Comment by Workaccount2 | Hacker News Reader

Comment by Workaccount2 | Hacker News Reader

Parent Full thread

OK, my day job is doing HV engineering, not transmission, but high energy stuff.

The author did something kind of equivalent to:

"If we scale a GPU clock to 75 Petahertz, we can make datacenters that fit in bed rooms! Here are the FLOPS calculations to prove it!"

This whole thing is so crazy I don't know where to begin. I applaud the author for jumping into a new subject, but there is _way_ more complexity here than laid out. HV is very difficult to penetrate too because there really isn't much info available online about it.

Those initial dielectric strength numbers are definitely off (maybe they used Wikipedia, which references a value from a 1920 physics book). As from what I can find fused silica has a dielectric strength around 50-100MV/m, which is taken from the AC figure and doubled to get the DC figure (which is fairly typical). Also these numbers are extrapolated, and dielectrics often have non-linear properties. The testers used to get these figures can be a little fickle, and HV is always fickle.

On top of that, in actual HV system design, you really need to be using 25% of the actual dielectric strength for any kind of reliability. So the practical strength of fused silica would ultimately be around ~20MV/m. Which pretty much kills the whole idea right there. Never mind that a single fracture or dielectric breakdown anywhere in the whole glass sheath would require the entire thing to be replaced. Spoiler: You cannot patch HV dielectrics. Trust me, I and many others have tried.

Some other hurdles would be dealing with the insane parasitics, which the author didn't even mention, but are one of if not the most limiting factor in transmission. HVDC lines can have up to 10% ripple, which for the author would be 1.4MV of high frequency ripple. And sea water is conductive! You are basically building a massive capacitor with sea water! The losses would be enormous.

And I don't even want to think about the electronics...14MV is so insane I cannot fathom anything that would be able to reliably handle it. 1MV is already nuts. 800kV is the highest in the world, and that is kinda just a flex.

Jumping on this bandwagon - these days I'm working in the submarine telco cable industry.

Considering a cable from singapore <> LA direct can run up $1.4bn USD. I think author needs a lot more research.

1. route planning takes a long time, the ocean floor moves (see: Fault Lines, Underwater Volcanos, pesky fisherman) 2. The ships do move _ a lot_ even with fancy station keeping and stabilisation. 3. cables get broken - a lot. Even now there's 10-15 faults globally on submarine cables. There are companies (See: Optic Marine) who operate fleets of vessels to lay and maintain cables. I'm sure the HVDC industry has the same.

Cool idea, I have been pondering it a lot myself, I figured maybe a ground return HVDC cable might be better for inter-country power grid links.

I know Sun Cable out of Australia want to build a subsea powercable to sell energy into ASEAN.

> HVDC lines can have up to 10% ripple

That's exactly why one uses a high switching frequency, MOSFETs and has a tiny ripple (perhaps 0.1%). This can be obtained cheaply with multiphase convertors.

Mosfets are now cheaper than IGBT's where you are paying for power losses and plan to run at full load for more than a few days to months. That's why nearly all EV's use MOSFETs - (and will use GAN MOSFETs at MHz switching rates when the patents run out)

Remember that the cable acts like a capacitor/inductor pair to ground. Ripple currents that are lost through it are not wasted money - merely wasted capacity and resistive losses in the cable. At these currents, you can assume earth is a perfect conductor, so no losses there either.

400V electric vehicles and 400,000V transmission lines play by different rules.

There are no MOSFETS anywhere in HV applications. IGBTs, but no MOSFETS. Most converters use thyristors and newer ones use IGBTs. No matter what, PN-junctions are king for HV silicon applications.

Also ripple is a function of filtering not switching. The reason higher switching frequencies generally have better ripple characteristics is because smaller capacitors can filter them and/or larger capacitors filter them better. So in a cost constrained/size constrained product you get more filtering for the same buck same size.

I also can't figure out what you are saying in your last line, apologies.

> 400V electric vehicles and 400,000V transmission lines play by different rules.

When stacked, they don't. Plenty of research on stacking both MOSFETs and entire power converters.

With stacking, the figure of merit (ie. Kilowatts per dollar, loss percentage) isn't a function of voltage (although the fact that you have to have an integer number in series and parallel could influence the design if you want to use off the shelf components)

Today's HV converter stations use IGBT's mostly because they used to be the best thing to use back in the 2010's when the design process for them started.

The reasons for using IGBTs is not only because BJTs withstand higher voltages, but also because their Vce(sat) can provide much lower loss than Rds(on) at high currents. I x V vs I^2/R.

Vce never really goes below 2 volts... Which for a 1000 amps means the running costs of the converter are 2000 watts * number of stages (~2800). 5.6MW of heat. That quickly dwarfs the purchase cost of those IGBTs.

Whereas the same calculation for MOSFETs [1] gives 4242 stages and an Rdson of 1.9 milliohms... = 8 Megawatts! Which sounds worse... But you can parallel the MOSFETs by spending double the money on them, reducing the loss to 4 megawatts... Or you can double it again to reduce the loss to 2 megawatts, etc.

When you run something 24*7, energy losses cost way more than capital costs - and MOSFETs let you make that tradeoff, whereas IGBTs do not.

[1]: https://www.infineon.com/cms/en/product/power/mosfet/silicon...

Well, SiC MOSFET do get used, but yeah. SiC JFETs are indeed better, lower lower with the same wafer technology, avalanche proof, high heat proof (the polyimide passivation hurts beyond ~220 C).

Much easier to drive when you stack them for HV.

That said, GaN is there for capacitive converters due to being able to run very efficient at >10 MHz switching frequency.

These converters in principle fit in very compact phase change coolant/insulator vessels, for example with propane. The capacitors at those frequencies get to be tiny, like, smaller than the transistor package by volume.

Thanks for the analysis!

I’m curious if there are any exotic materials that would be way better dielectrics?

Also are there ways to step down really high voltages? I can’t picture how the electronics would work without shorting?

>I’m curious if there are any exotic materials that would be way better dielectrics?

There are, but like glass they tend to be rigid crystalline structures, and not necessarily formable into what you need. There also is the problem that the dielectric needs to be perfect, as any imperfection becomes a pressure point and once you get even a microscopic breakdown, the whole thing is junk. Any practical repair is going to be very imperfect on the molecular level, so see what I said earlier. Also gaps are imperfections, so usually layering layers of dielectric is a non-starter too (but can be done, it's just very engineering intensive). The HV will "leap" from imperfection to imperfection until it finds it's ground. Insulating HV is a totally different world than your typical 240V, 480V, even 1kV insulation.

>Also are there ways to step down really high voltages? I can’t picture how the electronics would work without shorting

Yes, they basically use stacks of thyristors or IGBTs to actively switch the DC "phases" which get fed into a transformer to step down. Wikipedia has a surprisingly good article on it:

https://en.wikipedia.org/wiki/HVDC_converter

> Insulating HV is a totally different world than your typical 240V, 480V, even 1kV insulation.

Hell, even the difference between 600V (low voltage) THHN (thermoplastic) or XHHW (XLPE) insulation and a 2.4kV/5kV (medium voltage) cable is enormous.

Also note this image in the sibling reply's article

https://en.wikipedia.org/wiki/File:Pole_2_Thyristor_Valve.jp...

Which is part of a transmission station bridging islands in NZ and probably one of my favorite pictures on the internet.

That's the scale of the hardware you're looking at... for a voltage 40 times lower.

It's also the picture I had in mind when thinking about 14MV. The size of everything to space out the stages would need to be so vast I don't even know if it would be structurally possible.

Tokyo Electric Power has 1MV lines afaik.

Sorry, 800kV is the highest HVDC.

China has one 1100 kV HVDC line completed in 2018:

https://www.nsenergybusiness.com/projects/changji-guquan-uhv...

The Changji-Guquan ultra-high-voltage direct current (UHVDC) transmission line in China is the world’s first transmission line operating at 1,100kV voltage.

Reply on news.ycombinator.com