This part of thePower Supply Design Tutorial is dedicated to edge control for optimum radiated EMI.
Section 4-1 Agenda
- Parasitic capacitance and inductance cannot be avoided
- Noise energy sources depending upon switch type
- When ringing is normal (DCM)
- What are “spikes” and “ringing”?
- Gate resistors
- R-C Snubber selection Method #1
- R-C Snubber selection Method #2
This first section on switching edge control starts with the fact that even a perfect PCB has unwanted capacitance and inductance. Combined with the energy stored in the junctions of the diodes or MOSFETs or BJT power switches, transients and oscillations called spikes or ringing are generated. Before digging deeper, we’ll review a normal situation where an LC oscillation does occur, which is in discontinuous conduction mode or DCM. With that cleared up, the webinar will go on to explain why and how gate resistors can help slow edges and how to select the best values. The last part of section 4-1 introduces the R-C Snubber circuit for damping resonances, and presents two methods for picking the right values.
Parasitic Capacitance and Stray Inductance Are Unavoidable
Even the very best layouts will have parasitic capacitance and parasitic, or so called stray inductance. These are unavoidable physical properties of any real circuit. Add a source of energy to an LC tank and you’ll get resonance, especially when the L and C are mainly parasitic because these elements have very little damping. The energy comes from the power switches, be they MOSFETs or BJTs, fast rectifier diode or Schottky as they turn on and turn off. This resonance is the ringing I’ve referred to multiple times in the previous two parts of the seminar, and it falls firmly in the band of radiated emission limits for the majority of power supplies.
Controlled Switches: Pick Your Poison
One way to reduce the overshoot and undershoot voltage transience is to slow the switching edges. For the resonance that comes afterwards, several methods of damping the LC filters will be presented. If you watched part two on buck regulator design, you heard me recommend Schottky diodes as the recirculating element or low side switch whenever possible. That means up to 100 volts DC, with great pricing and plenty of pin compatible or clone products with the same part numbers from various manufacturers, and up to 200 volts with more limited selection. Reverse recovery charge is a fascinating topic, but I won’t dig much deeper into its origin.
It’s relevance here is that it’s a source of energy that excites LC tanks. Schottky diodes are practically free reverse recovery charge, but are by no means perfect, because they still have plenty of junction capacitance, and that stores energy.
Controlled Switches: Pick Your Poison
If your buck switcher is a synchronous device, then the power MOSFET replaces the diode as the low side switch. Most of the parasitic capacitance is in the value labeled Coss, and it’s notably lower than the capacitance of the Schottky diode from the previous page, but there’s another problem, which is the parasitic diode or body diode of the fit.
I put the word bad in quotations here because, like most anything, it really has pros and cons. The body diode turns on before the MOSFET does, producing switching loses, and that’s wonderful, but the body diode also has a lot of reverse recovery charge, and that’s terrible for the transient on the switching node as it turns off.
When “Ringing” is Normal
Non-synchronous converters (and some synchronous converters with “diode emulation”) ring when DCM
For anyone who hasn’t seen parts one and two of this seminar series, it’s important to distinguish between undesirable parasitic ringing and normal DCM ringing. DCM stands for discontinuous conduction mode, and this scope shot shows why. The inductor current, in green, is discontinuous. It falls to zero before the end of the switching cycle. This happens in standard, non-synchronous circuits of any typology, and it also happens with modern, more sophisticated synchronous control LCEs and regulators that control their synchronous switches to emulate a diode’s one quadrant behavior.
DCM ringing is normal, since the parasitic capacitance of whatever switches are used, forms an LC tank with the full inductance of the power inductor itself when the current falls to zero. Note the ringing period. It’s about 220 nanoseconds in this shot. That corresponds to a frequency of 4.5 MHz. One way to distinguish DCM ringing from parasitic ringing is the frequency. DCM is usually 10 to 20 times lower than parasitic ringing, and that makes sense because the leakage inductance is usually around five percent of the total inductance.
Another way to tell these two oscillations apart is the power level where they occur. DCM occurs only at minimum, or no load, whereas parasitic ringing is worse at maximum load.
Ringing Affects Even the Best Layouts
This scope shot was taken with a non-synchronous buck controller using a P-MOSFET for the control switch. I laid out the circuit board using all of the best practices for PCB outlined in part three of this seminar, but my customer changed their mind halfway through the design, and reduced the maximum output current by more than one half. As a result, even though hit was a P-FET, the gate driver and the control I see was way oversized for the minimal gate charge of the new, smaller FET that I selected.
An oversized driver pumps the gate extremely fast, and that makes the drain to source channel turn on very quickly, and the faster the edge, the more severe the transient and ringing. In this shot, taken at maximum load, there’s not only an EMI problem, but a potential over voltage problem. The peak ringing is about two X the input voltage, which is the theoretical maximum. A MOSFET rated 20 volts, drained of source, could easily fair due to over voltage. You’d never even reach the EMC lab.
Ringing Affects Even the Best Layouts
Here’s the same circuit again, but this time we’re looking at the output voltage, AC-coupled, in order to focus in on the ripple. The reading, at about 200 millivolts, peak to peak, is so pronounced that you might not even notice the switching frequency ripple, which is about 10 or maybe 15 mV peak to peak. This scope shot also shows the market difference between the rising edge ripple, when the control switch turns on, and the falling edge ripple, when the control switch turns off. That’s also when the diode turns on.
There’s a bit of ringing there, but nothing compared to the rising edge. That’s typical for switchers. In just about every topology, one edge will generate more noise than the other. You might be wondering just how this high frequency noise got through to the output, considering that the output inductor sits between the switching node, where this noise originates, and the output, and it ought to filter high frequency signals. The main answer to that is the frequency. At above 100 MHz, where this noise exists, the output inductor is no longer inductive, but is actually capacitive, so the ringing sails right on through.
I always start with gate resistors because they’re so simple and so cheap. A surface mount 0603 or 0805 chip resistor costs less than one penny. I also always include footprints for gate control when I use a controller and discrete power MOSFETS. If I don’t need those footprints, I just stuff them with a 0 ohm jumper, which also costs less than one penny, and maybe two cents or so to assemble. A Gate Resistor
- Slows both rising and falling edges of switch node voltage
- Slowing edges reduces ringing – BUT: too much slowing down increases switching loss
Gate resistors only work when you have external power-MOSFETS. That’s one of the reasons I still prefer controllers over integrated FET ICs or regulators. If you’re not using controllers and external FETs, skip ahead to the R-C Snubber section. When you put resistance in series with a gate, you limit the current that charges or discharges the gate capacitance. By the way, most of the time we consider only the gate to source capacitance, but the gate to drain capacitance, even if it’s smaller, also exists and much also be charged and discharged.
Two slides back we saw that the rising edge, or turn on edge for the control FET generated a lot more ringing than the turn off. That means that, for bucks, slowing that switch node rising edge is a lot more important than slowing the falling edge. In fact, most of the time you don’t want to slow the falling edge at all, because that increases the switching loss, but usually does little to improve the falling edge ringing. Selecting a gate resistor is all about finding a good compromise, so slowing the rising edge also increases switching loss, and you can bring up your FET in a hurry if you’re not careful.
Gate Resistor Selection
- Set up low inductance switch node measurement
- Set up a current meter to measure input current
- Set a 10-turn, 20 or 50Ω potentiometer to zero ohms
- Solder the pot in series with the gate of the MOSFET
- Slowly dial the pot, increasing resistance
- Watch both the ringing on the scope and the input current
- When the ringing stops getting smaller, but input current continues to increase, stop
- Back up, finding the best compromise
- Check resistance of pot
- Replace with the closest 1% SMT resistor
I like to say that my method for selecting gate resistors is highly scientific. I put a multi-turn trimmer potentiometer of 20 to 50 ohms, one of those blue ones with the brass knob, like I show here in the corner, and I solder pin three to pin two. That supposedly helps with noise, but it also ensures that the resistance increases when I turn the knob clockwise and vice versa. If you prefer the opposite, just solder pin one to pin two. In either case, clip on a multi-meter and make sure that the pot is set to zero. That’s usually about zero point one ohms in practice. Some pots click when you turn them past zero, or past full resistance, but others don’t. Plus, it’s a soft click, and labs are noisy when you play with power. Too many cooling fans in action.
Another practical thing I like to do is to solder a 10 megaohm resistor in the gate resistor footprint, and then I solder the pins of the pot on top. This is purely a mechanical thing. I’m always accidentally ripping up the pads and the tracks of the PCB. If you’re testing a circuit that didn’t have a footprint for a gate resistor, then you probably cut the PCB trace and scrapped off some of the solder resist. In that case the 10 meg resistor is even more important.
Severing the gate drive line doesn’t always destroy the rest of the circuit, but it certainly can. When watching the oscilloscope, it’s very important to use the low inductance spring tip, which guarantees that most of the wave forming on the screen is true conducted signal, not radiation picked up by the ground pigtail. Details about this are shown in parts two and three of this seminar, as well as later on in slides 24, and 25 of this section.
Properly Tuned Gate Resistor Works Wonders
At zero ohms there should be no change to the ringing. Then, as you increase the resistance, the amplitude of the initial spike, along with the number of oscillations and the ring they form before it decays will come down, dial slowly. That’s the main reason they suggest those 10 term potentiometers, because when you go too far, and you always will, several things happen at once. The rising and falling edges rapidly lean in, becoming less vertical, and at the same time the input current will increase quickly. Both are signs of too much slowing. It’s time to back off.
Unfortunately, a lot of the times this increased switching loss burns up the MOSFET. I always keep at least 20 spares of any power FET that I’m using. You won’t always get results this dramatic. This was a case of a driver that really was overpowered for the little P-FET. Still, one very cheap 18 ohm, thick film resistor reduced the reading to more or less the same amplitude as the switching frequency ripple. 18 ohms is a lot. Expect most gate resistors to be five ohms or less for a reasonable match between the strength of the gate driver and the size of the power MOSFET.
One nice thing about this process is that when you replace the pot with an actual resistor, the results are always slightly better. This is because the resistor, especially small 0805 or 0603 devices, have a lot less inductance than the potentiometer itself.
R-C Snubber, Method 1
There are plenty of cases where just to get resistor isn’t enough, and that’s when the next weapon in your arsenal against ringing and radiated noise comes into play, the Snubber filter. The basic idea of a Snubber is to damp the unwanted LC tank, absorb the high frequency ringing noise, and convert it into heat. This must be done without absorbing much energy of the switching frequency. That would burn up the Snubber resister, as well as ruin the power efficiency of the converter.
Many different methods exist for picking the optimum values, and I’m going to present two that I like, and that have worked well for me. The first way is based upon setting a budget of power loss that you’re willing to sacrifice.
R-C Snubber Across Low-Side Switch
- Tuned filter dissipates high frequency ringing while letting lower frequency switching pass through
To reduce transience and ringing on the rising edge of a buck convertor, the Snubber filter is placed in parallel with the low side switch, be it a diode or a MOSFET. If you watched part three, you heard me insist that this filter must be placed in the smallest area, lowest inductance loop possible, or else it won’t be effective. The filter will operate when the control FET turns on, and the diode enters reverse blocking mode.
Snub the MOSFET for Falling Edge Ringing
We saw earlier that the ringing on the falling edge in most buck converters is small compared to the rising edge, but if it causes radiated EMI problems, the same type of series R-C filter, with the same tight loop, is placed in parallel to the control switch. Slowing the falling edge must be done with great care in synchronous bucks especially, because of shoot through, that most fatal of conditions where both the top and bottom switches are on. Even a few nanoseconds is enough for lots of amp of current to flow and kill a MOSFET. I’ve seen plenty die this way. An overly slowed, falling edge can take so long that the top FET turns on, and then well, poof.
Ex. Sync Buck with Paralleled MOSFETs
Here’s an example from a fairly high-power buck that I did. This one has a peak output power of 600 watts, so you can be sure that there will be some noisy, radiated emissions here. Notice in the layout how the stopper components have been placed as close as possible to the switch that they are filtering. This is a good time to talk about another way to reduce the rising edge ringing and bucks, placing a Schottky diode in parallel with the synchronous MOSFET. If you can get it close enough, it will turn on before, or only slightly after the body diode of the FET.
Recall that the body diode has lots of reverse recovery charge, whereas Schottkys have lower forward voltage drops, and little more than junction capacitance. My advice is to have footprints for the parallel Schottky and the Snubber. The Schottky diode can be physically small. It only carries the current during a few hundred nanoseconds, so it doesn’t burn much power. The newer, low inductance packages are great for this job, like the ESMP or power DI types.
Ex. Dual Sync Boost
For our boost converter the output switch usually gets the Snubber treatment. This is also true with SEPIC, buck boost and flyback regulators. This is a synchronous design, so the output elements are MOSFETs. Once again, the layout shows how each R-C branch goes in a tight loop to minimize inductance. I didn’t use them in this design, but parallel Schottky diodes, placed in nice and close to the output MOSFETs, without the same benefits as with the buck, less energy to create ringing in the first place if they can turn on before the body diodes of those output FETs.
R-C Snubber Selection, Part 1
A low inductance voltage measurement is absolutely essential for this kind of testing. You must be sure that the signal you’re working with is as true as possible to the actual conducted signal. You can find plenty of detail on making low inductance voltage measurements in section three dash two, slide 17 through 19, in the subsection titled, “Tips and Tricks.”
Power and energy are easy to calculate when you know how much capacitance, how much voltage, and frequency. Frequency is just time inverted after all. The main advantage of this Snubber design method is that you know the power dissipation ahead of time, so you can pick the resistor size. Standard 1206, for example, will usually dissipate a maximum of 250 milli-watts. It’s generally cheaper to parallel multiple, standard thick-film 1206 resistors than it is to use larger ones, or higher powered 1206 devices.
If your calculations come out to 10 pF, or 100 nF, something went wrong, so check again. You should be in the range of 100 pF to 3 nF or so.
For the resistance, I start with a fairly scientific method, measuring the rise time of the problematic edge. Here, a rise time of 10 nanoseconds is blazing fast. I expect designs with a reading under control to rise in the range of 20 to 50 nanoseconds. Now, I like a 10% to 90% metric to determine rise time. Another note, it might be easier to stop the scope or use a single trigger, as opposed to a free running trigger. You want as clean a rising edge as possible.
This time I start with a 10 turn pot, set to the resistance calculated in step seven, repeating my techniques with the 10 mega-ohm placeholder, and the PCB trace saver. Also, as before, I watch both the oscilloscope and a multimeter that shows me the input current. A lot of bench top lab supplies now have a reasonable current meter built in, so you might not need that multimeter. Next comes the not so scientific phase, which is turning the pot up and down, slowly, to see if you can get a better response. By better, I mean the lowest initial overshoot, and the shortest period of ringing. The target is critical damping, one ring with an exponential decay.
If you dial the pot’s resistance too high, you effectively disconnect the Snubber capacitor, so nothing happens. If you dial the pot’s resistance too low, you basically put the Snubber capacitor in parallel with the parasitic capacitance of the switch. This is easy to see because all of a sudden the ringing frequency changes. You should also see a rapid increase in input current, because you are charging and discharging the Snubber capacitor with each cycle. This is if you multiply the capacitance of the low side switch by several times.
Response with Tuned R-C Snubber
Here, the gate resistor did most of the work, and the Snubber basically turned the remaining ringing from an un-damp to a damp oscillation with just one ring. The high frequency ripple is now about the same peak to peak value as the switching frequency ripple. This is a good target to try and get to, especially if your load is a digital processor, since their input power specs usually give a maximum ripple regardless of frequency.
A few high quality catalog power supplies will specify both the switching frequency and the high frequency ripple, but many give a combined spec that I suspect is vague on purpose.
R-C Snubber Method 2
Lately this second method has become my favorite. Rather than setting a power budget, you focus on achieving that critical damping. Mathematically, it’s the same principle as the big, glossy, aluminum electrolytic input capacitors I recommended in part two of this seminar for damping input filter oscillation, and the formulas are the same. This method is good because the power budget method might not give you enough capacitance for true, critical damping. The downside is that you don’t know how much power we needed for critical damping, so you have to guess about the Snubber resistor size.
Here’s a picture of a voltage probe set up for low inductance testing. Find that accessory pouch and get the little spring that fits around the ground connection at the end of the barrel. I take 2.54mm pitch breakaway hitters, as you can see in this photo, and then make a test picture with them by snapping three or four section pieces, and cutting off the central pins. Taking advantage of extra parallel MOSFET footprints on the backside of this PCB, let me solder this text fixture in very solidly, and that’s really important. I’ve destroyed quite a few PCBs by accidentally knocking into the probe and ripping up the tracks, the components, et cetera.
More often than not, a sudden open circuit because of a torn up PCB trace makes something blow up. If you lost your spring tip ground accessory, strip the insulation off 10 cm of a 0,5 mm diameter solid wire, and wrap all but 2 cm around the probe, making sure it contacts ground. Then, bend the last 2 cm in 1 cm long folds at 90°, and you’re ready to go.
This is not the same circuit as before, and it doesn’t ring nearly as high nor as long. The power MOSFET is better matched to the driver, which explains part of the difference. This MOSFET also has a different output capacitance, the output inductor is different, so it has a different leakage inductance. All of these are reasons why edge control circuits can’t be set up ahead of time. Simulations can’t help much either because it’s just about impossible to accurately estimate all these parasitic values. You just have to set aside some time in the lab to adjust and set Snubber filters.
Here comes the trial and error part of this method. You need to start adding in capacitance directly in parallel with the low side MOSFET. I always start with 1 nF, since that’s right in the middle of the typical range. This will definitely increase the switching loss, so you might want to add in a fan, or blow some air across your PCB for this phase. Without digging too deeply into the math, let me say that the frequency of an L-C tank drops in half when either the capacitance or the inductance increases by four times. Again, if you watch the discussion of L-C filter damping in section two, or if you’re familiar with Dr. Middlebrook’s work on this subject, then you’ll recall that four X the capacitance is needed for critical damping. If 1 nF is too much or too little, I suggest going to 470 pF, or 2,2 nF respectively, rather than the minimum steps to 1,2 nF, or 820 pF.
Our friend, the 10-turn potentiometer is back in this photo. You can probably guess that my lab is full of these guys. If you look closely, you can see that I accidentally melted the corner of this one with my soldering iron. No problem, it still works. This time, take a 50 or 100 ohm part and dial it to the maximum resistance. Stand in the capacitor you had soldered in across the low side switch on one end. This is a good time to mention that using through hole capacitors makes the increase in T-ring larger than it should be due to their parasitic inductance.
Therefore, use only surface mount MLCCs, and in the tightest loop possible. That detail caused me a lot of frustration while trying to damp circuits with too little damping capacitance. An alternative to starting with the maximum resistance would be to calculate the resistance needed for critical damping, but I’ve found that it’s simply faster to start high and dial down.
Another important note, these trimmer type potentiometers are usually [whetted 00:19:23] to one half watt, so they can usually handle the power in these tests for a few minutes, but I’ve burned my fingers in a few high power designs, so take care.
In this example I wasn’t able to achieve as great an improvement as I did with the previous one, but I got the transient down to about half of its initial value, and eliminated one ring cycle. As before, the good news is that when I replace the potentiometer with an actual thick film resistor, the result will be better because the resistor has far less inductance.
Here’s the final result with a 1.8 ohm resistor. The result of the power calculation and step 7 here is 270 mW, so two standard 1206 size resistors in parallel, or one 0.5W 1210 size device would be good options.
Next up: Section 4-2 Switching Edge Control for EMC
Section 4-1 examines edge control and regulars with internal power switches but external inductors, presenting the bootstrap resistor method and application of R-C snubber filters to both high side and low side switches. A non-synchronous buck is used as an example, and a brief explanation of how a synchronous buck regulated differs is also presented.
Part 4-2 of our Power Supply Design series will be available as of April 16, 2018.