More buck converter design

A project log for GPS Disciplined xCXO

A DIY GPS disciplined 10 MHz reference clock

nick-sayerNick Sayer 05/03/2016 at 16:326 Comments

When we last left off with the 15V buck converter, I was designing an external-switch MC34063 based version. It turns out there are a lot of details that all have to be juggled at once. I haven't gotten the boards back yet, but I think I now have something that's more likely to work than not.

First, I noticed that the existing MOSFET (SI3443CDV) has a 20V absolute max Vds. Well, that's going to be a problem if we feed it with a 24 volt supply, unfortunately. This also winds up impacting all of the FE-5680A breakout boards I've sold so far. Oops. So they should probably be operated with 18 volt power supplies only. Sorry about that.

Well, I've already picked the SQ1431EH to replace it. It has a Vds of 30 volts, so no problem there. It turns out, though, that it's just as well I'm contemplating replacing the LM3485. It's gate output terminal only swings between Vin and Vin-5V. The SQ1431EH wants a gate swing of more like 10 volts (or at least close to that). It's Vgs absolute max is 20 volts.

Well, we can put a turn-off resistor between the gate and source, and a turn-on resistor between the gate and the switch output of the MC34063 (one other detail is that you want to wire the driver and switch transistors in the 34063 as the Darlington configuration, meaning you tie DVC and SWC together and use those as the switch output, and tie SWE to ground. The DVC+SWC pair are effectively an open collector). With those two resistors acting as a voltage divider, we can insure that the gate voltage at turn-on is low enough, but not too low.

But now we face another issue - we need to balance the current through the two resistors against switching speed. Naively you could try using 10k resistors for both, but if you do that, then the switching speed will be way, way too low. The majority of the MOSFET's power dissipation occurs not when it's switched on or switched off, but during the periods when it's moving from one state to another. The faster that happens, the less power the MOSFET has to dissipate. But the opposite problem hits the resistors. The lower the resistor values, the more current flows through them when the switch is on. Not only does that stress the resistors, but it's also wasted energy, which takes away efficiency.

I think the sweet spot is something around 330 Ω for each. At a Vin of 24 volts, that's just under a half watt (and you can get 1/2W 0805 parts) on each one (assuming that there's no voltage drop across the 34063's transistor pair, which won't be the case). At 17 volts, there's still likely to be 8+ volts of Vgs to turn the MOSFET on. If you look at the datasheet, at 4 amps of drain current, there isn't a great deal of difference between a Vgs of 8 volts as opposed to 10. In principle, another way to go would be to add a 10V zener diode across the turn-off resistor and use a 220 Ω turn-on resistor. That's 890 mW, but it won't be a 100% duty cycle, so you could figure that the average dissipation will be lower - particularly because while the supply needs to supply 2A at startup, the steady-state consumption is just over a quarter that, meaning the switching duty cycle steady-state should be quite low.


Nick Sayer wrote 05/03/2016 at 19:07 point

Hmm The TPS54231 looks attractive. It's the 2A version of the 54331 you mentioned. I think we can get away with running nearer the maximum spec because the 2A is just a startup load. The FE-5680A only pulls that briefly before it's warmed up and then drops down to less than an amp steady-state.

I'll have to investigate it. I may try and design a new 5680A breakout board as a test platform.

  Are you sure? yes | no

K.C. Lee wrote 05/03/2016 at 18:08 point

Looks like SQ1431EH is a good part with low gate charge.

Q2, D2, R4 can be used as an active pull up to open drain (Q1) and offers much faster turn off speeds without the power dissipation of a turn off resistor.

Have you consider using more modern regulators with built-in switches?  They are easier to use and much more efficient.

  Are you sure? yes | no

Nick Sayer wrote 05/03/2016 at 18:14 point

The one downside to that is that it appears likely that it will violate Vgs (max 20V) at Vin=24V. Naively, I'd fix that with a zener across the gate, but that Zener will rather quickly lose the magic smoke, me thinks.

  Are you sure? yes | no

K.C. Lee wrote 05/03/2016 at 18:29 point

Much of this is academic as there are plenty of decent regulator chips out there with built-in switches that would solve the problem.  The active pull up circuit is a voltage follower, so it is up to you to change your pull down to limit Vgs.

Just inserting a Zener in series with the Q1 collector would have done the trick. or level shift the pull down transistor.  

  Are you sure? yes | no

Nick Sayer wrote 05/03/2016 at 18:25 point

I keep floating through DigiKey periodically looking for better switch controllers, but all of them have much higher external parts counts, and the trend is for two MOSFETs (the second one replacing the Schottky diode in the buck architecture). I haven't gone through the switching regulators yet (that is, switching controllers with embedded switching), but I fear that asking for 30W @ 15V is a tall order. On the other hand, the good news is that the input spec for the FE-5680A is only 100 mV P-P of noise and ripple on the +15 rail. That's not nearly as high a bar as the OH300 was (3W @ 3.3v with something like 5 mV P-P).

  Are you sure? yes | no

K.C. Lee wrote 05/03/2016 at 18:45 point

Look for one with integrated switch.  2A isn't that much of a current.

SC4525ESETRT  28V Max input, 3A output
TPS54331DR 28V input, 3A output

some of TI/National easy switcher series would do the job too.

I can get cheap MP2307 modules from China, but they might not have enough margin for you as the Vin is 26V absolute max.

  Are you sure? yes | no