What is the maximum voltage that silicon can handle?
You can get 8 kV rated (at several thousand amps) thyristors for use in HVDC converters. The gate is optically coupled for the obvious reasons and also because, when used in tandem on HVDC links, the gate driving speed differences between series connected thyristors is important and optical is a little bit more clear cut speed-wise: -
Stack a few together in a tray with the various extras you need to control them safely (snubbers etc) and you get one of these: -
Then you build a monument to the gods of Megavolt by stacking the trays like so: -
Notice the little guy at the bottom.
Regarding power I've read that it takes 40 grams of silicon to control 20 MW of power and a lot of these installations are literally a thousand MW or more.
And it could be even better to rectify 10kV medium voltage line, as it usually come to village transformer.
Ah but you don't get safe isolation that is reliable - one breakdown and 10 kV in your house wiring is not good. Plus, the break-even point on a HVDC link versus a regular AC link is many, many miles.
Where are 3-phase 380v to 12V PSUs?
Well there is a technical snag that is inherent to the circuit used for many years in the "standard" 3 phase rectifier circuit: -
The problem is how they switch and power factor correction. In the good old days nobody cared but these days PF and supply cleanliness is paramount in many countries. And this is the problem with the standard 3 phase rectifier - it cannot be PF corrected because diodes can't conduct from 0 volts through to 0 volts (throughout one-half cycle) because of the blocking effect of the other phases and their diodes. The pulsing current taken from the 3 phase supply is really bad.
The solution is to use three single phase (and PF corrected) supplies all contributing power to a common DC bus. So, the modern 3 phase switching supply is in fact three single phase supplies.
How do the HVDC thyristors do it you might ask? They use filters as big as small houses to quench the harmonics generated.
Notice the relative size of the harmonic filters compared to the "valve hall" where all the thyristor "valves" are. All manner of double and single tuned filters are used just to remove those harmonics and, if the same technique were used on more ordinary standard 3 phase switching supplies (the ones that will never meet modern legislation) then guess what; the cost of the filtering is more than the added cost of individual supplies with PF correction built in.
Could you provide a link to model name, or at least name the product series?
Infineon thyristor discs rated at up to 8 kV and 4800 amps.
But what is voltage limit silicon keys (mosfets) can survive without breaking through?
There's virtually no limit; if your voltage exceeds the breakdown voltage of a component, well, put two in series.
There's silicon semiconductor-based rectifiers for high voltage DC power transfer. These work around 800 kV or higher.
Still, it'd be stupidly expensive to try to use multiple kV as input to a power supply that in the end generates voltage three orders of magnitude smaller. Also, it's incredibly dangerous to handle multiple kV within home installations, to plain impossible (isolation can easily get thicker than cable openings).
Mitsubishi IGBT hybrids with FET input BJT outputs can now switch Megawatts and very high voltage 15kV and are also used in smart power inverters and 600V GTI’s in arrays for redundancy to smaller GTI's such as Huawei's 2000S 50kW units.
Below is a Mitsubishi hybrid IGBT which has many patents for exceptionally high switching energy and extremely low internal driver ESL and ESR. (inductance and resistance) I believe they working on their 8th generation now.
TI also has great design info on their IGBT's