Oscillating regulators with ceramic caps

[Chris Jones]

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

[Jan Panteltje]

On a sunny day (Sun, 29 Mar 2015 02:45:34 +1100) it happened Chris Jones
<lugn...@spam.yahoo.com> wrote in <5516cca2$0$12771$862e...@ngroups.net>:

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.

[John Larkin]

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

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

[Tim Williams]

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
Website: http://seventransistorlabs.com

"Chris Jones" <lugn...@spam.yahoo.com> wrote in message
news:5516cca2$0$12771$862e...@ngroups.net...

[Jim Thompson]

On Sat, 28 Mar 2015 12:31:04 -0500, "Tim Williams"
<tiw...@seventransistorlabs.com> wrote:

>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.

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.

>
>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

...Jim Thompson
--
| James E.Thompson | mens |
| Analog Innovations | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| San Tan Valley, AZ 85142 Skype: skypeanalog | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

[Jim Thompson]

On Sat, 28 Mar 2015 11:48:12 -0700, Jim Thompson
<To-Email-Use-Th...@On-My-Web-Site.com> wrote:

>On Sat, 28 Mar 2015 12:31:04 -0500, "Tim Williams"
><tiw...@seventransistorlabs.com> wrote:
>
>>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.
>
>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.
>

[snip]

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

<http://en.wikipedia.org/wiki/Sziklai_pair>

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

[Spehro Pefhany]

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"
sp...@interlog.com Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog Info for designers: http://www.speff.com

[Jim Thompson]

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 ;-)

[John Larkin]

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

[Jim Thompson]

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 ;-)

[Phil Hobbs]

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
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

[David Eather]

On Sun, 29 Mar 2015 01:45:34 +1000, Chris Jones <lugn...@spam.yahoo.com>
wrote:

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

[Tim Williams]

"Jim Thompson" <To-Email-Use-Th...@On-My-Web-Site.com> wrote
in message news:juudhatdo01q2av4k...@4ax.com...

> Took me awhile to remember how to spell Sziklai ;-)..
>
> <http://en.wikipedia.org/wiki/Sziklai_pair>
>
> In the LM337 case the PNP and NPN positions are reversed.

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:
http://s.zeptobars.ru/LM1117-HD.jpg
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?

[Chris Jones]

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

[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
<20nV/rt-Hz noise and a startup time of a few microseconds. A few times
I have wished I could use one of my LDOs on a circuit board but sadly
they are only available with a cellular radio attached, they need some
SPI commands to turn on the regulator, and the datasheet is under NDA.

At least the ADI AnyCap ones do seem stable in my experience of them.

Chris

[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

[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)

[Chris Jones]

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

[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

[k...@zzz.com]

On Sun, 29 Mar 2015 15:15:54 +1100, Chris Jones

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

[John Larkin]

The ESR of wet aluminum electrolytics skyrockets below 0C. Using them
is a good way to build a temperature-activated oscillator.

Polymer alums are good over temperature, but their ESR may be too low
for some regulators. Tantalums are just right for 1117 type regs.

[Tim Williams]

<k...@zzz.com> wrote in message
news:ql3ghap3vo2e589mn...@4ax.com...

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

They'll be better off than the regs -- 100C-ish cutoff, and that's Tj on
top of whatever power they're dissipating, so I hope this thing isn't
actually needing an ampere of capacity and/or an ambient over 60C.

[M.Randelzhofer]

>
> 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
>
>

like that low ohmic resistor R for debugging and measuring current too, but
as i understand the R is after the cap, so the mlcc problem for LDOs still
exists.

I like to put a current measure R between the cap and the feedback voltage
divider, so the V drop is regulated out.
Disconnecting the load for debugging is done by putting the feedback divider
on another pad connected directly on the cap output, additionally put on the
layout.

MIKE

--
www.oho-elektronik.de
OHO-Elektronik
Michael Randelzhofer
FPGA und CPLD Mini Module

[Phil Hobbs]

No, the resistor goes between the regulator and the cap, but the voltage divider is on the regulator side. Otherwise the output voltage would go nuts.

Cheers

Phil Hobbs

[David Eather]

On Sun, 29 Mar 2015 21:00:28 +1000, Chris Jones <lugn...@spam.yahoo.com>
wrote:

That does sound an awful lot like what happens with LDO's - there is a
range of output cap that will oscillate. Never heard of it for a 317
though. A re-spin or an LDO in drag?