Led by Tesla, EVs drive chip industry's shift beyond silicon

I am far from a subject matter expert on this, but I have been doing my best to read up on this lately. It seems like we’re globally diverging on fabrication expertise at a process level i.e. nanometer scale [0]. The bulk of leading edge chip manufacturing (currently around 7nm or below) seems to indeed be slipping away to Southeast Asia, but as you said the U.S. is holding their own at larger process levels (Cree is mentioned in the article as well) and still producing at a volume that may be surprising to many doomsayers. It’s worth noting at this point that, according to the article I linked below, it is estimated that five larger process chips are required for every leading process chip produced. So this doesn’t seem to entirely be an accident on the part of U.S. companies. Sure, some companies like Intel have simply failed to innovate in the past few years, but others seem to have made a calculated decision to avoid massive R&D expenditures that may or may not pay off as we begin to approach the limits of Moore’s law. The reality is that leading edge chip fabs are extraordinarily expensive. I read a podcast transcript recently with a prominent professor who is an expert in this field [1]. He mentioned that he did some consulting for the U.S. Govt., and when they asked him for his estimate for how much it would cost in subsidies to catch up with or outpace the rest of the world in fabs his response made the guy who asked the question practically fall out of his chair. Have we reached a point where quantity (at an average level of quality) is now more important than having the fastest and the smallest IP?

[0]: https://semiengineering.com/can-the-u-s-regain-its-edge-in-c... [1]: https://www.theverge.com/2021/8/31/22648372/willy-shih-chip-...

> To see how fast this technology is moving, you can watch this video [1] where sandy munro needs a helper to wrestle with the ford mach-e inverter on a table.

My take away from watching that clip wasn't him being confused with the function of the inverter but rather being puzzled by how every step of the assembly eschews the principles of Henry Ford and even the rules he said he wrote while at Ford.

Maybe you linked to the wrong time?

>There is another little known US company, ON semiconductor,

I chuckled. Only on a Software developer focused forum would ON being called a little known US company. They are really well respected in semiconductor industry. If I remember correctly, ON along with Freescale came from Motorola.

Correct me if I'm wrong but I don't believe Cree manufactures any chips in the US?

Thank you for answering everything I ever wondered about SiC. Are there any other semiconductors that would be even better than SiC in all four points?

This is the key observation - hitting breakpoints in efficiency that let us remove active cooling, silicon area or heatsinks reduce weight, cost & complexity.

This isn't quite right...

In a car which can accelerate up to speed and brake again every 30 seconds while going through a city with a lot of stop lines, the efficiency of the power components really matters.

If you loose 5% in the motors, 5% in the power electronics and 5% in the battery in each direction, then each accelerate-brake cycle is losing you ~30% of the energy you started with. At town speeds, friction and air resistance losses are tiny, so that 30% is probably actually most of the energy you are using!

Add that to the fact that certain motor control techniques such as field weakening involve every Amp of real power that comes from the battery passing many times though the inverter, since there is a higher circulating current between inverter and motor coils.

You mean ESC, not ECU.

Generally cooling ESCs is not that big of an issue, just stick them in the same cooling system as the batteries.

You are absolutely right there. I first read that in an article an EE wrote. He said what you care about with power inverters isn't the efficiency but the inefficiency. Small gains in efficiency you wouldn't care about translate in to large reductions in heat and smaller cheaper power devices. Power devices don't follow Moore's Law and aren't cheap.

> If we go by this figure, Hyundai's EVs need to waste 10% of its energy on Si chips.

Most of it is not lost in the silicon, but in the interconnects (wiring and connectors) and motor(s).

SiC is better (more robust) than Si with high voltage, high current, and/or high temperature. Specifically, kilovolt-level, kiloamp-level, and temperatures above 150 celsius.

If you double your battery voltage from 400V (current standard) to 800V, you halve your resistive losses in the "wires" for the same electrical power demanded. Silicon MOSFETS that can withstand a kilovolt and a kiloamp are very big and very expensive.

You get more power delivered for the same power leaving the battery.

Size: SiC's very high thermal conductivity means less of a margin of safety for secondary breakdown. SiC tolerates higher current density (amps/square mm) than Si, so smaller devices are possible, saving mass.

Its temperature stability allows it to be sintered to the package heatsink--further improving heat removal. Tolerance for high temperatures means you need less mass and volume for heat dissipation structures.

Other efficiencies: With very high currents (kiloamp) and medium frequency switching, SiC IGBTs provide more benefits, because rather than being directly proportional to current as with MOSFETs, losses are proportional to the log of current.

Alternatively, doubling the battery voltage again to 1.5kV (and so halving the currents) might allow Hyundai to use more aluminium for the "wires", with a cost saving over copper.

>Applications for HV SiC

>Data center distribution - MV AC or DC to the rack, (13.8kV distribution to match gensets

What the hell are they smoking? MV to the row, maybe, but are they seriously proposing MV terminating to a power supply in every rack??? That seems like it'd be an installation and maintenance nightmare. I feel like it'd be much more sensible to at least limit in-rack power to 480V max. Stick with something that doesn't require DC techs to use a hot stick to plug in a new rack.

It's the irony of the push for more efficiency. In theory you could have the equivalent of a 90's PC now with a fraction of the power consumption, but in practice any gain is used to create more powerful chips and do more calculations. There's a saying that was more about financial budgets, but tl;dr expenses will fill up the budget.

In my country there's irony as well. On the one side, power companies and the government invest a lot in green power production, offshore windmill farms and solar farms and the like. But at the same time, companies like Microsoft and Google come in and have datacenters built that use up all the energy that those 'green' alternatives produce. We would've been carbon neutral years ago if it wasn't for the increase in power consumption that follows production.

There’s a huge, big incentive to increase Wh/kg though which will reduce EV mass significantly. Tesla’s 4680’s will make a decent step forwards there.

The cars getting bigger and bulkier are ICE SUVs and trucks largely, things like Cadillacs, Jeeps, fully loaded trucks being used as commuters, etc. Tesla actually started larger with S/X then moved downwards in size/cost towards 3/Y. Most of the other makers have moderate sized EVs, like Priuses, Volts, Bolts, Leafs (leaves?), they're all modestly sized.

For electric cars though, you make more money as a carmaker if their batteries are smaller, which happens when your consumption is lower, so it encourages slicker designs, unlike ICE cars where somehow Americans still think 20mpg is acceptable in 2021, and the carmaker makes as much money or more selling you a 20 or a 50mpg car.

You're using the wrong units. 1.6MW global power consumption has no meaning on its own. 1.6MW is the power that two sports cars can generate.

Power consumption (e.g. energy) requires time, so 2W per car at 2 hours of daily use would be 4Wh per car per day or about 1460Wh per year per car or 1168MWh for 800k cars.

That's the electricity consumption of ~106 average US households [0].

I think that gives a much better picture of the amount of energy you're talking here.

[0] https://www.eia.gov/tools/faqs/faq.php?id=97&t=3

It's worth noting that there's also gallium nitride to consider and while it isn't as good as silicon carbide in high temperature applications or areas needing higher thermal conductivity it's better than silicon carbide as far as breakdown voltage and low resistance are concerned. Maybe eventually we'll see GaN in server power supplies and SiC in VFDs for fans, cooling, and DC infrastructure. There might even be a justification to use SiC solid state transformers to bridge between utility power, generators, battery banks, etc.

It's very important to note that 50% efficiency gains might not be as huge as it sounds given that we're just talking about losses in MOSFETs. This isn't going to be turned into your next CPU or show the same kinds of efficiency gains in logic circuits.

As far as I understand it, these SiC chips are things like discrete MOSFETs designed to work at hundreds of volts and handle huge power running through them. I think the likelihood of this scaling down to millivolts is quite unlikely. The more likely applications for this are power stations and hooked up to solar power plants rather than in your iPhone I would expect.

Silicon Carbide is extensively used in Power Electronics. It's just that there is a very small group of people that are having to deal with kilovolts on a daily basis, so I suppose it is not perhaps widely known.

All EVs are already highly efficient, any further increases is going to be extremely marginal.

> If we can get these chips manufactured at a reasonable cost it would really help EVs get longer range.

How much longer though?

The motors that move the wheels consume way more energy than the microchips on board the car. Even when a car is standing still, the AC or the heating system are also quite energy-hungry compared to a few CPUs.

> And it might even give us multi-day use out of our smart devices (phones, watches, etc.)

Not going to happen, they'll just find a way to use more power instead. You probably noticed that for years now, phones and the like have had a fixed battery lifetime of about a day of medium use, even though battery and chip technologies have improved. I'll admit that in terms of efficiency - performance per watt - things have increased, but the goal of the phone manufacturers is not to make things last longer, it's to min/max and find a balance between battery life and power.

With current day technology you can easily make a phone that lasts for a month on a single charge, but it'd be bulky compared to what you can do with it.

> And it might even give us multi-day use out of our smart devices

I don't think you understand how software development works. Q: How much hardware do you need? A: How much you got?

The latest generation only gets the battery longevity because software developers still have to support the previous generation. As soon as enough e-waste happens, the gains disappear.

I might be a cynic but there sure seems to be a lot of empirical data to support my disorder.

SiC is not a good fit for anything under ~100-200V, it's just not what it is designed for.

GaN makes more sense for consumer electronics, that's why it's popping up in a lot of devices, even if prices right now are too "high end" for most users.

As for phones, watches, etc..:

1- These technologies won't be as dense as standard silicon for quite some time (if ever.

2- We've had big swings in both efficiency and battery capacity several times already, but it works in the same way internet speed works. Power-hungry features grow with increased efficiency in the same way websites grow in size with increased internet bandwidth.

Presumably you mean to rather than by?

I'm really surprised you don't see the inverter and the motors closer coupled in Tesla's designs.

For example, by having MOSFETs spliced into the wire wound onto the stator. That minimises wire length (and therefore resistance). The heavy motor windings are a great heatsink for brief acceleration, and both will need liquid cooling for prolonged 100 mph driving up a hill.

The article mentions GaN not widely used because it is integrated with Si. Cree’s process is GaN on SiC, so you’d get better heat dissipation than in Si, but it costs more. There are also done processes integrating a diamond heat spreader under the GaN channel.

@madengr it seems you were secretly banned (shadowbanned)

https://ibb.co/NK6Yhj5

So obvious you didn't read the article.