On this question, let's take a look at some of the answers from our peers:
Engineer A: The problem is actually... To put it bluntly, everyone uses it, or it works most of the time.
In fact, if we read more information about power integrity, we can find a lot of discussions. The frequency band of decoupling capacitor of 100nF is generally considered to be from a few MHz to 20 or 30 MHz. Together with bulk capacitor on the board, it can achieve a lower power impedance in the whole range from very low frequency to 20 or 30 MHz, which is sufficient for most applications. On the other hand, because people have formed the habit of using 100nF to decouple, such capacitors usually have a large inventory, and the purchase price is also cheap, so there is no need to be unconventional to choose other capacity values, such as 82nF, the effect may be similar, but you use this way, the purchasing colleagues will certainly find trouble for you.
On the other hand, a lot of high speed ics, especially high speed digital ics, are now integrated with high frequency decoupling capacitors, so only some relatively low frequency decoupling off the chip is enough, which is usually done with a bunch of capacitors of several hundred nF level. Moreover, the effect of the high-frequency decoupling capacitor integrated on the chip is much stronger than that on the board. The limit of the frequency band of the decoupling capacitor on the board is about 200 or 300 MHz, which mainly plays a comfort effect for the clock frequency up to hundreds of MHz now.
Other capacitors are also available. Low frequency analog circuits may use uF capacitors, PLL or clock drivers may use pF capacitors. This is a case-by-case analysis, but 100nF is fine in most cases. I recently encountered a problem that using a decoupling capacitor of 1nF works better than using a decoupling capacitor of 100nF because the frequency band needed is around 200MHz.
Engineer B: It is a rough design to put a lot of 100NF. The reasonable way is to do PDN simulation and control the impedance in all frequency bands of the power plane to meet the requirements.
100nF capacitors have a low cost, so if you don't do a detailed analysis, you can put more of them, and they can provide a low AC impedance between 1 and 100MHz, so there is generally no problem.
Regardless of the cost, the larger the capacity under the same package, the better the filtering effect is. The filtering effect of 1uF 0402 is better than that of 100nF 0402 in all frequency bands.
Engineer C: 100nF capacitors can better remove high frequency components, small capacitors usually have low ESR, low lead and internal inductance, high self-resonant frequency, and low impedance at higher frequencies. Therefore, 100nF is very important in the PCB board of high frequency digital circuit, and then as a design habit, this way has been applied to many digital circuit design, become a norm. However, in the past, when debugging many boards, this decoupling effect may not be quantified and intuitive, is a performance improvement in extreme environments.
The capacitance is not an exact value, because the situation of each circuit is different, and does not necessarily need 100nF, and the commonly used 100nF is 5% precision, the precision requirements are not high, but 100nF capacitance, packaging can be small, easy to manufacture, low cost, convenient procurement. The premise is that 100nF performs well for most digital ics.