MOSFET selection for boost converter problem

I've been working on a power supply for antique Teletype selector magnets. These need 120VDC 60mA for the first

2ms or so of each bit time, but the sustain voltage needed is only 3-4V. So I've built an isolated boost supply and interface board, all powered from a USB port. The whole project is on Github:

formatting link

where there's an explanation of how it works which doesn't require downloading the whole KiCAD project.

Schematic as image:

formatting link

LTSpice model:

formatting link

I've had a board made, and it produced 120VDC briefly, until the IRLB8721 MOSFET blew out. I just used that because it's a common MOSFET. Not surprising - the circuit shows 40V across source and drain, and that's a 30V part.

So what to put in there? I've been trying various parts in LTSpice. The AP9465GEM works great in LTSpice, but it's a surface mount part and only 40V, so there's no headroom on voltage. Almost nothing with a rating of 50V or better works in simulation. Turn-off is too slow, or capacitance is too high, and the resulting output voltage peak on the primary is too low. Try loading up the LTSpice model and using different MOSFETs to see this.

I don't really know what I'm doing here. I'm aware that this MOSFET application depends strongly on the transient properties of the device. Turning off that transformer primary to get the inductive kick is tricky. I'm trying to struggle through the Art of Electronics chapter on MOSFETs, while hammering on the problem with LTSpice.

(I'd like to use a TO-220 form factor, or at least a through-hole part, and get this first board working. After that, I can get a new board made.)

(I was asking previously about low duty cycle

555 timer circuits. That part is working fine now. Thanks.)

John Nagle

Reply to
John Nagle
Loading thread data ...

One thing I notice is that it looks like you're using a bipolar 555 from a 5 volt supply to drive the '8721, which has a max Vth of ~2.4 volts. However the bipolar 555's output high voltage is going to be quite a bit lower than the supply voltage - I've seen it as much as 2 volts lower than Vcc in practice. And maybe I'm reading the schematic wrong, but it looks like you've got a 680 ohm resistor in the 555's power line. That's not going to help things.

Unfortunately I think that using a 555 as a gate driver from a 5v supply is a poor choice here. Needs to be driven harder.

Reply to
bitrex

Also I think 100kHz is a pretty high frequency to run a standard 555 at and expect it to source/sink much current from a load during the switching cycles (which needs to be done.) The output BJTs don't turn on and off too quick.

Reply to
bitrex

There's no flyback clamping or snubbing on that fet. The drain probably spikes up a hundred volts or more. Your sim has no leakage inductance, so doesn't model the drain spikes.

Use a higher-voltage fet and add some sort of flyback absorber, an R-C or a zener or something.

And isn't the fet switching, basically, into a shorted load most of the time?

Why all that isolation? Aren't the keyboard and magnet isolated already?

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

Looks like power to the flyback is coming from the TPS2030 high-side switch, OP says he needs 120V @ 60 mA for a few mS, but from the '2030 data sheet it says that the switch current-limits at 300mA. Could be a problem

Reply to
bitrex

Seems unnecessary, could just use a gate driver IC with an enable pin.

Reply to
bitrex

It all looks very complex to me. More stuff to go wrong.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

The Iron Rule of the 555: It's a very useful IC, which will always be the wrong choice for whatever application you have in mind.

Reply to
bitrex

Well, given that an MC34063 is 18 cents and an LM555 is 10, the 555 is wrong for this one, sure. But CMOS 555s are useful sometimes, e.g. for chopping LEDs. I use them all the time in protos, though not recently in a finished design.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

used that because it's a common MOSFET. Not surprising - the circuit shows

40V across source and drain, and that's a 30V part.

but it's a surface mount part and only 40V, so there's no headroom on voltage. Almost nothing with a rating of 50V or better works in simulation. Turn-off is too slow, or capacitance is too high, and the resulting output voltage peak on the primary is too low. Try loading up the LTSpice model and using different MOSFETs to see this.

I made some suggestions in your previous post, so I am familiar with what you want to do. So, where to begin?

As a reference, here is the LTSpice simulation you provided:

formatting link

Some issues: (1) D1, D2, and D3 have no model defined (probably not a problem) (2) Your source is 4.6V with 5 ohms resistance - that severely limits performance (3) As mentioned, no leakage inductance for L1-L2 is specified. (4) Also as mentioned, 100 kHz is very high, and the 555 is not a very good gate driver under best circumstances (5) The AP9465GEM is not commonly available. (6) The IRLB8721 is only a 30V part, and it is really characterized for gate drive of at least 3 volts and preferably 4.5V. Your power supply is already crippled with 4.6 volts, and the NE555 has a typical output high voltage of only 3.3V at 100mA. Even at 1 mA the output will be 1 volt below the rail, so just barely 3.6 volts with a 4.6 volt supply. (7) I don't see any bypass capacitors on your actual schematic on the nominal 5V supply. (8) You have a TPS2030 USB switch for the enable. This part may be obsolete, but probably not a problem. (9) There is U5 that looks like "IEQ524S" but I can't find it. Looks like an interface to the optos.

Enough for now.

Paul

Reply to
P E Schoen

Others have pointed out potential problems, including the lack of a snubber. In the interest of getting this going on the bench fast put a zener-based snubber in there, like here:

formatting link

A 7-12V zener to your 5V supply, fast Schottky diode in series, from the drain. Use a big zener of at least 1/2 watt if you have one. That should prevent it from blowing up. Of course, a 555 is too wimpy to drive this FET but this should essentially get the circuit going. I never use 555 but maybe you can find a CMOS version that will at least swing all the way to 5V? Longterm I'd get rid of the 555 as was already suggested.

We were all there once and many of us have the scars to prove it :-)

When you set the coupling coefficient in SPICE to, say, 0.98 you'll get the leakage inductance. Or just put a small non-couple inductance in series with the primary. The sim will now slow way down but you'll see the ugly stuff brewing at the drain node. Lightning, peals of thunder, some smoke.

[...]
--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

I found a major error in your simulation. The Coilcraft DA3022 has a ratio of 10:1, so with a primary inductance of 10 uH, the secondary is 100 times that, or 1000 uH. I also replaced your MOSFET with a commonly available TO-220 logic level device rated 100V and 15A IRL530NS_L

formatting link

So here is the simulation:

formatting link

Note that the peak drain voltage on the MOSFET is less than 20V, and the output now easily reaches 120V. But something else that may be a problem is that the voltage on the selector coil drops to -120V when turned off. This might be corrected by using a standard diode in place of D7 (but this will cause current to flow for a longer time). Other methods as discussed in the previous thread may point to a better solution.

Thru-hole components are becoming scarce for some devices, especially newer ones. But it is really quite easy to solder larger packages such as the power flat pack, and even SOIC-8 and SOT parts are not bad, even on the usual proto-boards. Don't let SMT keep you from choosing better components.

Paul

Reply to
P E Schoen

Why not dig up a '70s edition of the ARRL Radio Amateur's Handbook, and see what they did for an interface?

--
Never piss off an Engineer! 

They don't get mad. 
 Click to see the full signature
Reply to
Michael A. Terrell

555 reasons to not use this anymore ;-)

Good point, that's why first initial chops on an empty cap load should be very short.......

The snubber alone will not be enough, the selection of the transformer is OK. What is missing is gate drive and THE-control.

When MOSFETs had been poor some spikes above the max drain voltage instantly killed them - like the old BUK Philips ones, what a crap this was....

You can sense and regulate by the flyback pulse the output and limit thereby also the flyback pulse down to non dangerous amplitudes. Used this method in my TL594 controlled analogue ballasts many years ago.

Now even available in one chip from LT:

formatting link

Regards

Joerg Ni

--
Joerg Niggemeyer on RISCOS Raspberry Pi2 

http://www.led-temperature-protection.com 
 Click to see the full signature
Reply to
Joerg Niggemeyer

I tried that. Then I put in a tiny resistor (0.01 ohm) between the output of the 555 timer and the gate of the MOSFET so I could get LTSpice to measure current. That shows a peak current of 350mA for about 200ns. The 555 timer can output a maximum of 200mA. I'm using the "idealized 555 timer" model, and it can apparently deliver more current with no voltage drop.

In the real world, the output from the 555 timer drops to about 2V and stays there for the entire ON portion of the cycle. I'm not sure how the gate of a MOSFET behaves when you underdrive it. I thought the high current draw was just during turn-on. Is the problem that, without enough drive, the gate current remains high? Or is something else wrong?

Here's the scope trace again:

formatting link

Incidentally, the other waveform

formatting link

shows data being sent through the board at 45 baud. Functionally, roughly the right stuff is happening.

Someone suggested using an MC34063 instead of a

555 timer, and I'll have to look at that option. But it's a lower switching frequency. More later.

John Nagle

Reply to
John Nagle

That's a worst case test. USB guarantees only 4.5V, although it's nominally 5V.

Now using the IRL530, which is a 100V part. The 555 timer doesn't have enough output power to drive it properly. Here's the scope trace at the gate of the IRL530.

formatting link

It's only reaching 2V. The overall result is that the circuit runs but the final output is only 45V instead of 120V.

formatting link

So the big problem is getting enough drive for the power MOSFET's gate.

Good point. I'm relying on the 220uF filter cap too much.

It's dated, but available in DIP.

IE0524S.

formatting link

That's a 5VDC to 24VDC converter used to provide voltage for the keyboard side of the system. That's independent of all the custom switching stuff.

Thanks.

John Nagle

Reply to
John Nagle

the cycle. I'm not sure how the gate of a MOSFET behaves when you underdrive it. I thought the high current draw was just during turn-on. Is the problem that, without enough drive, the gate current remains high? Or is something else wrong?

I could try to analyze what's going on with your most recent efforts, but I found more problems that are more serious. First off, it's not going to work with a 5 ohm power source resistance. A USB will supply close to 5VDC (maybe as low as 4.8) and it will do so until it approaches the standard 500 mA limit, at which it may throttle back or shut down completely. So I used your

4.6V with a 0.5 ohm resistance. That will still mean a drop to about 4.3 volts at 500 mA, but still workable.

BTW, you can read current without adding a resistor. Hover the cursor on the pin (such as the gate of M2). It will show an icon for a black clamp-on meter and in this case it will measure Ig(M2). You may also choose it from a list of available data by right-clicking on the plot, choosing "Add Trace (Ctrl-A)", and a list will appear. You may also manually define what you want to plot, such as a voltage ratio, or voltage times current (power).

Next, I changed the input diode to a Schottky, which drops only about 0.3 volts rather than almost 1 volt. That was a huge improvement. I also found a better MOSFET, the IRL3915. It seems that it is actually an IRLR3915. There are probably even better ones available, but this is only a dollar and is easily available:

formatting link

It is 55V, 30A, with 14 mOhm RdsOn, and will work with as little as 2V on the gate. I also cleaned up your simulation, getting rid of parts not needed, and added a commutating snubber that totally eliminates negative spikes and ringing. I dropped the frequency to about 40 kHz. Here is the result:

formatting link

Note that this seems to draw only about 300 mA from the supply, or about 1.6 watts. Fine for a USB source.

Here is the netlist:

formatting link

Hope this helps.

Paul

Reply to
P E Schoen

I thought you might like to se the details of the waveform, particularly the gate current Ig(M2), drain current Id(M2), and the drain voltage Vd, actually V(n004).

formatting link

Paul

Reply to
P E Schoen

What is the FET P/N that you are measuring?

Reply to
John S

If I'm reading the schematic correctly it looks like you're feeding the flyback from the '2030 high-side switch, which the datasheet has a continuous current limit of 300mA.

Looking at the waveforms here:

formatting link

it looks like the mean MOSFET drain current may be over that.

Reply to
bitrex

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.