Oscillating regulators with ceramic caps

I've just built up a batch of PCBs with several LM317 and LM337 regulators on them. As this thing will be in a hot environment, I wanted to avoid electrolytics so I used 10uF MLCC ceramic capacitors on the inputs and outputs of the regulators. Now that I have built it, all of the regulators oscillate!

I had thought that it was only LDOs (with the drain or collector of the pass device connected to the output) that would oscillate with low-ESR capacitors, but I now know that this also applies to some non-LDO linear regulators. Having seen the oscillation, I looked in the LM337 datasheet and sure enough it says it isn't stable with low-ESR capacitors. It does not even say what range of ESR it IS stable with. In the case of LDOs, I already had the policy of refusing to design in the ones that require some ESR without specifying what range is acceptable.

By experiment, for the LM337LM I have found that somewhere in the range of 50mOhms to 100mOhms added between the regulator and 10uF MLCC seems to just barely stop it from oscillating, without any other decoupling capacitors connected. I would use a somewhat higher value to get more phase margin. Adding these resistors is a pain because I have a lot of other MLCCs decoupling the same rails, so there is no nice place to hack in the resistors, especially since the board was not routed with this in mind. If I put the resistor between the regulator and the power plane, then there will be some interaction between the different circuit blocks as they would then share a significant common supply resistance at DC. Perhaps an inductor? Meh.

I guess there is a good reason why ADI can charge more for their AnyCap LDOs. I wonder if any of them is pinout compatible with the surface mount LM317MDCYRG3 and LM337LM that I have used.

Chris

Reply to
Chris Jones
Loading thread data ...

On a sunny day (Sun, 29 Mar 2015 02:45:34 +1100) it happened Chris Jones wrote in :

I have used many many LM317, and never had one oscillating. but I normally use tanatalum caps, say 1 to 10 uF very close.

Did you try LM317 from a different source, it is not ebay fake I hope? LM317 is also stable with electrolytics AFAIK. Maybe it is simpler to replace the caps with tantalum than add series resistors.

Duno about the 337.

Reply to
Jan Panteltje

I generally use tantalums, 10 to 22 uF, derated at least 2:1 on voltage. One big tantalum plus a bunch of smaller ceramics (bypass caps scattered around the board) seems to always work.

--
John Larkin         Highland Technology, Inc 
picosecond timing   laser drivers and controllers 
 Click to see the full signature
Reply to
John Larkin

LM337 is a typical LDO configuration, so is sensitive to load much the same as a LM1117 or etc. That was a limitation of the technology back then.

LM317 should be OK, but maybe your layout has other problems (chains of LC resonators?).

It's generally unwise to use just ceramic caps (or aluminum polymers). The ESR is much lower than sqrt(Lstray/C), so you get resonant peaks everywhere. You always want a bulk R+C somewhere to dampen the lumped equivalent transmission line that is the supply network. Use a cap value several times the total capacitance on the rail, with ESR (MLCC or Al Poly

  • external resistor; or tantalum or electrolytic of known ESR) equal to sqrt(Lstray/Cbyp). If Lstray is the average inductance between bypass caps, then use Cbyp = value of those individual caps. Or use Lstray = total supply inductance and Cbyp = total bypass. Same thing.

Preferably this should be placed at the start and end of a chained (linear, a>b>c>...) network, or at the end of each leaf in a branched network (a>b(>d>e)>c>...), since it serves as a termination resistor for the lumped equivalent transmission line.

Note that ferrite beads and filter chokes change the values of this damper substantially; design accordingly.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
 Click to see the full signature
Reply to
Tim Williams

Not quite. While the output primary current carrier is NPN, the configuration is actually PNP follower, except that the PNP is "synthetic", a weenie PNP drives a NPN Darlington (the LM317 is NPN follower). So the output impedance is low. The trouble with that configuration is that the PNP is a lateral structure... poor fT _plus_ a transit time excess phase issue... that's where the potential instability comes from.

...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

[snip]

Took me awhile to remember how to spell Sziklai ;-)..

In the LM337 case the PNP and NPN positions are reversed. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

Use resistors in series 1-3 ohms is more like what a tantalum or normal aluminum electrolytic does. You can put a 0.1uF MLCC in parallel.

Lots of LDOs **require** this, some of the older ones neglect to mention it clearly on the datasheets because cheap 10uF ceramic capacitors were the stuff of science fiction back when the data sheets were written. Watch out also for the really good low-Z electrolytic caps.

--sp

Best regards, Spehro Pefhany

--
"it's the network..."                          "The Journey is the reward" 
speff@interlog.com             Info for manufacturers: http://www.trexon.com 
 Click to see the full signature
Reply to
Spehro Pefhany

It would be nice if some company would actually apply some engineering... it's not exactly rocket science to make a stable LDO without all these series-R shenanigans... I do it all the time on-chip... I even had one chip design that has +5V input, with LDO's at 3.3V, 2.5V and 1.8V ;-) ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

The LM8261 (c-load opamp) architecture does it right. The dominant pole isn't buried inside, it's hung on the output pin.

--
John Larkin         Highland Technology, Inc 
picosecond timing   laser drivers and controllers 
 Click to see the full signature
Reply to
John Larkin

In CMOS it's almost impossible not to gain a pole inside the regulator as well as the one due to the load... solved by magical application of a load-current-sensitive zero in the regulator structure ;-) ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

I'm usually happier putting ~100 milliohms in series with the regulator output, after the reference divider. It's nice for debugging, too.

Cheers

Phil Hobbs

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

Thanks for the post. It's very timely. I was just about to do the exact same thing.

Reply to
David Eather

"Jim Thompson" wrote in message news: snipped-for-privacy@4ax.com...

Yup. And the LM1117 isn't terrifically unstable, as far as I know. Though it's also not a true LDO, in the sense that dropout ~= 0 for low currents.

Can't find one of the 337, here's the 1117:

formatting link
The PNP area seems larger compared to darlington structures (e.g., LM317, MC34063), which corroborates the poor hFE of most.

Was 1117 done with lateral as well, or did they have a [nearly] symmetrical process for that?

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
 Click to see the full signature
Reply to
Tim Williams

Yes, if I knew in advance then that is what I would have done. In order to prevent interactions between blocks (at least at DC), it is nice to put a separate series resistor feeding each circuit block, rather than one resistor common to all blocks. Sadly my existing layout is not very amenable to this.

Chris

Reply to
Chris Jones

Exactly what I was thinking. I have designed LDOs for cellular radios, and one of the requirements was that it would be stable with an ideal capacitor or anything that approaches one. Another requirement was

Reply to
Chris Jones

I also usually used tantalums or electrolytics. I was trying out MLCCs because I would prefer to avoid electrolytics at the operating temperature that I am expecting, and I wanted to avoid tantalums for a number of reasons. I did not realise that the 317 and 337 require ESR. Also even if I changed the bulk 10uF caps to tantalum, I suspect it might still be close to instability as there are a lot of other MLCCs distributed around the board.

Chris

Reply to
Chris Jones

The parts are from element14 and RS. I only tried one version of each part as these are the only ones that I have that will fit the PCB footprint.

I have built up separate breadboards with just the regulators and caps and some MOSFETs to turn on and off a pulsed load. The 337 oscillates like a banshee, the 317 just rings like a bell. So the 317 does not really oscillate, but it doesn't have significant ripple rejection around its ringing frequency. On the real board both regulators are powered from an isolated DC-DC, and perhaps the 317 was just resonating with the ripple coming from the DC-DC rather than oscillating as such. (240mVpp ripple at input, 170mVpp ripple at output - not much of a regulator)

Reply to
Chris Jones

I hope your PCB will need less bodge wires than mine.

Reply to
Chris Jones

Some more experimental results:

If I leave the 10uF MLCC directly connected to the regulator output, but also connect in parallel an additional 100uF tantalum with measured ESR of 0.17 Ohms, this makes both the LM317M and LM337LM show very stable-looking response to a load transient (current step) with no ringing. Similarly, a 220uF aluminium electrolytic with 0.14 Ohms ESR produced a very nice load transient response.

Adding these capacitors in parallel with the 10uF MLCC rather than adding resistors in series with the LDO output does give better load regulation and transient response, but of course the large value capacitor is undesirable due to size and long-term reliability, or cost (if it is made of MLCCs with series resistors added).

Chris

Chris

Reply to
Chris Jones

What temperature? Note that ceramics have a limited temperature range, too. X5Rs, for example are only rated to 85C. X7R = 125C.

Reply to
krw

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.