Estimating MOSFET switching losses means higher performance ...

1、estimating mosfet switching losses means higher performance buck converters the relentless trend towards lower voltage, higher current and faster load slew rate poses a constant challenge for power supply designers. a decade ago, a typical power supply would have 5v and +/- 12v outputs. obtaining re

2、asonable efficiency and transient response within +/- 5 percent limit was rather straightforward. relatively mature technologies of aluminum electrolytic capacitors and schottk'y rectifiers were not leaving much room for error (or creativity). as the voltage required for a new generation of semi

3、conductors had gone below 3.3v changes started to accelerate. diodes were replaced by mosfets, several new types of high performance capacitors became available and new multiphase structures gained popularity with the advent of highly integrated controllers. the structure of the power systems shifte

4、d from a central to a distributed one. new categories of power supplies like point of load (pol) and vrm entered the mainstream market. the most typical set of features of those new power supplies are as follows: - high current, low voltage, single output - close proximity to the load - non isolated

5、 buck topology - fast transient capability with narrow regulation band - high level of integration with sophisticated pwm controllers at the same time the availability of integrated controllers allowed engineers with just a basic power supply background to design high performance converters quickly

6、and efficiently. on the surface, bringing to life a pol converter is simpler than an isolated telecomm brick or an ac/dc switcher. however, high and ever changing performance requirements, variety of new components and the relentless pressure on cost and cycle time make an optimization process a non

7、 trivial task. with the simplification of the structure, the bulk of the converter's losses is now located in the synchronous switching cell. the optimization of this part of the circuit is critical for achieving balance between the cost, size, efficiency and transient response capability. sever

8、al factors require a more detailed look at the phenomenon of switching losses: - the fast transient requirement pushes us towards possibly high switching frequencies, which in turn increases the relative importance of the switching losses - switching processes of a synchronous cell are more complex

9、than with a diode - the selection of components is non-obvious (bigger does not have to be better) and has a critical impact on the performance let's take a look at several possible approaches to the task of estimating the switching losses in a synchronous buck converter. method 1: just do it. y

10、ears ago it was possible to achieve acceptable results without giving much thought to switching losses at all. with slowly changing technologies and fewer choices, the selection of typical mosfets with typical drivers and at a typical switching frequency could give us a solution not far from the one

11、 obtained after meticulous optimization. i have seen designers with a good track record working as follows: calculate the conduction losses, add 50 to 100% provision for switching losses and be done. this is undoubtedly the fastest and the cheapest of all methods. currently obsolete, because nobody

12、knows what typical components and typical frequency means. method 2: fall and rise time.switching losses are created as a result of a simultaneous exposure of a mosfet to high voltage and current during a transition between the open and closed states. therefore, it is sufficient to know the duration

13、 and the type of a transition (for example resistive or inductive) and the calculation is straightforward: figure 1: idealized switching waveformsidealized resistive switching (psw = 1/6 * fs * vds * id * tsw) idealized inductive switching (psw = 1/2 * fs * vds * id * tsw) true, the real waveforms a

14、re a bit more complicated, but those errors can be within acceptable limits, providing that we are able to properly predict the rise and fall time of the drain to source voltage and the drain current. those parameters can be found in the manufacturers' datasheets. there are however numerous diff

15、iculties with obtaining accurate results in this way: - components are described at one set of typical conditions, without an obvious method of conversion to another one - some parameters are characterized at unrealistic conditions because manufacturers do not have an equipment to perform measuremen

16、ts reliably in life-like setup (di/dt of the reverse recovery of a body diode with a low inductance layout can reach several thousand a/us, as opposed to 100 in datasheets) - components are tested differently by various vendors, making direct comparison difficult - timing is usually defined for the

17、resistive switching which is convenient for manufacturers but has limited relevance for the inductive, diode clamping commutation taking place in the synchronous buck - data is presented in a confusing way - for example fall time is defined as a time needed for drain to source voltage to rise from 1

18、0 to 90% - last but not least some datasheets simply contain strange numbers if we cannot obtain rise and fall times from manufacturers, we have to find them ourselves, hence the third approach: method 3: measure itthe basic premise of this approach is that the switching phenomenon are just simply t

19、oo complex to be accurately predicted. therefore, we have to capture relevant waveforms in the working model and calculate the switching loss. this task used to involve a bit of computational effort, but with new oscilloscopes the whole math will be done for you. just hook up the probes, sit back an

20、d relax. not so fast: first: even very good current probes introduce a substantial delay in the waveforms. in the past 10ns or so was not so important, but in the case of switching losses in a modern, fast dc/dc converter, capable of traversing from rail to rail in few nanoseconds it will lead to a

21、completely false results. this delay can be estimated and proper correction introduced, but for fast switching circuits the error will remain huge. second: introducing a current probe into the circuit without disturbing a switching transition is sometimes very difficult. in the case of pol buck conv

22、erters this act must increase the parasitic inductance of the switching cell significantly, thus distorting waveforms and making measurements inaccurate. sometimes the converter will refuse to work completely. third: if anybody is not yet convinced of the limited usefulness of the measurement method

23、, please consider this: in order to implement this method we have to have a working model of the power supply. if we get to this point we may just as well measure the overall efficiency and the temperatures of key components as a way of verifying of the design. after all, how important is the exact

24、split of losses between switching and conduction, if the whole thing just works fine? as we can see optimizing mosfets of a synchronous buck is not an easy task. engineers do not have a luxury of time for detailed studies but trial and error method may turn out to be even more costly. a trade off be

25、tween the accuracy of losses prediction and the amount of time and effort invested is necessary. presented below approach is an attempt to balance accuracy with practical simplicity. method 4: calculationupon closer examination the switching processes of the synchronous buck converter are surprising

26、ly complex. we can easily distinguish 9 major components of these losses: conduction and gate charge losses for both switches, turn-on, turn-off and output capacitance losses for the forward mosfet and body diode conduction plus body diode reverse recovery for the freewheeling mosfet. below, each of

27、 these components is analyzed in a greater detail and accompanied by the calculation for a "typical buck". for simplicity let's omit the second order impact of the ac inductor current. parameters used in the example: table 1 figure 2: basic structure of a synchronous buck converter for

28、 the freewheeling (synchronous, lower) mosfet: 1. conduction losses: 0.60 wthe losses caused by the flow of the current through the on resistance of the device during the freewheeling part of the switching cycle. conduction process of the freewheeling mosfet is a good place to start the calculations

29、. it is the main component of the losses in this device and is relatively independent of the rest of the design. pcdmd = iout2 x (1-d) = rdson = 0.60 w where d - duty cycle (d vout/(vin*efficiency) rdson - on resistance (consider the target temperature and gate voltage) typically the freewheeling mo

30、sfet should have the rdson as low as possible within the space and budget constraints. for high frequency and silicon reach designs a balance must be achieved with the limitations imposed by an excessive gate charge. 2. gate charge losses: 0.27 wenergy lost due to periodic charging of the gate capac

31、itance:pgd = fsw ( qgtot ( vgate free = 0.24 w (at assumed vgate free = 6 v) qgtot - the total gate charge it is sometimes mentioned in the literature that gate charge losses of the freewheeling mosfet should be calculated without the miller charge (approximated by the horizontal part of the gate ch

32、arge chart). this is incorrect. the loss associated with the miller charge is still present. it just occurs when the drain to source voltage swings, not when the driver changes its state. figure 3: gate charge chart with qgd - the miller charge the energy used to drive the gate of the device is lost

33、 unless a special recovery circuit is employed. this loss is divided between the gate driver and the mosfet depending on the impedance ratio of those components. new generation mosfets typically have gate resistance in the 1 to 2 ohm range, much less than a mid size gate driver, so most heat ends up

34、 in the driver. frequently the ability of the driver to dissipate the heat will be a limiting factor for the selection of the freewheeling mosfet. the speed of switching, which is strongly affected by the size of the mosfet's die is less critical because of the assistance of the body diode (exce

35、pt for the extreme case, when a very sloppy transition leads to an overlap with a drive signal of the upper mosfet and shoot-through). typically the lower rdson, the bigger is the capacitance of the gate, the slower is the switching and the more loss is incurred for driving. new devices are also opt

36、imized for driving with lower gate voltage. driving our freewheeling mosfet to 12v would produce over 1 w of heat, exceeding the conduction component! if a lower gate voltage is not possible, a higher rdson part may be necessary to keep the gate charge losses down. 3. body diode conduction losses: 0

37、.45 wlosses incurred during a brief period just before or just after the switching transition, during which the freewheeling mosfet conducts the current with zero gate voltage. this forces the current to flow in the internal body diode and has a significant impact on the efficiency because of a much

38、 higher voltage drop across the device during this period (and subsequently due to the energy loss for the reverse recovery): pbd = fsw * vbd * iout * (tdead on + tdead off) = 0.45 w vbd - forward drop of the body diode (0.7 v in this case) tdead on, tdead off - dead time in the gate driver (assume

39、100 ns total for on and off) figure 4: turn-on waveforms the dead time is mostly determined by the selection of the gate driver. it is not unusual to see 1 or even 2% difference in the overall efficiency due to the duration of the dead time. most manufacturers prefer to insure ample dead time, becau

40、se it minimizes the danger of cross-conduction (overlapping the on state of both devices) regardless of the mosfet selection and layout. cross-conduction may lead to a component failure, while too much dead time only to a slightly worsen performance. in order to avoid proliferation of parts with a v

41、ariety of timing a conservative approach is preferred. this unfortunately leads to an under optimized design. in fact, operation with no dead time or even with minimal overlapping (cross-conduction) is the most efficient mode of operation. the idea of switching at the boundary of cross-conduction (w

42、hich is just a less pejorative word for shoot-through) requires a bit of explanation. the gain is threefold. first, the period during which the lossy body diode conducts is reduced. second, as the conduction of the body diode is getting very short, the reverse recovery process is getting easier. it

43、is because there is not enough time to fully saturate the junction with the excess carriers. third, with further reduction of the dead time some of the current never gets to the body diode from the mosfet's channel making the reverse recovery even more easy. to achieve this mode of operation the

44、 gate drive timing must be very precisely adjusted to a particular design. what is even more difficult, it must be also adjusted with the line/load conditions and variations in the component properties. for this reasons such designs have been rare in the past. currently new drivers, that actively se

45、nse the dynamics of the switching transition and adjust the dead time accordingly are becoming available. it should give good results providing that the operating conditions do not vary too rapidly from one cycle to another. therefore we should approach this option carefully in the designs that have

46、 to change the duty cycle quickly due to a very fast transient response. a widely known trick aimed to alleviate the body diode pains is the addition of a schottk'y in parallel with the freewheeling mosfet. this is meant to eliminate the current flow in the sloppy body diode and provide the bene

47、fit of almost instantaneous (lossless) reverse recovery of the schottk'y. unfortunately it works only at lower switching frequencies or with the schottk'y integrated in one device with the mosfet. the reason is simply the stray inductance between the mosfet's channel and the schottk'

48、y causes a substantial amount of time to commutate the current. the body diode even though with a higher voltage drop is much closer. let's assume that an average body diode has a voltage drop of 0.8v, schottk'y only 0.4v and the inductance between them is only 4nh (an optimistic assumption)

49、. this means that the current will be shifting from the internal body diode to the external schottk'y at the rate of 1a per 10 ns. in our example we would have to wait for complete transition for 150ns. this additional time would produce a lot of extra losses due to the body diode conduction whi

50、ch would cancel the benefits. also the design of the controller may become complicated. a schottk'y integrated in one package steals the space for the mosfet, limiting its rdson and therefore is applicable only for low current designs. an external schottk'y without a dead time extension is j

51、ust a dead weight.4. body diode reverse recovery losses: 0.54 w (0.18 w in the freewheeling mosfet)losses incurred during reverse recovery of the body diode of the freewheeling device. the dynamics of this process are determined by the properties of the body diode, turn-on of the control mosfet and

52、the stray inductance in the triangle: forward mosfet, freewheeling mosfet and input capacitor (depicted as lstray in figure 2). a body diode is an inherently slow device and a substantial amount of reverse charge must be delivered during the recovery process (see the big current spike on the figure

53、3). this charge is delivered under the voltage approximately equal to vin and results in a large amount of power loss. it also extends the duration of the turn-on transition of the forward mosfet causing further degradation in efficiency. the calculation of these losses is difficult to conduct preci

54、sely because of poorly characterized properties of the body diode in the manufacturer's data. they also vary significantly with the temperature, forward current and the duration of the recovery process. an error introduced by reverse recovery is usually the biggest factor behind inaccuracy of th

55、e overall prediction of the efficiency. it is however necessary to estimate the magnitude of this source of losses because of its significant contribution. if manufacturer specifies the reverse recovery charge qrr: prr = fsw * qrr * vin = 0.54 w if only reverse recovery time at given di/dt is specif

56、ied qrr can be then estimated as follows: irrpeak = 0.6 * di/dt * trr = 3.3 a - peak reverse recovery current irrav = irrpeak * 1/2 = 1.65 a - average reverse recovery current qrr = irrav * trr = 91 nc - reverse recovery charge it is difficult to determine how the heat generated during the reverse r

57、ecovery process splits between both devices. lets say 1/3 of the power is dissipated in the body diode (freewheeling mosfet), 1/2 in the forward mosfet and 1/6 in the rest of the circuit, but it will vary from case to case. see more information about optimizing the reverse recovery losses in section

58、 describing turn-on of the forward mosfet. 5. turn-on losses 0 w turn-on losses are here understood as a simultaneous exposure of the component to a significant voltage and current experienced during turn-on of the device. there are no turn-on losses of the freewheeling mosfet because it turns-on at

59、 zero voltage. the commutation process is forced by the energy stored in the choke. 6. turn-off losses 0 wturn-off losses are here understood as a simultaneous exposure of the component to a significant voltage and current experienced during turn-off of the device. similarly turn-off losses are zero, because of the assistance of the body diode. 7. output capacitance losses 0 woutput capacitance losses are the losses resulting from dissipating the energy stored in the output capacitance of the mosfet during turn-on. ou