Dave looks at his TOP 5 (plus change) Jellybean Voltage Regulators and References, and explains why you need to know them.
00:00 – Jellybean Voltage Regulator & References
01:52 – 78xx Linear Voltage Regulator
08:01 – Adjustable Voltage Regulator
13:52 – 1117 Low Dropout Regulator
15:21 – LDO Stability
20:01 – LM4040/4041 Voltage Reference
24:38 – Using a Reference as a Regulator
30:53 – TL431 Voltage Reference
38:58 – Use as a PSU regulator
40:43 – Beware of Stability
41:43 – REF01 a better Voltage Reference
MC33063? I think that should've made this list. Also a list of components never to use.
Fun fact, I have a quote from some Chinese supplier, where it is 2 cents a piece on a full reel.
Since another reply mentioned the MC33063, out of curiosity I checked on it at Mouser. Except for the PDIP version it is virtually unavailable, not as bad at Digikey. Really brings home how bad the chip shortage is now. Same for MC34063 (lower temperature range). I would guess that most of the 12V car jack in to USB out devices use these chips.
1. 723 voltage regulator.
2. 78xx/LM340 positive and 79xx/LM320 negative fixed voltage regulators. These were also the first jellybean power ICs.
3. LM317 positive and LM337 negative adjustable voltage regulators. The LM317 was also available as the 3 amp LM350 and 5 amp LM338 which reached jellybean status. The LM317 integrated output transistor became available as the LM395 Ultra Reliable Power Transistor, but unfortunately it never reached jellybean status.
4. TL431 variable shunt reference.
5. 78S40/MC33063 integrated switching regulator, because Arch in IRC.
No regulator list is complete without the venerable LM723 ! 
I agree. Any list of jellybean regulations which lacks the 723 is incomplete. The 723 is *the* jellybean regulator.
Since another reply mentioned the MC33063, out of curiosity I checked on it at Mouser. Except for the PDIP version it is virtually unavailable, not as bad at Digikey. Really brings home how bad the chip shortage is now. Same for MC34063 (lower temperature range). I would guess that most of the 12V car jack in to USB out devices use these chips.
Do not forget that the MC33063 is a cut down 78S40 without the uncommitted operational amplifier and power diode, which at one time was a jellybean part. It was the first popular integrated switching regulator.
Two remarks from my side:
1) If memory serves me well the LM317 / LM337 also have decent noise specs. Not ultra low noise but if you have a noise and cost sensitive application they can be a good choice.
2) I'm using the 1117 regulators as well but they come in many flavours. Including ones with lower drop-out voltages, higher input voltages, lower current consumption and versions that are stable with ceramic MLCCs.
1) If memory serves me well the LM317 / LM337 also have decent noise specs. Not ultra low noise but if you have a noise and cost sensitive application they can be a good choice.
They have better noise because the adjustment terminal can be bypassed. The equivalent point is not brought out on fixed regulators unless they have a 4th pin, which was sometimes done.
2) I'm using the 1117 regulators as well but they come in many flavours. Including ones with lower drop-out voltages, higher input voltages, lower current consumption and versions that are stable with ceramic MLCCs.
There are all kinds of flavors of the 1117 from different manufacturers and I think this excludes it from jellybean status. You are in a maze of twisty 1117 regulators, all marked alike.
1) If memory serves me well the LM317 / LM337 also have decent noise specs. Not ultra low noise but if you have a noise and cost sensitive application they can be a good choice.
They have better noise because the adjustment terminal can be bypassed. The equivalent point is not brought out on fixed regulators unless they have a 4th pin, which was sometimes done.
2) I'm using the 1117 regulators as well but they come in many flavours. Including ones with lower drop-out voltages, higher input voltages, lower current consumption and versions that are stable with ceramic MLCCs.
There are all kinds of flavors of the 1117 from different manufacturers and I think this excludes it from jellybean status. You are in a maze of twisty 1117 regulators, all marked alike.
... maybe I got cheap clone stuff, but my experience on LM317 and especially LM337 was that the noise was quite large, and bypassing made it better but not by much. Neither ST, TI, ON, ... advertises these a slow noise. 10Hz to 10kHz RMS is 0.003% of VOut, which is 30uV at 10V out, about 8 times of this value in peak. Not impressive, IMHO.
My favourite is the TL431, for it’s only 3 pins, cheap, accurate and supports lots of variations: window detector, comparator, reference, capacitance multiplier...
... maybe I got cheap clone stuff, but my experience on LM317 and especially LM337 was that the noise was quite large, and bypassing made it better but not by much. Neither ST, TI, ON, ... advertises these a slow noise. 10Hz to 10kHz RMS is 0.003% of VOut, which is 30uV at 10V out, about 8 times of this value in peak. Not impressive, IMHO.
I do not consider the 317/337 to be particularly low noise, but they are lower noise than the 78xx/79xx series where the adjustment point is internal and cannot be bypassed.
Let's see ... 317 wideband input noise is about 170 nV/sqrt(hz), which is consistent with the adjustment pin current (from a common base stage) and a bandgap voltage reference. That is actually very good considering that a bandgap reference is used, but they hardly need to operate at micropower current levels. I do not see a problem. Most bandgap references are several times noisier than that.
For comparison, a 741 has an input noise of about 23 nV/sqrt(hz). When I have made low noise regulators using the 317 or 7805 as a pass element, I have used a 5 nV/sqrt(hz) operational amplifier like an OP27 or LT1007, and a buried zener reference like the (ancient) LM329 which operates down in the 14 nV/sqrt(hz) range when you include its 6.9 volt reference voltage, suggesting that an OP27/LT1007 is overkill and an OP07/LT1001 would be more suitable. So a realistic "high performance" regulator circuit should be about 20nV/sqrt(hz), or only 8 times better than a 317, which itself is about 8 times better than an 7812 if the adjustment pin is bypassed.
AFAIK the LM329 is a bit more noisy: LT gives something like 60 nV/sqrt(Hz). It is still a reasonable choice, as it was cheap (got expensive the last few years) and the 1/f noise is not so bad. The troublesome noise is anyway the 1/f low frequency part. Here the possibility for a cap at the feedback is limited.
One can get lower noise in the higher frequency range with filtering.
The noise of the 78xx also varies. I have an ET7805 (some Chinses manufacturer ?) with rather low noise. They even give a noise number in the dataseet:
40 µV for 10 Hz to 100 kHz and thus 125 nV/sqrt(Hz) if this would be white noise only and likely a little less for the white noise part, as the 1/f noise likely contributes a little at 10 Hz.
AFAIK the LM329 is a bit more noisy: LT gives something like 60 nV/sqrt(Hz).
It is, but for a fair comparison with a bandgap reference the 6.9 volts is attenuated down to 1.25 volt, so the 60 nV/sqrt(Hz) noise becomes 11 nV/sqrt(Hz). I started with 75 nV/sqrt(Hz) yielding 14 nV/sqrt(Hz).
40 µV for 10 Hz to 100 kHz and thus 125 nV/sqrt(Hz) if this would be white noise only and likely a little less for the white noise part, as the 1/f noise likely contributes a little at 10 Hz.
The 1/f noise is much more significant since the broadband noise can be filtered out, but I only had good numbers for the white noise. There is much less data available describing low frequency noise of the 317 and 78xx.
I don't see any buried zener references from Texas Instruments. Linear Technology has the noisier LT1021, but maybe with better low frequency noise. That leaves the LM329 and LM399 from Linear Technology which are pretty expensive compared to the cheap LM329s I used to buy. I do not see anything else on Mouser.
Getting low noise supply voltages in Systems on Chip type ICs is a difficult issue, many ourselves included opted for distributed regulators throughout the chip rather than try and supply all the circuitry with a single on or off chip regulator.
When we were developing SOTA RF/MW/MMW SoCs many years ago and required very low phase noise on chip frequency references for receive/transmit and clocks, getting close in low phase noise results were obscured by the supply low frequency noise levels which modulated the VCOs in the frequency synthesizers used for various clocks and RF signals. This lead to a development effort for an ultra-low noise fully on-chip LDO which would be stable with any load capacitance, culminating in patent 8692529.
https://patents.justia.com/patent/8692529Recall we achieved a measured noise level of ~12nv/rt Hz for a 2.5V reference which included the bandage reference and scaling amplifier. The LDO was a somewhat unique architecture to allow low noise, stable transient response, and achieve overall stability with any load capacitance.
Much later while at a IEEE ISSCC conference a young intel engineer presented a almost identical regulator that was employed in the latest processors, this regulator was used in on-chip clock distribution for filtering for the various VCOs. Had a nice discussion with the author and showed how they could significantly improve the low frequency noise with the addition of a single pinch resistor/fet.
Anyway, I've found during my career that low noise high DR designers typically don't consider the regulators and supply rails until the end of the design cycle rather than much earlier, and likely only after extensive lab measurements which were eventually traced to the regulators
Best,
The TL731/732 is the definition of jelly bean, because it is actually the IC produced in highest numbers on earth, every year.
One trap :
The NCP431/432 (I like to use those because of low consumption) has inverted pinout!
TL431 -> NCP432
TL432 -> NCP431
How lame !!
Do the various "unknown" manufacturers of chips like the LM7805, LM317, and LM1117 all correctly implement the various over-xxx protections?
I keep hearing about people letting the magic smoke out of the regulators on their Arduino Clones, usually from using an input voltage such that the power dissipation is too high for the (pitiful to non-existent) heat sink.
I would have thought that the "thermal shutdown" should have saved the part, but it has occurred to me that perhaps the no-name generic "1117-like" regulators "cheat" on things like those protections to get the lowest cost...
I don't think there is a significant saving from not including the thermal protection. There is however quite some savings from not testing the chips if the thermal prtection actually works at the right temperature. Non working thermal protection is more likely just out tolerance temperature and not fully missing protection. Even if working, the high temperarture is stress to the part and may lead to a higher failure rate. Besides the thermal protection, the SOA can be another problem. There can be variations between units and not all will fail, but some.
Cheaper parts may not be checked so often and in some cases reels with otherwise reject parts may show up on the gray market.
I doubt the generic regulator chips from no-name sources are any different design than the original design, but the fab process maybe be older/less controlled and lack of any detailed testing is highly likely. We found this to be true with the jellybean OP07, many no-name sources were generally good but showed more variability than quality known sources, and long term reliability was always an unknown with the no name sources.
If your product is in high quantity production, staying with a known good chip source for any "jellybean" component is probably a wise decision, since a chip failure or parameter drift later could easily cost well more than the "apparent" initial cost savings. Quality products demand quality components which usually demand a cost premium. Cheap stuff is well cheap and you roll the dice
Best,
Do the various "unknown" manufacturers of chips like the LM7805, LM317, and LM1117 all correctly implement the various over-xxx protections?
In the past the major manufacturers of these parts have sometimes let a bad design get into production. It would not surprise me if that still happens occasionally, especially with low quality manufacturers.
Do the various "unknown" manufacturers of chips like the LM7805, LM317, and LM1117 all correctly implement the various over-xxx protections?
Short answer is no, they don't.
http://www.kerrywong.com/2015/07/24/lm2596-dc-dc-converter-module-testing/See the first comment after the article. Switching frequency way wrong, no thermal protection. Unscrupulous bastards will do anything to save a few fractions of a cent, that's why I never buy anything I want to rely on from grey market. They're almost all fake and have chunks of functionality ripped out.
"You get what you pay for" + "if it seems too good to be true it probably is"
The DCDC converter test is a different thing. The problem there seem to be fake chips from the gray market, not chips from lesser known manufacturers. Having a switcher with a different frequency suggests a different type and lack of testing at the PCB level.
The 3rd source chips from other manufacturers can still be different in fine details, like the tendency to oscillate and the temperature where thermal protection sets in. Even with the mayor manufacturers the details of the chips may change over time, e.g. from process changes and possible new masks.
The 431 is a nice component and there is also a funny FOD2742 variant with opto coupler integrated, good if you want to keep the component count down
