1. Home
  2. Electronics Design
  3. The 12V level shifter saga

The 12V level shifter saga

Hi,

I am working on a high-current 4-switch buck-boost converter. The switches are psmnr70-40sshj, which -- with their 202nC gate charge -- makes the 12A gate driver MCP14A1202 a perfect spouse. The problem is with passing the control signals to the upper switches. I want to have true 100% duty cycle operation capability, which excludes AC coupling techniques such as a signal transformer. The high-side switches float at most 15V above the GND. Options:

  1. Use a 1.5kV digital isolator. There are units capable of 7ns propagation delay, which makes a total end-to-end delay of about 33ns. Not bad, but I feel it is heavily overengineered. In total that would make 4 gate driver ICs
  • 2 high-side isolators + 1 2-channel isolator to match the timing between the high- and low-side paths. 7 ICs or 6, is the delay equalizer is based on an RC delay + a Schmitt trigger.

  1. Similar to the above, but use a half-bridge driver such as the LM5101 just as a level translator. 6 ICs in total and the propagation delays are equalized out of the box. Disadvantage: the part is not stunningly fast with its own 25ns propagation delay. The fastest HB gate driver I know of, the LMG1210, has 10ns delay. OK, I tested the 5101 variant, works nice up to 2MHz, but has 58ns end-to-end propagation delay... well, meh.

  2. A *simple* solution based on discrete components. Total failure in practice, really simple common-base units have ~400ns delay; the best I could make was beyond the pain threshold complexity level and was capable of 78ns, not including the output stage.

OK, I have a working solution (2), can go from 25 to 16ns translator delay by replacing the LM5101 with MP18021 or switch to (1) and get

7..10ns, but I have a feeling that something is very wrong, because (3) should be the solution.

Why is making a fast level shifter so notoriously difficult? One interesting observation so far is that isolators are so much faster than their non-isolated counterparts.

Best regards, Piotr

Reply to
Piotr Wyderski
Loading thread data ...

Dunno if you really need 12A gate switching, that gives a full 10V gate drive in 10ns, way faster than you need.

--
 Thanks, 
    - Win
Reply to
Winfield Hill

202nC/12A=16.8ns. How did you get to a figure 68% better?

OK, going lower: the 1EDN7550 contains a level shifter, but its drive strenght is 4A/8A. The 8A sink looks fine, but 4A source is 50.5ns, i.e.

1% of the switching period. Scope confirms the math, this is basically the dV/dT as seen on the gate. This is a hard switching converted due to its (relative) simplicity, so the switching losses will be plenty of watts.

Best regards, Piotr

Reply to
Piotr Wyderski

I've been using discrete GaN fets (the EPC BGA parts) as mosfet and SiC drivers. They are cheap and crazy fast and easy to drive.

A bit of series gate inductance can help too sometimes. It does the classic thing, add a bit of delay but improves edge times.

--

John Larkin         Highland Technology, Inc 

The cork popped merrily, and Lord Peter rose to his feet.  
"Bunter", he said, "I give you a toast. The triumph of Instinct over Reason"
Reply to
jlarkin

Just how fast do you think you're going to be running this thing?

You at least need isolated drivers for the simple reason that you'll explode the bootstrapped type in one cycle. Assuming you're running those bastards anywhere near their ratings.

Tim

-- Seven Transistor Labs, LLC Electrical Engineering Consultation and Design Website:

formatting link

Reply to
Tim Williams

200kHz is the assumed frequency, but since the control is going to be digital anyway, I hear the mermaid singing "Variable Frequency Drive". At least in several steps. This is to handle the low load conditions with a a nice CCM. 50+25ns offered by the 4A floating driver looks disapointing.

Already prototyped, tested and eager to source power. This solves the

100% duty cycle as well, as pass-through mode should be quite a typical case.
50-80A in one direction, 10A in the other.

Best regards, Piotr

Reply to
Piotr Wyderski

Would these be any use to you:

Infineon 1EDN7550, 8A/4A, true differential input with 150V common mode range.

MK

Reply to
Michael Kellett

A PV optocoupler or a cheap dc/dc can furnish a small amount of isolated power to keep a bootstrap alive.

--

John Larkin         Highland Technology, Inc 

The cork popped merrily, and Lord Peter rose to his feet.  
"Bunter", he said, "I give you a toast. The triumph of Instinct over Reason"
Reply to
jlarkin

Yes, this was my starting point. Works as advertised, i.e. the 4A current looks a tad too low. The 8A rail is fine.

Best regards, Piotr

Reply to
Piotr Wyderski

A simple one-BJT booster fixes that.

Vcc | |/ .------| | |>. | .3 |

| | '---|

Reply to
dagmargoodboat

Reply to
Piotr Wyderski

Reply to
dagmargoodboat

I don't want the full allowed gate swing for efficiency and reliability reasons. Just the bare minimum, which is 0-10V typically.

Exactly, with a minor remark that the floating gate driver supply is already there, powering 6 SiC transistors. So adding two more windings to the transformer is not an issue. The (now rectifying) diode and the capacitor need to be there in the bootstrapped configuration as well, so half a meter of 0.2mm TIW wire it is all it costs.

I don't consider this exchange of ideas a game, as I can't come up with a satisfying winning criteria. My findings I wanted to share are:

Surprise #1: there are no (or I am unable to find, which is the same in practice) fast non-isolated level translators that can operate in the

30V VDD range. Just lots of (useless) glue logic translators with 6V VDD_MAX.

Surprise #2: If a HB MOSFET driver is used to do the translation, the existing isolated translators can be 2-3 times faster.

Surprise #3: making a discrete level translators out of a small number of cheap parts is very difficult at the required level of frequency and dV/dt. For me it had been a no-brainer till I had to make one. This adds a lot of respect to John's picosecond circuits and sheds some light on the baroque complexity thereof.

Best regards, Piotr

Reply to
Piotr Wyderski

Another highside driver trick, if you don't have outrageous swing, is to use a few resistors to turn a cheap LVDS line receiver into a high cmrr line receiver, to ride on the source of, and drive, the highside mosfet, or SiC, or whatever. You still need some isolated power, although it might work bootstrapped.

We pay 40 cents for DS90LV012. If someone changed the part number and sold it as a

Reply to
jlarkin

+10V at the gate doesn't work if, as you describe, you want the source to output +15V. You need 25V swing. +15v >---. .-+--------> Vout | V | - - - Q1 .-----. | drive >-------'

But if you floated the driver using +15V for GND and for Vcc, you'd still have to drive the gate to 0V in order to turn Q1 off. That's a mess.

So, the only choice is to run the driver from a floating supply referenced to Vout. And then you need a logic-level translator.

Which isn't that hard, but it's going to be pretty nigh unto impossible to satisfy your desire to do it with zero parts and zero cost.

If that's the case, why not make a floating bootstrap supply in lieu of the usual bootstrap cap -- thus meeting your d.c. operation spec. -- then find an integrated driver to supply the logic-level translation?

If you're dead-set on 12A drive but can't find it, you might consider paralleling drivers. That gets you lots of gate drive with no extra discretes, a fairly clean solution.

Cheers, James Arthur

Reply to
dagmargoodboat

From the very beginning I have wanted a floating driver, not the solution you refer to. E ~ C*V^2 and I have just too much charge to transfer even at the mere 10V already. Going to 25V would make the situation 5 times worse.

Yes, this has always been the way I wanted it.

I beg to disagree. Nanosecond-level of rise/fall time and delay below

20ns make a discrete level translator very difficult to design. The best I have done has 78ns propagation delay, which makes a description based on the geological time scale relevant.

Cost is a secondary issue, but part count is not, as it directly impacts reliability.

As far as I know, there are no integrated drivers able to work with sufficiently high common mode input voltage AND capable of 10A+ source/sink. To be specific, there is exactly one such a driver (UCC5390SC), but it is for IGBT and has UVLO of 12V, which is beyond the required 10V operation with an option to lower that, if performance is satisfactory.

Indeed, it is an option. But then it means a) doubling the number of drivers (8 ICs in place of 6) and b) propagation time spread: 4A + 4A doesn't mean 8A (simultaneously). Have a look at 1EDN7550, for instance:

38 to 55ns propagation delay range -- that is whopping 17ns, equal to the total switching time at 12A. Better to use a single power stage like the Microchip's 0902 or 1202 with no dispersion issues.

Best regards, Piotr

Reply to
Piotr Wyderski

Over the past few years I've designed and am still actively extending a line of pulse generators that produce kilovolt pulses with nanosecond rise and fall times from logic-level inputs. So I feel at least somewhat familiar with the issues attending high-speed level translation.

I'm confident your problem could be solved with discretes. It would take a bit of doing. But what I don't know is how to do it more simply than the single BJT booster you've already rejected for being too complex.

Cheers, James Arthur

Reply to
dagmargoodboat

I bought some GaN FETs for that high-side driver I was toying with a few years ago, but I've still not ever powered them up. The BGA looked intimidating at first, but I've since hand-reworked some chip scale / die packaged parts ... (that was a huge pain, but after many tries it did finally work).

It's truly amazing that our craft is still advancing so rapidly on so many fronts, so many years on.

Cheers, James Arthur

Reply to
dagmargoodboat

The EPC parts are hard to handle but otherwise wonderful. They need just a few volts of gate drive, switch super hard and fast, and have near zero Cd-g. Incredibly, they don't offer eval boards, so we had to make our own just to connect to them.

Electronics as such is over 100 years old, and we get a continuous and nicely paced series of new toys and tools.

But where are my diamond transistors?

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com
Reply to
John Larkin

A couple of years ago, I was the defence's expert on the patent infringement side of the Waymo v. Uber case ($2.3B ask). In the process I got to take apart and deliberately blow up some of Uber's best lidar TX sections.

Like most such, they use a GaN FET to dump a few-nanofarad cap into an Osram stacked laser. Doing some modelling based on my measurements, the total inductance of that discharge loop was less than 400 picohenries, about half in the drain circuit and half in the source. Not bad a-tall.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC / Hobbs ElectroOptics 
Optics, Electro-optics, Photonics, Analog Electronics 
Briarcliff Manor NY 10510 

http://electrooptical.net 
http://hobbs-eo.com
Reply to
Phil Hobbs

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.