PCB – Part 4 of X. Decoupling capacitors

Hi,

I got some questions regarding my recommendation to use 330pF in parallel with 100nF. Do they really work well together and what happen when connecting to different values in parallel?

No capacitor is an ideal component. There is an equivalent model that include capacitance, inductance and a serial and a parallel resistor. The actual values of these parameters depend on package, dielectric material and the capacitance value. So what I did to find of what is really happening when using a 330 pF 0603 capacitor closest to the IC (5mm signal length) and then 100nF (0603, X7R) with 2mm signal in parallel? I used this schematic:

DecouplingSCM

The driver impedance is set to 25 ohm. The board stackup is a four layer board with signal layer 1 and ground on layer 2. Symmetric stackup is used. This kind of stackup with FR4 material gives ~100 ohm impedance for transmission lines on layer 1. Line 1-4 in the schematic are transmission lines (signals on the PCB). C1-R1-R4-L1 is the model used for the 330pF 0603 capacitor. An AC sweep simulation with start frequency 100 kHz and stop frequency 10 GHz results looks like this:

Decoupling

We can see that the two capacitors work quite nice together. The decoupling can be said to be effective from ~1 MHz. The models used in this simulation is a bit simplified, but I think they are good enough for this example. It shall thus be noted that all parameters in of a complete design are not included, like capacitance between power and ground planes.

Do you want some further reading?

http://web.mst.edu/~jfan/slides/Archambeault1.pdf

http://www.cvel.clemson.edu/Presentation_Slides/PowerBus-Decoupling.pdf

/Anders

I got a question that it would be interesting to compare 330pF // 100nF with just 100nF. So here it is.

DecouplingDecoupling100nF

We can note two things. Between ~100-250 MHz the attenuation is less with 330pF // 100nF. Is there some kind of resonance going on? I actually don’t know… But over from 250MHz > there is an improvement. So if your design is fine with up to 100 MHz stay with 100nF. But if you use buffers/outputs that generate frequencies significantly more than 100 MHz I would go for 330pF // 100nF. An example of functions generating high frequencies are clock buffers that typically have rise/fall times of 500-700ps,  meaning frequencies of 500-700 MHz. However, there might be some optimization possible here…

/Anders

The answer on the question if there are resonance going on is yes! For more information:

http://www.electrical-integrity.com/Quietpower_files/Quietpower-2.pdf

Thanks Stefan, for the link.

/Anders

 

 

One Comment

Leave a Reply

Your email address will not be published. Required fields are marked *