Groundplane under SMPS power inductor

[John Devereux]

I have seen designs where there is a cutout in the groundplane under the
power inductor in e.g. a buck converter. Inductor is nominally a
"shielded" one.

I assume it is to prevent some kind of "shorted turn" effect? What do
you think?

Might a continuous plane be better? It could help to shield any field
leakage and reduce emissions and circuit noise.

I am asking generally, but say 1A, 500kHz.

Thanks,

--

John Devereux

[rickman]

Chip makers like Linear Tech often provide excellent information in
their data sheets and app notes. I seem to recall the idea is to
minimize injected currents into the ground plane to reduce noise, but it
has been a while since I've done any SMPS design.

--

Rick

[John Larkin]

That's not a lot of power.

A poorly shielded inductor could squirt some grould-loop voltages into
the ground plane, which might matter in some systems. I doubt that any
"shorted turn" effects would be significant.

We mix switchers and low-level stuff, but we keep them as far apart as
possible, and don't cut the ground plane under inductors.

Take a candidate inductor and measure it free-space, then up against a
piece of copperclad. See if anything changes.

[George Herold]

On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote:
> I have seen designs where there is a cutout in the groundplane under the
> power inductor in e.g. a buck converter. Inductor is nominally a
> "shielded" one.
>
> I assume it is to prevent some kind of "shorted turn" effect? What do
> you think?

I saw the shorted turn effect with an unshielded open ended inductor
used in a resonant circuit.
The ground plane under the inductor reduced the Q from ~35 to ~20.

I'm not sure about SMPS... I guess it depends on how much of the loss is
in the inductor.

George H.

[John Devereux]

John Larkin <jjla...@highlandtechnology.com> writes:

> On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
> <jo...@devereux.me.uk> wrote:
>
>>
>>I have seen designs where there is a cutout in the groundplane under the
>>power inductor in e.g. a buck converter. Inductor is nominally a
>>"shielded" one.
>>
>>I assume it is to prevent some kind of "shorted turn" effect? What do
>>you think?
>>
>>Might a continuous plane be better? It could help to shield any field
>>leakage and reduce emissions and circuit noise.
>>
>>I am asking generally, but say 1A, 500kHz.
>>
>>Thanks,
>
> That's not a lot of power.
>
> A poorly shielded inductor could squirt some grould-loop voltages into
> the ground plane, which might matter in some systems. I doubt that any
> "shorted turn" effects would be significant.

You think it might be removed to reduce induced noise currents? I was
thinking the other way.

>
> We mix switchers and low-level stuff, but we keep them as far apart as
> possible, and don't cut the ground plane under inductors.

Me too so far.

> Take a candidate inductor and measure it free-space, then up against a
> piece of copperclad. See if anything changes.

I can try that, good idea!

Yay, an excuse to fire up the old 4192A :)


--

John Devereux

[mako...@yahoo.com]

On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote:
> John Larkin <jjla...@highlandtechnology.com> writes:
>
> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
> > <jo...@devereux.me.uk> wrote:
> >
> >>
> >>I have seen designs where there is a cutout in the groundplane under the
> >>power inductor in e.g. a buck converter. Inductor is nominally a
> >>"shielded" one.
> >>
> >>I assume it is to prevent some kind of "shorted turn" effect? What do
> >>you think?
> >>
>

if it is a magnetically shielded inductor, then it should not matter

if it is an open magnetic circuit inductor, then it could matter.

Mark

[DecadentLinuxUserNumeroUno]

On Wed, 4 Nov 2015 14:48:32 -0800 (PST), mako...@yahoo.com Gave us:

More likely to maintain minimum creepage/gap between circuit segments.

Pot cores... no effect. Toroidal... no effect. Therefore it must
be for protection rules/reasons.

[John Larkin]

People take great liberties with the word "shielded", but most such
surface-mount inductors leak more field out the top than the bottom.



--

John Larkin Highland Technology, Inc
picosecond timing precision measurement

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

[Bill Sloman]

Pot cores do have some flux leakage around the gaps left to let the wires come in and out, and there's a - small - current loop between the pins that terminate the windings.

Non-progressively wound toroids don't have any leakage fields, but the usual way to wind a toroid, with the winding progressing around the toroid, creates a single turn in the plane of the toroid, which will leak flux into an adjacent ground plane, and induce a circulating current.

--
Bill Sloman, Sydney

[krw]

On Wed, 04 Nov 2015 18:11:36 -0500, DecadentLinuxUserNumeroUno
<DL...@DecadentLinuxUser.org> wrote:

>On Wed, 4 Nov 2015 14:48:32 -0800 (PST), mako...@yahoo.com Gave us:
>
>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote:
>>> John Larkin <jjla...@highlandtechnology.com> writes:
>>>
>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
>>> > <jo...@devereux.me.uk> wrote:
>>> >
>>> >>
>>> >>I have seen designs where there is a cutout in the groundplane under the
>>> >>power inductor in e.g. a buck converter. Inductor is nominally a
>>> >>"shielded" one.
>>> >>
>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do
>>> >>you think?
>>> >>
>>>
>>if it is a magnetically shielded inductor, then it should not matter
>>
>>if it is an open magnetic circuit inductor, then it could matter.
>>
>>Mark
>
> More likely to maintain minimum creepage/gap between circuit segments.

Not likely.

>
> Pot cores... no effect. Toroidal... no effect. Therefore it must
>be for protection rules/reasons.

It's most likely an old wives tales. Some inductors aren't closed
magnetically (or aren't done well) so there is some chance of inducing
a current in any metal under the core. One of the engineers where I
work does things like this but I find there's more danger in
compromising the ground plane than any current induced in ground. The
only place I delete planes is where it's specified by the part's
manufacturer.

[John Devereux]

krw <k...@nowhere.com> writes:

> On Wed, 04 Nov 2015 18:11:36 -0500, DecadentLinuxUserNumeroUno
> <DL...@DecadentLinuxUser.org> wrote:
>
>>On Wed, 4 Nov 2015 14:48:32 -0800 (PST), mako...@yahoo.com Gave us:
>>
>>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote:
>>>> John Larkin <jjla...@highlandtechnology.com> writes:
>>>>
>>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
>>>> > <jo...@devereux.me.uk> wrote:
>>>> >
>>>> >>
>>>> >>I have seen designs where there is a cutout in the groundplane under the
>>>> >>power inductor in e.g. a buck converter. Inductor is nominally a
>>>> >>"shielded" one.
>>>> >>
>>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do
>>>> >>you think?
>>>> >>
>>>>
>>>if it is a magnetically shielded inductor, then it should not matter
>>>
>>>if it is an open magnetic circuit inductor, then it could matter.
>>>
>>>Mark
>>
>> More likely to maintain minimum creepage/gap between circuit segments.
>
> Not likely.
>>
>> Pot cores... no effect. Toroidal... no effect. Therefore it must
>>be for protection rules/reasons.
>
> It's most likely an old wives tales. Some inductors aren't closed
> magnetically (or aren't done well) so there is some chance of inducing
> a current in any metal under the core.

If it does, I think generally I *want* it do to that, so that the metal
acts to confine the field.

> One of the engineers where I
> work does things like this but I find there's more danger in
> compromising the ground plane than any current induced in ground.

That's what I was thinking...

The other thing is whether to do the thing with a separate island of
ground for the local switcher return currents, and connect this island
to the main circuit ground plane at a single point.

[ SMPS Parts]
----------------- local GND
----------------''---------------------- main GND

> The
> only place I delete planes is where it's specified by the part's
> manufacturer.

--

John Devereux

[John Devereux]

John Larkin <jjla...@highlandtechnology.com> writes:

> On Wed, 4 Nov 2015 14:48:32 -0800 (PST), mako...@yahoo.com wrote:
>
>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote:
>>> John Larkin <jjla...@highlandtechnology.com> writes:
>>>
>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
>>> > <jo...@devereux.me.uk> wrote:
>>> >
>>> >>
>>> >>I have seen designs where there is a cutout in the groundplane under the
>>> >>power inductor in e.g. a buck converter. Inductor is nominally a
>>> >>"shielded" one.
>>> >>
>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do
>>> >>you think?
>>> >>
>>>
>>if it is a magnetically shielded inductor, then it should not matter
>>
>>if it is an open magnetic circuit inductor, then it could matter.
>>
>>Mark
>
> People take great liberties with the word "shielded", but most such
> surface-mount inductors leak more field out the top than the bottom.

Yep, hence my wording "nominally sheilded". It is easy to probe the
leakage field around the system, with a small wire loop on the end of a
coax and a scope. (Someone suggested a SMT inductor instead of the loop,
might have been you).


--

John Devereux

[DecadentLinuxUserNumeroUno]

On Thu, 05 Nov 2015 07:56:48 +0000, John Devereux <jo...@devereux.me.uk>
Gave us:

>If it does, I think generally I *want* it do to that, so that the metal
>acts to confine the field.

Copper does not contain a magnetic field.

A faraday cage does for AC generated emissions, but merely attenuates
it.

mu metal is needed to "steer" a magnetic field.

A copper plane on a PC board? No.

Again, it is likely to gain a proper "nearest distance" thing, and
could keep stray currents from being injected into a power or ground
plane. Note that traces are likely managed around such elements as
well, and said ground plane exclusion probably has a lot of vias ringing
the area as well.

[John Devereux]

DecadentLinuxUserNumeroUno <DL...@DecadentLinuxUser.org> writes:

> On Thu, 05 Nov 2015 07:56:48 +0000, John Devereux <jo...@devereux.me.uk>
> Gave us:
>
>>If it does, I think generally I *want* it do to that, so that the metal
>>acts to confine the field.
>
> Copper does not contain a magnetic field.

Yes it does at high frequencies. I have made measurements, quite fun and
not too hard to do actually. Small coil on the end of a 50ohm coax going
into a sensitive scope. Trigger the scope from the SMPS, wave the coil
around the board. You can see the signal, and also the field direction
(by orienting the coil differently).

> A faraday cage does for AC generated emissions, but merely attenuates
> it.
>
> mu metal is needed to "steer" a magnetic field.
>
> A copper plane on a PC board? No.

At 50Hz, sure, but a copper plane makes a big difference at 500kHz+.

> Again, it is likely to gain a proper "nearest distance" thing, and
> could keep stray currents from being injected into a power or ground
> plane. Note that traces are likely managed around such elements as
> well, and said ground plane exclusion probably has a lot of vias ringing
> the area as well.

--

John Devereux

[bloggs.fred...@gmail.com]

At the frequencies and performance in use today it very well could be a precaution against proximity effect and the associated loss.

[John Larkin]

On Thu, 05 Nov 2015 07:59:36 +0000, John Devereux

<jo...@devereux.me.uk> wrote:

>John Larkin <jjla...@highlandtechnology.com> writes:
>
>> On Wed, 4 Nov 2015 14:48:32 -0800 (PST), mako...@yahoo.com wrote:
>>
>>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote:
>>>> John Larkin <jjla...@highlandtechnology.com> writes:
>>>>
>>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
>>>> > <jo...@devereux.me.uk> wrote:
>>>> >
>>>> >>
>>>> >>I have seen designs where there is a cutout in the groundplane under the
>>>> >>power inductor in e.g. a buck converter. Inductor is nominally a
>>>> >>"shielded" one.
>>>> >>
>>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do
>>>> >>you think?
>>>> >>
>>>>
>>>if it is a magnetically shielded inductor, then it should not matter
>>>
>>>if it is an open magnetic circuit inductor, then it could matter.
>>>
>>>Mark
>>
>> People take great liberties with the word "shielded", but most such
>> surface-mount inductors leak more field out the top than the bottom.
>
>Yep, hence my wording "nominally sheilded". It is easy to probe the
>leakage field around the system, with a small wire loop on the end of a
>coax and a scope. (Someone suggested a SMT inductor instead of the loop,
>might have been you).

A lot of the surfmount "shielded" inductors are drum (H-shaped)
ferrite cores, wound and then dropped into a ferrite tube. That makes
a ring-shaped air gap at both ends and allows lots of field leakage.
That's symmetric, except that the board end sometimes has a plastic
base which, with the solder joint, spaces the part a bit above the
board. I tried one, sitting on a PCB with a layer 2 ground plane. It
lost a couple per cent of its free-space inductance on the board in
its normal mounting position, maybe 5% if pushed onto the board upside
down.

I think there are surface-mount power inductors with better shielding.
Pot cores and toroids are more expensive than shielded drum cores.
Unshielded drums leak huge fields.

We don't usually worry about it. On the other hand, it's really
convenient to buy little potted dc/dc converters, cheap and all done.
They sit above the board and usually have toroidal inductors inside.
That can rip a lot of parts off your BOM. I sometimes use them even
when I don't need the isolation.

[John Larkin]

On Thu, 05 Nov 2015 06:57:59 -0500, DecadentLinuxUserNumeroUno
<DL...@DecadentLinuxUser.org> wrote:

>On Thu, 05 Nov 2015 07:56:48 +0000, John Devereux <jo...@devereux.me.uk>
>Gave us:
>
>>If it does, I think generally I *want* it do to that, so that the metal
>>acts to confine the field.
>
> Copper does not contain a magnetic field.

It sure can block an AC field.

[whit3rd]

On Wednesday, November 4, 2015 at 4:18:30 PM UTC-8, krw wrote:
> On Wed, 04 Nov 2015 18:11:36 -0500, DecadentLinuxUserNumeroUno
> <DL...@DecadentLinuxUser.org> wrote:

> >>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux
> >>> > <jo...@devereux.me.uk> wrote:
> >>> >
> >>> >>
> >>> >>I have seen designs where there is a cutout in the groundplane under the
> >>> >>power inductor in e.g. a buck converter.

> > Pot cores... no effect. Toroidal... no effect. Therefore it must
> >be for protection rules/reasons.

> It's most likely an old wives tales. Some inductors aren't closed
> magnetically (or aren't done well) so there is some chance of inducing
> a current in any metal under the core.

The familiar old slug-tuned inductor-in-a-can that used to dominate radios, had
no other barrier to flux leakage downward, and it'd be a good idea to
give it a groundplane. For a power inductor, though, any nearby conductive plane
will change the leakage flux (and that can have deleterious effects, if the internal
flux geometry changes as a result). I'd leave out the copper pour unless the
manufacturer suggests it.

So, the reason to put more copper in: block leakage flux (i.e. shielding)
The reasons to put less copper in: energy lost to heating, ground plane induced voltages,
and interference with 'nominal' magnetization in the core

[krw]

On Thu, 05 Nov 2015 07:56:48 +0000, John Devereux

Copper isn't going to do much to confine a magnetic field.

>
>
>> One of the engineers where I
>> work does things like this but I find there's more danger in
>> compromising the ground plane than any current induced in ground.
>
>That's what I was thinking...
>
>The other thing is whether to do the thing with a separate island of
>ground for the local switcher return currents, and connect this island
>to the main circuit ground plane at a single point.
>
> [ SMPS Parts]
> ----------------- local GND
>----------------''---------------------- main GND

I always do that, with the input and output capacitors, and switches
connected to the island. Note that the ground plane isn't compromised
at all.

[Kevin McMurtrie]

In article <87wptxu...@devereux.me.uk>,

Nearby copper changes the ringing in discontinuous mode by quite a bit.
Maybe two semicircles of copper dampens a certain unwanted frequency.

--
I will not see posts from astraweb, theremailer, dizum, or google
because they host Usenet flooders.

[dagmarg...@yahoo.com]

On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote:

I inherited a SMPS a few years ago that was blanking out nearly
the entire F.M. band.

Part of the problem was the main inductor's gap, which faced down
into a ground plane and impressed a very impressive signal across
same.

This inductor was "shielded," a drum core cemented inside a 5-sided
ferrite box.

I substituted a gap-up inductor, and (that part of) the problem was
solved.

Cheers,
James Arthur

[John Devereux]

Hey you beat me to it, thanks. I found ~3% with a SRU1048 22uH. If I
drill a 10mm hole under it the change is about 1.5%. So it makes a bit
of a difference but does not eliminate the effect.

This was all at 500kHz.

The indicated Q went from 24 to 22 as it was brought up to the plane.

> I think there are surface-mount power inductors with better shielding.
> Pot cores and toroids are more expensive than shielded drum cores.
> Unshielded drums leak huge fields.
>
> We don't usually worry about it. On the other hand, it's really
> convenient to buy little potted dc/dc converters, cheap and all done.
> They sit above the board and usually have toroidal inductors inside.
> That can rip a lot of parts off your BOM. I sometimes use them even
> when I don't need the isolation.

--

John Devereux

[John Devereux]

Interesting, thanks.



--

John Devereux

[RBlack]

On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmarg...@yahoo.com)
said:

Interesting. I have a very similar problem at the moment, a board with
two synchronised 90W boost converters, using the type of inductors you
describe. They're made by Würth, who are pretty good IME, but I've
never tried this inductor construction before.

The board is 4 layers with a solid ground plane under the inductor, and
no traces or components on the opposite side to the inductor.
Originally I thought any currents induced in the ground plane would have
a small enough loop area, and the plane itself would be a low enough
impedance, that this wouldn't be a problem. Clearly this analysis
doesn't hold up at this sort of power level...

I will try and find a gap-up inductor with the same specs which fits the
PCB pads. I suspect the problem I will then have, is that there is an
array of right-angle connectors on an adjacent board, whose PCB pins
pass within a few mm of the top of one of the inductors. The boards are
in a metal housing and the connectors and cabling are shielded, but
still not ideal.

[John Devereux]

Looking at it pessimistically, you could think of it as a voltage
transformer. 12V on the "primary" and xxV around the "secondary" which
is a circular region of ground plane under the inductor.

If the inductor has 6 turns say that would be 2V induced with perfect
coupling. If it only couples 5% that is still 100mV which is not nothing
when it appears at RF on a cable or in a sensitive circuit.

The rest of the ground plane tries to short this out but there could be
a lot of current available, the same ratio that steps down the voltage
steps up the available current.

(Sorry for the above travesty of transformer theory and I know it's all
a bit handwavey but I can see how it could be an issue!)

> I will try and find a gap-up inductor with the same specs which fits the
> PCB pads. I suspect the problem I will then have, is that there is an
> array of right-angle connectors on an adjacent board, whose PCB pins
> pass within a few mm of the top of one of the inductors. The boards are
> in a metal housing and the connectors and cabling are shielded, but
> still not ideal.

Unfortunately I never see the leakage mentioned anywhere as an inductor
specification. Except that it might be "low".

The dirt-cheap multilayer inductors are quite good here, I guess the
turns are fully embedded in ferrite. But they only seem to go to low
power levels and are more lossy than other types.


--

John Devereux

[dagmarg...@yahoo.com]

I pulled up the old artwork for a look and thought it worth sharing
some more details of my experience.

It was a buck SMPS, 12V input, 5V output @2.4A, ~460KHz.

As received,
Layer 1 was needed to connect the inductor,
Layer 2 had non-critical signals,
Layer 3 was solid ground plane.


Layer 1
.---------------------------//
.------/-----------.
| |-----. | .-+++-. +5V output
| | | | | C1 |
| '-----'-----|-|- - -|-------//
| .--|-|- - -|---------->
| | .| '-xxx-'. . . . .
| L1 |. | . . . . ..------->
| | .|. . . . . |
| .-----. |. | . GND . .|
| | | | .|. . . . . |
| '-----' |. | . . . . .|
'-------\----/--|--'. .TP1. . |
| | /. . . (). . .|


TP1 is a solid through-hole metal post, connected to the GND trace and
to the ground plane on layer 3.

You can see that the original ground run ran under L1, parallel
to L1's winding progression, creating a transformer-coupling in the topside
GND above TP1 (a solid metal grounded test post at the lower bound) and
prior to the bulk filter cap C1's negative terminal.

So, no doubt there was significant coupling into that trace despite a
solid plane two layers deeper.

That rendered C1 (and additional bypasses) ineffective at r.f. The
top-right portion of the GND trace supplied a ribbon connector, providing
a path for wicked radiation and conduction of the induced GND signal.

The revised layout routed layer 1's GND well around L1 rather than under,
and nailed it to layer 3's ground plane at both ends of the run, and in-
between.

The replacement inductor(*) was also a drum cemented into a ferrite box
like the first unit, however, the new unit's down-facing gap was tighter,
and filled with what appeared to be a ferrite-loaded cement. (The original
unit had a larger downward-facing gap, air-filled, IIRC.)

* Bourns SRR1240-150M

The improved gap configuration wasn't obvious on the datasheet--I had to
get a sample in hand.

So, I can't say for sure that an inductor directly over a solid ground plane
would be problematic from this experience.

I can say there was significant transformer coupling due to fringe flux
from the downward-facing gap.

Cheers,
James Arthur

[John Larkin]

On Fri, 06 Nov 2015 07:59:52 +0000, John Devereux

Cool. That might matter if you need a super-efficient switcher.

[dagmarg...@yahoo.com]

Since the Q fell by roughly the same factor as the inductance, I take that
as indicating minimal added dissipation. Q = xL/R, etc.

That's encouraging, actually.

Cheers,
James Arthur

[John Devereux]

dagmarg...@yahoo.com writes:

> On Friday, November 6, 2015 at 8:34:36 AM UTC-5, John Devereux wrote:
>> RBlack <ne...@rblack01.plus.com> writes:
>>
>> > On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmarg...@yahoo.com)
>> > said:
>> >> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote:
>> >> > I have seen designs where there is a cutout in the groundplane under the
>> >> > power inductor in e.g. a buck converter. Inductor is nominally a
>> >> > "shielded" one.
>> >> >
>> >> > I assume it is to prevent some kind of "shorted turn" effect? What do
>> >> > you think?
>> >> >
>> >> > Might a continuous plane be better? It could help to shield any field
>> >> > leakage and reduce emissions and circuit noise.
>> >> >
>> >> > I am asking generally, but say 1A, 500kHz.

[...]

I guess the very low impedance is what makes the cap ineffective.

> The revised layout routed layer 1's GND well around L1 rather than under,
> and nailed it to layer 3's ground plane at both ends of the run, and in-
> between.
>
> The replacement inductor(*) was also a drum cemented into a ferrite box
> like the first unit, however, the new unit's down-facing gap was tighter,
> and filled with what appeared to be a ferrite-loaded cement. (The original
> unit had a larger downward-facing gap, air-filled, IIRC.)

I suppose I need to buy a few types.

> * Bourns SRR1240-150M
>
> The improved gap configuration wasn't obvious on the datasheet--I had to
> get a sample in hand.
>
> So, I can't say for sure that an inductor directly over a solid ground plane
> would be problematic from this experience.

Thank you very much for the detailed write-up James.

> I can say there was significant transformer coupling due to fringe flux
> from the downward-facing gap.

I think if I build the whole SMPS section on a local ground "island"
over the overall groundplane this will work OK. Any magnetic leakage
through the island is only going to be at lower frequencies. As you
pointed out my "Q" measurement indicates that efficiency should not be
affected significantly.

Thanks,


--

John Devereux

[krw]

The ground island my help here but it's not the reason to do a ground
island. I almost always use the powdered metal type of inductors
(similar to Vishay IHLP series) but also ground islands to keep the
loop currents local to the SMPS. Only the input and output capacitors
and low-side switch/inductor get connected to this local plane,
though. The feedback, compensation, and any "analog" ground pins
connect to the over-all ground. The ground connections of the input
and output capacitors should be close to each other on this island,
too. Most SMPS recommended layouts follow this strategy.

[John Devereux]

krw <k...@nowhere.com> writes:

> On Sat, 07 Nov 2015 13:10:43 +0000, John Devereux

[...]

>>I think if I build the whole SMPS section on a local ground "island"
>>over the overall groundplane this will work OK. Any magnetic leakage
>>through the island is only going to be at lower frequencies. As you
>>pointed out my "Q" measurement indicates that efficiency should not be
>>affected significantly.
>>
> The ground island my help here but it's not the reason to do a ground
> island. I almost always use the powdered metal type of inductors
> (similar to Vishay IHLP series)

Nice, they don't seem to have the air gap problem (a bit like the
multilayer chip inductors I mentioned, windings totally surrounded by
ferrite). Rather more expensive than the "glued bobbin in a ferrite
sleeve" type but I will get some in.

> but also ground islands to keep the loop currents local to the SMPS.

I appreciate this is the usual reason, but I think the "ground island"
should also work nicely both to stop any residual field and to eliminate
the effect of circulating induced currents from same.

> Only the input and output capacitors and low-side switch/inductor get
> connected to this local plane, though. The feedback, compensation,
> and any "analog" ground pins connect to the over-all ground.
> The ground connections of the input and output capacitors should be
> close to each other on this island, too. Most SMPS recommended
> layouts follow this strategy.

Thank you (and that matches the datasheet very well).


--

John Devereux

[dagmarg...@yahoo.com]

> > | |-----. | .-(+)-. +5V output

> > | | | | | C1 |
> > | '-----'-----|-|- - -|-------//
> > | .--|-|- - -|---------->

> > | | .| '-(-)-'. . . . .

> > | L1 |. | . . . . ..------->
> > | | .|. . . . . |
> > | .-----. |. | . GND . .|
> > | | | | .|. . . . . |
> > | '-----' |. | . . . . .|
> > '-------\----/--|--'. .TP1. . |

> > | | /. . . (O) . .|

> >
> >
> > TP1 is a solid through-hole metal post, connected to the GND trace and
> > to the ground plane on layer 3.
> >
> > You can see that the original ground run ran under L1, parallel
> > to L1's winding progression, creating a transformer-coupling in the topside
> > GND above TP1 (a solid metal grounded test post at the lower bound) and
> > prior to the bulk filter cap C1's negative terminal.
> >
> > So, no doubt there was significant coupling into that trace despite a
> > solid plane two layers deeper.
> >
> > That rendered C1 (and additional bypasses) ineffective at r.f. The
> > top-right portion of the GND trace supplied a ribbon connector, providing
> > a path for wicked radiation and conduction of the induced GND signal.
>
> I guess the very low impedance is what makes the cap ineffective.

My interpretation was that passing C1's ground lead under L1 created a
parasitic transformer winding coupled to L1.

L1
---. . . .---------+-------+--->+5V
'-'-'-' | |
=======~~~=== --- C1 --- C2..n
.-. --- ---
GND trace ~>| | | |
under L1 | '---+-------+---> GND' to
| connector
(o) TP1
|
===
GND plane, layer 3

Since C1's ground now carries the same signal meant to be filtered from +5V,
it (and bypasses [C2..n]) cannot function as intended to attenuate +5V ripple
and spikes.

+5V and GND', which are cabled off-board, are then laden with said spikes,
hence massive RFI.

> > The revised layout routed layer 1's GND well around L1 rather than under,
> > and nailed it to layer 3's ground plane at both ends of the run, and in-
> > between.
> >
> > The replacement inductor(*) was also a drum cemented into a ferrite box
> > like the first unit, however, the new unit's down-facing gap was tighter,
> > and filled with what appeared to be a ferrite-loaded cement. (The original
> > unit had a larger downward-facing gap, air-filled, IIRC.)
>
> I suppose I need to buy a few types.
>
> > * Bourns SRR1240-150M
> >
> > The improved gap configuration wasn't obvious on the datasheet--I had to
> > get a sample in hand.
> >
> > So, I can't say for sure that an inductor directly over a solid ground plane
> > would be problematic from this experience.
>
> Thank you very much for the detailed write-up James.
>
> > I can say there was significant transformer coupling due to fringe flux
> > from the downward-facing gap.
>
> I think if I build the whole SMPS section on a local ground "island"
> over the overall groundplane this will work OK. Any magnetic leakage
> through the island is only going to be at lower frequencies. As you
> pointed out my "Q" measurement indicates that efficiency should not be
> affected significantly.
>
> Thanks,

Sure. Even if it's possibly ultimately okay I'm still not 100% comfortable
with the prospects of inducing large currents in a ground plane, both for
signal integrity and loss (efficiency) reasons.

--quote--

>Hey you beat me to it, thanks. I found ~3% with a SRU1048 22uH. If I
>drill a 10mm hole under it the change is about 1.5%. So it makes a bit
>of a difference but does not eliminate the effect.
>
>This was all at 500kHz.
>
>The indicated Q went from 24 to 22 as it was brought up to the plane.

--/quote--

I read that too quickly, mistook 3 *per cent* reduction in L for 3uH out
of 22uH. d(L)=3% is quite a bit smaller than Q dropping from 24 to 22
(a drop of 9%), so there may indeed be significant added loss, I'm not sure.

Inductor loss is just a fraction of total SMPS loss, but big enough that
I wouldn't want to increase it unnecessarily.

Cheers,
James Arthur

[krw]

On Sun, 08 Nov 2015 09:17:18 +0000, John Devereux
<jo...@devereux.me.uk> wrote:

>krw <k...@nowhere.com> writes:
>
>> On Sat, 07 Nov 2015 13:10:43 +0000, John Devereux
>
>[...]
>
>>>I think if I build the whole SMPS section on a local ground "island"
>>>over the overall groundplane this will work OK. Any magnetic leakage
>>>through the island is only going to be at lower frequencies. As you
>>>pointed out my "Q" measurement indicates that efficiency should not be
>>>affected significantly.
>>>
>> The ground island my help here but it's not the reason to do a ground
>> island. I almost always use the powdered metal type of inductors
>> (similar to Vishay IHLP series)
>
>Nice, they don't seem to have the air gap problem (a bit like the
>multilayer chip inductors I mentioned, windings totally surrounded by
>ferrite). Rather more expensive than the "glued bobbin in a ferrite
>sleeve" type but I will get some in.

Our volumes are typically in the millions, so the cost delta isn't a
lot, if any. We usually get the top-tier volume price on any value in
the series because they're used so much. Vishay is pricey but there
are others in the market now. For smaller devices (under .5" or so)
Toko/Murata is much cheaper. Delta/Cyntec is also in this market.
There are probably several others now.

These parts have soft saturation characteristics, too. One doesn't
have to worry as much about saturating the core in an overload, for
instance. They're really easy parts to work with.

Toko/Murata also makes MLC inductors (2012 and 2520) in powdered
metal. Nice parts!

>
>> but also ground islands to keep the loop currents local to the SMPS.
>
>I appreciate this is the usual reason, but I think the "ground island"
>should also work nicely both to stop any residual field and to eliminate
>the effect of circulating induced currents from same.

Sure. The idea is to keep the loop area small but the island also
keeps any voltage induced in the ground to a minimum.

>
>> Only the input and output capacitors and low-side switch/inductor get
>> connected to this local plane, though. The feedback, compensation,
>> and any "analog" ground pins connect to the over-all ground.
>> The ground connections of the input and output capacitors should be
>> close to each other on this island, too. Most SMPS recommended
>> layouts follow this strategy.

I forgot to mention that the island and the larger plane usually get
connected under the chip, particularly if there is a power pad. This
helps power dissipation, too.

>
>Thank you (and that matches the datasheet very well).

I had to go out to lunch, on our TI FAE, many times to learn why
things are done the way they are. ;-)

[RBlack]

On Fri, 6 Nov 2015 11:19:49 -0800 (PST), (dagmarg...@yahoo.com)

Thanks for the write-up.
The inductor I am using is a different construction again from the two
you have mentioned, the windings are on a core which goes parallel to
the PCB surface, which is then cemented into a ferrite box. The cross
section looks a bit like:

+----------------+
| |
| +------------+ |
| | | |
| | oooooooooo | |
| +------------+ |
| |
| +------------+ |
| | oooooooooo | |
| | windings | |
| | | |
+-+ +-+
-------------------------
PCB
-------------------------

So the gap, such as it is, is the (presumably) lower-mu material which
joins the core rod to the inside of the box. I'd expect this
construction to have a significant downwards fringe field as well, due
to the assymmetry.

Another poster has mentioned the Vishay IHLP series, I will check these
out. I'm hoping a different inductor design will help here, having to
re-spin this PCB will not make me popular...

[dagmarg...@yahoo.com]

On Tuesday, November 10, 2015 at 4:08:48 AM UTC-5, RBlack wrote:

> Thanks for the write-up.
> The inductor I am using is a different construction again from the two
> you have mentioned, the windings are on a core which goes parallel to
> the PCB surface, which is then cemented into a ferrite box. The cross
> section looks a bit like:
>
> +----------------+
> | |
> | +------------+ |
> | | | |
> | | oooooooooo | |
> | +------------+ |
> | |
> | +------------+ |
> | | oooooooooo | |
> | | windings | |
> | | | |
> +-+ +-+
> -------------------------
> PCB
> -------------------------
>
> So the gap, such as it is, is the (presumably) lower-mu material which
> joins the core rod to the inside of the box. I'd expect this
> construction to have a significant downwards fringe field as well, due
> to the assymmetry.

Man, that's ugly. That's not an inductor, that's a rail gun / field
projector!

Here's a cross section of the units I used:
.--..-------..--.
|.-'|_ _|'-.|
|| oo| |oo ||
|| oo| |oo ||
|| oo| |oo ||
|| oo| |oo ||
|| oo| |oo ||
|'-..-' '-..-'|
'--''-------''--'

I guess they can't do this...
.--..-------..--.
|.-'|_ _|'-.|
|| oo| |oo ||
|| oo| |oo ||
|| oo| |oo ||
|| oo| |oo ||
|| oo| |oo ||
|| .-' '-. ||
|| |_______| ||
|'-------------'|
'---------------'
....because they need some way to get the leads out the bottom.

> Another poster has mentioned the Vishay IHLP series, I will check these
> out. I'm hoping a different inductor design will help here, having to
> re-spin this PCB will not make me popular...

If you posted a part number, the intrepid s.e.d. crowd-sourcing crew might
conjure up a substitute...

Cheers,
James Arthur

[RBlack]

On Tue, 10 Nov 2015 09:04:44 -0800 (PST), (dagmarg...@yahoo.com)
said:

> On Tuesday, November 10, 2015 at 4:08:48 AM UTC-5, RBlack wrote:
> > On Fri, 6 Nov 2015 11:19:49 -0800 (PST), (dagmargoo...@yahoo.com)
> > said:
>

[snip]

The part is Würth Elektronik 7443320330 - 3.3 uH, 17 amps, 4.4
milliohms, in a cubic SMD package roughly 12mm on a side.

Würth are normally my go-to supplier of magnetics, I will have to cast
the net a bit wider this time I think.

[Bill Sloman]

If you need a tighter spec on leakage inductance, get somebody to wind something for you. The people who make ferrites also make formers onto which you can wind what you want.

There aren't as many coil-winding shops around as there used to be - in part because somebody in China will always under-cut them - but it's not a particularly expensive option.

EPCOS has particularly nice manuals and application notes.

http://en.tdk.eu/tdk-en/1190522/products/product-catalog/ferrites-and-accessories/epcos-ferrites-and-accessories

--
Bill Sloman, Sydney

[John Devereux]

Is leakage inductance the same thing as the magnetic leakage we are
discussing?

> There aren't as many coil-winding shops around as there used to be -
> in part because somebody in China will always under-cut them - but
> it's not a particularly expensive option.

Maybe it's possible, but we are talking about rather small, surface
mount parts here.

> EPCOS has particularly nice manuals and application notes.
>
> http://en.tdk.eu/tdk-en/1190522/products/product-catalog/ferrites-and-accessories/epcos-ferrites-and-accessories

--

John Devereux

[Bill Sloman]

> >> The part is Würth Elektronik 7443320330 - 3.3 uH, 17 amps, 4.4
> >> milliohms, in a cubic SMD package roughly 12mm on a side.
> >>
> >> Würth are normally my go-to supplier of magnetics, I will have to cast
> >> the net a bit wider this time I think.
> >
> > If you need a tighter spec on leakage inductance, get somebody to wind
> > something for you. The people who make ferrites also make formers onto
> > which you can wind what you want.
>
> Is leakage inductance the same thing as the magnetic leakage we are
> discussing?

No. I was using the wrong term. I really should have said magnetic leakage . The magnetic leakage of interest is that bit of the fringing field that can induce current in a convenient ground plane.

Leakage inductance is what you've got left when you short one winding on a transformer - or put on a shorted turn.

dV1/dt= L1.dI1/dt + M.dI2/dt

dV2/dt= L2.dI2/dt + M.DI1/dt

where M is the mutual inductance of L1 and L2. In and ideal transformer M=L1.L2^0.5 and the extent to which it falls short of that is the leakage inductance.

It's flux which doesn't thread all of both coils so it really isn't the same idea.

> > There aren't as many coil-winding shops around as there used to be -
> > in part because somebody in China will always under-cut them - but
> > it's not a particularly expensive option.
>
> Maybe it's possible, but we are talking about rather small, surface
> mount parts here.

I haven't been playing close attention to what EPCOS and FerroxCube are selling at the moment, but they certainly did have some surface mount formers when I last looked.

I don't know whether they would be small enough for the OP's job.

A quick look at the element 14 range of 324 cores reminded me that OP could go for a printed winding, in the PCB, and glue the core halves onto either side of the board

http://www.farnell.com/datasheets/1719196.pdf

At least he'd know exactly where his windings were.

<snip>

--
Bill Sloman, Sydney

[dagmarg...@yahoo.com]

http://www.digikey.com/product-detail/en/7443320330/732-2141-2-ND/2175557

Yow! (But two of those on opposite sides of a PCB might make a nice isolation
transformer. :-)

> Würth are normally my go-to supplier of magnetics, I will have to cast
> the net a bit wider this time I think.

Close, but close enough?
http://www.digikey.com/product-detail/en/7443551370/732-1127-1-ND/1639188

This sort of thing really bears measurement. Your original part's geometry
looks terrifying, but measurement might show it's completely fine. Or not.

Cheers,
James Arthur

[John Larkin]

On Wed, 11 Nov 2015 07:42:29 -0800 (PST), dagmarg...@yahoo.com
wrote:

That first (furst?) Wurth part is interesting. They wind the wire on
some sort of rod, then stick it sideways into the square ferrite box.
It is sort of shielded.

At rated current, 17 amps, the L is down about 40%, which means the
shielding won't be so good there. And the temperature is up 50C!

John

[John Devereux]

Speaking of measurement...

I switched my 4192A to show the series resistance part of the impedance
and did some more measurements.

SRU1038 22uH, using no-leads (leaded component test fixture).

no plane Groundplane
f L R L R

10k 20.41 0.082 20.30 0.091
50k 20.30 0.151 19.91 0.231
100k 20.19 0.319 19.67 0.439
200k 19.99 0.79 19.41 0.930
500k 19.52 2.59 18.93 2.73
1M 19.10 5.36 18.50 5.56
2M 18.90 9.80 18.29 10.10

Not sure what to make of these TBH



--

John Devereux

[RBlack]

On Wed, 11 Nov 2015 07:42:29 -0800 (PST), dagmarg...@yahoo.com
wrote:

> On Wednesday, November 11, 2015 at 5:27:46 AM UTC-5, RBlack wrote:
> > On Tue, 10 Nov 2015 09:04:44 -0800 (PST),
(dagmarg...@yahoo.com)
> > said:
> > > On Tuesday, November 10, 2015 at 4:08:48 AM UTC-5, RBlack wrote:
> > > > On Fri, 6 Nov 2015 11:19:49 -0800 (PST),
(dagmargoo...@yahoo.com)
> > > > said:
> > >
> > [snip]
> >
> > > >
> > > > Thanks for the write-up.
> > > > The inductor I am using is a different construction again

from the tw=

> o
> > > > you have mentioned, the windings are on a core which goes

parallel to=

> > > > the PCB surface, which is then cemented into a ferrite box.

The cros=

> s
> > > > section looks a bit like:
> > > >
> > > > +----------------+
> > > > | |
> > > > | +------------+ |
> > > > | | | |
> > > > | | oooooooooo | |
> > > > | +------------+ |
> > > > | |
> > > > | +------------+ |
> > > > | | oooooooooo | |
> > > > | | windings | |
> > > > | | | |
> > > > +-+ +-+
> > > > -------------------------
> > > > PCB
> > > > -------------------------
> > > >
> > > > So the gap, such as it is, is the (presumably) lower-mu

material whic=

> h
> > > > joins the core rod to the inside of the box. I'd expect this
> > > > construction to have a significant downwards fringe field as

well, du=

check the=

> se
> > > > out. I'm hoping a different inductor design will help here,

having t=

> o
> > > > re-spin this PCB will not make me popular...
> > >
> > > If you posted a part number, the intrepid s.e.d. crowd-sourcing

crew mi=

> ght
> > > conjure up a substitute...
> > >
> > > Cheers,
> > > James Arthur
> > >
> >
> > The part is Würth Elektronik 7443320330 - 3.3 uH, 17 amps, 4.4
> > milliohms, in a cubic SMD package roughly 12mm on a side.


>
http://www.digikey.com/product-detail/en/7443320330/732-2141-2-ND/21755
57


> Yow! (But two of those on opposite sides of a PCB might make a nice

isolati=
> on
> transformer. :-)

Funnily enough I was going to try that, I have a spectrum analyser
with tracking generator. Switching frequency is 370 kHz - Anyone care
to lay bets what the insertion loss will be? :)

> > Würth are normally my go-to supplier of magnetics, I will have to

cast=

> > the net a bit wider this time I think.


> Close, but close enough?
>
http://www.digikey.com/product-detail/en/7443551370/732-1127-1-ND/16391
88


> This sort of thing really bears measurement. Your original part's
geometry
> looks terrifying, but measurement might show it's completely fine.
Or not.

Watch this space...

[dagmarg...@yahoo.com]

ISTM the furst is the wurst. Most of the flux goes thru the rod and up
the box thru the top, but potentially a great deal goes down the box walls to
the bottom rim of the box.


^ >>>>>>>> v
^ v
^ oooooo v
+ <<<<<<<< +
v oooooo ^
v ^
v ))))(((( ^

The air gap at the bottom looks like an induction heater, with a low-z shunt
across the top magnetic circuit.

The field is 90 degrees compared to vertically-wound inductors; I'm
still mulling whether that's better for minimizing circulating currents
coupling-wise or not... I think containing the flux in the first place
is more important / matters more.

> At rated current, 17 amps, the L is down about 40%, which means the
> shielding won't be so good there. And the temperature is up 50C!

Cheers,
James Arthur

[dagmarg...@yahoo.com]

The apparent resistance is increased by the following factors:
50K 1.52
100K 1.38
200K 1.17
500K 1.05

I expect those are at least in part real increases in dissipation caused by
stray flux inducing circulating currents in the ground plane.

Both inductor and ground plane e.s.r.s will show skin effect at higher
frequencies. I suspect we're seeing that too, and that the inductor's
added e.s.r. dominates over ground plane losses at higher frequencies (because
the inductor winding path length is sooo much longer than the ground plane's
circulating current's path length).


Cheers,
James

[RBlack]

On Wed, 11 Nov 2015 21:35:55 +0000, RBlack (ne...@rblack01.plus.com)
said:

> On Wed, 11 Nov 2015 07:42:29 -0800 (PST), dagmarg...@yahoo.com
> wrote:
> > On Wednesday, November 11, 2015 at 5:27:46 AM UTC-5, RBlack wrote:

[snip]

> > > The part is Würth Elektronik 7443320330 - 3.3 uH, 17 amps, 4.4
> > > milliohms, in a cubic SMD package roughly 12mm on a side.
>
>
> >
> http://www.digikey.com/product-detail/en/7443320330/732-2141-2-ND/21755
> 57
>
>
> > Yow! (But two of those on opposite sides of a PCB might make a nice
> isolati=
> > on
> > transformer. :-)
>
> Funnily enough I was going to try that, I have a spectrum analyser
> with tracking generator. Switching frequency is 370 kHz - Anyone care
> to lay bets what the insertion loss will be? :)
>
>

> > This sort of thing really bears measurement. Your original part's
> geometry
> > looks terrifying, but measurement might show it's completely fine.
> Or not.
>
>
> Watch this space...
>

OK, this is how I set up the test, two of the parts facing each other
across a piece of (mostly) bare FR4:

http://www.rblack01.plus.com/images/IMG_20151113_093946933.jpg
http://www.rblack01.plus.com/images/IMG_20151113_093915919.jpg
http://www.rblack01.plus.com/images/IMG_20151113_093930839.jpg

(the holes on 0.1" pitch are not plated through)

SMB connector to each one, I mirror-imaged everything to keep the E-
field coupling down as best I could. One part connected to the spectrum
analyser input, the other to the TG output, set to 0dBm.

Results:

100 kHz - 1 MHz :
http://www.rblack01.plus.com/images/Inductor-leakage-spectrum-lo.png
1 - 10 MHz :
http://www.rblack01.plus.com/images/Inductor-leakage-spectrum-mid.png
10 - 100 Mhz:
http://www.rblack01.plus.com/images/Inductor-leakage-spectrum-hi.png

Not much to see at low frequencies, but a nasty big peak at 40ish MHz,
which corresponds to the SRF of the part. We have a big, broad
emissions peak at 65 MHz, I suppose the presence of the ground plane in
the real product could be shifting the resonance.

Thanks everyone for the suggestions for alternative parts, I will try
some of them and post back here in a week or two.

[Jan Panteltje]

On a sunny day (Fri, 13 Nov 2015 11:09:04 -0000) it happened RBlack
<ne...@rblack01.plus.com> wrote in
<MPG.30afd52fe...@reader80.eternal-september.org>:

>OK, this is how I set up the test, two of the parts facing each other
>across a piece of (mostly) bare FR4:
>
>http://www.rblack01.plus.com/images/IMG_20151113_093946933.jpg
>http://www.rblack01.plus.com/images/IMG_20151113_093915919.jpg
>http://www.rblack01.plus.com/images/IMG_20151113_093930839.jpg
>
>(the holes on 0.1" pitch are not plated through)
>
>SMB connector to each one, I mirror-imaged everything to keep the E-
>field coupling down as best I could. One part connected to the spectrum
>analyser input, the other to the TG output, set to 0dBm.
>
>Results:
>
>100 kHz - 1 MHz :
>http://www.rblack01.plus.com/images/Inductor-leakage-spectrum-lo.png
>1 - 10 MHz :
>http://www.rblack01.plus.com/images/Inductor-leakage-spectrum-mid.png
>10 - 100 Mhz:
>http://www.rblack01.plus.com/images/Inductor-leakage-spectrum-hi.png
>
>Not much to see at low frequencies, but a nasty big peak at 40ish MHz,
>which corresponds to the SRF of the part. We have a big, broad
>emissions peak at 65 MHz, I suppose the presence of the ground plane in

> real product could be shifting the resonance.

Nice pictures.

The basic solution to prevent magnetic coupling is:
'put the coils at right angles to each other'.
that would require a small extra rising PCB in this case.
But the second rule:
'as far away from each other as possible'.
So why put these on opposite sides of the board?
As you mention adding a ground-plane it seems you are making a new layout.
In that case move the coils further away from each other.

[RBlack]

Yes, the parts were deliberately placed in the 'poorest' configuration.
This was a test to see how magnetically 'leaky' this particular inductor
design is. The answer seems to be 'very' - I will repeat the test on
the replacement inductors proposed upthread, once I have samples of
them.

[Jan Panteltje]

On a sunny day (Fri, 13 Nov 2015 14:36:59 -0000) it happened RBlack
<ne...@rblack01.plus.com> wrote in
<MPG.30b005e84...@reader80.eternal-september.org>:

Ha, my misunderstanding
Yes coupling coils this way makes a good rotary transformer,
say 2 potcore halves...