BJT vs MOSFET for linear power supply output

[little_carlos]

Hey guys, i was wondering, wich is better for output control of a linear power supply? i made a lm317 simple 1.5-24v 3 amps psu, the lm317 is meant only for voltage reference, the actual power is driven by two 2n3055, its been working great for 3 years but its kinda big, and i want it to take less space, so i thought on changing the npn for mosfets, like irf540, will i have the same results or should i stay with the npn's?
btw dont ask me to change over switch mode psu's i already have one, i want to know how effective mosfets are for linear aplications
thanks for your time

[moya034]

The MOSFET will have to dissipate just as much as heat as the BJT to produce the same results in a linear supply, so you wont really be saving any space at all. I'd just leave it like you have it now.

[Kleinstein]

MOSFETs in linear applications are a little tricky. The SOA limit for modern the MOSFETs usually does not permit high power at more than about 10 V drop. Often it is not eve specified for DC analog operation. There are a few, mainly older MOSFETs that can work, but it allready gets difficult to have 2 in parallel. So for a powersupply for more than  about 10 V, I would stick to BJT.

There are also BJTs in smaller cases just like the MOSFETs. Heat disipation is essentially the same.

[fubar.gr]

I made the same question some time ago:

https://www.eevblog.com/forum/beginners/7805-with-mosfet-as-pass-element/

One of the main problems is that transistors get fully switched on with just 0.6-0.7 volts. A mosfet, even a logic level one, needs at the very least 3 volts.

That increases dropout voltage and therefore makes the power supply more inefficient

[Kleinstein]

It depends on the type of circuit, wether the higher control voltage is a problem. Some circuits use a seperate supply (or just an extra cap) for the driver.

The real problems with MOSFETs are:
- generally poor SOA at voltage above about 30 V
- special types with tested SOA are expensive
- difficult to parallel (need higher resistance or extra circuit)
- nonlinear control and slow at low current -> so higher minimum load or rather slow

The only real advantage I see is that there is essentiall no DC gate current and thus current can be measured at drain or source.  Still the driver needs to be capable to deliver quite some current to make is resonable fast.

[T3sl4co1l]

BJTs are more traditional for linear applications under 200V.  The low and consistent voltage drop is perhaps the best reason for this.

Above 200V, it is hard to make BJTs for linear operation (minimizing 2nd breakdown), and MOSFETs are preferable.

As mentioned, in both cases, a type made for linear operation is required.  Devices optimized for switching operation (whether BJT or FET) exhibit 2nd breakdown and are generally unsuitable.  If it's not announced in the datasheet's front page text, check the SOA (safe operating area) graph: if there is no DC curve, or it doesn't extend far enough to the right (i.e., there's a steep section cutting into power dissipation at voltages near the rating), it's probably not suitable.

(If your application is within the DC SOA anyway, it's probably fine, even if the SOA isn't particularly good all its own, and the device isn't advertised for linear operation.)

Tim

[bktemp]

As mentioned, in both cases, a type made for linear operation is required.  Devices optimized for switching operation (whether BJT or FET) exhibit 2nd breakdown and are generally unsuitable.  If it's not announced in the datasheet's front page text, check the SOA (safe operating area) graph: if there is no DC curve, or it doesn't extend far enough to the right (i.e., there's a steep section cutting into power dissipation at voltages near the rating), it's probably not suitable.

Some additions, because there can be several traps:
When looking for the SOA in datasheets, look for the forward bias safe operating area not the reverse bias or turn-off or switching SOA, because they are only valid for switching applications.
If the DC curve in the datasheet looks like a straight line, showing no derating at high voltages, be careful, because many manufacturers do not measure the forward SOA for switching transistors and simply plot the (deceptive) curves based only on the maximum power dissipation (many Infineon IGBT datasheets have this type of SOA curve).

Even if you use a preregulator, you still have to look for the peak power dissipation when the output voltage is set to the maximum value and the output gets shorted. Then the transistor sees a full voltage and a high current until the current limit kicks in, limiting the current and the preregulator is reducing the input voltage and all the capacitors in front of the transistors are beeing discharged. This is very important for a reliable power supply.

[T3sl4co1l]

Yeah, IGBTs are right out -- they're made to harness the best of both BJTs and MOSFETs, specifically for switching applications.  The current density is massive, many times higher than MOSFETs'.  Which means the power density (at a given voltage drop) is greater by the same amount, which exacerbates the development of hot spots on the die that much more!

Of the few IGBTs I've seen with fully specified SOAs (I forget if they included DC, but several pulse widths anyway, e.g., 1us, 10us, 100us, 500us..), the limits drop very sharply with time.  Which means, it's *very* sensitive to developing hot spots, and subsequent 2nd breakdown.

(Funny, I think vacuum tubes have similar limits too, though for very different reasons.  Cathode current is limited due to wear -- something about space charge, electric field at the surface, interface chemistry, and who knows what other poorly-understood things.  Plate voltage is limited statically due to arcing around the glass or between electrodes, but dynamically also, due to ion bombardment (reduces cathode life even further), sputtering, generation of x-rays and so on.  There are only three types of vacuum tubes designed for high voltages: those with low temperature cathodes, made for pulsed operation (sweep tubes, radar modulators), where the low duty cycle gives acceptable life; those with low temperature cathodes, including ion traps (such as some CRTs) and low currents, to deal with wear; and transmitter types with more robust, high temperature cathodes, intended for quiescent operation at 30kV and beyond, but usually not for high current pulsed applications, or very high frequencies, which are usually served by the first class.  There are also gas-filled e.g. thyratrons, but the higher gas pressure actually reduces cathode wear, allowing much greater current flow -- at lower voltage drop, which is key.  But you also can't use it as a linear amplifier.)

Tim

[bktemp]

Here is an interesting application note about linear operation of IGBTs (it is also applicable to mosfets):
http://media.digikey.com/Resources/Microsemi/microsemi-make-linear-mode-work.pdf
It shows some destructive test results using 600V 158A 682W IGBTs. At 500V they fail at only 110W. So the useable FBSOA is limited to around 50W at 500V.
You can use almost any switching transistor for linear operation, but for reliable operation you have to do a lot of destructive testing to measure the FBSOA curve if the manufacturer did not measure the curve and the results may be disappointing.

[bktemp]

Even if you use a preregulator, you still have to look for the peak power dissipation when the output voltage is set to the maximum value and the output gets shorted. Then the transistor sees a full voltage and a high current until the current limit kicks in, limiting the current and the preregulator is reducing the input voltage and all the capacitors in front of the transistors are beeing discharged. This is very important for a reliable power supply.

Thats a given. Any modern SMPS controller will have some means of detecting a short either through direct current sensing or detecting the collapse of the output voltage. On detecting a short they ussually either enter burst-mode or latch-off. Adding OCP wouldn't be overly difficult for that matter. You should for a bench PSU be monitoring the temp of the pass element anyway.

The switching preregulator is a different story. The point I wanted to emphasis is the linear post regulator. Let's assume the power supply can deliver 50V at 5A. The preregulator is designed to generate a
3V difference, so it will supply 53V to the post regulator. The post regulator transistor has to handle a power dissipation of only 3V*5A=15W. If a short at the output happens, the current limit in the post regulator kicks in. The post regulator transistor now has to dissipate 53V*5A=265W and even more if the current regulation is a bit slow. The preregulator will quickly ramp down, but you often have capacitors with several 1000uF after the switching regulator. This takes some time and the transistor has to withstand the high power dissipation.

[Kleinstein]

Having so many MOSFETs available does not really help - somthing like 99% of them are not suitable for llinear operation at more than about 20 V. For MOSFETs specified for high power linear operation are rare and relatively expensive.

The warning about false SAO diagramms is also needed - some just show the maximum power and don't take into acount the developement of hot spots. If the maximum power does not go down at higher voltages this is a bad sign, if the FET is not expicitely specified for linear operation.