Welcome. My name is Ambreesh, and I am a system and applications engineer at Texas Instruments. Today, we are going to talk about various techniques that can be used to design low-EMI power converters for industrial and automotive systems. Here is the agenda of presentation.

We will talk about automotive and industrial-related wide input DC/DC converters, followed by techniques to mitigate EMI in SMPS design. During the course of the discussion, we will talk about optimizing PCB layout, as well as spread spectrum and fluid control technique. This will follow with the input filter design calculation. And to conclude, we will be talking about implementation of input filter calculator, which are available online now on Texas Instruments WEBENCH Tool.

Here is a range of wide Vin DC to DC regulated solutions optimized for automotive and industrial needs. In the integrated synchronous switches segment, TI offers regulators which are capable of operating at 40 volt input and with up to 2.1 megahertz switching frequency. As we all know, switches switching above EM band is quite attractive for automotive needs as it avoids EM band altogether and also results in integration.

Some of the switches have present current in the tune of 10 to 20 microamps, which are critical for always on automotive, as well as industrial [? crop ?] systems. Here, I would like to introduce certain class 40 volts 2.1 megahertz switches [INAUDIBLE] LM536xx, which can deliver [? 30 ?] milliaps to 3 and 1/2 amps of current. LM536xx3's also offer low-EMI solution through spread spectrum and package, as well as [? turn ?] out optimization. Also, we do have some in band switching regulators, LMR23610 in 1 amp, 2 and 1/2 amp, and 3 amp bracket.

Also, TI offers 100 volt synchronous switches, especially designed to support wide Vin industrial and telecom segments. These 100 volt switches are [INAUDIBLE] part controller converters, part [? Vin ?] start and on time, designed to support wide input output ratio, and also, offer excellent performance when considered in fly-back mode. LM5160 is 100 volt. And LM5161 is 65 volt synchronous buck converter, which has been designed widely in industrial system in both isolated and non-isolated designs.

We are currently working on ultra [? low-IQ ?] 65 volt synchronous converters in 150 milliamp and 500 milliamp states. In the power module segment, we have integrated inductor modules with input voltage range to 60 volt and proven EMI performance. [? LMZ34202 ?] is 42 volt input. And LMZ36002 is 60 volt input capable of delivering 2 amp of output current.

In the boost and buck-boost solution designed to support [? higher ?] power system, TI offers both synchronous, as well as non-synchronous option with wide input voltage range of 75 volt. The boost controller, like LM5122, supports multi-phase and not stackable. LM5175 is TI's first synchronous full switch wide range buck-boost controller, which offers smooth transition from buck to boost mode and vice versa. We are also releasing LM5170, which is coming soon, and has deferred 48 to 12 volt bidirectional controller.

On the buck controller segment, we have 100 volt input capable device, LM5116, which supports emulated peak current mode control. Further, we have released out first 2.1 megahertz dual buck controller, LM5140. And it is designed to support wide input volt range and load [INAUDIBLE] current operation.

You may have be familiar with TI's smart diode solution LM74610-QN. This is designed to replace a [? voltage ?] reverse protection diode with NFET. It requires no ground connection and results in efficient solutions for systems, which need reverse terminal protection.

Now, let's move on to the EMI. So what is EMI? What is the Electromagnetic Interference? Basically, EMI is an unwanted coupling between two electrical systems and is a major adverse effect caused by the application of switch mode power supply.

EMI in some spheres is classified into two forms, the conducted EMI and radiated EMI. They are differentiated by the manner in which electromagnetic field propagates between circuits. For conducted EMI, as you can see in the picture here, noises coupled by conductor are through parasitic impedance power and ground connections while the [? variated ?] EMI and [? mounted ?] noise is coupled by a radio transmission. You can see here, radiated EMI, which is coupled via radio transmission.

In switching power supplies, EMI noise is unavoidable due the switching action of the [? silicon ?] connected devices. And resulting discontinuous current, that is what you can see here in the picture. Voltage ripple generated by the discontinuous current can be conducted to other systems via physical contact of the conductor without control. Excessive input or output voltage ripple can compromise operation of the source, load, or also, the adjacent system, which is connected to the supply.

The goal for EMI control in power supply design is basically, to minimize the noise operation, basically, to minimize the generator of unwanted emission through conduction of radiation. And the second is to ensure that the system performs properly with the presence of EMI generated internally or externally. So the goal is to reduce the noise generation and to increase the noise compatibility.

So as a power engineer, we need to know how to mitigate the EMI noise in the system. Noises start from the source, and they are coupled to the other circuits via various means. So we shall start with identifying the source of the noise and then figure out where this part on the PCB that is critical for noise generation and transmission. Then use the proper [? IC ?] selection and PCB layout to minimize noise generation and transmission. If they are not good enough, we can also add input or output EMI filters, or add number of circuits or additional shielding to further protect the susceptible systems. As you can see towards your left, we can add EMI filter and the shielding to lower the EMI performance.

These are the high level steps needed to be taken for better EMI performance. The initial step, as per the previous slide, [INAUDIBLE] identifying the nodes that has high di by dt signals, and then the steps in the PCB guideline for reducing the loop area, and therefore, parasitic loop inductance. This relates to [? VL ?] di by dt. So if you have higher parasitic inductance, you will see higher overshoot and hence, the higher noise. Lastly, how we go about protecting these sensitive nodes again through PCB layout, we'll be going in details of that.

Now, again, to minimize the critical path loop area, place switching components, which are these MOSFETs here and high frequency bypass capacitor as close as possible to each other. As you can see, bypassing of the input output containing the pulsating current is very, very important. Pay most attention to the return path from the [? load ?] sides at ground to the ground north of the bypass capacitor. So the return path of the load sides at ground and the return path of this high-frequency bypassing capacitor should be tied as closely as possible.

The in cap, basically, the input cap must be taken to the P ground as closely as possible. It needs to be on the same layer. P ground is the noisy ground. And the A ground is the quiet ground. So the bias cap, as an example, would be places from bias to the A ground. So basically, separate the noisy ground path from the quiet ground path.

Now, here, just to show the effect of high di by dt path and bypass capacitor placement, we did an experiment. We have one buck converter [INAUDIBLE]. And we used different input capacitor placement for the comparison. In the first case, the capacitor is moved to be little bit further away from the [? IC, ?] which contain the switching component inside. So we measure the switch node at the peak current, the [? VL ?] peak-to-peak ripple, and the conducted EMI data. And the EMI screen shot shown is for the conducted EMI in the range from 30 megahertz to 108 megahertz.

Now, in the second case, the capacitor is moved closer to the switching element, basically, closer to the [? IC, ?] effectively reducing the loop area by about 2 and 1/2 times compared to the first case. And the measurement we did, we see that the maximum switch node voltage is reduced by 2 volts. The output voltage ripple is reduced by half, as well as the peak, the EMI peak is also reduced by 5 dB microvolt. You can see the reduction here compared to the earlier slide. The peaks were up higher by 5 dB microvolts at this particular point. And it's exactly the same volt. And you can see the reduction here at this particular point by 5 dB microvolt.

So TI also offers switches with pinout optimized for EMI reduction. We have taken LM43603 as an example. Now, what you can see is Vin and the P ground pins are next to each other. Basically, this allows more bypass capacitor as close as possible to the place and on the same layer with no need of Vout. Similarly, Cboot right next to the switch pins, because of the pinouts can result in a smaller [? high ?] [? side ?] driver current path.

Now, let's move onto the feedback pin. What you can see here is the feedback pin is away from the switch node. So node switching noise can couple to the [? switch back ?] node, as well as it is placed next to the A ground, which result in shielded and clean feedback divider network. Also, the ground plate can continue to be underneath the [? IC ?] for best terminal performance.

In LM53635 automotive switcher, we have introduced two parallel parts for input, which further reduce the parasitic inductance and results in superior EMI performance. [? Same ?] can be seen for output to for the ground. Further, LM5335 is designed in [? F ?] [? call ?] package, that is, [? switches ?] on lead screen, which further optimizes the EMI performance by avoiding switch node [? stringing. ?]

Here is the example of overall good PCB layout based on LM5165 synchronous bulk converter. Now, let's see the recommendation. Keep the input capacitor close to LM5165, We have already dealt with this particular rule in detail to minimize high di by dt loop.

Let's move on to the second recommendation here. The second recommendation here is to remove copper between inductive path to reduce the parasitic capacitance, and hence, avoid ringing. A full layer ground plane under converted topside layout provides magnetic field cancellation. Also, see how the feedback components are place away from the switch node here. So the switch node needs to be placed close to the feedback and the ground pin.

Now, we have already talked about critical path loop reduction. Now, I want to also cover how to design the ground plane and shielding, and how that is going to help the EMI problem. So an unbroken ground plane could provide the path, the least impedance path for a counter trace on the top or on the bottom to return to the current source. And if the current it right on top of each other, like as shown in the case in the figure towards the left, EMI generated by this rule is as small as possible, and the magnetic field can be canceled as well.

The ground plane is cut in the case shown in the right side. You can see a cut here on the ground plane. Then the return current has to find a longer path, a much longer path back to the current source. This, the area enclosed by the current is much bigger, and the EMI problem is going to be much worse.

Now, move on to the complete EMI picture. So there are basically three components that are needed to paint EMI picture. The first one is the input noise. The second one is the output noise. And the third one is the switch node noise, which is very, very critical. Now, let's go into the details of the input noise.

For automotive systems, CISPR 25 class 5 needs, we can calculate back the maximum allowed noise level on the input, which is closely around 40 degree microvolt, which corresponds to 140 microvolt peak-to-peak input ripple. This means at the input of the buck converter, a maximum of 9 millivolts input ripple noise limit is allowed.

And what is the reason why I'm saying this? Basically, because if you divide this 9 millivolts by 140 microvolts, you would get around like 40 dB of attenuation, which can be attained through a low path, through an [? else/if ?] filter, which is shown here. If we have 9 millivolt input ripple noise on the [? IC, ?] then we can filter down to 140 microvolt after the filter.

Now, let's move on to the output noise. That can be a significant contributor in the case if we are dealing with a very long cable, for example, USB charging cables. In this case, cable may be acting as an antenna. In most cases, this could be improved by, for example, by using [INAUDIBLE] on output to reduce output ripple to maximum allowed 140 microvolt peak-to-peak.

Now, let's say you have already 140 microvolt peak-to-peak at the input and the output cable, but you can still fail the EMI limits. And why is it so? It's basically because of switch node noise The largest ripple noise is the switch node noise. It is rectangular, very large, 12 volts with a lot higher dB by dt. Keep this perspective in mind.

You can have like a 1.2 millivolts peak-to-peak output ripple noise. But in the switch node, you are dealing with 12 volts rectangular signal, which is 10,000 times higher. Plus to make things worse, at such node signal is nice rectangular shape with nearly infinite harmonics, compared to ripple noise that is bit more triangular and directly filtered by ceramic input and output capacitors. Such a rectangular noise will cause a lot of harmonics from the [INAUDIBLE] dB by dt of the rectangle shape.

So the 2 megahertz is not what we have to worry about alone. It is higher megahertz or up to gigahertz spectrum where the potential to pick up these harmonics. What you can see is the switch node is being coupled [? back ?] from the input signal, and also, to the PCB to near [? electric ?] [? field ?] coupling and resulting into much higher than expected EMI peaks.

Now, how do you build with the switch node noise? So basically, the first and foremost thing is to shield the field and short it immediately to ground on PCB. Also, we can use the common-mode filter at the input, which will make the measurement better. But the electric field is still there and can disturb the nearby electronic system.

Now, we were talking about the switch node noise. Now, let's talk about switch node ringing, which is a major source of EMI, which are caused by these parasitic inductance and the switch node capacitance. And the parasitic inductance can be of PCB trace, as well as of MOSFET packages, both inductance and as well as parasitic capacitance.

At such higher frequency, that deferential [? circuit ?] is not attenuating. And why? Because in anything above 5 megahertz to 10 megahertz, the effect of LC filter, the differential filter, would be negligible. It will be only attenuating by 30 dB per microvolt, which is not enough. And hence, different techniques are needed to avoid the emissions at higher frequency.

So we were talking about parasitic inductance and capacitance of PCB trace, as well as of MOSFET package. Here, the MOSFET package parasitic inductance and capacitance are shown. Ringing waveform results from the parasitic inductance and the switch node capacitance, primarily consisting of output capacitance [? COS ?] of the [? load ?] side MOSFET.

Now, how to avoid that ringing? Trigger here shows high current driver output with independent [? source ?] and independent current sync, then for its slew rate control. The slew rate control enables the user to adjust the switch node rise and fall time. So you can adjust independently device, as well as the fall time, which can reduce the EMI in the FM radio band, but clearly higher frequency band from 30 megahertz to 108 megahertz.

The slew rate controls simplifies the compliance with [INAUDIBLE] and automotive [? EMI ?] requirements. Now, see the difference in the switch node waveform, which have no ringing due to slew rate control. And in this particular design, there was no snubber used. So you can see the difference here with the slew rate control and the waveform without the slew rate control. And this high frequency actually coupled back to their input and the PCB and result into a much higher EMI noise.

TI offers LM5140 2.1 megahertz DC/DC controller with slew rate control to adjust both the switch rise time and fall time independently. The comparison is shown on the bottom here with the design having slew rate control and not having the slew rate control. So this is the picture with no slew rate control. And this is a picture without the slew rate control. You can see here due to switch note ringing, the design not having the slew rate control has very high peaks while the other one, one which has a slew rate control, has a lower peaks, physically, a peak which is lowered by 20 degree microvolt because of the slew rate control.

Now, we have already talked about one of the major techniques to reduce noise at the higher frequency, which is slew rate control. Now, let's talk about the spread spectrum. Spread spectrum is a means of reducing EMI interference by dithering the switching frequency. This has the effect of widebanding the noise spectrum and reducing the fundamental energy, as shown in the figure.

Now, what you can see is the energy was [INAUDIBLE] the fundamental frequency. You have wideband that frequency, so the peak is reduced. Again, load frequency conducted emission from the first few harmonics of the switching frequency can easily be filtered. The reason is your [INAUDIBLE] filter is can easily attenuate those signal.

A more difficult design criteria is reduction of emission at higher harmonics, which falls in the FM band. Now, these harmonics often couple to the environment through electric field around the switch node, as discussed earlier. LM536xx series, the series which I've already talked about, which is automotive synchronous DC/DC converter use a plus minus 3% spread of frequency. Plus minus 3% frequency is spread here, which spread [INAUDIBLE] smoothly across the FM band. But it is small enough to limit the subharmonic oscillation emissions below its [? specific ?] frequency. Peak emission at the path switching frequency are only reduced by slightly less then 1 dB while the peaks in the FM band are typically reduced by more than 6 dBs, which is really difficult to achieve by any other techniques.

Now, let's see. What we have done is we have combined the slew rate control and the spread spectrum. We have earlier talked about LM536xx converters with the integrated FET. And now, I would like to introduce our first 2.1 megahertz bulk controller, which can be designed to support higher power systems.

The device has the wide input voltage range from 3.8 volt to 65 volt. Comes in 6 output voltage version of 3.3 volt and 5 volts, and also, comes in adjustable form from 1.5 volt to 15 volt output. It has a fixed 2.2 megahertz and also, 440 kilohertz oscillator. Frequency can be shifted form the fundamental via our resistor setting.

There's a lot of EMI feature in this particular device. We have spread spectrum. We've already talked about spread spectrum. We have the slew rate control. It can be synchronized to external clock. So what is advantage? Basically, you won't see any [? beat ?] frequency. And also, you can have the full PWN at the lightened node.

We have already seen that the switch node power supply has two wires to connect to the supply and to the load. So any current that goes in has to go through the other wire. So the sum of two current is always zero. There is no common-mode conducted EMI involved in non-isolated SMPS design. So we only consider differential-mode conducted EMI here.

For the deferential-mode conducted EMI, in such mode power supply, it involves the normal operation of the circuit because you are dealing with the discontinuous current. It's more of the current phenomenon. And you can sense it from the fact that when the input voltage is reduced, then the input current is going to be much higher and the differential-mode conducted EMI problem is going to be worse. So why do we care about the conducted EMI? The conducted EMI can go back to the power supply or to the load and could compromise their operation.

There are two basic requirements for the conducted EMI filter. We need to meet the noise attenuation requirement. It can be CISPR 25 for automotive system and CISPR 22 for industrial system. And we need to not interfere with the normal operation of the SMPS converter. And this is very important. We need to adhere to is the fact the input filter will interact with our control loop of the buck converter. And we are going to make sure that the output impudence of input filter is much lower than the input resistance of the buck converter.

Now, here the example is shown is a buck regulated conducted EMI performance without any input filter. And you can see that it fails CISPR 25. That's by regulation limit. Now, the question is, how do we estimate how much filter attenuation is needed for this particular example?

Now, there are two methods to calculate the required input filter attenuation. The method one is the estimation using oscilloscope measurement. Basically, in the [? time ?] domain amplitude of the input ripple at the fundamental switching frequency can lend a good estimation of the attenuation. Measure the input peak-to-peak ripple with a wide bandwidth oscilloscope and subtract it from the Vmax, which is the EMI standard that we are trying to meet.

So this is the Vin ripple that is measured from the oscilloscope. And you compare it with 1 microvolt to get the dB microvolt. And you subtract it with the dB microvolt. And you can get this value from the standard. Like CISPR 22 will have a different Vmax than the CISPR 25 standard.

Then the method two is the estimation using the first harmonics of input current. The current at the input can be modeled as a square wave, assuming small single approximation. Simply [? extend ?] the first harmonic of the current from the [INAUDIBLE] series of the input current waveform. And when multiply that by the input capacitance impedance, Cin, you can see how it has been multiplied. So this is your first harmonic, first current harmonic, and you multiply it by this input capacitance.

Now, you can calculate the required attenuation the similar way you did in the method first. You compare it with 1 microvolt to get a dB microvolt value. And you subtracted this value by the allowed maximum ripple voltage in dB microvolt for the particular CISPR limits.

Now, filter design is basically from the right hand side to the left hand side when looking at the power source and the device under testing. Basically, it is the low path LC filter. Now, let's go to the step to choose this LC filter. identify the noise level at the fundamental frequency using method one and two, which we discussed earlier, then calculate the required attenuation. Now, select the value of the inductance of the LC filter from anything from 1 microhenry to 10 microhenry. Sometime, also for the higher current design or high power design, you would see a lower [? inductance. ?] You may select a lower inductance value.

Now, the important thing is to calculate the capacitance value here. Now, you can see that they're like two capacitance values, Cfa and Cfb. Now, Cfb is the very important. It basically ensures proper attenuation of EMI filter. You can see the attenuation factor here. You divide it by 40 dB, which is basically the maximum attenuation you can achieve from the LC filter.

Also, we need to ensure the resonance frequently of EMI filter is at least 1 [INAUDIBLE] below the switching frequency. And this has been ensured by the Cfa figure. Now, we need to choose the higher value that we get from Cfa and Cfb. Once we have chose the CF, that is, the value which is higher than the Cfa and Cfb, we then calculate the Cd, which is the damping capacitor and is usually [? electrolytic ?].

And why we need it? The output impedance of the filter at the input stage of the device must be suitably small enough so that the loop [INAUDIBLE] of the path converter is not affected. The peak of the impedance at the resonance of the filter is usually dictated by the LC parasitic. When the output impedance is too high, damping is required. Using an electrolytic with a high ESR reduces the peak impedance of the filter resonance. You can see how the series can be calculated with these two equations here.

Now, here, the conducted EMI performance of LM53603 is compared with and without the input filter. The limits are based on automotive mode stringent CISPR 25 class 5 regulations. So the top one here is the peak limit. And the bottom one here is the average limit.

The design is switching at 2.1 megahertz. And how we can say that? You can see the harmonics at 2 megahertz, 4 megahertz, 6 megahertz. And as expected, you can see the peak at these harmonics. And also, you could see that some frequency components, some other frequency component, basically because of high di by dt of the switching action.

On the right hand side, you see conducted EMI performance where peaks are pretty much all filtered out. And the peaks at the switching frequency is also greatly reduced. So with the filter, this converter actually passes the EMI limits.

Now, here is, again, example of a DC/DC converter EMI-optimized design with input filter at front end. The design is based on LM5156. This is a pretty useful [INAUDIBLE] for industrial applications. And what you can see is the LC filter at the front end and the [INAUDIBLE] the transient voltage suppressor to suppress any transient in the input supply. Now, this particular design can take up to 65 volt input and can deliver 150 milliamp output at 5 volt.

LM5165 is widely used in industrial subsystems. And here, the collected EMI performance is shown on industrial CISPR 22 Class B limits. [? Again, ?] with insertion of the input filter, the peaks are greatly reduced, and the design path delimits easily for all the frequency range. Now, what you can see is the device was switching at 230 kilohertz. And without the input filter, you can see all these peaks, which are above the limits. Once you have this LC filter of 22 microhenry and 10 microfarad, you can see that the peaks are greatly reduced and it passes the EMI limits easily.

Now, to make the thing simpler, we have also released a new EMI filter designer in TI's WEBENCH online tool. It is the input filter design calculator for switching regulators to satisfy CISPR conducted EMI [INAUDIBLE] specs. So [INAUDIBLE] circuit attenuates conducted EMI noise generated by [INAUDIBLE], who meet target CISPR EMI standard. It also ensures proper damping to prevent undesired oscillation at the input of the power supply. So what you can see here is the green one shows the EMI performance without the input filter. And the blue one here shows the [? amount ?] performance with the input filter.

Now, you can either auto-calculate this input LC filter, or you can have your custom input filter and see by how much the peak is getting attenuated. You can also choose the EMI limits. But the most important thing is you can do these impedance analysis to ensure stable operation. And this is the uniqueness of this particular online calculator design. So effectively, you would ensure that the LC filter is not affecting the loop performance of your DC to DC converter. And with this, we have reached to the end of my presentation. Thank you for watching.