Maker Pro
Maker Pro

Digital and Analog power decouplings

L

linnix

Jan 1, 1970
0
Considering the three uCs:

1. Lmi - no decouplings
2. Avr - power decouplings
3. Nxp - power and ground decouplings

Question #1:
Can I assume that Lmi has internal decouplings between Digital and
Analog circuits?

Question #2:
Is ground decoupling (in addition to power) really necessary for 10
bits Analog?

Question #3:
If I have to choose one, would it be better to:
1. Power to Digital and filter to Analog
2. Power to Analog and filter to Digital
 
J

Joerg

Jan 1, 1970
0
linnix said:
Considering the three uCs:

1. Lmi - no decouplings
2. Avr - power decouplings
3. Nxp - power and ground decouplings

Question #1:
Can I assume that Lmi has internal decouplings between Digital and
Analog circuits?

I don't know this series but how would they internally decouple? On a
chip you can't create any caps to write home about. I guess they assume
that the circuit designer maintains good bypassing.

Question #2:
Is ground decoupling (in addition to power) really necessary for 10
bits Analog?

No. I never split ground unless it must be done for safety reasons such
as patient isolation. But this is my own humble opinion and almost
guaranteed to set off a huge political debate within the next few
milliseconds ;-)

Question #3:
If I have to choose one, would it be better to:
1. Power to Digital and filter to Analog
2. Power to Analog and filter to Digital

#1 is the prevalent method. But it depends a bit on what consumes the
most. So if you had analog stuff that uses 500mA and one uC that
consumes 2mA then it would be #2, of course. Actually, I am doing that
right now on a circuit that is 95% analog but needs a bus interface that
consumes very little. On this one the analog rails must be super squeaky
clean.
 
L

linnix

Jan 1, 1970
0
I don't know this series but how would they internally decouple? On a
chip you can't create any caps to write home about. I guess they assume
that the circuit designer maintains good bypassing.

Unless they have very good fab process to build L & C on chip.
Or they didn't think it was necessary.
Or a design oversight.
This could potentially push me over to the NXP.
No. I never split ground unless it must be done for safety reasons such
as patient isolation. But this is my own humble opinion and almost
guaranteed to set off a huge political debate within the next few
milliseconds ;-)


#1 is the prevalent method. But it depends a bit on what consumes the
most. So if you had analog stuff that uses 500mA and one uC that
consumes 2mA then it would be #2, of course. Actually, I am doing that
right now on a circuit that is 95% analog but needs a bus interface that
consumes very little. On this one the analog rails must be super squeaky
clean.

Thanks.
 
J

Joerg

Jan 1, 1970
0
linnix said:
Unless they have very good fab process to build L & C on chip.


L and C on a chip both cost oodles of real estate and for decoupling
purposes they would be prohibitively expensive. Usually the only L and C
you can find on chips would be on RF circuits in the GHz range and then
you'd be talking about very few nanohenries and a few picofarads (but
usually even femtofarads).

Or they didn't think it was necessary.


Well, I wouldn't expect stand-alone performance from an AD converter on
a uC.

Or a design oversight.


Most certainly not. Typically more of a compromise because people often
want the lowest possible pin count and also because a larger package
will cost more.

This could potentially push me over to the NXP.



Thanks.

Always welcome.
 
L

linnix

Jan 1, 1970
0
L and C on a chip both cost oodles of real estate and for decoupling
purposes they would be prohibitively expensive. Usually the only L and C
you can find on chips would be on RF circuits in the GHz range and then
you'd be talking about very few nanohenries and a few picofarads (but
usually even femtofarads).


Well, I wouldn't expect stand-alone performance from an AD converter on
a uC.

Without power filtering, 10 bits might be meaningless.
Most certainly not. Typically more of a compromise because people often
want the lowest possible pin count and also because a larger package
will cost more.

But there are four power pins and four ground pins, they could have
spare one for analog power. Avr has 2 digital power pins, 1 analog
power pins and 3 ground pins.
 
J

Joerg

Jan 1, 1970
0
linnix said:
Without power filtering, 10 bits might be meaningless.

With good bypass caps I've seen 12 bits work just fine.
But there are four power pins and four ground pins, they could have
spare one for analog power. Avr has 2 digital power pins, 1 analog
power pins and 3 ground pins.

Then they could indeed have split the power pins. Who knows, maybe the
converter is in an area on the chip where it wouldn't have helped much.
But that's hard to say without mask pictures.

[...]
 
E

Eeyore

Jan 1, 1970
0
Joerg said:
I don't know this series but how would they internally decouple? On a
chip you can't create any caps to write home about. I guess they assume
that the circuit designer maintains good bypassing.


No. I never split ground unless it must be done for safety reasons such
as patient isolation. But this is my own humble opinion and almost
guaranteed to set off a huge political debate within the next few
milliseconds ;-)

It's virtually essential in audio gear where clocked digital circuitry is also
present. You simply can't allow the digital supply current to flow in your nice
clean audio ground.

Graham
 
J

Joerg

Jan 1, 1970
0
Eeyore said:
Joerg wrote:




It's virtually essential in audio gear where clocked digital circuitry is also
present. You simply can't allow the digital supply current to flow in your nice
clean audio ground.

Even there I never did it and I have designed rather noise sensitive
Doppler systems where you mix right down to baseband for I/Q audio
detection. I did encounter a few system that others had designed with
split grounds. Most didn't perform. Others miraculously did (not perfect
but the marketeers liked it enough) but then failed the EMI cert, big time.
 
R

Rich Grise

Jan 1, 1970
0
It's virtually essential in audio gear where clocked digital circuitry is
also present. You simply can't allow the digital supply current to flow in
your nice clean audio ground.

Just from lurking here for some years, I've come to realize that the
consensus seems to be that split planes aren't needed, and sometimes
more trouble than they're worth.

What's needed is reasonable care when designing the power circuitry. Call
it "intelligent design", maybe. :) What you need to do is watch where
your return current paths go, and see to it that they don't actually share
any copper (or as little as possible).

Sometimes just hanging out can be an education. :)

Cheers!
Rich
 
K

krw

Jan 1, 1970
0
linnix wrote:


L and C on a chip both cost oodles of real estate and for decoupling
purposes they would be prohibitively expensive. Usually the only L and C
you can find on chips would be on RF circuits in the GHz range and then
you'd be talking about very few nanohenries and a few picofarads (but
usually even femtofarads).

L's cost oodles. C's are done on chip all the time. All of our
whitespace was filled with decoupling C's. Of course caps are added
to the packages too. When you're switching many amps...

The analog PLLs had separate AVdd pins too.
 
J

Joerg

Jan 1, 1970
0
Rich said:
Just from lurking here for some years, I've come to realize that the
consensus seems to be that split planes aren't needed, and sometimes
more trouble than they're worth.

What's needed is reasonable care when designing the power circuitry. Call
it "intelligent design", maybe. :) What you need to do is watch where
your return current paths go, and see to it that they don't actually share
any copper (or as little as possible).

Sometimes just hanging out can be an education. :)

Especially right before the next elections :)))
 
J

Joerg

Jan 1, 1970
0
krw said:
L's cost oodles. C's are done on chip all the time. All of our
whitespace was filled with decoupling C's. Of course caps are added
to the packages too. When you're switching many amps...

But you won't likely get enough pF to write home about. When a bunch of
comparators are clanging around in an ADC there needs to be some serious
capacitor close by to take the brunt.

Once when I redesigned a pre-amp on an RF chip I could have really,
really used a fraction of a pF to get rid of a spike. So I went to the
guy that was in charge of chips at that client and politely asked
whether I can have that little corner lot over there. "Nope".

The analog PLLs had separate AVdd pins too.

That is a smart thing to do.
 
J

Jim Granville

Jan 1, 1970
0
linnix said:
Considering the three uCs:

1. Lmi - no decouplings
2. Avr - power decouplings
3. Nxp - power and ground decouplings

Question #1:
Can I assume that Lmi has internal decouplings between Digital and
Analog circuits?

Question #2:
Is ground decoupling (in addition to power) really necessary for 10
bits Analog?

Question #3:
If I have to choose one, would it be better to:
1. Power to Digital and filter to Analog
2. Power to Analog and filter to Digital

Filter to the lower energy/smaller node, which normally means option 1.


When you say decouplings, do you mean Separate Analog/Power Pins ?

What's needed depends also on if you use the internal Vref, or external
Vref, or do relative to Vcc.

A well specified device will give Power Supply Rejection ratios, for
the analog sections.
Some spec differing LSB errors, on the physical pin locations -
expect the channels nearest GND pins to be better.

More important, is does the vendor specify a MAX, or only typicals?
( and at what Vcc's )
The better suppliers will spec their ADCs properly, the companies that
'add the tickbox', have lower standards.

-jg
 
L

linnix

Jan 1, 1970
0
Filter to the lower energy/smaller node, which normally means option 1.

When you say decouplings, do you mean Separate Analog/Power Pins ?

Yes, need separate pins to have external filter.
What's needed depends also on if you use the internal Vref, or external
Vref, or do relative to Vcc.

No external Vref either.
A well specified device will give Power Supply Rejection ratios, for
the analog sections.
Some spec differing LSB errors, on the physical pin locations -
expect the channels nearest GND pins to be better.

More important, is does the vendor specify a MAX, or only typicals?
( and at what Vcc's )
The better suppliers will spec their ADCs properly, the companies that
'add the tickbox', have lower standards.

I think they just forgot to hire an analog consultant before making
the chip.
 
K

krw

Jan 1, 1970
0
But you won't likely get enough pF to write home about. When a bunch of
comparators are clanging around in an ADC there needs to be some serious
capacitor close by to take the brunt.

Lotsa nFs (don't remember exactly) in reverse biased junctions. The
slew rate of the comparitors will be less so you would need quite a
bit more C.
Once when I redesigned a pre-amp on an RF chip I could have really,
really used a fraction of a pF to get rid of a spike. So I went to the
guy that was in charge of chips at that client and politely asked
whether I can have that little corner lot over there. "Nope".

There was no white space on the chip. Any whitespace we had was
automatically wired as decoupling. Warnings were generated if C got
below some threshold.
That is a smart thing to do.
Learned (and relearned) the hard way. Some parts called out a low-
pass filter on AVdd. I could never figure out why each part was
different (other than different engineers).
 
V

Vladimir Vassilevsky

Jan 1, 1970
0
I have to agree with Eeyore. The common current paths and the ground
loops can be a big problem in the audio. It requires careful attention,
and it doesn't take much to ruin the performance. One solid ground plane
can't be used as the power and signal return and the zero reference at
the same time, unless all of the signals are differential.

Even there I never did it and I have designed rather noise sensitive
Doppler systems where you mix right down to baseband for I/Q audio
detection. I did encounter a few system that others had designed with
split grounds. Most didn't perform. Others miraculously did (not perfect
but the marketeers liked it enough) but then failed the EMI cert, big time.

There is a technique to place the most noisy (or the most sensitive)
parts of the schematic on the isolated island of the ground plane. All
connections are done through the one bridge to this island. Thus the
EMI, the noise currents and the ground bounce are restricted to the
area. I had to do it that way a couple of times; however this technology
is the overkill in most of the practical cases, and it can do more harm
then benefit if used unproperly.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com
 
V

Vladimir Vassilevsky

Jan 1, 1970
0
John Larkin wrote:

I do NMR gradient amps with nanovolt shunt amps, uP, fpga's, switching
regs, display drivers, and power output stages all on a single ground
plane. Noise is a few PPM from 0 to 50 KHz, and the noise is all
thermal.

This is not about what is right or wrong. This is about the numbers. The
stray impedance is in mOhm/nH range. This can or can not be a problem,
depending on many other issues at the system and board levels.
I just did a combination gradient amp and thermocouple-based
temperature controller on a single board, single ground plane, and it
worked fine first time. Passed UL/CE/FCC, too.

I have the application where I have to pick the narrowband signals from
~50nV with the dynamic range up to 150dB. There is also DSP, LCD, SMPS
and audio output. It would be difficult to accomplish that with just one
ground plane.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com
 
J

John Larkin

Jan 1, 1970
0
It's virtually essential in audio gear where clocked digital circuitry is also
present. You simply can't allow the digital supply current to flow in your nice
clean audio ground.

Graham

I do NMR gradient amps with nanovolt shunt amps, uP, fpga's, switching
regs, display drivers, and power output stages all on a single ground
plane. Noise is a few PPM from 0 to 50 KHz, and the noise is all
thermal.

I just did a combination gradient amp and thermocouple-based
temperature controller on a single board, single ground plane, and it
worked fine first time. Passed UL/CE/FCC, too.

John
 
R

Rich Grise

Jan 1, 1970
0
It's virtually essential in audio gear where clocked digital circuitry is
also present. You simply can't allow the digital supply current to flow
in your nice clean audio ground.

I do NMR gradient amps with nanovolt shunt amps, uP, fpga's, switching
regs, display drivers, and power output stages all on a single ground
plane. Noise is a few PPM from 0 to 50 KHz, and the noise is all thermal.

I just did a combination gradient amp and thermocouple-based temperature
controller on a single board, single ground plane, and it worked fine
first time. Passed UL/CE/FCC, too.
[/QUOTE]

Yeah, but you're a Leprechaun, and so have magickal powers. ;-)

Cheers!
Rich
 
Top