How to choose capacitors for filter decoupling?

What is filter decoupling? 

Simply put, it stores energy when the chip  does not need current, and replenish energy in time chip when needs  current.

You may think that this responsibility is for DCDC or LDO? Yes, they can  be done at low frequencies, but high-speed digital systems are  different.

Let's take a look at the capacitor first. The function of the capacitor  is simply to store charge. We all know that a capacitor should be added  to the power supply to filter, and a 0.1uF capacitor is placed on the  power supply pin of each chip to decouple.

Why the capacitors next to the power pins of some board chips are 0.1uF or 0.01uF. Is there anything to pay attention to?

To understand this, it is necessary to understand the actual  characteristics of the capacitor. The ideal capacitor is just a storage  of energy, that is, C.

The actual manufactured capacitors are not so simple. When analyzing  power integrity, the commonly used capacitor models are shown in the  figure below.

In the figure, ESR is the series equivalent resistance of the capacitor,  ESL is the series equivalent inductance of the capacitor, and C is the  real ideal capacitor.

ESR and ESL are determined by the manufacturing process and materials of  the capacitor and cannot be eliminated. What impact will these two  things have on the circuit?

ESR affects the ripple of the power supply, and ESL affects the filter frequency characteristics of the capacitor.
We know that the capacitive reactance of the capacitor Zc=1/ωC, the  inductive reactance of the inductor Zl=ωL, (ω=2πf), the complex  impedance of the actual capacitor is Z=ESR+jωL-1/jωC=ESR+j2πfL-1/j2πfC.

It can be seen that when the frequency is very low, the capacitor works,  and when the frequency is high, the effect of the inductance cannot be  ignored. When the frequency is higher, the inductance takes the leading  role and the capacitor loses its filtering effect.
So remember, the capacitor is not a simple capacitor at high frequency.  The filter curve of the actual capacitor is shown in the figure below:

As mentioned above, the equivalent series inductance of the capacitor is  determined by the manufacturing process and material of the capacitor.  The ESL of the actual SMD ceramic capacitor ranges from a few tenths of  nH to a few nH. The smaller the package, the smaller the ESL.

From the filter curve of the capacitor above, we can also see that it is  not flat. It is like a 'V', which means that it has frequency selection  characteristics. Sometimes, we hope that it is as flat as possible  (previous board-level filtering)， sometimes it is hoped that the sharper  the better (filtering or notching).

What affects this characteristic is the quality factor Q of the  capacitor, Q=1/ωCESR. The larger the ESR, the smaller the Q and the  flatter the curve. On the contrary, the smaller the ESR, the larger the Q  and the sharper the curve.

Usually tantalum capacitors and aluminum electrolysis capacitors have  relatively small ESL, but large ESR, so tantalum capacitors and aluminum  electrolysis have a wide effective frequency range, which is very  suitable for the board level filtering for power supply.

That is, in the input stage of DCDC or LDO, a larger capacity tantalum capacitor is often used for filtering.

And put some 10uF and 0.1uF capacitors close to the chip for decoupling,  ceramic capacitors have very low ESR. Whether we put 0.1uF or 0.01uF  close to the pins of the chip, the following are listed for your  reference:

So, don't put 0.1uF capacitors in every circuit in the future. In some  high-speed systems, these 0.1uF capacitors won't work at all.