How (Not) to Decouple High-Speed Operational Amplifiers

[Ruach]

A good article for bedtime reading:

http://focus.ti.com/lit/an/sloa069/sloa069.pdf

[DSP_Geek]

Good app notes are great to find; TI is worth mining these days, much like National Semiconductor was in the days of disco. They still have Bob Pease's [1] columns on their site:

http://www.national.com/rap/

[1] You want to know this guy if you're doing opamps.

[Geoff-AU]

that first link is another reminder that the more I learn, the more I realise there is to know..

[Guido Tent]

Originally posted by Ruach
A good article for bedtime reading:

http://focus.ti.com/lit/an/sloa069/sloa069.pdf

Nice reading but from page 12 onwards the keyboard started slipping......(NEVER use power planes as these resonate as well, similar to capacitors).

Pitty they don't pay attention to tyhe supply impedance, but only the decoupling path (as some may know, I like ferrite beads).

best

[Extreme_Boky]

A good article for bedtime reading:

http://focus.ti.com/lit/an/sloa069/sloa069.pdf

Nice read

Thanks,
Extreme_Boky

[bocka]

Tuning series resonance with a cap for decoupling purpose is a very bad design practice. Hopefully the guys have measured and build once the things (as I did) they're suggesting. I completely disagree with them. Using very small caps in the pF range for power supply blocking definitely does not work. The impedance of such small caps is some ohms (at about 1 GHz) to some hundred or 1000 ohms (about 1 to 10MHz) in the relating frequency area. As well as the recurrent advisings of using different caps for different frequencies. The impedance of a cap for f >> 1/(2*Pi*sqrt(Le * Co)) is deteminated by its parasitic inductance only:

Z = jwLe

For a typical SMD 0603 chip capacitor Le is about 0.4nH, giving an impedance of about 0.3ohms @100MHz and 3ohms @ 1GHz. If someone would increase this impedance by adding an additional capacitive component, go on.

If f << 1/(2*Pi*sqrt(Le * Co)) the impedance is

Z = 1 / jwCo

with C1 = 30pF and C2 = 100nF the larger cap gives 70dB more noise filtering than the smaller one. Only if f = 1/(2*Pi*sqrt(Le * Co)) small band decoupling is sufficiant. Improving power supply decoupling can only be done with lowering the parasitic inductance of the caps, i.e. using smaller cases like 0402 or 0201.

This leads to very simple rules:

1. Use the smallest possible case
2. Use the highest possible value for decoupling
3. Minimize the area for the current loop.
4. Use power supply tiles

I won't get starting a discussion for power supply transmission lines here, but for high frequencies in the GHz range explaining theory with lumped elements is adventurous...

[bocka]

Something I've forgotten: I've the designers have designed their chips wrong (like using pins 4 and 8 for an SOIC-8 case), nothing can be done for improving power supply decoupling, as the intrinsic impedance of the power supply loop is deteminated by its dimensions. For best decoupling results power supply pins MUST be located in direct proximity.

[Guido Tent]

Originally posted by bocka
For best decoupling results power supply pins MUST be located in direct proximity.

Hi

Agree in the case you want to minimise the voltage across these pins. Some (big) chips have on-chip decoupling.....

best

[maxlorenz]

Improving power supply decoupling can only be done with lowering the parasitic inductance of the caps,

What do you think (or what have you mesured) about BG's
super-E-cap configuration? That is non polar BG in antiparallel arrangement. They claim supression of inductance through mutual cancelation.

My limited experience has been favorable, soundwise. (safe one occasion in wich I got ringing on PS decoupling going from 100nf to 100+100nf in superEcap). Why I go into things that I don't understand anyway

[jan.didden]

You can expect lower ESL and ESR by paralleling, that is standard electrical engineering. But cancellation? That is hard to swallow!

As an example, ESL in wirewound resistors is cancelled to some 80 or 90% by bifilar winding, i.e. winding with two wires close together and then connecting them in opposite polarity. That appears not possible in caps.

Jan Didden