> A related trick is to put ferrite beads on MCU output lines
You must be extremely, extremely careful with this. In general, ferrite beads are hard to apply correctly and great care is required if you want them to work out. Do it wrong and you'll probably be making things worse, instead of helping. See for example:
https://incompliancemag.com/article/ferrites-to-kill-ringing...
https://incompliancemag.com/article/all-ferrite-beads-are-no...
>The datasheets ... The manufacturer is
Honestly, the datasheets are usually just wrong. The writer is an intern and the material is cargo-cult copy-pasted. If it works it's by accident, or because it wasn't critical in the first place. Read the datasheet, understand what it's trying to do, then go ahead and achieve that end in the most sensible manner.
> It is true that at very high frequencies — hundreds of megahertz — the capacitor’s residual inductance becomes a limiting factor. At that point, combining multiple different capacitors can offer somewhat better wideband noise suppression.
Nope. It's still wrong and dumb to do this with MLCCs unless you have simulated the hell out of the whole thing. You should practically never parallel different values of MLCC. Instead follow EMC wizard Henry Ott's advice: pick the smallest package you're willing to deal with, then the largest capacitor you're willing to pay for in that package, and just use that everywhere. This is called "big V" decoupling by Ott and decoupling master Bruce Archambeault and it is not the best way to do things, but every better way is much, much, much harder to do. "Big V" will work for everything that doesn't involve underfilled BGAs, and even most of those.
> tantalum polymer
I actually kind of hate these guys, I don't find much use for them outside DC-DC converter output capacitors. They have too little ESR to damp things that need it, too much ESL for high frequency use, and are just too expensive for general use. They're not bad or anything, they just don't really have a sweet spot. MLCCs plus a few cheap high-ESR aluminum electrolytics (often found these days as the high-temp long-life parts) is a really effective combination. Maybe a few solid tantalums for intermediate bypass if appropriate.
> Y5V
Thank the heavens that these are basically extinct. Good riddance.
Okay, enough comments, you might then ask, how the hell do I decouple things in production designs?
First, put one bulk capacitor, minimum, on every rail. Aluminum electrolytic if the rail leaves the board ever, maybe tantalum if it doesn't (or maybe not, solid tantalums have... reputations). My go-to series is Rubycon YXM or YXJ for through-hole electrolytics, Nichicon UCB/UCW or Chemi-Con MLE/MLF for surface-mount, and AVX TAJ for tantalums. These can physically be located anywhere.
Then sprinkle down one 1uF 0402 per part for the small parts, or one per power pin for the big digital chips. Place these at the power pins, no exceptions. Things like MCU analog rails usually don't need ferrites but might get pi filter type structures. It depends on how important they are, really. If it's a big or dense board, toss in another tantalum or two physically near the chips or chip clusters to help keep the electrolytics honest.
You can decouple 500MHz processors and pass radiated EMC at Class B with this approach. It's not hard to do, it's cheap (but not cheapest, this isn't going to get you to Shenzhen-special COGS), and it works great.