EEVblog #1294 - LLC Resonant Mode Converter Design

[EEVblog]

A brief look at how LLC resonant mode converters work and their advantages. A spin-off from the Rohde & Schwarz NGP800 power supply video.
Application note: https://www.infineon.com/dgdl/Application_Note_Resonant+LLC+Converter+Operation+and+Design_Infineon.pdf?fileId=db3a30433a047ba0013a4a60e3be64a1

NOTE: There appears to be quite some debate as to whether this NGP800 PSU is actually a resonant converter topology, and yes, I need to take a closer look at this. But just go with the flow and pretend it is for the purposes of this video.

[johnlsenchak]

Watched the video  twice,   because it's  like watching a  "Signal  Path" video    where it's mind  blowing   all the way through 

I think  a simpler  way  of  saying it ,  and tell me if I'm wrong here, is that the  L.C.C.   resonant  circuit  attenuates the   square  wave  at a  certain  frequency  that the Pulse width  modulator ,  runs  at to bring the  oscillator  frequency   down  to more  of a   sine  wave to improve  efficiencies in the main  switching  transformer

It  has  to do  with that  "Q" thing in  a inductor  / capacitor  in a A/C  circuit , it's  starting to come  back to me now

Q factor

The winding resistance appears as a resistance in series with the inductor; it is referred to as DCR (DC resistance). This resistance dissipates some of the reactive energy. The quality factor (or Q) of an inductor is the ratio of its inductive reactance to its resistance at a given frequency, and is a measure of its efficiency. The higher the Q factor of the inductor, the closer it approaches the behavior of an ideal inductor. High Q inductors are used with capacitors to make resonant circuits in radio transmitters and receivers. The higher the Q is, the narrower the bandwidth of the resonant circuit.

The Q factor of an inductor is defined as, where L is the inductance, R is the DCR, and the product ωL is the inductive reactance:

    Q = ω L R {\displaystyle Q={\frac {\omega L}{R}}} Q={\frac {\omega L}{R}}

[filssavi]

Transformer utilisation is a factor, however the main benefit how soft switched topologies is the reduction in switching losses.
The best way to put it is that the resonant frequency and Q of the tank circuit are chosen in order to have zero voltage (ZVS) or zero current (ZCS) when the transistor has to switch, thus having almost no losses

The main challenge with these type of supplies is that the tank circuit is optimised for a certain (usually fairly narrow) range of operating conditions, once out of those (typically due to load variations) the operation reverts back to normal hard switching, with the associate challenges

With complex loads (especially loads that can occasionally back-feed energy the regular loops and protection circuitry must be fast, as exiting soft switching region at full load is guaranteed to result in a very exciting couple of seconds 

[ali_asadzadeh]

Dave, doing some SMPS design tutorials covering various topologies would be nice, specially the MCU based ones

[David Hess]

You really can't do this with bipolar transistors because the drive requirements are too much.

Shown below is the schematic to the resonant half-bridge switching power supply for the Tektronix 7704 oscilloscope which was released in 1969 and uses bipolar transistors.  After the 7704, the controller implemented with discretes was replaced with an ASIC and this power supply design was used until 1991 in the 7000 series oscilloscopes and some other instruments.

Resonate mode operation is especially beneficial for reducing switching losses of bipolar transistors.  Low impedance driving requirements are accommodated with a transformer which also provides the proper phasing.  In this case the half-bridge is self oscillating with the output providing drive to the transistors through a transformer however the control circuit blocks the drive with a separate winding providing phase modulation allowing regulation; this is not a burp regulator or open loop inverter.

I suspect the aversion to using bipolar transistors in circuits like this was more from a lack of understanding than difficulty of providing proper drive.  Schooling tends to emphasize working with voltages instead of currents and I think this shows in an aversion to other design techniques like current switching and current feedback.

[chris_leyson]

The 7000 series resonant inverter is a nice design. Transistor base drive is proportional due to the single turn current sense winding.

[David Hess]

The 7000 series resonant inverter is a nice design. Transistor base drive is proportional due to the single turn current sense winding.

It is an amazing design.  I like how the current sense transformer provides current limiting, phase information, and powers the controller.

[filssavi]

The power BJT’s are not used anymore for several good reasons

1) switching them fast and hard is very, very difficult if not impossible
To avoid excessive on state losses the transistor must be turned on very hard, unfortunately this leads to extremely high carrier concentrations that must be extracted from the base before the transistor can turn off, all techniques used in small signal applications to enhance switching behaviour can’t be applied  (baker clamps of various topologies and high gate resistances) as they would result in massive on losses
2) the NTC characteristic of BJTs makes parallel configurations very prone to thermal runaway, and thus dangerous
3)current gain of power BJT’s is very low (in the order of 10/20), requiring very high drive currents/powers (this is very annoying especially for the high side where every transistor (two in a full bridge, three in a VSI) needs its own isolated DC/DC converter

However bipolar devices are actually the backbone of current power electronics, in the form of IGBTs, while they somewhat retain the first problem from BJTs (the tail current) with turn off times that can be as high as several microseconds (that is glacially slow) they resolve the other two, making it much more of a compelling choice, especially for high voltage

[T3sl4co1l]

Sony was fond of resonant supplies as well, using current feedback to solve base drive, via a saturable reactor to control oscillation frequency and thus power into the resonant tank.  Trinitrons for example often used this.

The common CFL does the same thing, without control (or if it's controlled, it's by fixed saturation in the drive transformer).

1) switching them fast and hard is very, very difficult if not impossible
To avoid excessive on state losses the transistor must be turned on very hard, unfortunately this leads to extremely high carrier concentrations that must be extracted from the base before the transistor can turn off, all techniques used in small signal applications to enhance switching behaviour can’t be applied  (baker clamps of various topologies and high gate resistances) as they would result in massive on losses

What's fast?  For the time, fast was 10s of kHz, then 100s.  MOSFETs took over at exactly the same frequencies; I wouldn't at all say it was difficult!

Base drive isn't hard to arrange, but it's really only practical through transformers, which I suspect is one of the main reasons they've fallen out of style.  Transformers are componentized labor, and labor costs are continuing to rise.

You don't avoid storage time, but doping profiles are optimized for rapid fall time once t_s has passed.  Typical late generation 800/1.5kV line output transistor did 2us storage and 150ns rise, not a bad ratio at all!

Alternately, you can use Baker clamp style drive methods, but they're really only practical with bootstrap drivers (yes, they've been used with BJTs as well!), especially in integrated circuits (where the dozens of additional transistors are practically no cost, and very complex drive methods can be arranged, perfectly tuning Vce(on), hFE(on), rise and fall).

A number of LT switching regulators are actually such things!  You can tell from the huge voltage range (say 3-40V operating), low required bootstrap voltage, and the prominently placed NPN in the block diagram.

And also, higher voltage drops aren't much of a problem at higher supply voltages, for example an old fashioned SimpleSwitcher (or, lord forbid, even a MC34063 and its ilk!), with its Darlington output, isn't much of a problem with a 24V supply where the voltage drop only amounts to a maximum say 90% efficiency.  Boo hoo!   No, certainly not going to compete with a modern synchronous regulator pushing 90-95% [overall, not just conduction], but they were competing with linear regulators of 50-70%, an easy win at the time.

2) the NTC characteristic of BJTs makes parallel configurations very prone to thermal runaway, and thus dangerous
3)current gain of power BJT’s is very low (in the order of 10/20), requiring very high drive currents/powers (this is very annoying especially for the high side where every transistor (two in a full bridge, three in a VSI) needs its own isolated DC/DC converter

Of course, they always made BJTs in whatever size you needed, so this was never a problem.  To this day, you can get another bipolar device, the SCR, in entire-wafer models.  No worries about matching there.

The biggest BJTs I've personally handled, were just some triple-Darlington style inverter bricks, good for around 10kW I think it was.  Was a kinda-oddball motor drive, IIRC its contemporaries were all SCR and it was the one model that wasn't?

Tim

[filssavi]

A number of LT switching regulators are actually such things!  You can tell from the huge voltage range (say 3-40V operating), low required bootstrap voltage, and the prominently placed NPN in the block diagram.

And also, higher voltage drops aren't much of a problem at higher supply voltages, for example an old fashioned SimpleSwitcher (or, lord forbid, even a MC34063 and its ilk!), with its Darlington output, isn't much of a problem with a 24V supply where the voltage drop only amounts to a maximum say 90% efficiency.  Boo hoo!   No, certainly not going to compete with a modern synchronous regulator pushing 90-95% [overall, not just conduction], but they were competing with linear regulators of 50-70%, an easy win at the time.

...

The biggest BJTs I've personally handled, were just some triple-Darlington style inverter bricks, good for around 10kW I think it was.  Was a kinda-oddball motor drive, IIRC its contemporaries were all SCR and it was the one model that wasn't?

Tim

Yes sorry my bad, I should have specified, that i'm talking about what a usual EE (not someone like me XD) would class as high power supplies, that is to say 2-3kW and more and at least EU line voltage, so 300V upwards at few amps at least.

the systems I'm working with right now are supposed to be in the hundreds of amps at 600-700V dc link

Historically you are correct Power BJTs were the first transistors used in a land dominated by SCRs, however they have been gradually phased out in favor of MOS and IGBTs

As to baker clamps, I wasn't saying they are technically not feasible (which they of course are) It's just that when reasonably high currents are involved to avoid massive losses the bjt's must be saturated very hard, not doing so would mean loosing few percent efficiency (even losing 0.5% on a machine drive is a massive hit)

[David Hess]

Sony was fond of resonant supplies as well, using current feedback to solve base drive, via a saturable reactor to control oscillation frequency and thus power into the resonant tank.  Trinitrons for example often used this.

The common CFL does the same thing, without control (or if it's controlled, it's by fixed saturation in the drive transformer).

Motorola published application notes on using bipolar transistors in CFL ballasts which were at least competitive if not better than MOSFET designs.  But of course they all required that one extra low impedance winding on the transformer.

The advantage of the bipolar transistors was economic if you could get past the more complex drive requirements.  Like IGBTs, they support a higher current density and higher voltages do not require as large a die as a MOSFET for a given on resistance although I guess MOSFETs have improved considerably in that respect.  I suspect just economy of scale of MOSFETs over bipolars has ameliorated the die size disadvantage of MOSFETs.

[filssavi]

Sony was fond of resonant supplies as well, using current feedback to solve base drive, via a saturable reactor to control oscillation frequency and thus power into the resonant tank.  Trinitrons for example often used this.

The common CFL does the same thing, without control (or if it's controlled, it's by fixed saturation in the drive transformer).

Motorola published application notes on using bipolar transistors in CFL ballasts which were at least competitive if not better than MOSFET designs.  But of course they all required that one extra low impedance winding on the transformer.

The advantage of the bipolar transistors was economic if you could get past the more complex drive requirements.  Like IGBTs, they support a higher current density and higher voltages do not require as large a die as a MOSFET for a given on resistance although I guess MOSFETs have improved considerably in that respect.  I suspect just economy of scale of MOSFETs over bipolars has ameliorated the die size disadvantage of MOSFETs.

MOSFETs being field effect devices will always have different characteristics with respect to bipolar devices like BJTs and IGBTs, as pointed out they require more silicon for a given current, however unlike bipolar they have a linear characteristic which  while useless at full load, it generally results in lower losses at partial load

Bipolar devices, in power electronics application, where losses are the main limiting factor and figure of merit, bipolar devices generally switch slower than unipolar devices (MOSFETs and HEMTs) by usually at least an order of magnitude.

By contrast bypolar devices are usually more rugged

Also power MOSFETs have come a long way since the seventies and eighties, the super junction structure especially makes for much better mosfets

[David Hess]

MOSFETs being field effect devices will always have different characteristics with respect to bipolar devices like BJTs and IGBTs, as pointed out they require more silicon for a given current, however unlike bipolar they have a linear characteristic which  while useless at full load, it generally results in lower losses at partial load.

Of more relevance is that MOSFET die size is proportional to the square of the voltage.  So bipolar transistors and IGBTs have a larger advantage at higher voltages where their smaller die size makes them less expensive.

[T3sl4co1l]

That used to be the biggest drawback, yeah.  The price increase just from 900V to 1200V FETs for example, of equivalent ampacity, was huge.  SuperJunction broke that barrier, so that switching area (VAs) goes as die area.  Even very high voltage MOSFETs (up to 4kV) are available now, with quite impressive specs.

Tim

[JPortici]

Dave, doing some SMPS design tutorials covering various topologies would be nice, specially the MCU based ones

Mcu based ones: at its basic eveything's the same, but in the digital domain (get vfb/ipeak -> apply compensator -> output new duty cycle)
some of the bonuses are integrated diagnostic, arbitrary frequency response of the compensastor, strange topologies or different compensators per different load points, combination of functions such as DC Link boost converter followed by a three phase motor driver

[David Hess]

That used to be the biggest drawback, yeah.  The price increase just from 900V to 1200V FETs for example, of equivalent ampacity, was huge.  SuperJunction broke that barrier, so that switching area (VAs) goes as die area.  Even very high voltage MOSFETs (up to 4kV) are available now, with quite impressive specs.

None of my big power semiconductor books are recent enough to cover superjunction MOSFETs so I am not sure what is going on with them.  I do know however that older technology MOSFETs have come down in price showing that economy of scale has made up for the difference anyway at least up to medium voltages.

It seems like area would still be proportional to voltage squared but I assume superjunction MOSFETs have a higher figure of merit.

[T3sl4co1l]

And old generation FETs are still often the same die size as ever, no shrink.  They must be about the cheapest way to buy high purity silicon in small quantities...  The die area means they're still desirable for linear applications today.

I imagine most of the cost reduction is from the vast production quantity, and wholly written-off masks.  I also wonder if they might be making them on >= 12" wafers to get just that little bit extra economy, or if those are mainly being used for high tech stuff.

SJ, in addition to having proportional scaling, also has a monstrous C(V) curve, where Crss and Coss drop sharply, starting at say Vds/30 and dropping by 100x or more by Vds/10.  A lot of datasheets show Coss rebounding slightly, but I wonder if that's a measurement error or what.  Also, my measurements have shown a less aggressive curve than datasheets show, which I wonder if it's methodological or real (and which one is in error..).

Tim

[poorchava]

Doing a FB-LLC for 250W seems like a huge overkill. I mean complex inductor design, full bridge drive etc. I'd say this power range calls for something like HB, 2-S Forward or even plain old push-pull.

If you added a zero to that 250W, that would make sense for FB-LLC (or PSFB-ZVS). LLC is also a bitch when you need adjustable voltage.

As for SJ fets, SiC eats them raw, and SiC is getting quite affordable. I recently completed a 400VDC to 400VAC@1.5kW/550VDC@3kW converter which achieves about >=93% (measured with PFC front end) and the entire secondary side is SiC. The Si FETs with reasonably low RDs for stuff like regenerative clamping or synchronous rectification are often more expensive than equivalent SiC parts (plus you can run SiC hotter and faster that Si.

For primary side SiC doesn't make sense yet at 3kW, I just used the CFD7 (honestly when it comes to state-of-the-art FETs, there is Infineon and the is the rest, chasing Infineons tail). SiC would make sense for better efficiency, since a PSFB-ZVS like that draws about 30-40W @ 50kHz without load, just because of parasitic capacitanes.