Monolithic Power Systems, Inc. (MPWR) Earnings Call Transcript & Summary

Hi, everyone. Thanks for joining us today. We still see that people are kind of pouring in, so I'm going to give it about 30 more seconds, and then we will get rolling. Okay. Again, thanks for joining us today. I'm Kelly Curd, Senior Marketing Manager here at MPS. And I'll be joined today by Tomas Hudson. He is one of our application engineers from our MPS Barcelona office. Tomas is going to walk you through a variety of topics related to COT, so he's going to cover Constant-On-Time Control versus current mode control. He'll also address multiphase COT to enable fast transients for large current steps and take us through MPS' advanced power blocks for powering FPGA core rails. Before I hand it off to him and get us into the meat of the presentation, I did want to do a little bit of housekeeping. We are recording this session. We often get asked this as I think people are diligently taking notes, but we record this session. We will have it available on demand within the next couple of days. We do have to process it, and then the presentation itself will be included as well. So we will send you out an e-mail with all that material here once it's ready to go. You can also always go to monolithicpower.com/webinar, where you can see all of our past webinars. [Operator Instructions] So without any further ado, I will hand it off to Tomas.

Thank you very much, Kelly. Thanks for the presentation. My name is Tomas Hudson. I'm an applications engineer here at MPS. And today, I want to talk to you about Constant-On-Time Control for powering FPGAs and other high-current loads that require high-current, fast-transient power supplies. So just so we have a bit of an overview of what we're going to be seeing. We're going to start off discussing a more commonly known control method, which is Peak Current Mode Control. And then we're going to look at why Constant-On-Time has been adopted for these sorts of power supplies for these sorts of applications, such as FPGAs and ASICs. And going over the advantages that Constant-On-Time presents and also the challenges that we need to get over. Then I'm going to dive a bit deeper into the way the MPS implements our Constant-On-Time in these -- in our power modules designed for supplying FPGAs and a bit into what we have to offer and how we can help ease the design process. So without further ado, we can start with, as I said, a more commonly known way of controlling power supplies, which is current mode control. In this case, we're going to be talking about Peak Current Mode Control. So as I'm sure many of you are familiar with this control method, where we have 2 control loops. The first one is the voltage loop, which is what we call also the slow loop. This senses the output voltage. We use a resistor divider, and it goes through this linear compensator that outputs this comp signal that you can see here in blue. The second loop that we use in this form of control is the current loop or the fast loop. Here, we're sensing the inductor current, and we use these 2 signals to generate the PWM signal that we then send to the drivers that control the switches in our back converter. So inside the PWM generator, we have a comparator, and we have a clock signal. The clock signal indicates when the drivers should close this high-side switch, in this case to kept it simple, so we just have control over this high-side switch. So the clock is going to indicate when the drive signal goes high. And then by using the comparator and this PWM latch, once the inductor current reaches the same value as this compensator signal that we've gotten from the voltage loop, the latch is reset, and it's -- the signal for the PWM signal to go low. This is how we then generate this PWM -- that's going to be able to modify the duty cycle when we have changes in the required output voltage or the required load current that we need to supply to the output. This method, as I said, is very commonly used.It's very often part of all the syllabuses in college and university, and it's something that's been widely used in the industry as a way of controlling DC-DC power supplies. However, as the load that we try to power evolve, so do the control methods. And in this case, with the advent of -- or the increase in power requirements from FPGAs and ASICs, we need to find faster control loops or faster control methods. This is because, of course, FPGAs have a large dynamic current variations, depending on the circuit that we implement in the FPGA, we can have very sudden current shifts. And the additional requirement here is we need to deliver the energy to our load as quickly as possible without seeing a drop in the output voltage. So if we reduce this to the basic requirement that we need a faster transient response, then we need to analyze where we can increase our response speed. So first of all, we see 2 basic things we can resolve: One is we can improve our response by not having to wait between one clock signal and the other, clock period to deliver the energy, this is going to limit how much energy we can deliver to the load instantaneously; secondly, we have a delay in our control. And the main culprit that we can look at is what we already call the slow loop, the voltage loop. This delay is mainly due to the linear compensation that we have in this loop. So if we want to increase speed, then the first thing that we can do is we can just scrap our compensator and directly connect the output voltage sensing to our comparator. In this way, we don't -- we can eliminate this delay. And then the second aspect, we -- I'll go into that later on. But of course, without the compensator, we don't have this compensation signal that we were seeing here. So we substitute this for a fixed voltage reference. The second aspect that we can shorten the response time with is by eliminating the clock signal. So instead of using the clock signal into the latch, we use a one-shot timer. The one-shot timer delivers a fixed pulse width, so fixed on-time, thus the name Constant-On-Time Control. And what we're changing here is not so much the width of our pulse, but we're shortening or lengthening the total switching period. So we're changing the frequency. This means that in steady state, we can have a, more or less, steady switching frequency. Once the voltage begins to drop and drops below our reference, then we deliver an additional pulse of -- so allowing energy to flow from the input to the output or to the load, and increasing the output voltage again until it reaches the reference. However, when we see a sudden rise in the output current, then, instead of having to wait for the clock, we can instantaneously deliver as many pulses as necessary in order to supply that energy through the output -- to the output capacitor and therefore, maintain this output voltage high and be able to meet the requirements of steady voltage delivery to targets such as the FPGA. So this would be our general view of the Constant-On-Time Control. The advantages of the Constant-On-Time Control: First one is evident. It's the main purpose. We want to be able to increase our low transient performance. So we want to make it as fast as possible and reducing the -- driven the output voltage as much as possible. The additional advantages of using Constant-On-Time is that we simplify the architecture due to the elimination of the compensator. And we also -- it also allows us to seamlessly transition between light-load operation and heavy-load operation as well as not requiring an internal oscillator due to the lack of a clock signal. I'll go into more into these advantages more in depth in the following slides. But I also want to mention that changing to Constant-On-Time also comes with its own set of challenges that we must address. First of which is that the feedback voltage is what we're using to -- as the input signal to our control. And therefore, we require to -- we require a certain periodicity in the feedback voltage signal. I'll go into more detail on this later on. But it's going to be dependent, somehow, on the ESR of our output capacitor. Secondly, the switching frequency is, therefore, going to depend not only on the load current, it's also going to depend on the ratio between the input and output voltages. And therefore, this can mean that we can have variations in our switching frequency, dependent on, for example, variations in the input voltage. This can be problematic because we, as engineers, want to know exactly what frequencies are present in our circuit, and that's already hard enough to do when we have a fixed switching frequency. So variations in the switching frequency is something that we want to minimize. So going back to the advantages, as I said before, when we have this change in the load, instead of having to wait for our clock signal and waiting on the delay from the [ RM ] Amplifier. We can immediately start delivering pulses with Constant-On-Time and therefore, try and deliver the required energy from the input to the output as fast as possible without relying so much on the capacitors discharging to deliver this energy instantaneously. This allows us to deliver sudden high-current steps with reduced drops in the output voltage. In a real example, we took 2 equivalent converters, one using Peak Current Mode and the other using Constant-On-Time Control. And we used them in the same operating conditions, so 12 volts in, 3.3 out, and did a 3.5 amp load step, which, in many cases, is pretty small to what we're seeing, for example, in the core rails for some FPGAs. We use the same inductors and the same output capacitors. And just in this small load step, as I said, we can already see a very big difference between the Peak Current Mode and the Constant-On-Time transient performance. With a Peak Current Mode, we were achieving about an 8% variation from our reference voltage, our desired output. And during the transitions in the load current, we saw a peak-to-peak difference of 0.5 volts, more or less, or an 8% variation. When we switch to the Constant-On-Time, the fact that we don't have to wait for this clock signal means that we can deliver the energy faster, so this energy doesn't have to come so much from the capacitors, so we reduced this peak-to-peak voltage difference down to point -- 2.5%. This value is pretty important because, when working with FPGAs, very often, there's a very strict stability requirement, especially in core power rails. So for example, this is from our reference design, I'll talk more about reference designs later on in the presentation. But for example, the core power rail here is a 1-volt, 20-amp rail. And in this case, we can see -- very easily, we can see variations from, say, 2 or 3 amps up to 16, 17, very sharp, very high DV/DTs, DI/DTs in this load current changes. And the requirement in the stability of this is often around 3%. So at 1 volt, this means that we can only vary about 30 millivolts when doing this load step. In this reference design, for example, we used the 3695-25 power module. MPS power modules -- just as a quick introduction or reminder, in our power modules, we integrate not only the converter, so the power [ fats ] and the control, we also integrate the inductor and some of the other passive components in the circuit into a single package. So with these 3 modules, we're delivering the main rails required for this FPGA. And the 3695-25 can deliver up to 20 amps of continuous output current as well as supporting digital communication for -- both for monitoring and also for programming different parameters such as the output voltage or switching frequency, and using, as I said, Constant-On-Time Control for delivering these very fast transients. Second advantage that I want to touch on for Constant-On-Time is that the fact that we're not requiring a compensator or we're eliminating the compensator from the compensation loop, means that we are eliminating essentially the delay that comes from the [ RM ] amplifier as well as the delay that comes from our RS latch that we're using for the PWM. The reason why -- one of the reasons why COT is so fast is precisely because of the elimination of these elements in the compensation. But this also reduces the time that we spend designing our control loop. It simplifies the design because, I'm sure many of you are familiar with the time it takes when you're testing and simulating or if you are then testing out on the board, modifying and testing different values for these resistors and capacitors is a very time-consuming process. So in Constant-On-Time control, we only have a high-speed comparator, and that directly goes into generating the PWM signal. Third advantage I wanted to mention is the transition from light load to heavy load. When we use more traditional control methods, such as current mode control, when we operate in light-load conditions, frequency can be greatly reduced. So we try to increase our efficiency by using alternative control methods when operating at light load. For example, in MPS, we have advanced asynchronous mode, which basically implements pulse skipping. With a Constant-On-Time, as I said, the pulse is only activated once the output voltage drops below the reference. This means that, as the current increases, the flow from operating in light load to heavy-load operation doesn't require a change in the way that we control the circuit, which can occur in current mode control. This then generates -- in certain operating points, it can cause a droop in efficiency due to the converter moving between these -- both these control methods. Whereas here, as you can see, it's a very smooth progression in the efficiency. And example of this that's implemented in our products is in the MPM3804, 14, 24 and 34. These are our low-current modules designed for some lower-current power rails in FPGAs. And these come in very small packages. For example, our MPM3834 is a 3-amp power module in just 2.5x2.5 package. As I said before, we integrate both the control, the power stage and the inductor in one single package. All we need in this case, is our output voltage divider, our output capacitors and input capacitors. And as you can see, we maintain a high efficiency very stably through very light loads up to the whole current range. The next step, after talking about the advantages, is, of course, we need to mention the challenges of using Constant-On-Time Control. So the first one that I wanted to mention is the variability of the frequency. So as I said before, not only are we going to depend on the output current to set the frequency, so when we increase the current, the voltage drops faster, therefore, we need to deliver these bursts of energy at a higher frequency. It's also going to depend on the operating point or the conversion ratio between the input voltage and the output voltage. This means that, even if we have a steady load current, if we have variations in the input voltage, this can make the steady-state frequency vary in a certain margin. This can be problematic for certain applications. As I said before, we want to have our switching frequency as stable as possible at least in steady state. We won't be able to avoid variations due to the load current, but we want to at least be certain that when we're operating in steady state, we have a good idea of what that switching frequency is. So the way that we have resolved this is using adaptive Constant-On-Time Control. And adding into the one-shot timer the input and output voltages, so that we can, therefore, maintain steady switching frequency during this steady state. So even if there's variations or noise in the input voltage, this doesn't have an effect on the switching frequency of the converter. This allows us to have a better idea or be more at ease of what frequencies we will be seeing in a steady state. The second challenge in this is the one that we see most when working with Constant-on-Time converters is that the output -- the stability is largely dependent on the ESR. So as I said before, we are using the output ripple as the main source for our -- or as the main input for our control loop. So we are dependent on the shape of this ripple in the stability of our control. So let's take a look at what 2 components make up this put output voltage ripple. First, we have the ESR current. ESR, or the equivalent series resistor, of the capacity is a parasitic element, which is present in all capacitors. The only difference is in the type of -- certain types of capacitors tend to have more ESR and others have smaller ones. So as current flows through this ESR, as the capacitor is being charged and discharged, this generates a voltage drop. And we can see the shape of this voltage in green here in this graph. The good thing about this ESR ripple is that because it comes from the inductor current, we can see that the shape that it has is very similar to the one that we saw when we were sensing the inductor current in our current mode control. So we can see that the peak of our ESR ripple occurs at the time where we switch off the PWM signal. So as the inductor charges, this voltage increases. However, the second element is the capacitor voltage itself. And the problem with this is that, as you can see here in the blue, the peak of the capacitor ripple occurs much after the duty cycle or the on-time is finished. So by superposing both of these ripples, we get the total ripple, which has not usually, or ideally, shaped much like this one. So the problem here is if we have an ESR -- so if we have a ripple that's dominated by the ESR ripple, then our ripple voltage will look like this, which is periodic, it allows us to have a very stable control and very fast transient response. However, if our ESR is very small because we're using, say, MLCC capacitors, which is generally what we mainly use as our output bulk capacitant, or what we aim to use because they're also the most cost-efficient way of implementing this output capacitor, they have very small ESRs. And therefore, you run the risk that the dominating component of our output voltage ripple is not the ESR but the capacitor voltage. This can lead the control loop then to subharmonic oscillation and much in the way that we can find subharmonic oscillation in other forms of control, such as current motor control. This can lead to unstable operation in our control loop and therefore, not being able to deliver a well-regulated output voltage. So this poses a big problem to resolve when wanting to implement Constant-On-Time. Luckily, there's several solutions. One, which is pretty obvious, is using output capacitors with high ESR. This is, as I said, a pretty obvious solution, but also problematic because it means that we have a reduced -- a reduction in the efficiency. It also -- using high ESR capacitors is often -- they often take up much more space. And as I said before, we want to be using MLCCs as much as possible in our designs. So the second option we have is for resolving the subharmonic oscillation in a pretty similar way that we resolve it in Current Mode Control, which is adding a amp voltage to our feedback signal. This will ensure that we have periodicity in our control signal. And therefore, we will be able to avoid entering subharmonic oscillation. However, adding this amp voltage can be a struggle. So what we do instead, and this is especially implementable in our modules, is we add this RC circuit in parallel to our inductor, which generates this amp voltage by -- and therefore, we have this large signal entering our control loop. And therefore, regardless of the ratio between the components of the output voltage, the ESR and the capacitor ripple, we will always have a tendency towards this slope voltage, which has the same shape as our ESR level. This will mean that regardless of the capacitors that we use in the ESR of these capacitors, we are much more certain that we will be able to have a proper periodic shape of the total output voltage ripple and therefore, guarantee a stable operation of our converter. So into an implementation of one of these, I wanted to talk a bit about the 3695-100. This is our 100-amp power modules. As you can see here on the top right, it comes in a BGA package, which is 15 millimeters by -- 15 millimeters by 30 millimeters. So this integrates 4 phases inside. So it uses a multiphase operation in order to deliver these 100 amps. So as I said before, we integrate all 4 phases, the control, the power conversion and the inductor, into this one package. And these 4 phases inside our module, the modules can also be paralleled to deliver up to 800 amps in a -- on a multiphase operation. This, I wanted to touch on because even though, as I said before, the frequency is going to depend on the load current, we can still get the advantages of operating in multiphase regardless of operating in Constant-On-Time, so we can further get improvements on our transient performance. And the fact that we can deliver the energy -- the required energy to [ load ] as fast as we can using Constant-On-Time, means that we are less dependent on these output capacitors to deliver this instantaneous energy to our load. This means that we reduce the required capacitants of our output capacitor and therefore, can save a lot of board space and a lot of BOM cost by using Constant-On-Time and using these Constant-On-Time multiphase modules. The other aspect that I wanted to show is that they also require a very few external components and the layout is also very simple, so we further reduce time in our design cycle. And also, when doing our prototyping and doing our designs, many of our modules, especially these high-current modules, come with digital programmability. This allows us to modify parameters very easily, and it will allow us to very quickly be able to trim and modify parameters such as the output voltage or the switching frequency or certain protections and as well as being able to monitor the upper current, the upper voltage and the temperature of our modules, which is very important in critical rails such as the core power rails of an FPGA. So just to give a bit more of information on this, also interleaving Constant-On Time, as I said before, we can put up to 8 of these modules in parallel. This evaluation board is our 800-amp evaluation board, which is available on the website for ordering. We also have a 400-amp version and a 200-amp version and a single IC version as well that are all available. But the -- if many of you have worked, I imagine, with multiphase converted in the past, so you know that the advantages of using multiphase converters is that we can deliver energy much faster to our load and we can also reduce the output ripple significantly. However, it also entails certain added complexity to our design, being, first of all, making sure that we are properly operating in multiphase distributing the phase differences equally amongst the different converters as well as balancing the output current because, for example, if we want to deliver 200 amps to the outputs, we're working with just 2 of these modules, we want to make sure that each one of these is delivering 100 amps and not one's delivering 50 and the other 150 because that is going to lead us to, potentially failure of the device. So that's why MPS modules have active current balancing. So we use the current monitoring, that you can also see via the [ ICE ] QUAD-C to make sure that all the modules are delivering the same amount of current. Second aspect that I wanted to explain is we also aim to simplify the use of multiphase operation by using the auto interleaving. The auto interleaving works by automatically detecting a master module and, as I said, up to 7 slave modules. Very simply, we have 3 pins, the TAKE pin, the PASS pin and the RUN pin. And then if we connect the TAKE pin to VCC, for example, this automatically selects this module as the master, which is then going to deliver the appropriate PWM signals using the PASS or the RUN pins to the next module. And then by knowing how many modules are connected in parallel, we can equally divide the phase shifts amongst the modules and make sure they are all working in balance, both in current and in phase. And we have -- so we have these 3 pins, the PASS, the RUN and the TAKE, and this is because it also allows us to connect the phases inside the module in different ways. Basically, we have 2 options. If we use the RUN, connect it to the TAKE pin, we can have basically these 2 modules operating in power. So there's only 4 phases in both of these. So at any given time, we're using 2 converters in the same in phase. This allows us to deliver more current to the load. So we will be able to deliver higher peak currents and therefore, increase our transient speed. However, it is going to also have a slightly larger ripple than option 2 using the PASS and the TAKE pin because here we're using effectively an 8-phase operation. So the difference between the phases is only 45 degrees. This is going to mean that at any given moment, we only have one converter operating, but it will allow us to have a smaller ripple. So it all depends on our requirements and our system requirements. When we connect up to 8 modules, for example, in the 3695, we can use a combination of either option 1 and option 2, so we can have a combination of these when we are using in 4 phases or in 8 phases. The other element that I wanted to mention of the 3695, which I touched on before, is the easy -- the ability to program very easily using our software Virtual Bench Pro 4. Most of our parts that have digital programmability are available on this software. And this allows us to very easily both monitor the parameters such as the output voltage or the output current or the temperature. And also, it allows us to modify parameters on the fly such as basic parameters. We can see the output voltage. We can modify the switching frequency. We can modify -- well, the switching frequency, we can modify the minimum on time -- the minimum off time, I mean, sorry. And then we can also switch some phases on and off and enable and disable these converters via PMBus or [ ICE ] QUAD-C. And in some of our other modules, more advanced modules, there are advanced parameters that we can modify, such as those that we use -- or those that we've inherited from data center or telecom applications, which are already used to operating in these very high current, very fast load step applications, such as active voltage positioning or load line configuration or modifying parameters for high-speed proprietary communication protocols such as SmartVID or AVSBus. And we can also modify protections and we can modify current monitoring, so we can meet the different requirements from different manufacturers on how we protect our -- both the FPGA and the power supply from potential issues or -- overcurrent or general current monitoring to increase efficiency. And we can also modify some aspects of the PMW control, so the PWM control, the amount of phases that we want to operate, the delays between the banking times and many other parameters, just all in this symbol interface. And the other interesting aspect of this is that, once we finished with all of our tweaking and all of our modifications, we can save the register map. And then by working with your contact at MPS, we can even develop specific suffixes for your memory. So this programming can already be done from production, and you don't have to worry about programming these ICs in-house. One last challenge that I wanted to mention that we need to face when using Constant-On-Time is the DC offsetter. As I said before, we're looking at the output voltage comparing it to a reference voltage. And then once they meet in the comparator, we trigger the one-shot timer and therefore, recharge the output. However, this means that our average output voltage is going to be slightly higher than our reference. In order to resolve this, what we do is we add an error amplifier to the reference voltage. This means that we can reduce our average output voltage to the required or to the desired voltage by reducing the value that we send to the comparator. This eliminates this steady-state error in the output voltage but doesn't affect the ability of the control loop to respond to changes in V out. So just to recap, we have increased transient performance, much faster transient to be able to deliver the energy to our load. We also simplify the architecture. We can operate more seamlessly between light load and heavy load. This is good for applications which have sudden bursts in power requirement. And we've also been able to resolve some of the challenges such as being able to maintain the stability of the converter, even if we're using capacitors with low ESR. The other aspect is that we can maintain a quasi-stable frequency, so we can have a better idea of the frequency our converter will be operating in steady state. And we also simplify, for these very high current applications, the development of multiphase converters by implementing auto interleaving capabilities in our power modules. So just as a very quick overview of the power models that we offer for these FPGA core current rails, we have a wide range of output currents, going from 10 amps to over 100 amps. They also come in the option of single output. For example, this 10-amp module or the 100-amp module that I showed before. But we also have multiple output versions such as the 3690 family, which can deliver dual 13 amps, 18 amps or 25 amps. These are also completely analog. So if they need to operate in an application that doesn't require serial communication, these only require the output -- the resistive divider for the output voltage, same as the 3683 family. And then the other ones do have [ ICE ] QUAD COPM bus capabilities. And the 3698, which is one of our newest modules that can deliver up to 120 amps or an 80-amp and a 40-amp rail, also allows for these high-speed current -- sorry, high-speed communication protocols that are required in some of the newer higher-end FPGAs. And this 82504 is our quad-output 25-amp module, so the outputs can be joined together as required, so we can have either 425 outputs or 125-amp output and 3, so 175-amp outputs or whatever is required. This further increases the flexibility and the capability of using the same part from multiple power rails in complex designs. The other aspect that I wanted to mention is the -- our reference designs. So on our web page, you will be able to find our reference designs for the main FPGA manufacturers, such as AMD Xilinx or Intel, used to be Altera, or Lattice Semiconductor. We have a very wide range of reference designs for most of the commonly used FPGA models. And generally, we have a block diagram that you can use to get a start on your power tree design. And for some, we also have schematics, we have boards and other information that can be of use when designing your power tree. If you have any questions about any of these, you can also reach out to your local MPS contact, and we'll be happy to clear any doubts up. And finally, I didn't want to finish before giving you a very quick overview of the key advantages of using power modules. As you know, MPS has been delivering power converters, Monolithic Power converters, so we integrate the control loop and the power [ fats ] in the single package. And now the next step of this integration is to also add some passive elements into the same package. So in our modules, we integrate the converter and the inductor and some of the other passive elements in the circuit such as the bootstrap capacitor or such as the decoupling input -- some of the decoupling input capacitors as well. This not only reduces the solutions as significantly, as you can see in this comparison here, it also greatly simplifies board layout, reduces the amount of iterations on board design. For example, you don't want somebody mistakenly placing the wrong -- routing the wrong -- routing in the wrong place and placing a signal that's very sensitive to noise underneath the inductor and then you need to redo the whole board. This is great -- the risk of this is greatly reduced by using our power modules. And also, by adding all of these elements into a single package, we reduced the size of the BOM, which also simplifies the manufacturing process and the procurement process as well. And for those of you who work in applications that require a lot of emphasis on EMI, a couple of things to mention is that, for example, for minimizing radiated noise from EMI, this very often comes from, for example, the switch node. By integrating the inductor within the module, we are effectively minimizing the switch node as much as possible, implementing it straight on to the lead frame and, therefore, reducing greatly the amount of emissions that come from this. Also, as I said before, we integrate the small input capacitors on some of our modules. This greatly reduces the size of the hot loops and, therefore, greatly, greatly improves the EMI performance. Most of our most of our module portfolio is designed for industrial applications, but we even have AEC-Q100 automotive-grade modules in our portfolio as well. And some of our modules come with full qualification as a complete power supply for -- with Class B emissions. So that is all that I wanted to share with you today. Kelly, we can, I think, open the floor to any questions that there might be.

Great. Lots of good info today, Tomas. [Operator Instructions]. We already have quite a few in there, and one of the -- ones that I saw buried in there was, is this presentation going to be available? And yes. For all of you attending today, we will send you an e-mail by the end of the week with a link to a video recording of this session and -- as well as a PDF of the presentation. So that's coming your way. And you can always find our past webinars at monolithicpower.com/webinars. All right. And we did have a bunch of questions already in here pretty early, Tomas. First one was, I think right before the [VERTEX slide came in, what is minimum current flowing if it is COT?

I'm not sure, I really -- for what current, the induct current?

Yes, I'm not sure. That's all I got in the question.

Can you repeat it, please?

What is minimum current flowing if it is COT?

In the COT control, well, the minimum current we can deliver is very low. I think it's going to be very similar to the Peak Current Mode. It also depends on what -- because our COT can also operate because -- right. So we implement something also called forced-CCM mode. This is what we use to ensure that we have very low ripple, even operating at light load. And in this case, in forced-CCM mode, the minimum current could be negative because we're operating in forced CCM, so we can go into the negative instantaneous current. However, in normal COT, it's DCM, so it would be, well, zero -- pretty much zero plus the leakage current.

Okay. And I did -- sorry, he added a follow-up to that one of -- for example, if a car is in parking mode without operations. So that was kind of the scenario he wanted to address.

Right. For -- yes, for these sorts of applications that are dependent on the battery, when operating in idle mode, the COT can maintain very high efficiency, as I said. It depends on -- as I said, if we operate a normal COT, we will have a slightly larger output voltage ripple in this idle mode because we're waiting a long time between the -- between pulses. And if we want to maintain lower ripple, we can implement forced CCM. The efficiency will be lower at light load, but we maintain this low ripple even at light load. So it's going to depend on what we implement for operating at light load. But COT is definitely something that -- is definitely recommendable, even for these applications where we have periods of idle states in the load.

Great. Thank you. Next one was, how is stability guaranteed in a COT control?

Stability -- well, stability is never guaranteed, is it really? So the way that we -- as I mentioned before, avoid the issue of ESR or the lack of ESR affecting our stability is we use this RC circuit -- let me see if I can find the right slide. By using the RC circuit, we generate a slope voltage. And this allows us to not be so much dependent on the ESR ripple for the stability of this converter because if the capacitor voltage is the main component of our total output voltage ripple, then we can enter the subharmonic oscillation, which is going to potentially bring us into instability. This is something that was also -- which was also an issue for peak Current Mode Control when operating in duty cycles above -- around 50% or 60%, where any variations or any noise can lead to disarray in the control circuitry and generate instability. So the way that we avoid this is by adding this slope voltage to the feedback and maintaining this periodicity, dependent on the inductor current.

Great. Let's see. I do understand that you have an advantage in load increase for fast response, but how does it increase the performance in load release, I believe?

Well, there's -- so when there's a load increase, the -- so when there's a load increase, we can deliver the power faster to our converter. The problem is, if we -- when there is a load decrease, then the output capacity needs to re-regulate. So the main advantage and the main -- but the main issue in all of this is when we have that sudden increase in the load requirement. So the main pain point in all of this is that voltage drop when we increase the load current. That's the main aspect that COT aims to improve for these sorts of increase -- for this sort of applications, where we have this step-up in current. For the step-down in current, I can't really answer right now off the top of my head, I apologize.

Okay. If fast transient response is requested, why not use hysterical control? The advantages of COT compared to hysterical control.

Yes. Well, hysterical control is pretty similar to COT. It uses a compensator. I think -- and it's been a while since I took a look at hysterical control, but from the best of my knowledge, Constant-On-Time offers more stability in the fact -- so hysterical control can be -- offers less of a stability in the steady state. It's more prone to being jittery. And with the COT and being able to modify the one-shot timer, we can maintain that fixed output voltage and reduce the error in the steady state to make it less dependent on noise in the circuit.

All right. Next one up was, making the converter response faster by eliminating the compensator must be, I guess, in or at the expense of decreasing the power factor and higher THD, isn't it? And then there was kind of a follow-up of, could you show the input current wave form? Did you get all that one, Tomas?

Yes. I don't have any plots of the input current right now that I can think of. But it's a good question. If -- can you repeat the first part, please? Because it's the elimination of the compensator affecting the THD, right?

Yes. That's pretty much it. Making the converter response faster by eliminating the compensator must be at the expense of decreasing the power factor and higher THD.

I mean, the THD in this sort of application, it's not something -- it's something I have to ask the team. I can't give you a good answer right now, I'm afraid.

All right. And [ Risa ], if you're still with us, you see Tomas' e-mail address there on the screen right now, why don't you fire him an e-mail? And he'll make sure that he shares that out to the team and gets it answered.

Yes, please.

Let's see. Next one we got. Compliments on your presentation, Tomas. And then, I once had issues with a COT control interfering with a fixed current limit. Is there any current limit built in with your devices? Is it adjustable?

So we do have -- yes. We have current limit, and it is adjustable, depending on the module, depending -- and some of them are programmable via the [ ICE ] QUAD-C as well. But -- yes.

Great. Then a pretty specific one about the MPM3695. What is the inductor inside it? It's probably in the data sheet, isn't it?

The value of the inductor inside is on the data sheet, in the block diagram. I can't tell you off the top of my head. But that's available on the data sheet.

Yes. So just go, look at the product online, and there will be a button there to access the data sheet, and I'm sure that will help address your question. Let me get caught up here. Let's see. Is the COT time, and in brackets, PWM on-time, adjustable?

In some of our products, yes. What is also adjustable is the off-time. So then you can also limit the sort of maximum and minimum frequencies that you want to be able to operate in. This is especially useful for when you want to avoid emitting in certain in certain bands, frequency bands, that you want to not go into, for example, for EMI purposes. Yes, you can modify the on-time and the off-time. It depends on -- it's on a module-by-module or controller-by-controller basis.

And then another follow-up on the inductor. Are they off-the-shelf discrete inductors? Are they integrated in the PCB? Are they coupled?