13
\$\begingroup\$

I'm working on a linear power supply design leveraging the uA723 regulator and I need power transistors to meet the current requirements (~2-3A). When looking looking through the regulators datasheet, they provide examples (see figures 4 and 5) for external NPN and PNP output transistors for achieving currents > 150mA limit of the IC.

Image 1

Image 2

Source of images: Texas Instruments - uA723 datasheet

My natural inclination is to go with a high side NPN transistor like in figure 4 but I've noticed this second high side PNP technique leveraged in other designs I've reviewed.

As I understand it, power PNPs are harder to manufacture resulting in less available options.... so why would I go with PNP over NPN? What am I missing?

\$\endgroup\$
1
  • 7
    \$\begingroup\$ You're really using the µA723? Why not consider the 317 (for positive voltages) or 337 (for negative voltages) these days? \$\endgroup\$
    – user117592
    Jan 7 at 21:21

4 Answers 4

19
\$\begingroup\$

Without fully exploring the topic (which could go on forever), just some points to highlight:

  • An external-transistor regulator like '723 may have lower voltage drop with PNP. Notice, as long as base bias can be supplied, VO can be arbitrarily close to VIN (as close as VCE(sat), minus any current-sense resistance of course).
  • PNP also has voltage gain, and additional phase shift, so therefore incurs a stability penalty. A direct substitution (as shown here, but without knowing anything else about the feedback network) is likely to cause oscillation. Check stability and adjust compensation as needed.
  • Conversely, NPN emitter followers tend to be lenient on stability, and have good PSRR (ripple on VIN is isolated by the collector impedance (Early effect)).
  • PNPs aren't harder to manufacture, they're made inversely to NPNs -- starting with p- instead of n- substrate (collector), epitaxy/diffuse n/p base layer, diffuse p+/n+ emitter layer and (degenerately doped) ohmic contacts, done. They may be more difficult in IC process, where either the collector is hard-wired to the substrate, or additional isolation wells have to be manufactured. The additional process steps then cost more up-front (masks, production tuning) and per-part (more steps, somewhat lower yield).
  • PNPs perform slightly poorer than NPN, on the order of about 10% in terms of overall conduction parameters IIRC. They actually perform better in some respects, or can, such as hFE being a little bit higher, but overall figure-of-merit tends to favor NPN. Basically, nothing that's strong enough to matter here; regulators are a very modest application.

If you've been reading a lot from contemporary sources (contemporary of µA723 that is; it's ancient, and there is no reason or need to use it today, unless perhaps as a requirement for homework assignment or something), then it was true back then that the NPN process was most amenable to consumer production parts, and thus lateral PNP was the go-to method to create them, when they were required at all. (Lateral PNP have very poor hFE, so they aren't easy to use as output devices.)

Such contemporary books are still good information, by the way -- just beware whatever advice is either dependent on available devices (all they had back then were simple regulators and controllers like '723, op-amps like '741, etc.), and changes over time (linear regs give way to LDOs and SMPS nowadays), or based on assumptions that were relevant at the time (speaking of SMPS, a big one that comes to mind is "minimize stray inductance" -- with many devices today, it's fundamentally impossible to minimize it to the degree required by what that rule of thumb is actually trying to achieve, and stating that underlying purpose makes much more sense). We also have much better means of simulation and calculation now -- compare signal filter tables to from-scratch analytical and Monte Carlo simulation techniques, and, just, the broad availability of SPICE and other simulators (which have been reasonably available since the 90s, but if we're talking 723s here, we're probably talking books from the 70s, maybe even 60s?).

  • MOSFETs are a highly viable option these days, from CMOS regulators (integrated) to external-device types (few of the linear type, but they are out there..!). The negligible gate current can improve efficiency, the RDS(on) can improve saturation (dropout), and devices are available with wider SOA than BJTs -- but on that same note, beware that modern MOSFETs can be as prone to 2nd breakdown as BJTs can be, and always check the datasheet for DC SOA, for an application like this.

  • MOSFETs do suffer from a large difference between N- and P-channel types. In silicon, P-ch is about 2.5 times worse in terms of RDS(on) and capacitance; basically the ratio of electron to hole mobility in the channel. Thus, there's no meaningful complementary pairs in MOS, as there are for BJTs; typically a compromise is chosen, such as P-ch twice the width of the N-ch to get a modest 25% higher RDS(on) and only double the capacitance, for CMOS gates for example. "Complementary" duals are available for power switching application as well. For a linear regulator, again we aren't generally too concerned about this figure-of-merit, and either type is usable; the main downside is, for N-ch, we need a gate drive voltage higher than the output, which makes low-dropout design awkward. (There are, in fact, LDOs where a higher bias voltage can be applied; thus you can regulate, say, 3.3V down to 2.5V, with help from a 5V supply, but without incurring the penalty of dropping directly from 5V to 2.5V.)

  • Compensation is probably more difficult with MOS, as the capacitances, and transconductance, all vary strongly with voltage, especially at low voltages (near dropout). BJTs do vary strongly as well, but capacitance is more gradual, and transconductance only near VCE(sat).

\$\endgroup\$
3
  • 2
    \$\begingroup\$ It's a little unclear what you mean by "contemporary" here. When I first read the phrase "contemporary sources," I thought that would mean "recent sources" (2018-2023, say), but now I'm thinking that it means "sources from around the time of the µA723" (1970-1975, say). \$\endgroup\$ Jan 7 at 20:37
  • 1
    \$\begingroup\$ That may be a little unclear, yeah; I'm imagining, at least, through some reason or another the 723 was selected, and maybe that that reason includes books etc. contemporary with it. \$\endgroup\$ Jan 8 at 6:06
  • 1
    \$\begingroup\$ In your first paragraph, don't you mean that Vc can be close to Vin? Or that Vce(sat) can be close to zero? \$\endgroup\$
    – SteveSh
    Jan 9 at 12:28
11
\$\begingroup\$

Some circuits are just more convenient with PNP due to where the transistor is required to be in the circuit.

It is usually simplest to provide a base signal to control BJT relative to one of the voltage rails. That means it is most convenient to have the emitter be tied to the voltage rail. So your signal reference matches the reference pin.

Because when this is not the case, then you start needing to add extra circuitry to float things around since now your emitter is not a fixed potential and can move around and that means the base signal that is applied also has to have its reference follow to match.

I realize that sounds really broad, but that's just the way it is. When you are trying to make a circuit, you'll know the moment you encounter a case where PNP is simpler than NPN. None of the voltages will line up properly with simple circuitry for where you want the BJT to be, and that's when you know it's probably the other one. I wouldn't worry too much about coming up with cases beforehand.

A lot of PNP circuits are special, rather strange, niche applications where it's super obvious that one approachis wildly more complex than the other. I can't think of too many where you have almost equal choice between PNP and NPN. The only one that comes to mind at the moment is PNP vs NPN linear regulators, just like you posted. You can look at those because in the difference in circuit complexity is small enough that it is overshadowed by performance criteria.

But ICs obfuscate a lot of things and can hide a lot of complexity to make the external solution look simple when it is actually more complicated on the whole. So it's best to look at simpler circuits where everything is laid bare.

You can see here that the NPN is more complex than the PNP, but not overly so that an NPN would be ruled out in most cases on the basis of complexity alone. The increased complexity of the NPN is to be expected here since for most linear regulators we want to regulate relative to the lower rail, which means the BJT needs to be hanging off the high rail. That's a location most suited for the PNP whose emitter is directly tied to the high rail which means control signals appled to the base can be reference directly to that same high rail making things simpler.

enter image description here https://www.edn.com/get-the-best-from-your-low-dropout-regulator/

\$\endgroup\$
7
\$\begingroup\$

One reason to use a PNP as the pass transistor is lower dropout voltage. This is because the NPN is an emitter follower and can only have its base pulled up to the supply voltage, so the supply has to be more than a diode drop higher than the output voltage, in order to saturate, while the PNP being common emitter only needs the base to be pulled down a few volts from the supply voltage to saturate.

Here's a simulation of each showing the output voltage, dropout voltages, and a voltage representing the base current, which I've put at around 19 mA for each. enter image description here

You can see that the dropout voltage for the NPN is nearly four times that of the PNP.

Some downsides of using PNP can be lower bandwidth and higher output impedance which can cause problems with capacitive loads. See this article.

\$\endgroup\$
1
\$\begingroup\$

Since this is a linear supply, there is no tradeoff in general. Both transistors will be dissipating the same heat power, given the same input and output voltage and load current. The NPNs may have a higher gain, but that can be dealt with in the driver. So use whatever is convenient.

Do note that in order to fully turn on the NPN pass transistor in low line, low temperature conditions, the base may need to be higher than the collector. With an NPN pass transistor, the 723's power supply needs to be higher than the supply to the pass transistor. This can be done with a small auxiliary transformer, an extra winding on the main transformer, or a boost converter.

So, with a PNP the 723 doesn't need a boosted supply voltage but will be usually driving the base harder than it'd need to for the NPN. The 723 will be dissipating way more heat because of that. So that's the tradeoff, mostly.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.