voltages are definitely much easier to measure, not to mention cheaper. Every oscilloscope or multimeter I’ve ever seen came with at least one set of voltage probes, but I’ve never seen even the most premium equipment come with a current probe. I never ever saw current probe capable of measuring DC signals until I worked at an academic research lab, but voltages are a poor choice to represent signals, and voltage probes are easily fooled. Take a look at this simple buck converter. Any analysis of the circuit will refer almost exclusively to various voltage nets, Vin, Vswitch, Vout, and of course, ground.
Signals Are Currents
My former colleague, Jeff, was the first person I ever heard insist that signals are actually currents. A current is a genuine physical thing, a flow of electrons going backwards, as we see in this drawing. You can touch and feel electrons, but please take care when doing so. Voltages are just potentials. It often helps me to try and visualize actual electrons, or if you prefer, holes, moving along in the circuit and on my circuit boards. When you start to imagine these little guys moving around, it can really help to understand why good PCBs are laid out the way that they are.
Same Circuit with Current Loops
Let’s go back to the simple buck converter from before, but this time, we’re going to analyze it from a signal current perspective. When the control MOSFET turns on, current flows as shown by the orange arrows, notice that the input source in the far left provides some of the current, and the input capacitor just to the right of the source also contributes. In a perfect world, the source provides only DC current, and the capacitor supplies all the currents and all the frequencies above zero hertz.
The current turning on time flows through the output inductor, where we know from the laws of physics that it cannot rise or fall rapidly. It then flows through the load and returns to the common reference ports of the circuit, meaning of course ground. When the MOSFET turns off and the diode turns on, the current, in blue, continues to flow in the inductor. After all, that’s why we have the inductor there. The current flows in the ground connection for part of the way, but then complete its loop to the diode. During this time, the source recharges the input capacitors as well.
Now, let’s see the circuit with the currents in the two switching states. There are two things to look for in this type of analysis. One, any paths that show only one color, this is a dead giveaway that a heavy switch current flow is in this section of the circuit. Another way to say heavy and switched would be to say high di/dt. Regardless of how you say it, this means the current going from zero to a high value and back again, and that means fast edges in the current signal, the vertical edges in fact. Hopefully, it’s no surprise that the path through the control FET and through the recirculating diode see high di/dt currents. The critical section here is the portion of ground extending from the anode of the diode over to the negative of the input capacitor.
Two, any paths where the current flows in one direction during one portion of the switching cycle, and switches direction in the other portion of the switching cycle. The input capacitor is the main victim of this type of high di/dt current signal. The output capacitor sees a current that changes direction too, but the output conductor reduces that delta, so that the output capacitors suffer far less than the input capacitors. Now that you have identified these sections where heavy current exists, minimizing the inductance of those sections is critical. Remember that V = L * di/dt, and that V is purely voltage noise, or electromagnetic interference.
Self Inductance of a Loop
Okay, you’re thinking, reduce the inductance, but how? The answer is mainly by reducing the loop area. I don’t actually calculate the inductance of the loops in my PCB, and I don’t suggest that you do it either. What I do want you to do is place the components that carry heavy current, those we call power path components, in such a way that any paths with high di/dt, that is heavy switch current, have as little area as possible. That will mean compromise, basically, that some loops will be larger. Make sure that the larger loops have lower current, slower edges or both.
Self Inductance and Geometry
Thicker wires or thicker PCB traces do reduce inductance, but on an inversely logarithmic basis. Your PCB traces, or even better, copper shapes or polygons need to be thick enough to carry their currents without overheating, but making them very, very wide does reduce inductance all that much. Instead, focus on loop area, with its direct proportionality to inductance.
The first time that I saw Jeff, the guru of signals as currents present on PCB layout, I was intrigued, because he had gone to the effort of dragging a bunch of lab equipment into the conference room. If I could only come to each viewer’s office or living room, I’d repeat his experiment, because seeing really is believing.
Give an Electron a Choice…
As we watched Jeff assembled a signal generator, three current probes, by the way, the nice thing about working at a big semiconductor company is their equipment budget, the oscilloscope and the length of the coaxial cable about one meter long bent into a loop, and a bunch of stuff sodded onto the ends.
We all got up to look more closely and saw approximately what I’ve drawn here on the slide, the signal generator with its 50 ohm output impedance was connected as I’ve shown. We all agree that the electrons had no choice but to travel down the center conductor drawn in green. Then they all flow through a nearly matching 51 ohm load. That’s where things got interesting, because once they got to the negative end of that 51 ohm load resistor, the electrons had a choice. A short thick went to the shield of the coaxial cable. Another short thick wire went back to the negative of the source.
I’ve only got two current probes, so I added a one ohm resistor across which I measured a voltage that’s proportional to the complete input current, IN. The electrons could either go back to the shield of the coaxial cable, traveling another meter or so, or they could take a more direct path of about five centimeters. “Which path will carry more current?” Jeff asked. Now I ask the same question of you. Where would the electrons go? If you smell a trick question, you’re on the right path.
A cautious engineer always responds in the same way to such questions, because this answer is never wrong. The cautious engineer says, “It depends,” and it always does. At low frequency, the yellow path is far less resistive, so most of the current goes that way. As frequency increases, resistance becomes less and less important. Inductance begins to dominate the impedance. This is a fundamental property of electrons. They’ll always find the path of lowest impedance. Impedance means the dependency upon the frequency.
The yellow loop here has much higher area, so it has much higher inductance. The hatched blue line show us the area enclosed between a central conductor of a coaxial cable and the shield. This type of cable specifically designed to carry high frequency signals has very low inductance on purpose. Compare that to data of the whole loop and yellow hatched lines, and you can see that the blue path has far less inductance.
This is the impedance experiment. Signal starts at the generator, follows that cable, and it’s injected on the board here by the red and black connectors, and it goes down the semiconductor of this coaxial cable, and a long loop, a little bit over one meter long, lots of inductance in this bug area loop. When that signal gets back to the circuit board here, it goes through a 51 ohm load resistor, and then it has a choice. It can go through either the yellow cable or the blue cable. I’ll clip a current probe around both of those. The yellow cable goes back to the shield of the coax cable, so the signal would have to go all the way back around to get to the source. The blue cable goes almost directly back to the source, the signal generator. It just passes to a one ohm resistor underneath this low impedance test fixture, and then back to the source.
Here we have a plot of the impedance experiment at 10 hertz, and it’s easy to see that almost all the current flows through the blue path. Remember that right now, that’s the path of least resistance. I’m going to increase the frequency slowly up to 100 hertz while adjusting the oscilloscope, and we’ll see the change is not much. 100 hertz is effectively a low frequency. Here is the system again starting at one kilohertz, and I’m going to go about increasing at one kilohertz increments.
Now we really start to see, right? Somewhere right around here is where the impedance of both path is about equal, right? That means that the combination of the resistance and the inductance has come together to make overall more or less equal impedance. This plot starts at 10 kilohertz. I’m going to go about increasing it in 10 kilohertz increments, and now we quickly begin to see that it’s the path of least inductance that is now the path lowest impedance, and much more of the current flows through the yellow path.
Finally, here is the same system operating at 10 megahertz. At this point, it’s pretty much reversed. Barely any current that’s flowing back directly through the shield, almost all the current is going back through the coaxial part, and that’s because now inductance completely dominates the impedance.
Busting Myths – RF interference is rarely the culprit
If you haven’t seen it, MythBusters is a TV show where some crazy ex-stuntman blow things up in order to challenge to commonly held assumptions. In power electronics, you definitely need to blow some things up in order to get a real feel for how big voltages and big currents work, but the main assumption I want to bust here is that radio frequency interference is the source of most problems.
The Only Four Ways That Noise Couples
- Conducted: two currents in the same copper
- Near-field Magnetic (inductors and transformers)
- Near-field Electric (capacitors)
- Far-field Electromagnetic (Radio)
There really are only four ways that electromagnetic interference, EMI, gets from a source to its victim. For switching regulators, the vast majority of problems exist due to reason number one, conducted interference.
When a noisy current and a sensitive current share the same conductor, those electrons gets mixed up. You might already suspect that ground planes and ground connections are commonplace for this to happen, and you’d be correct. The positive power rails or power supply voltages are another shared connection where conducted noise can also propagate. In descending order of how much trouble they create come near-field magnetic noise, the straight lines of magnetic flux, the couple, and the conductors passing nearby inducing noise currents.
In descending order of how much trouble they create come near-field magnetic noise, those straight lines of magnetic flux, the couple, and the conductors passing nearby inducing noise currents. Then comes near-field electric noise, which is when unwanted capacitance develops between something noisy, like the train tab of a power MOSFET, and something quiet, like the safety earth connection in an AC to DC power supply.
Last, and in this case, least is far-field electromagnetic noise, or radio frequency noise. This is the myth that I want to bust, because many engineers seem to default that RF interference is the problem, quite possibly because it’s the least understood. Unless you’re building a truly huge power supply, the distances within your PCB are too small for RF to be the problem. Focus on conducted noise from shared connections.
Near-Field Coupling, Magnetic
Let’s get the EMI sources of modern influence out of the way before concentrating on conductor problems. Inductors and transformers are sources of magnetic fields, mainly from the straight fields, meaning the lines that I show here that don’t close completely within the magnetic path. I’ve drawn a non-shielded pot core inductor here, and as the name implies, this type of inductor tends to emit more EMI than shielded inductors. I would say now that shielded inductors are much more common in off-the-shelf catalog parts, but a non-shielded device is often cheaper, so they’re still around.
Any conductor passing within the lines of straight magnetic flux will have a current induced, and since voltage equals current times impedance, it’s the high impedance sectors of your circuit that need to be kept away from the magnetics. For power supplies, the high impedance lines of the voltage and current circuits, which are usually fed into off-amps or transconductance amplifiers.
Magnetic Field Shielding
The bad news here is that magnetic shielding is expensive. Ask anyone who has ever built or paid for a chamber for radiated emissions testing. Very few power supply budgets include money for ferrite-impregnated shielding sheets. The way to reduce straight magnetic flux is to self-shield. Looking at these examples, the gray dots represent currents coming out of the screen, and the red Xs represent currents going in. Then the lines around them represent the magnetic flux.
My “excellent example” here is nothing more than coaxial cable, where nearly 100% of the flux of the current going out is canceled by the equal but opposite flux of the current going in. If we could build our switchers using some sort of high current coaxial cable, they would be great for EMI, and would have a fascinating 3D look too, but that’s not very practical.
The terrible example is what you get when you have a big loop. Inductance is once again snuck back into our discussion. That brings us to the “compromise example.” This means knowing how the signal got to the load from the source, and minimizing the area between the send and return paths. We can’t make coax with the PCB. We can run the send and return as differential pairs, or use ground and power planes like strip line cabling.
Near-Field Coupling, Electric
Electric field coupling is less likely to cause your power supply design to malfunction, but much more likely to drive you crazy if you have to do conducted EMC testing at the input or the output of your design. To put it bluntly, big things with quickly changing voltages, that is to say, with high dV/dt. Couple that noise capacitively into anything that happens to be placed parallel to them. The list of big things is quite short. Inductors or transformers are the biggest followed by power switches.
The typical victim is earth ground, which is always present during EMC testing. One of my favorite tricks shown here is become obsolete due to newer packaging. In the days of the TO-252 or the TO-263, you could stand those devices on their legs, pins one and three, and connect the solid wire for pin 2. Once the suspected noise emitter and noise receiver were perpendicular, the capacitance between the two of them would drop off dramatically. Now, you can still do this with most simple inductors, but a transformer with 12 pins is not easy to stand on end.
Electric Field Shielding
The good news here is that you can shield against electric field noise much more easily. Just about anything with a lower impedance back to the source of the noise will do, and not just ground. If you watch part two of this webinar about the buck converter, then you saw slide after slide talking about how the input voltage and output voltage to switchers need a blend of different capacitors to provide low impedance over a wide range of frequencies. Now you can take advantage of that low impedance, and use Vin or Vout as a shield. Even the output of a good off-amp can sometimes be used. Route your low impedance connection between the noise source and the victim to short out those electric field lines. As a final note, the typical victim is the same as with magnetic fields, anything that’s high impedance.
Next Up: Section 3-2 – PCB Layout for EMC, Part 2
If you do come back for more, we’ll start in Section 3-2 by looking at conducted noise, and see an example of how to determine if the voltage noise you’d see at a given node is due to electric or due to magnetic fields. Then we’ll look at how holes and slots in otherwise solid ground planes or power planes can create intendance that radiate unwanted noise, a review of best practices for measuring both currents and voltages in the lab is next, followed by slides on how to get the most out of filter capacitors, especially for higher frequencies. The last part of section 3-2 will discuss functional isolation, meaning the distances required to keep your circuit boards both functional and not arching.
Part 3-2 of our Power Supply Design series will be available as of March 26, 2018.
Link to previous section: 2-3 The Buck Regulator, Part 3