SMPS - Why are my FETs cooking?

where's the power inductor?

Dear friends,

Once again an inductorless schematic is presented and the circuit doesn't work. There seems to be no end to these no-inductor circuits. An SMPS operates on the following fundamental principal. A power transistor (usually MOSFET) is turned on and current is delivered from the input supply to an inductor, storing energy in the inductor's magnetic field. Also, in forward topologies, the output cap is being charged and the output is being supplied while the FET is on. When the FET turns off, its energy is transferred to the output cap and load. As the input voltage varies, the PWM controller adjusts the duty cycle to keep the output voltage constant. At light loads, pulses are skipped to prevent overcharging the cap. Without an inductor it is hard to maintain constant output voltage over varying input voltage conditions as well as varying output current conditions. The inductor also filters the switching pulses very well. For some reason a lot of folks don't believe in using an inductor, and I just don't understand it. Best regards.

so he needs to add an inductor for the circuit to work??

actually, two inductors

In order to make it work with good tight regulation, lowest noise, and highest efficiency, yes, inductors are needed. Two are needed in this case since there are two outputs, plus and minus. To improve cross-regulation, the two inductors should be magnetically coupled. Texas Instruments has good application notes regarding this. Also, the IRF540 MOSFETs are 100 volt devices. These are much more than required in terms of voltage. The automotive system voltage (12-14 volts) limits are 9 to 16 volts. A push-pull circuit places twice the voltage on each MOSFET, or 32 volts max. A 40 volt rated part provides some safety margin. If one wants bigger margins a 55 volt part is more than enough. The 100 volt IRF540 has much higher on resistance than a 55 volt part, the IRF1010, for example. THe IRF1010 is pin compatible, and should drop right in. The IRF1010 should run much cooler. Heat sinks should be used. Best regards.

Bosium said:

I have breadboarded a smps

I tried increasing dead-time by increasing the timing cap like it says in the app notes of the sg3524, but still nothing. Have I wound the transformer wrong?
Gareth

Click to expand...

In addition to the comments which have been made already:

First, breadboarding SMPS isn't a good idea. You've got to lay down a board -- i believe that in the case of the LM3524 National Semi provides you with a PCB template. There are all sorts of nasties running around and you have to pay close attention to layout. www.expresspcb.com has just reduced the price of their miniboards and improved their software, btw.

Dead Time instead try decreasing the duty cycle -- the Nat Semi apnote tells you how to do it. after all, the chip does have a voltage reference and an error amp on it -- this is what they are for.

the push-pull transformer supply, one of the 5 supply designs which Claude has referenced in the past, spreadsheet can be found at
http://www.tech-diy.com/smps_xfmr.xls
this should give you an idea of the peak currents, etc. the equations were cobbled together from Linear Tech's application notes.

jack

Your core is probably saturating.......You can't just take any core and use in these things .........

core saturation would be my guess as well

I would agree that core saturation is very likely the culprit. With a push-pull converter, the transformer core is being subjected to a volt-second (flux) ramp in both directions. If the two FETs possess unequal on resistance, or slightly differing pulse widths due to delays in the PWM controller, the volt-seconds in one direction will be unequal to that in the other direction. The core magnetic flux will creep positively or negatively into saturation. One method to prevent this is to use a series blocking cap. Better yet, a current-mode controller would sense and limit both volt-seconds and amperes, preventing saturation. Unfortunately, the LM3524 is a voltage-mode controller. Also, if the core is from a power inductor, designed to carry dc, it more than likely has a gapped core. This is not desirable in a transformer because the magnetizing current increases as a result, increasing conduction losses. Also, power is lost due to the gap. The right core geometry should be used (low leakage inductance), and the right material should be selected (ferrite, or at 50 kHz, a low frequency for an SMPS, maybe tape-wound nickel-iron is suitable), and no gap should be used (other than the small incidental gap which is inevitable when mating two halves of core material). The windings should be constructed for minimum leakage inductance, and low skin effect. Also, the full bridge rectifier is not used in the push-pull, but rather, the full wave center-tapped, followed by a power inductor, followed by the filter cap, for each output. App notes cover this. I hope this helps. Best regards.

the transformer

remember to orient the beginnings and endings of the transformer correctly.

I drew in the error amp circuitry. if you make the drawn in resistor adjustable you can adjust the duty cycle of the LM3524.

I guess I have the inductors where Claude wants them.

3300uF is probably overboard -- I suppose that since you haven't used the "regulator"portion of the LM3524 there may be a reason for this. at any rate, such a high capacitance value just swamps the ability of the error amp to resond to changes in line voltage, etc.

Without the whole circuit, what are you doing to simulate the feedback? When fully operational you should not see anything near a squarewave with a light load. Just narrow pulses and some ringing............Looking at the the drains

i should have specified that the resistive voltage divider going from V+ to the LM3524's error pin should provide enough voltage for the duty cycle you want. I use these chips for driving a bigger H-Bridge for DC motors so use an adjustable pot in this position.

if you have a scope you should also make sure that the FET's are really turning off

My impression is to wind them in opposite directions so that the magnetic flux in the core does not try to cancel from each polarity. I am not sure if if that is right though. Or just experiment to see what works better.

I have successfully designed SMPS for car audio both with AND without output inductors. They're not necessary expenses for this "low power" application. I've found that IRF540s are not good parts for SMPS switchers as their Rds(on) is too high. An IRFZ44 or 48V or N is great for this app. Even an IRFZ34 is OK, but the 48s are pretty inexpensive these days. Also, the speedup transistors are just NOT necessary for two phases of two MosFets, even for fast switching. Just take them out and add 75-100ohm gate resistors. For more than two per phase, use the speed-up circuit...but it's just complicating things at this point. If you are concerned with shoot-through, move up to a SG3525 regulator which includes shutdown logic to prevent shoot through upon failure of the chip/oscillator.

For your overheating problem - I'll bet your core is saturating. OR the windings are incorrectly wound ie: out of phase. OR, the FETs are entering a linear mode (ie: oscillation). Usually this is due to stray capacitance - put those gate resistors close to the Gate pins and keep your traces to and from the FETs reasonably large and short. You can even pull the strands off the toroid and connect within mm's of the FET - no long traces!


Hitsware: feedback is not necessary in this free running SMPS. Check out a few hundred car amps over the past ten years and get back to me on that. Most have NO feedback comp, though I disagree with that method.

EnvisionAudio said:

I have successfully designed SMPS for car audio both with AND without output inductors. They're not necessary expenses for this "low power" application.

Hitsware: feedback is not necessary in this free running SMPS. Check out a few hundred car amps over the past ten years and get back to me on that. Most have NO feedback comp, though I disagree with that method.

Click to expand...

Nope, you need the inductors. Perhaps you should take a look at the output on a scope. There are spikes generated because of the inductance leakage in the transformer in a push-pull design. The equation is:
L=0.5Vo / (I*f)

Feedback -- I showed voltage feedback to the error amplifier -- you should remember that in an automotive environment you can have some significant swings -- obviously a good amplifier design will accomodate variation in the supply voltage, but since this design calls for more than 4X stepup through the transformer, any variation in primary voltage is multiplied by the same factor. You can damp the resonse of the error amplifier.

There is a current limit function on the LM3524 -- not the "Current Mode" which Claude referred to, but if it's there, why not use it.?

Here's a graph snipped from Pressman's book ("Switched Power Supply Design") on the inductance leakage spikes:

The core used in coupled inductors in AT and ATX power supplies are made of iron powder

They have very low permeability and are intended to make power-storage inductors to be used in buck or boost regulators [or in the secondary side of the transformer if you want regulated output]

These cores are not suitable to be used as transformers in simple magnetic coupling circuits, for that purpose you must use a ferrite core

MOSFETs heat a lot simply because they are conducting lots of current due to low core inductance but very little power is drawn from the 12V supply as current is cycled in both directions from/to core and supply capacitors

Simple magnetic-copupling SMPS need no inductors in the secondary side, both sides are simply coupled like primary and secondary in a 120V or 230V AC transformer with its output rectified

Altough inductors are necesary only if supply has to be regulated, a pi filter is recommended to filter the ripple caused by diode switching and capacitor discharge during deadtime on the output

The core is saturating. If you are using a iron powder core for the transformer, you are definitely using the wrong material. When you design the transformer follow faraday's law. You must know the DC input voltage that you are switching, the operating frequency, the cross sectional area of the core, and from that you will calculate the number of turns and choose the operating flux. At 50kHz operating frequency and some sort of push-pull topology, you should design the operating flux to be around 800-1000Gauss so that the core losses are minimized. It is also possible even with a soundly designed core to saturate it over time. By using a topology that traverses the core's B-H curve in the first and third quadrants, the core can "walk" up the curve if each half cycle is not an equal and opposite Volt-second product. You can get in touch with me if you want help in designing the transformer. I also have some ferrite toroids that I could sell to parties interested in the USA.

BeanZ

BeanZ said:

The core is saturating. If you are using a iron powder core for the transformer, you are definitely using the wrong material.

Click to expand...

OK, I am going to assume that we are going to use a proper core -- the Amidon website can help here --

I put a spreadsheet on my website for calculating the turns ratio, inductance of the primary, making assumptions such as the input voltage, output voltage, current, permissible ripple. It does work, by the way!

the physics can be found on Linear Tech's website for the LT1683,

http://www.tech-diy.com/smps_xfmr.xls

>Hitsware: feedback is not necessary in this free running SMPS. Check out a few hundred car amps over the past ten years and get back to me on that. Most have NO feedback comp, though I disagree with that method.

I didn't follow the circuit close enough to tell if it required the loop to be closed or not. I know switchers can be configured without regulation, BUT ....... Most of the car amps I've seen do use it. I guess I've looked at a differant 100 than you have