VIDEO SERIES
Multifase 101
En el ámbito informático actual, las unidades centrales de procesamiento (CPU), las matrices de puertas programables en campo (FPGA), los circuitos integrados para aplicaciones específicas (ASIC) e incluso los periféricos son cada vez más complejos y, en consecuencia, también lo son sus requisitos de suministro de energía. A fin de satisfacer las demandas más altas, los reguladores multifase son cada vez más frecuentes en las placas base de muchas áreas de la informática. Integrar estos reguladores en los diseños es más desafiante que el uso de conmutadores convencionales y reguladores lineales, pero las ventajas que brinda la multifase superan la complejidad de las aplicaciones de potencia de alto rendimiento. Nuestra cartera de reductores multifásicos incluye reguladores de núcleo de CPU Intel, así como otros reguladores reductores multifásicos ASIC/Procesador/FPGA, con o sin interfaces al procesador. Esta serie de capacitaciones está diseñada para proporcionar las ecuaciones y la orientación necesarias para poner en marcha un nuevo diseño multifase y listo para la validación. Después de una descripción general de las ventajas de la multifase, se presenta un ejemplo de diseño en profundidad de un regulador reductor multifásico para un carril de núcleo ASIC.
An Introduction to Multiphase Buck Regulators
Presenter(s)
Medios
[MUSIC PLAYING]
Hey, everyone. My name's Carmen, and I'm a Power Management Applications Engineer here at TI. Today I'll be giving an introduction to multi-phase buck regulators. In this video, I'll briefly discuss what a multi-phase buck regulator is exactly, what applications they're suited for, and some of the challenges associated with implementing them.
This video follows along with a new application note I released earlier this year, which goes into more detail on each of these topics. If you're interested in learning more than what's covered today, please check it out. So to start, what is a multi-phase buck regulator? Essentially, it's a parallel set of the buck regulator topology we're all familiar with, each phase with its own inductor and power MOSFETs.
Collectively, these components are called the phase, and all the phases share the same set of input and output capacitors. Each phase runs at the duty cycle of a standard single-phase buck during steady state, but the controller will fire off each phase at a spaced interval, so that at any given point, only one phase is active. You'll want to consider a multi-phase buck over a single-phase one for high-performance applications that pull a large amount of output current, with tight transient response requirements, and strict thermal design targets.
These type of regulators are typically found on CPU motherboards, powering processors, top of the line FPGAs, ethernet switches, and various other high-powered ASICs. Today's designs most commonly use anywhere from two to eight phases, unless it's a [INAUDIBLE] application with extreme power requirements. As a rule of thumb, individual phase currents should be kept to 30 or 40 amps. You could push that number higher, but it's recommended to do a thorough study of your design's efficiency targets, budget, and available cooling methods to understand all the trade-offs before committing to a design.
So now that we've discussed what a multi-phase regulator is and where they're most commonly found, we can move on to the biggest advantages versus a single-phase regulator for the same design parameters. Reduced input capacitance, reduced output capacitance, better thermal performance and efficiency at high loads, and finally, better over- and under-shoot during load transience.
You don't need to use as many input capacitors for a design, because adding additional phases reduces the RMS input current flowing through the decoupling capacitors, and thereby reducing the voltage ripple on V In. Looking at a two-phase example, the total input current wave form of a two-phase buck versus a one-phase buck has both a lower peak value, and an RMS current value. This not only reduces C In requirements, but also provides less stress on the upper MOSFET of each phase, making it easier to pick a transistor as more components are available to choose from.
Just how much benefit do you get from adding more phases to a design? Let's take a look at a graph of a normalized RMS current flowing through the input caps, versus the duty cycle of the regulator as phase count increases. The addition of even one more phase can drop the RMS current by a factor of 2 or more, depending on the duty cycle. At several points on the graph, the RMS current even drops to 0 as the individual ripple currents of each phase cancel one another out.
It's important to note, though, that while mathematically you can design to operate at one of these zero-current points, in reality, this isn't possible. You're always going to need input capacitors, whether it's due to noise in the system, line-in load transience, or the natural variation in the duty cycle from both the controller architecture, and the processor demanding different output voltages for different modes of operation.
Moving on to the second benefit of fewer output capacitors, let's refer back to our basic multi-phase diagram. Each phase is operating as if itself were a single standalone regulator. So it has the same inductor ripple current as a single-phase buck would for the same operating conditions. However, because the phases are tied to a common output node, the inductor currents all sum together, and the output capacitors are being concurrently charged and discharged at the same time, producing a lower overall ripple current.
Take a look at this two-phase example, showing how the inductor currents sum together. The I Sum current has a lower peak-to-peak value than the individual phase current, which produces less ripple flowing through the output capacitors. And it also reduces the amount of capacitance needed to keep the DC ripple on V Out within spec.
Looking at the normalized ripple current flowing in the output caps versus duty cycle, you can see that once again, adding more phases to your buck regulator decreases the ripple current significantly. Operating purely at a zero-ripple point and not using output caps is once again impossible in the real world, for the same reasons that prevent you from not using input caps.
However, since the output capacitors are typically clustered into one small area of the PCB, instead of distributed evenly between the inputs of each phase, you can see more capacitance reduction for C Out than C In. The third major benefit multi-phase bucks offer is better thermal and efficiency performance at high loads. This is because all the output power is no longer concentrated into just one set of FETs in a single inductor, which causes a sharp drop in efficiency, and a sharp peak in component and board temperatures.
The power is instead distributed evenly between all the phases, which makes sourcing components less stressful and cheaper. Finding a set of FETs to handle 33 amps DC for a three-phase design is easier than trying to find a 100-amp DC rated FET with acceptable package size, cost, resistance, gate charge, et cetera, et cetera. And that's without getting into high-current inductor selection at all.
Looking at an example of efficiency versus load current, it's easy to see the multi-phase advantage. As we increase the amount of current the load pulls, at some point, there comes a point when the losses in a single phase become too great, and the efficiency starts to decrease. That's when the controller turns on an additional phase, and the efficiency of the regulator can remain higher than it otherwise would. Likewise, as the load current decreases, the controller will turn off phases, and even allow a single phase to enter discontinuous conduction mode to maintain optimal efficiency.
For low currents, fewer phases are used to minimize switching loss in the power FETs, and reduce the current draw associated with the gate drive circuitry. Conversely, at the high-load currents, conduction losses in the FETs and inductors dominate the switching losses, and more phases are needed. Most modern controllers allow for these edge route points to be tuned by the designer, so efficiency can be optimized for each individual application.
The last big advantage multi-phase bucks have over their single-phase counterparts is improved transient response. During a transient event, the multi-phase controller can overlap active phases during a load step, or turn all the phases off during a load release. This effectively puts the inductors of all the phases in parallel with one another, reducing the equivalent inductance seen by the output by a factor of n, where n is the total number of phases.
With the smaller equivalent inductance, charge can more quickly be supplied to the output caps to reduce under-shoot. This also means that less excess charge gets stored in the inductors of each phase to transfer to the output caps when the phases are all turned off, reducing the over-shoot. Since there's no such thing as a free lunch, we've now got to talk about some of the challenges associated with the multi-phase buck regulator.
The first challenge is the trade-off between your design cost and the amount of PCB space needed for the regulator. Adding phases means the price for more FETs and inductors has to be weighed against the cost of more robust components needed to implement a lower-phase count design. The reduction in input and output capacitors can help tip the scales in favor of a multi-phase solution, but there are diminishing returns as phase count increases, which we'll see later on in the video.
Real estate on a PCB isn't cheap High-layer-count complex motherboards are expensive, and to prevent your design costs from ballooning out, a balance between current capabilities and thermal performance versus phase count has to be found. The biggest challenge that comes from moving to a multi-phase design is the issue of phase management.
In order to get the highest performance out of the regulator, lowest currents must be distributed evenly between all active phases. Without good current balancing, you lose the benefits of ripple cancellation on the input and outputs, which can throw your input voltage or output voltage out of spec, potentially causing stability issues. Current balance is also responsible for keeping the FETs and inductors from being thermally stressed, and possibly damaged by excessive current flowing through them.
A phase that's only expecting 30 amp max may just get a bit hotter than expected at 35 amps, but could blow up at 40 if an inductor saturates, or a FET exceeds its current limit. Now, there's a number of ways to balance phase current that we won't dive into too deeply for this video, but each has its own pros and cons associated with it.
Two common current sense methods are DCR sensing and resistor sensing. DCR sensing uses the parasitic DC resistance of the inductor to generate a current sense signal that's fed back to the controller. Resistor sensing places a small sense resistor in series with the inductor, instead of using the DCR. Both methods require a passive network around the sense components to provide filtering, and must be replicated for each phase of the system.
A differential pair for each phase must then be routed back to the controller, which increases the overall layout complexity. DCR sensing requires that passive network contains an NTC thermistor to compensate for the positive temperature coefficient associated with the DCR of the inductor, which complicates the design somewhat. You also have to typically maintain a minimum DCR in order to have enough of a signal to accurately regulate the phase current.
Resistor sensing is simpler to implement, but it adds in additional loss components to the phases, which hurts your efficiency. If you'd like to avoid the headaches associated with tuning these components, smart power stages now exist that integrate the current sense circuitry into the driver MOSFET IC. When paired with a compatible controller, this eliminates a bunch of passive components from the design, while offering greater sense accuracy.
With these smart power stages, fewer, if any, differential pairs is needed to be routed across the PCB. Smart power stages are quite costly, however, and compared to other solutions, may not be the best fit for all applications. All right, so that will about do it for our Intro to Multi-Phase Power video. Thanks a lot for watching. Click here to see the full application note, and stay on the lookout for the next video in the series, where I'll be working through the initial design of a multi-phase buck regulator.