Help: ground plane under microcontroller.

[Ge0rge Marutz]

Can anyone recommend a good tech note explaining the merits of keeping
ground plane under a micro? Perhaps this article could demonstrate
examples of good and bad design. How much copper is enough? Where do
you need it, what does it have to cover? Does it need to be continuous
or can a trace run through it, etc.

For those of you who care why I ask (warning, I get wordy)...

I have an embedded design based on a 4 layer board. It has a single
micro clocked with a internal RC oscillator running at 1MHZ, 4 analog,
and 4 digital I/O. I've stuck with what I know to be proper and
recommeded design practices.

Top copper layer
http://img402.imageshack.us/my.php?image=topcopper7pl.jpg

Ground plane
http://img402.imageshack.us/my.php?image=groundplane9yl.jpg

This board has passed EMC test by a large margin when tested under
SAE J1113, GM, Volvo, and Ford standards.

I have a new customer who, for some unknown reason, is critical of my
design. His only problem with it is the fact that there is not a
continuous copper ground fill under the micro in the top layer of the
board. Traces and vias under the part break it up.

I tried to ensure him that this was acceptable by providing evidence
that the design has passed a number of manufactures EMC requirements,
is manufacturable, and has shown zero failures with over 60,000
products in the field. This is not good enough for him. He says
unless I can come up with a good reason why my grounding strategy will
work for him, I will have to redesign the board to meet his request.
He does not care, or understand, that I have a full ground plane
immediately underneath the top layer.

I can't for the life of me get my point accross. What makes it even
more difficult is that he admittedly knows very little about
electronics himself. So, I can't talk him through this. He is a high
level manager with a background in mechanical engineering. Someone
within his organization told him that my design would not work and he
believed them. They offered no explaination why they were concerned or
exactly what they were looking for.

By no means an I saying my design is perfect. I am a humble person
just trying to do the best I can. However, his concerns appear to be
unjustified. This is frustrating. I can't just change the board to
make him happy. There are a number of reasons why this can't happen.
Mostly due to timeline. We want to launch the product in two months.
I can't make all the necessary changes, build protoypes, and go through
a full EMC test in that timeframe.

Any words of advice? I would be deeply appreciative.

Ge0rge

 

[Tim Shoppa]

Can anyone recommend a good tech note explaining

the merits of keeping ground plane under a micro?

Not really, because it's not that critical. Power pins should be well
bypassed, and if you are clocking any high-current outputs through the
I/O ports you will want to make sure that you can source/sink those
currents without too much ground or Vcc bounce. Ground planes help
with all these but are not magic bullets by themselves.

Things can get hairy if you've got high-resolution analog ins/outs on a
microcontroller but often these drive you to split the ground plane
rather than requiring one big one.

Without any extenuating circumstances (fast clocked outputs, high
current loads being switched simultaneously) you probably don't need a
ground plane at all.

Tim.

 

[John Woodgate]

I read in sci.electronics.design that Ge0rge Marutz

Any words of advice? I would be deeply appreciative.

Walk away, if that's not suicidal. The irrational client stays
irrational no matter what. He obviously doesn't respect your technical
knowledge and that may extend to not paying up when you've satisfied his
whims. 'Oh, we gave you a lot of technical help with the ground plane,
so we're only prepared to pay you 70% of the fee.'

As a matter of fact, NOT walking away may be suicidal. I just now walked
away from one who wanted me to make a device that gave 30 dB more output
than is required. I explained that the source would not deliver enough
power to do it, even if it was necessary. No dice. 30 dB he wanted, and
30 dB he would have. Not from me!

 

[Ge0rge Marutz]

I wish it was that simple John. I don't think my management would
allow me to step away from a potential $30million in sales because of
my frustration. I know there is a way to get through to this guy. I
just need the right mix of data to back up my findings (presented in
laymans terms) and a sales guy willing to kiss ass. I already have
numerous sales guys willing to do that and more . Now I need to come
through on my end.

Tim mentioned things to watch out for:

*Proper bypass of power supply pins - check, got it.

*Not switching any high currents in main section of board. Have
isolated power and ground returns for high current switching.

*No high res analog. Just rough 8 bit measurements.

*Using low speed clock (2Mhz) with proper slew limiting on outputs to
reduce sharp clock edges and minimize harmonics.

etc...

Ge0rge

 

[Joerg]

Hello George,

He does not care, or understand, that I have a full ground plane
immediately underneath the top layer.

He probably isn't able to understand that having a full ground plane
underneath (a very good thing, BTW) is virtually the same as another
ground right on top of that. Understandable, considering that he is a
ME. Probably he relies on advice from someone else whom he trusts and
that advice might not be correct. If a friend of yours would say that
the hydraulics in a certain piece of machinery you want to buy from this
ME aren't optimal you'd probably be skittish as well.

Anyway, it sounds like the old adage "the customer is always right". If
he wants a ground there and the NRE for relayout are justified by the
order he places, I'd just do it. If someone wanted a purple circuit
board as a condition because it supposedly makes it perform better, I'd
do it as well. As long as they sign on that dotted line ;-)

Regards, Joerg

 

[Frank Bemelman]

Top copper layer
http://img402.imageshack.us/my.php?image=topcopper7pl.jpg

Ground plane
http://img402.imageshack.us/my.php?image=groundplane9yl.jpg

Any words of advice? I would be deeply appreciative.

Well, you have my deepest sympathy. If it is of any help, I
have rated both your pictures at imageshack as 10, no less.

But it is up to you, you either give him what he wants, or
you don't.

 

[Frank Bemelman]

"John Woodgate" <[email protected]> schreef in bericht

[snip]

As a matter of fact, NOT walking away may be suicidal. I just now walked
away from one who wanted me to make a device that gave 30 dB more output
than is required. I explained that the source would not deliver enough
power to do it, even if it was necessary. No dice. 30 dB he wanted, and
30 dB he would have. Not from me!

Let me guess, a hearing aid for JT.

 

[Spehro Pefhany]

Any words of advice? I would be deeply appreciative.

Ge0rge

This is one of those situations where there is some tiny difference,
but it's not really a rational request.

Can you re-order the layers in the board and get essentially the same
(tested) board with the ground plane visible on the top and the signal
layer buried? Will that satisfy him?


Best regards,
Spehro Pefhany

 

[Ge0rge Marutz]

HAH!!! Joerg, you hit the nail on the head. This customer fully
believes that any other color circuit board besides BLUE will not work.
So, I agreed to supply blue circuit boards instead of green.

I offered to make the design change once the small volume product
launch is underway and BEFORE high volumes kick in. Its going to take
4 months to complete the design cycle using the most agressive timing
possible. I have 2 months. He will not buy into this.

I understand he took comments by a trusted collegue. I do not blame
him a bit for this. I just need to find a way to get through to him.
There is more than one way to skin a cat. One method will work just as
good as another in this application. Butter knife vs scalpel.

Ge0rge

 

[John Woodgate]

I read in sci.electronics.design that Ge0rge Marutz

I wish it was that simple John. I don't think my management would
allow me to step away from a potential $30million in sales because of
my frustration.

Hence my remark about suicide. But I also mentioned the 'technical
advice' scam. Be prepared to settle for 21 million.

You could call his bluff. Just hit about walking away. 'Gosh, if this
guy is prepared to walk away from 30 million, maybe he's got a point.'

 

[KoKlust]

First of all: DO NOT promise things that you don't believe in. It is an
excellent recipe for disaster. The fact that there are mega-$$$ in the game
should only make you more cautious, as you are about to take a huge
responsibility (and liability...) .
I agree completely with John Woodgate on this.

Having said that, there is little time left for creativity.
You might agree with the customer that you shift a number of tests from
'before' to 'during' the small volume phase.

But then you say you need 4 months design cycle for this minor layout
change. Why?
- If there is money to spend, and there is a good supplier network,
production can be speeded up considerably.
- Focus on testing the deltas. For the EMC testing, first repeat the
near-field magnetic radiation tests. I guess that this test is most likely
to give different results. A comparative pre-compliance test can be carried
out on the bench, with a (homebrew) near field probe and a sensitive
oscilloscope.

Good luck,

Marco

 

[Joerg]

Hello George,

HAH!!! Joerg, you hit the nail on the head. This customer fully
believes that any other color circuit board besides BLUE will not work.
So, I agreed to supply blue circuit boards instead of green.

ROFL...

Seriously, once I was asked for purple, for molded parts. To me it
looked plain ugly but the client got purple.

I offered to make the design change once the small volume product
launch is underway and BEFORE high volumes kick in. Its going to take
4 months to complete the design cycle using the most agressive timing
possible. I have 2 months. He will not buy into this.

Four months seems a bit high. But maybe you are in a heavily regulated
field like med. Anyway, when clients insist on unrealistic goals I bow
out. Usually they come back with a bloody nose after they found someone
who said yes, couldn't deliver, a major tradeshow or other milestone was
missed and someone got chewed out at the board meeting.

I understand he took comments by a trusted collegue. I do not blame
him a bit for this. I just need to find a way to get through to him.
There is more than one way to skin a cat. One method will work just as
good as another in this application. Butter knife vs scalpel.

Sometimes an evening in town together helps. It can build trust.

Regards, Joerg

 

[Pooh Bear]

Can anyone recommend a good tech note explaining the merits of keeping
ground plane under a micro? Perhaps this article could demonstrate
examples of good and bad design. How much copper is enough? Where do
you need it, what does it have to cover? Does it need to be continuous
or can a trace run through it, etc.

For those of you who care why I ask (warning, I get wordy)...

I have an embedded design based on a 4 layer board. It has a single
micro clocked with a internal RC oscillator running at 1MHZ, 4 analog,
and 4 digital I/O.

I've had micros running faster than that work fine on single-sided pcbs !

Graham

 

[Pooh Bear]

What makes it even
more difficult is that he admittedly knows very little about
electronics himself. So, I can't talk him through this. He is a high
level manager with a background in mechanical engineering. Someone
within his organization told him that my design would not work and he
believed them. They offered no explaination why they were concerned or
exactly what they were looking for.

You just summed it up. A know-nothing PHB. His arse is exposed since his
equally badly informed coleague has drawn his atttention to a non-problem
and he now feels he has to cover said arse.

I'd simply go to *his boss* and tell him the guy doesn't know what he's
doing. But then I don't stand for shit.

Graham

 

[Pooh Bear]

I wish it was that simple John. I don't think my management would
allow me to step away from a potential $30million in sales because of
my frustration. I know there is a way to get through to this guy. I
just need the right mix of data to back up my findings (presented in
laymans terms) and a sales guy willing to kiss ass. I already have
numerous sales guys willing to do that and more . Now I need to come
through on my end.

Tim mentioned things to watch out for:

*Proper bypass of power supply pins - check, got it.

*Not switching any high currents in main section of board. Have
isolated power and ground returns for high current switching.

*No high res analog. Just rough 8 bit measurements.

*Using low speed clock (2Mhz) with proper slew limiting on outputs to
reduce sharp clock edges and minimize harmonics.

None of the above need a ground plane either.

Philips had some interesting stuff on EMI in a microcontroller handbook.

The real biggy wrt EMC is loop area. They had some interesting worked
examples for consumer stuff that even avoided screened enclosures.

I'll see if I can find the app note numbers.

Graham

 

[Pooh Bear]

"John Woodgate" <[email protected]> schreef in bericht

[snip]

As a matter of fact, NOT walking away may be suicidal. I just now walked
away from one who wanted me to make a device that gave 30 dB more output
than is required. I explained that the source would not deliver enough
power to do it, even if it was necessary. No dice. 30 dB he wanted, and
30 dB he would have. Not from me!

Let me guess, a hearing aid for JT.

There's none as deaf as those who don't wish to hear.

Graham

 

[Pooh Bear]

Hello George,


He probably isn't able to understand that having a full ground plane
underneath (a very good thing, BTW) is virtually the same as another
ground right on top of that. Understandable, considering that he is a
ME. Probably he relies on advice from someone else whom he trusts and
that advice might not be correct. If a friend of yours would say that
the hydraulics in a certain piece of machinery you want to buy from this
ME aren't optimal you'd probably be skittish as well.

Anyway, it sounds like the old adage "the customer is always right". If
he wants a ground there and the NRE for relayout are justified by the
order he places, I'd just do it. If someone wanted a purple circuit
board as a condition because it supposedly makes it perform better, I'd
do it as well. As long as they sign on that dotted line ;-)

I've never done a *purple* board ! I expect someone makes a resist in that
colour though.

Red looks quite wacky too. Done a few of them.

Hmmm - BLACK ! I bet the audiofools would love that ! Maybe with gold
legending.

Damn - that's given me an idea now. ;-)

Graham

 

[Joerg]

Hello Graham,

I've never done a *purple* board ! I expect someone makes a resist in that
colour though.

It's all a matter of money and they'll do any color you want. Remember
when you could have your car delivered in a color that matched your
wife's hair? I just wonder what happened to those cars in cases of divorce.

Regards, Joerg

 

[[email protected]]

HAH!!! Joerg, you hit the nail on the head. This customer fully
believes that any other color circuit board besides BLUE will not work.
So, I agreed to supply blue circuit boards instead of green.

Two possibilities.

1) The guy really doesn't know what's going on, and is taking advice
from someone who doesn't know either.

2) The guy has already selected a competitive product (because he gets
a better kickback or his brother works there or whatever) and is
looking for a reason to give the work to them instead of you. The
blue board was his first attempt, and when you got around that, he's
trying this. This could backfire, but you could go to him and say
you think you've worked out a way to make the ground plane change
in time, and ask if there's anything else. If another goofy reason
pops up, politely run the other way.

OTOH, for thirty megadollars, you can get creative. Tell him you are
going to paint the bottom of the micro with a metallic paint to serve as
a ground plane, and add a contact for it to the board, or run a line of
paint over to the ground lead of the micro, or whatever. After the
board is stuffed, have somebody in production use a paint pen to draw a
silver line on the bottom outside edge of the micro - "overspray" from
when the bottom was coated.

Matt Roberds

 

[riscy]

What interesting that while course taught us about electronics, maths
and so on, but never covered EMC and ground plane. Most datasheet and
application weblinks (ie analog device) make larger issue on ground
plane and that where we pickup up assuming analog device is right
becuase they experts on high specification op-amp, interface, DSP,
microcontroller.

I tend to find thier rules relatively generic for many reason and does
not alway successfully solve problem. One thing clear in term of physic
that larger sheet of copper has much low ESR than the trace, especially
important for decoupling caps, this absorb the gound (or PSU ripples)
bounce.

For precision analogue signal, you have to watch carefully how much
current is flowing on reference trace (ground or 2.5V) prior to the ADC
stage. Any current flow in trace will cause interference signal to
appear on ADC (unless there is common mode) due to ESR of the trace.

Having ground plane is way of isolating the electric field emission
between trace, this is positive benefit. That perhap why they seperates
analogue and digital ground. But it so damn hard to choose where to
place star ground link AGND and DGND(!).

For 8 bit resolution ADC, you may get away without needing ground
plane.

You can experiment between those two board layout, one with ground
plane and one without and then show it to your boss that they wrong(!).
I found experiment could effectively twist thier arms, because they
have no other proof that they wrong.

Good luck

Riscy