Filtering with bypass capacitors: how to check ICs' switching frequencies?

Question

When adding by-pass capacitors to a design, I have read that it is not enough to just place the typical 0.1 μF value and the 10 μF one; you have to check why you need those values. I mean, which are the frequencies in your system, in your PCB, that can disturb the design.

My design would be something like:

Signal 🡆 Voltage follower (op-amp) 🡆 DAQ

In the power supply lines of my op-amp I have to add by-pass capacitors to filter two noises: the one that comes from the switching frequency of the power line (I checked in the datasheet that it is 500 kHz) and the 'switching' frequency that comes from sampling time (DAQ).

In spite of this, I am at a bit of a loss in term of choosing good by-pass capacitors (capacitance, dielectric and package) and specially in how to check switching frequencies or frequencies to filter in the datasheets of the components of my circuit (which term do I have to search in the op-amps', DAQ and other components). Filtering for 500 KHz, for example, and checking real curves impedance-freq does not give a clue of which capacitor I have to use (they all filter for higher ones).

Apart from this, if I am filtering in order to achieve a stable behaviour of the buffer, would I need an RC circuit in the output before the DAQ? Which would be the difference?

I cannot find either a capacitor that could filter properly in my frequencies (500 kHz). Here it is a plot from KEMET capacitor simulator from one example.

Answer 1

Actually, that's a bit misnormer. Sure the frequencies of the system is important if you want to bypass at that frequency. But that's not all.

For example, if you take a simple square wave generator circuit. Even if you have it running at 0.1 Hz, that's really not the frequency you have to worry about as it's almost DC.

However even if the edge of the square wave happens very rarely, the edge of the square wave can be very fast. In theory the mathematical edge is infinitely fast so it requires infinite amount of energy in no time at all to switch.

In real life let's assume it can happen in 1 nanosecond. If you have a modest amount of wire and it drives some inputs, the chip would have to drive say 50pF of capacitance from 0V to 5V in 1 nanosecond. It needs a current pulse of 250mA for 1ns to achieve that. So that's a lot of current needed and it is needed fast, meaning there's high frequencies.

The point of getting that large current means that the wiring and capacitor must have low resistance, as resistance limits how much current is available.

The fact that the current must rise to 250mA in much less than 1ns means the wiring and capacitor must have low inductance, as inductance limits how quickly the current is available.

This gives a rough understanding why a 100nF plus 10uF capacitor may not have any effect at high frequencies, because you can't fit large capacitors close to chip, and you need capacitors with small physical size to reduce inductance, and small physical size caps also come with smaller capacitances, which means smaller capacitors (both in size and capacitance) are better at higher frequencies.

So, cap types.

If you want very large capacitance for bulk storage, it might be electrolytic, but those are not good at high frequencies, or cold.

If you want a simple bypass cap, large and small capacitances are available in many Class 2 dielectrics, such as X7R. They just have large tolerances, large temperature coefficiet, and even large voltage coefficient. A 10 uF 50V rated capacitor may have 10 uF capacitance at 5V, but it may have only 5uF at 40V.

Class 1 dielectrics are most stable and have good tolerances, but are small, from few pF to few nF. But for making analog filters or crystal load caps, you definitely want NP0 type.

Sure, other types of caps exist, such as plastic film, low ESR electromytics, polymer electrolytics, tantalum, and they all have their applications, but also each cap type have their own problems or quirks.