Today, in a race for efficiency, we have moved from transformers to switching power supplies. Almost all PSUs were designed for single-phase low-voltage operation (220Vac/310Vdc in my country). I've never ever seen 380V 3-phase 3+ kW ATX PSUs for PCs despite their efficiency and lower ripple noise. They would be very useful for stacks of GPUs. I think that it is mainly because electrolytic capacitors cannot survive rectified 660Vdc.

And it could be even better to rectify a 10kV medium voltage line, as it usually comes to the village transformer. But what is the voltage limit silicon devices (MOSFETs) can survive without breaking down?

  • 13
    \$\begingroup\$ Do you think there is much of a market for 380V, 3-phase, 3kW ATX power supplies? Is it possible that no one builds these supplies because there aren't enough buyers, rather than because of fundamental technical limitations? \$\endgroup\$ – Elliot Alderson Sep 11 '18 at 15:15
  • 9
    \$\begingroup\$ I would assume that not many people have 3-phase outlets readily available in their house. Those who do, may as well buy specialized equipment that might not come in an ATX form factor. \$\endgroup\$ – ilkkachu Sep 11 '18 at 17:47
  • 5
    \$\begingroup\$ @xakepp35 Requiring a high current output on a particular rail does not mean you require a three-phase PSU. You can easily get 1200W and even 1500W single-phase PSU's that are better than 95% efficient in the standard 80-90% load envelope,which is more than sufficient for devices that actually use an ATX form factor PSU. \$\endgroup\$ – Austin Hemmelgarn Sep 11 '18 at 18:15
  • 3
    \$\begingroup\$ A three-phase supply for device as low-powered as 3kW seems completely pointless. The standard wall sockets in UK houses can already supply 3kW each (240V 13A single phase) and there are 6 of those sockets in the room where I'm typing this comment! I assume that other countries' standard domestic electrical codes are similar. \$\endgroup\$ – alephzero Sep 11 '18 at 19:28
  • 2
    \$\begingroup\$ In the UK we typically have our sockets on 32A 240V circuits, so we rarely have to worry about putting too much on one circuit. Other countries tend to have lower rated socket circuits. \$\endgroup\$ – Peter Green Sep 11 '18 at 20:32

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: -

enter image description here

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: -

enter image description here

Then you build a monument to the gods of Megavolt by stacking the trays like so: -

enter image description here

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: -

enter image description here

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.

enter image description here

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.

  • \$\begingroup\$ I didn’t know what you meant by optically coupled (or what obvious reasons applied), so I did some reading on Wikipedia, which explained the situation well enough. However, I note that the Wikipedia article, though it does note the use and advantages of optical coupling, suggests that it’s still not common and electrical coupling is still more the norm. Would that article be out of date, then? Or maybe the 8 kV versions are the ones that are getting the optical coupling? \$\endgroup\$ – KRyan Sep 11 '18 at 22:03
  • \$\begingroup\$ @KRyan for sure, optically coupled LEDs are the reserve of the high voltage rated types I mentioned in my answer. \$\endgroup\$ – Andy aka Sep 12 '18 at 6:17
  • \$\begingroup\$ Ah, I spoke unclearly: Wikipedia was talking about “HVDC”—is 8 kV particularly high even in that category? \$\endgroup\$ – KRyan Sep 12 '18 at 13:20
  • 1
    \$\begingroup\$ I can answer for one of them. We stopped using optically coupled thyristors in the 80s. \$\endgroup\$ – winny Sep 12 '18 at 15:32
  • 1
    \$\begingroup\$ @Barleyman No, that won't work - think about it or draw it to see the problems. At the very least, the most active phase (the one that is peaking most positively) will reverse bias the bridges of the other two phases meaning there can be no current from them for a significant period of the cycle and this means that PF correction cannot work. \$\endgroup\$ – Andy aka Sep 12 '18 at 15:36

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).

  • \$\begingroup\$ Hmm, i was primarily interested in CPU\GPU supplies, it seems they need somewhere around 1 volt and infinetely many amps (more you have=more chips you can empower). So does there exists some.. device, that could convert 10kV 1amp to 1v 10kAmp? \$\endgroup\$ – user197437 Sep 11 '18 at 15:35
  • 1
    \$\begingroup\$ What would you put between 10kV input and your 10000 GPUs? A transformer 10k->380? Or does some powerful 10 kilovolt PSU exists? \$\endgroup\$ – user197437 Sep 11 '18 at 15:37
  • \$\begingroup\$ The 10KV would be stepped down likely 2 or more times to get to 400V or so, that could then be rectified into DC and switching power supplies take it down further as needed. \$\endgroup\$ – CrossRoads Sep 11 '18 at 15:53
  • 6
    \$\begingroup\$ 1v 10kA power supply is more like a large spot welder than anything sensible to have inside a computer. Given the resistive losses it seems sensible to distribute power at 240V and downconvert as close to the point of use as possible. \$\endgroup\$ – pjc50 Sep 11 '18 at 16:56
  • 2
    \$\begingroup\$ @xakepp35 the biggest PC supplies I've seen from a reasonably reputable maker are 2KW (ex FSP). I've seen marginally higher no-name units from China but wouldn't want to be in any building where one is energized. Probably >90% of their output is ultimately delivered around 1V, but within a few PCB inches of the consuming chips for obvious reasons. I wouldn't be surprised if some large blade enclosures/rack PDUs could hit 10kW, although I think they normally use a higher intermediate DC voltage than the 12V of ATX. \$\endgroup\$ – Dan Neely Sep 11 '18 at 18:26

They are actually building solid state transformers with greater efficiency and control, these run at 7.2kV

The workhorse switch of power electronics, the silicon-based insulated-gate bipolar transistor (IGBT) is a better fit. These devices have been used to build SSTs for rail applications in Europe. And they are certainly faster. But the most rigorous commercial devices can withstand voltages up to only about 6.5 kilovolts. While this breakdown voltage is perfectly fine for a range of power applications, it isn’t sufficient to handle the electricity that flows through distribution transformers; in the United States, a typical voltage at the low end of the spectrum is 7.2 kV.

They are using silicon carbide which has a bigger bandgap and is more tolerant to heating problems also:

Fortunately, silicon is not the only option. In the last 10 years, great strides have been made in the development of switches based on compound semiconductors—silicon carbide in particular. Silicon carbide has a range of attractive properties that stem from its large bandgap—the energy hurdle that must be overcome to switch from insulator to conductor. Silicon carbide’s bandgap is 3.26 electron volts to silicon’s 1.1 eV, which means the material can be exposed to significantly higher electric fields and temperatures than silicon can without breaking down. And because this compound semiconductor can withstand much higher voltages, power transistors built from it can be made more compact, which in turn allows them to switch much faster than their silicon-based counterparts. A faster switching speed also cuts down on energy loss, so silicon carbide transistors can carry more current for a given thermal budget.

Sources: https://spectrum.ieee.org/energy/renewables/smart-transformers-will-make-the-grid-cleaner-and-more-flexible

  • \$\begingroup\$ I find that hard to believe (7 kV). Aren't they stacked devices with each junction rated at 1200 V? \$\endgroup\$ – Peter Mortensen Sep 11 '18 at 17:59
  • \$\begingroup\$ AFAIK European high-speed trains run at 25kV. \$\endgroup\$ – MSalters Sep 12 '18 at 10:46
  • \$\begingroup\$ 25 kV is the supply voltage; it is transformed on the trains to around 1500V. \$\endgroup\$ – Michael Harvey Sep 12 '18 at 12:46
  • \$\begingroup\$ @PeterMortensen I believe the source article attempts to address that point (between the two paras quoted here). I'm not qualified to judge its validity. \$\endgroup\$ – Dan Neely Sep 12 '18 at 15:48

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. enter image description here enter image description here enter image description here

TI also has great design info on their IGBT's

  • 4
    \$\begingroup\$ Thanks! Could you provide a link to model name, or at least name the product series? \$\endgroup\$ – user197437 Sep 11 '18 at 16:27
  • 3
    \$\begingroup\$ do you think you can research yourself \$\endgroup\$ – Tony Stewart Sunnyskyguy EE75 Sep 11 '18 at 16:29
  • \$\begingroup\$ (-1 for the snarky "just google it" comment in defense of a really low effort answer) \$\endgroup\$ – dn3s Sep 17 '18 at 4:17
  • \$\begingroup\$ @dn3s I usually put in more effort than the question , it's not snarky , it's training learners how to be less dependent google.com/… \$\endgroup\$ – Tony Stewart Sunnyskyguy EE75 Sep 17 '18 at 6:01
  • \$\begingroup\$ that's not the kind of training this website is designed to provide. stackexchange's goal is to be a reference, not a forum. Anyways, glad to see you've improved your answer, i've removed my -1 (although it still seems more like an advertisement for a specific product while other answers more directly address the question itself) \$\endgroup\$ – dn3s Sep 17 '18 at 16:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy