EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: splin on August 11, 2016, 11:06:46 pm
-
I'd like to achieve > 96% efficiency overall for a 24V to 2.4V, 25A synchronous buck converter operating at 150kHz. Using low resistance planar inductors such as Coilcraft SER2915 or SER2918, it looks like core and copper losses of < 1% can be achieved at 150kHz. Dropping to 100kHz reduces the switching losses but then the inductors get more expensive.
I'm playing with a switching loss spreadsheet from Fairchild to see what efficiency might be achieved. The results looked good, better than 3% losses for the FETs, until I realised that the switching times were unrealistically fast with sub nano-second switching times for the sync FET when driven from a 10V, 1 ohm driver. The FET was a BSC0504NSI 3mOhms, 11.1nC Qg total at 10V and only 2.2nC Qsw). I realise this FET is only 30V so probably not suitable but I was looking to see what the best case could be.
So what sort of switching speeds might be achievable in reality? Is 4ns or less (Tpl-h and Tph-l) realistic without using exoctic devices or excessively complex drivers? I'm assuming that given enough gate drive the plateau part of the switching time can be reduced to <1ns but the rise and fall times are limited by the source and drain stray inductances? If < 4ns could be achieved would EMI problems become too severe?
Is 96% a realistic target? Other converter topologies might be more suitable given the relatively high input/output ratio but I like the relative simplicity of the buck converter and the relative ease of obtaining suitable inducters. As you can probably tell I've never built a switcher bigger than a few hundred milliwatts so I'm open to any suggestions.
-
Given that I could hit 95% efficiency in just a few minutes poking at LTpowerCAD, 96% on paper should be doable without too much trouble. Of course all these modelling tools leave out a lot of small stuff, stuff that stops being so small at 25A and very high efficiencies. And they never include layout, that little detail by which switchers thrive or fail. But your goals seem reasonable, even without resorting to exotics like GaN.
(FWIW, the LTpowerCAD setup I was looking at was the LTC3867 with SER2915L-152KL inductor and FDPC8012S dual FET. This is not a recommendation of these parts. They are just a combination that's somewhere in the right ballpark!)
-
my experience in the past is that reducing switching times into the range of <10ns exxagerates parasitic ringing, and it becomes increasingly difficulty to compensate it with snubbers. As this is a synchronous design, make sure that your driver has a short, even better adjustable dead time. Maybe make the switching asymmetric with a diode: fast off, slow on. During the dead time current flows through the low side reverse diode. You may want to add a schottky in parallel to the transistor. Besides the transistors and the PCB layout, the inductor plays a big role in parasitic ringing, check the parasitic capacitance of your part (can be derived indirectly from self resonance spec).
To the FET voltage rating: most modern FETs are self-protecting and act like a zener diode above their voltage rating. You just need to make sure not to generate too much heat in them. Look for repetitive avalanche rating, or UIS (unclamped inductive switching).
-
Given that I could hit 95% efficiency in just a few minutes poking at LTpowerCAD, 96% on paper should be doable without too much trouble. Of course all these modelling tools leave out a lot of small stuff, stuff that stops being so small at 25A and very high efficiencies. And they never include layout, that little detail by which switchers thrive or fail. But your goals seem reasonable, even without resorting to exotics like GaN.
(FWIW, the LTpowerCAD setup I was looking at was the LTC3867 with SER2915L-152KL inductor and FDPC8012S dual FET. This is not a recommendation of these parts. They are just a combination that's somewhere in the right ballpark!)
Thanks for that - your results are encouraging. I had overlooked LTpowerCAD and am now having a play with it. As you say, there is the question of how close real implementations get to the simulation results given the factors that aren't included in the simulation. There's no substitute for real world experience here which I don't have.
I used the LTC3867 model and selected, as closely as possible, FETs, inductor and output capacitors to match the L3867 demo board DC177A: RJK0305DPD, PSNM1R0-30, Vitec 59P9875N etc. (I added the 2mohm sense resistor of the demo board into the simulator's inductor DCR). With 12V in, 1.5V, 15A out @ 400kHz the simulation gave an efficiency of 91.9% which is almost 4% higher than the 88% typical claimed for the demo board. That doesn't give me much confidence that I could reach 95% in reality but I could easily have made a mistake.
There is also the problem of the very large typical to maximum spreads in many MOSFET characteristics such as gate resistance, capacitances etc. which mean the worst case efficiency of a converter may be quite a lot worse than simulated or the bench models built with parts which happened to have close to typical values.
-
You might also look at a multi-phase design which would permit you to reduce the current per phase and reduce the switching frequency for a given inductor size given the help of the ripple / noise cancellation. Of course the more complex drivers will have an efficiency penalty but it may work out favorably at the higher current levels. Linear has some nice multi-phase controllers.
Cost is an important consideration so multi-phase may be too expensive but the reduced EMI filtering requirments could justify it.
Of course the problem with high efficiency is whether you mean the peak efficiency over the operating region or the worst case efficiency which might occur at light loads in which case 96% will perhaps be unrealistic due to operational overhead if your minimum load is just a few percent of your maximum 25A load.
I only need high efficiency between approx 70% and 100% of full load. Transient response and ripple are not particulary important as this is for a battery charging application. Cost is naturally an important consideration so multi-phase may be too expensive but the reduced EMI filtering, if necessary, could justify it.
-
my experience in the past is that reducing switching times into the range of <10ns exxagerates parasitic ringing, and it becomes increasingly difficulty to compensate it with snubbers. As this is a synchronous design, make sure that your driver has a short, even better adjustable dead time. Maybe make the switching asymmetric with a diode: fast off, slow on. During the dead time current flows through the low side reverse diode. You may want to add a schottky in parallel to the transistor. Besides the transistors and the PCB layout, the inductor plays a big role in parasitic ringing, check the parasitic capacitance of your part (can be derived indirectly from self resonance spec).
I did think that sub 10ns switching was getting into difficult terroritry. I found this interesting article, 'Estimating MOSFET switching losses means higher performance buck converters'
http://www.eetimes.com/document.asp?doc_id=1225701 (http://www.eetimes.com/document.asp?doc_id=1225701)
Its dated 2002 so things will have moved on but it made me realise how much more complex switching actually is than I had appreciated and compared to the rather simplified model used, for example, in the Fairchild switching loss calculator I was using. The author makes some interesting assertions including:
Similarly like with the turn-off the speed with which the transistor travels through the active region will determine the amount of losses. Here however we do not have a full freedom to speed up this process as much as possible.
It is because the resistive voltage drop across the forward MOSFET and the layout inductance LSTRAY are the main factors limiting the speed of the reverse recovery process of the lower MOSFET's body diode. The optimal turn-on of the upper MOSFET is relatively slow and in most cases measures should be taken to avoid too speedy transition.
Which addresses my original question and he suggest 12ns to 15ns is a good starting point - or was in 2002 - I believe TI have released some MOSFETs in packages with very low parasitic inductance which might mean faster switching speeds can be achieved with less pain.
In order to avoid proliferation of parts with a variety 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.
Given the large spread in MOSFET characteristics (typical to maximums in the datasheets), I don't see how you could achieve this in practice unless you were prepared to select on test each product and even then changes with time and temperature would, I presume, force you to design in some deadtime for a reliable product.
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 benefit 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'y causes a substantial amount of time to commutate the current.
I admit I was assuming a Schottky would help - but at 150kHz or less that may still be the case. There's a lot more interesting material in the article but it will take me some time to get my head around it all.
To the FET voltage rating: most modern FETs are self-protecting and act like a zener diode above their voltage rating. You just need to make sure not to generate too much heat in them. Look for repetitive avalanche rating, or UIS (unclamped inductive switching).
That's interesting thank you - lots more reading to do!
-
For such high voltage transfer ratio (10:1), you have better luck with GaN devices or soft switching.
The easiest way: buy Vicor non-isolated three switch deep DCM soft switching modules. $40-ish each is not cheap, but that pretty much represents state of art for silicon devices.
I can't say I liked their website; I could only find PI3301-00-LGIZ and similar devices which achieve 91.5% and would require three of the 10A devices totalling $34.5. Are they what you were referring to?
With GaN devices at lower frequency (600kHz) and careful layout, 96% is not hard at all.
I got 96% simulated efficiency at 2.5MHz using my proposed layout (to be published in WiPDA 2016), and 84% efficiency at 10MHz.
Sounds impressive, but GaN MOSFETs are expensive and the bumped die package might be tricky to manage so probably not for me right now. But definately more research required on my part!
-
I'd like to achieve > 96% efficiency overall for a 24V to 2.4V, 25A synchronous buck converter operating at 150kHz. Using low resistance planar inductors such as Coilcraft SER2915 or SER2918, it looks like core and copper losses of < 1% can be achieved at 150kHz. Dropping to 100kHz reduces the switching losses but then the inductors get more expensive.
snip...
How do you arrive at your efficiency target of 96%?
If you achieve 96% and your load power is 2.4V x 25A = 60W
and your input power = 60/0.96 = 62.5W
You have 2.5W lost in the power supply and 60W lost in the load.
You might be better trying to reduce the load power consumption instead of trying to increase efficiency in the power supply.
If you size the inductor for 30% ripple current you need about 33mV across the sense resistor to a 10mV signal for current mode control. 33mV x 25A = 825mW. So you need DCR current sensing, where the resistance of the inductor is being used for sensing current.
The LTC3866 has sub-milliohm sensing capability.
Link: http://www.linear.com/product/LTC3866 (http://www.linear.com/product/LTC3866)
You do not want to over design the MOSFETs. You have to right size them. If they are over size them you are trading Gate Charge losses for conduction losses.
Using LTpowerCAD and the LTC3866 at 250kHz with BSC0504NSI at the top MOSFET and BSC011N03 as the bottom MOSFET and 0.66uH Inductor the efficiency approaches your target of 96%.
At 25A output every 1m Ohm of resistance between the power supply and the load is going to cost 625mW in dissipation or 1% efficiency.
Regards,
Jay_Diddy_B
-
GaN transistors go that fast, pretty easily.
Bigger problem is that your switching impedance is quite low (24V / 25A ~= 1 ohm), which will screw you on loop inductance, no matter how tight you build it.
The transistors will also have pretty serious Coss energy, which must be pushed around regardless of speed. This limits your switching loss pretty severely unless you can do resonant/ZVS. Which kind of stinks for a buck converter where inductor loss, size and cost is just as important.
Or you can think smarter, not bigger: use snubbers to harness the reactive energy, and direct it elsewhere (such as into a recovery rail, or 'stirred' back into the supply directly).
Distributing the current over several transistors, or using smaller channels, is the easiest way to go, and most conventional these days. And then you can put phase shifts between them, reducing input and output ripple as well, meaning you can save on those $$ polymer caps.
Tim
-
You might also look at a multi-phase design which would permit you to reduce the current per phase and reduce the switching frequency for a given inductor size given the help of the ripple / noise cancellation. Of course the more complex drivers will have an efficiency penalty but it may work out favorably at the higher current levels. Linear has some nice multi-phase controllers.
Cost is an important consideration so multi-phase may be too expensive but the reduced EMI filtering requirments could justify it.
Of course the problem with high efficiency is whether you mean the peak efficiency over the operating region or the worst case efficiency which might occur at light loads in which case 96% will perhaps be unrealistic due to operational overhead if your minimum load is just a few percent of your maximum 25A load.
I only need high efficiency between approx 70% and 100% of full load. Transient response and ripple are not particulary important as this is for a battery charging application. Cost is naturally an important consideration so multi-phase may be too expensive but the reduced EMI filtering, if necessary, could justify it.
Sounds like a single-phase design is a reasonable choice. 25A is kind of on the boundary where you could go either way on multi-phase designs. With a forgiving load, that probably tips the balance towards single-phase.
(I added the 2mohm sense resistor of the demo board into the simulator's inductor DCR). With 12V in, 1.5V, 15A out @ 400kHz the simulation gave an efficiency of 91.9% which is almost 4% higher than the 88% typical claimed for the demo board. That doesn't give me much confidence that I could reach 95% in reality but I could easily have made a mistake.
Did you add in the AC core losses? There's an entry box for them on the second tab of the calculator. LTpowerCAD can't calculate them itself as they are intimately dependent on the details of the particular inductor. Both Coilcraft and Wuerth have calculators on their websites, as do other manufacturers. At 150 kHz, AC losses aren't too bad (hence I didn't bother with them in my ballpark estimate), but in your comparison they are not negligible. A bit more screwing around last night got LTpowerCAD to spit out numbers in the 98% range after changing to a separate bias supply; 98% on paper might actually get you 96% in reality. Might.
You also have to decide what you want to trade off. Your suggested starting point is large inductors and low frequency, which is a very reasonable way to go if you can spare the space. I like LTC and TI power ICs, but they're not cheap (especially LTC); if cost is important they are not a great option. Et cetera.
At 25A output every 1m Ohm of resistance between the power supply and the load is going to cost 625mW in dissipation or 1% efficiency.
This is the sort of stuff that's going to make or break this design. Connectors, trace resistance, all of that -- it adds up at this power level, and LTpowerCAD or Webench won't help you here.
At the end of the day I think your goals are achievable without exotics or heavy R&D. You may go through a prototype or three but there's nothing pushing the limits of the state of the art here. All you need is a solid, top-notch design incorporating existing, well-documented stuff.
-
IMO, with step-down ratio this high (1:10), it becomes important to go multi-phase, especially thinking about the input caps. Wattage or output current alone is not as important - if you were doing a 24V to 12V converter, single phase would be just fine, even at high power or current.
Market is full of cheap integrated solutions for exactly this case of 1:10 buck converter, using multiphase; your problem is very close to a typical motherboard CPU supply design used for more than 10 years. They are always multiphase, because single phase is not a sensible choice.
This is not so much about efficiency, but size, cost, component lifetime and ripple. Multiphase may be better in all of these, including the cost.