Switching power supply ground plane cuts (again)

[Joel Kolstad]

(I know, this is a never-ending discussion...)

I came across this article today:
http://www.analog.com/library/analogdialogue/archives/41-06/ground_bounce.html
.... and if starts out pretty well, but I'm not sure I completely buy into the
guy's later advice. Specifically, in figure 10, if he's going to insist on
making a "surgical cut" in the ground plane, it seems as though putting the
ground pour "draw bridge" beneath the high-side switch would make more sense
than placing it on the far left as shown. Secondly, without any cuts in the
ground plane whatsoever, since much of the return current will still tend to
run underneath the signal traces anyway, I suspect there might actually be
less EMI without any cuts than the way shown.

Thoughts?

This seems like a case where the guys who advocate that leaving your ground
plane completely solid makes the design's performance as good or better than
strategically cutting it up does are probably right.

---Joel

 

[Joerg]

(I know, this is a never-ending discussion...)

I came across this article today:
http://www.analog.com/library/analogdialogue/archives/41-06/ground_bounce.html
... and if starts out pretty well, but I'm not sure I completely buy into the
guy's later advice. Specifically, in figure 10, if he's going to insist on
making a "surgical cut" in the ground plane, it seems as though putting the
ground pour "draw bridge" beneath the high-side switch would make more sense
than placing it on the far left as shown. Secondly, without any cuts in the
ground plane whatsoever, since much of the return current will still tend to
run underneath the signal traces anyway, I suspect there might actually be
less EMI without any cuts than the way shown.

Thoughts?

You just brought it up: The advocates of surgical cuts often haven't
heard the hollering in a board room after the team blew the EMC test and
now the ship schedule ist toastissimo.

This seems like a case where the guys who advocate that leaving your ground
plane completely solid makes the design's performance as good or better than
strategically cutting it up does are probably right.

I belong to the faculty of solid ground plane folks

 

[John Larkin]

(I know, this is a never-ending discussion...)

I came across this article today:
http://www.analog.com/library/analogdialogue/archives/41-06/ground_bounce.html
... and if starts out pretty well, but I'm not sure I completely buy into the
guy's later advice. Specifically, in figure 10, if he's going to insist on
making a "surgical cut" in the ground plane, it seems as though putting the
ground pour "draw bridge" beneath the high-side switch would make more sense
than placing it on the far left as shown. Secondly, without any cuts in the
ground plane whatsoever, since much of the return current will still tend to
run underneath the signal traces anyway, I suspect there might actually be
less EMI without any cuts than the way shown.

Thoughts?

This seems like a case where the guys who advocate that leaving your ground
plane completely solid makes the design's performance as good or better than
strategically cutting it up does are probably right.

---Joel

For EMI, a solid plane is probably the best. Probably.

But if circulating currents in a switcher could induce sheet
potentials into a ground plane, and sensitive stuff is nearby, a
couple of strategic slits (either near the supply or near the victim)
can fence off the currents and improve things.

Something like a thermocouple front end can be located over a ground
peninsula, inside a C-shaped cut, to keep various circulating currents
out.

John

 

[Harry Dellamano]

Especially around switching power supplies, I make strategic voids in the
planes to prevent large circulating currents from wreaking havoc with
sensitive signals, such as output sense. I go as far as to void all layers
beneath non-toroid inductors in some of my power supplies. I've seen the
most bizarre behaviour from plane/signal interactions which is exacerbated
at higher switching speeds, of course.

I would rather deal with increased EMI (and manage it) rather than the
supply not working because transient induced problems from the local plane
prevents the power supply from properly operating (I've seen it).

Solid planes have advantages, but so does shaping the plane a little for
specific circumstances. As always - 'It depends'

Cheers

PeteS

Totally agree. A few mV of noise at the control chip can drive you nuts.
Cheers,
Harry

 

[HVYMETAL]

Cutting the ground plane is a very reasonable method to control where
your high-frequency currents are flowing. For high-current switching
converters it is absolutely necessary. I design this into all of my
circuits. Last one I did, it met DO-160E cat L for conducted emissions
the first time tested; no white wires, no re-layout. The best thing
for control circuitry is to isolate the plane under this circuit and
make sure all signals in and out of this area only pass over the place
where the ground planes connect. I'm talking about high power
switching here, like 50A at 200kHz. For small currents you might not
have to worry.
Basically, you do not want to create a path for high-frequency, high
current through the plane which is under sensitive control circuitry.

 

[John Popelish]

Especially around switching power supplies, I make strategic voids in
the planes to prevent large circulating currents from wreaking havoc
with sensitive signals, such as output sense.

(snip)

I also use other layer ground paths, either fills or traces,
to provide specific places for large and/or fast, local
ground currents to go. But the first goal is always to
minimize areas surrounded by such currents.

 

[PeteS]

For EMI, a solid plane is probably the best. Probably.

But if circulating currents in a switcher could induce sheet
potentials into a ground plane, and sensitive stuff is nearby, a
couple of strategic slits (either near the supply or near the victim)
can fence off the currents and improve things.

Something like a thermocouple front end can be located over a ground
peninsula, inside a C-shaped cut, to keep various circulating currents
out.

John

Especially around switching power supplies, I make strategic voids in
the planes to prevent large circulating currents from wreaking havoc
with sensitive signals, such as output sense. I go as far as to void all
layers beneath non-toroid inductors in some of my power supplies. I've
seen the most bizarre behaviour from plane/signal interactions which is
exacerbated at higher switching speeds, of course.

I would rather deal with increased EMI (and manage it) rather than the
supply not working because transient induced problems from the local
plane prevents the power supply from properly operating (I've seen it).

Solid planes have advantages, but so does shaping the plane a little for
specific circumstances. As always - 'It depends'

Cheers

PeteS

 

[Joerg]

(snip)

I also use other layer ground paths, either fills or traces, to provide
specific places for large and/or fast, local ground currents to go. But
the first goal is always to minimize areas surrounded by such currents.

That's exactly the point, to pay attention to where stuff is returned to
ground instead of chopping up a ground plane. Most of this is related to
clever component placement. Luckily I've got a layouter whom I don't
have to prime on the subject. His first placement plot is usually almost
right on the money, same for the traces he puts down.

 

[John Larkin]

That's exactly the point, to pay attention to where stuff is returned to
ground instead of chopping up a ground plane. Most of this is related to
clever component placement. Luckily I've got a layouter whom I don't
have to prime on the subject. His first placement plot is usually almost
right on the money, same for the traces he puts down.

The other thing a ground cut can do is stop thermal gradients.

Sometimes I do a rectangular slit that is bridged by small traces in
several places. It adds useful electrical resistance (to reduce sheet
currents) and thermal resistance, but is still a pretty good RF short.

John

 

[Joerg]

The other thing a ground cut can do is stop thermal gradients.

Sometimes I do a rectangular slit that is bridged by small traces in
several places. It adds useful electrical resistance (to reduce sheet
currents) and thermal resistance, but is still a pretty good RF short.

Well, those are the higher levels of sophistication and sometimes
obfuscation. Just imagine a competitor wrestling with drift and their
CEO hollering "So why can Highland Technology pull that off and you
#%^*!! are to freaking &^^@#!! for that?"

 

[PeteS]

Totally agree. A few mV of noise at the control chip can drive you nuts.
Cheers,
Harry

Careful Harry - this is a subject of Holy Wars

I do appreciate the positive feedback, however

Cheers

PeteS

 

[Fred_Bartoli]

Joerg a écrit :

Well, those are the higher levels of sophistication and sometimes
obfuscation. Just imagine a competitor wrestling with drift and their
CEO hollering "So why can Highland Technology pull that off and you
#%^*!! are to freaking &^^@#!! for that?"

No, I once have been better: jump the slit with ceramics

 

[PeteS]

(snip)

I also use other layer ground paths, either fills or traces, to provide
specific places for large and/or fast, local ground currents to go. But
the first goal is always to minimize areas surrounded by such currents.

I fully agree. There are times when clever placement can completely
eliminate the need to play with plane cuts because the circulating
currents are neatly localised.

Some very high density designs are more difficult to accomodate, though.
It can be a little tricky, if not impossible, to run a few hundred
really high speed differential pairs (on a small board) and prevent the
return currents from appearing in other circuitry without the judicious
use of some shaping of the planes.

That's not advocating separate analog and digital ground planes, though.
I've seen (in the last month alone) 5 designs done that way that have
horrible problems because of plane differentials, that I am now being
asked to fix without re-laying the boards out (I'm doing the consulting
after the fact). Ugh.

Cheers

PeteS

 

[Joerg]

Joerg a écrit :


No, I once have been better: jump the slit with ceramics

And we regularly made slits of less than 10um width _into_ ceramic ;-)

 

[Fred_Bartoli]

Joerg a écrit :

And we regularly made slits of less than 10um width _into_ ceramic ;-)

So I suppose the trick is to put the ceramics where they are not needed ;-)

 

[Joerg]

Joerg a écrit :


So I suppose the trick is to put the ceramics where they are not needed ;-)

No, this had a technical reason. But I am not at liberty to say ;-)

 

[neon]

Some times you read something from an educated idiot and wonder? First of all the ines will dissipate as the square of the distance and where do you stop? the curren is not only there on the gnd but everywhere there is flow of massive currents If he had any practical experience he would have known abaout a little secret of 3 gnd planes all connected together at smart points only one carrying the big current.

 

[John Larkin]

No, this had a technical reason. But I am not at liberty to say ;-)

High voltage PC boards sometimes have fully-routed-through slits.

Then there's Agoston Agoston's sampling head patent, where a trench is
milled into a solid metal block, and the trench itself is a 3D
slotline connecting a SRD impulse generator to two channels of
traveling-wave diode samplers.

Got it around here somewhere. Sure looks like a dead short to me.
Slotlines just defy my intuition somehow.

John

 

[John Larkin]

That's not advocating separate analog and digital ground planes, though.
I've seen (in the last month alone) 5 designs done that way that have
horrible problems because of plane differentials, that I am now being
asked to fix without re-laying the boards out (I'm doing the consulting
after the fact). Ugh.

Yet the cult lives on.

John

 

[Joerg]

Yet the cult lives on.

And it bringeth us bread and butter :-D