Hacker News Reader

Hacker News

Comment by rasz | Hacker News Reader

Full thread

rasz

>Hyundai Group is pushing for internal chip production

HAHA, good luck even coming close to specification of Infineon parts.

>LACK OF ALTERNATIVES

no shit Sherlock, this is life on the bleeding edge

>It is also trying to shorten the delivery time by transporting chips from global semiconductor makers by air.

you mean they used ocean freight on tiny light components (maybe hundred kg per 10K cars) to cut costs? :)

>Infineon accounts for 12.7% of global automotive chip output, trailed by the Netherlands-based NXP and Japan’s Renesas with 11.8% and 8.4%, respectively.

automotive, but not EV. Afaik NXP makes zero IGBTs, they make gate drivers. Renesas makes some, but their EV grade parts were supposed to start shipping in 2021 and didnt https://www.renesas.com/us/en/products/automotive-products/a...

View on HN

If you have a consistent supply of something then a slow but lower cost delivery method is entirely sensible, as is switching to higher cost but faster methods when there have been interruptions.

The story also mentions that their internal plans are for lower spec items.

> If you have a consistent supply of something then a slow but lower cost delivery method is entirely sensible, as is switching to higher cost but faster methods when there have been interruptions.

It might make sense from an economical point of view, but not from a resilience point of view. Ignoring the chip shortages over the last two years, there have been major disruptions in ocean freight logistics. Given the relatively high value density and the crucial role of these components in the vehicle manufacturing process, making air-freight the default would have been a sensible choice as well. Despite the feasibility of alternatives

The problem isn't that a ship got delayed with the chips on board, they threw out months of production at the factory due to a flaw before shipping them. So switching to air freight now is just boring and sensible (and would also work if a ship got delayed too, meaning the cheaper option is basically risk free).

Even with consistent supply, for a high value low weight item, ocean shipping does not make sense simply from a time value of money perspective. Also, it makes planning that much more difficult, you have to adjust orders that much farther in advance.

Not really true - depending on volume.

In almost all certainty they have volume commitments out months - and have for at least two years.

JIT stopped being a reality a while ago.

If you know you're sitting on 2 months of inventory, and you're getting shipments that are equal to 1 month of capacity every month, and it takes 3-5 weeks by sea, it's a perfectly reasonable thing to do still.

Also, depending on their contracts, they may not pay the vendor until they receive the goods at their location, so the sea freight shipment doesn't have a cost for time value of money.

If you're sitting on a buffer, time value of money doesn't matter - and only gets worse by spending more money to get it sooner if you can only produce say 5k cars/week - doesn't matter if you get parts sooner you can't use.

With semiconductor manufacturing, you are in a dance where customers need products now but your lead times can be weeks or months. This has always been true, even pre-pandemic. Customer demand forecasting is a huge deal in this space, since they still pursue "lean planning" and "JIT production" as much as possible to out-compete on delivery times and volumes.

Inventory costs are necessarily minimized in this framework, with all the supply hiding further back in the chain. Once the products are done, they're packaged and shipped; there is basically no buffer. So you don't have resilience of that sort, though you do have a significant off-ramp (in terms of the products still in process) to help adjust to market shocks if necessary.

Instead of this kind of inventory-buffer plan common in Western economists' and analysts' assumptions, you have instead resilience in the relationships between you and your customer and the stability that confers. Because the supply chains are so long and complex between semiconductor manufacturers and their customers, it pays to keep good relations with your supplier (and with your customer). So account managers and supply planners are very careful not to cancel orders or adjust delivery dates so as to maintain that congenial relationship. This is the main lesson higher-ups learned from their time spent playing the Beer Game[1].

Source: worked in Infineon's logistics and supply chain department

[1] https://beergameapp.com/

Generally agree, but I think 2022 isn't the case of "consistent supply".

On point. Hyundai can bitch and posture all they want but those chips aren't some commodity parts they can multi source or make themselves not matter how much money they throw at the problem. You can't bring up decades of analog microelectronics design experience overnight.

So they have to deal with whatever issues Infineon has. It's part of the risks of being on the bleeding edge. That's why even luxury German car companies usually stay on older tech.

Power semiconductors aren’t the most bleeding edge. SiC technology is a recent game changer leveling all market players to the same tech level. IFX advantage with silicon is gone. The structures are big and Hyundai-Kia conglomerate can afford their own production for sure. They also have Samsung there. And they know their requirements best and can optimize for them instead of buying off-shelf parts from IFX. German premium manufacturers also don’t use 800V system with fast charging while Hyundai-Kia does.

Samsung doesn't design cutting edge analog chips.

Just because they fab NAND, DRAM, and cutting edge logic chips, don't mean they can switch to high performance analog chips on a whim.

There's a reason, Infineon, TI, Vichay, and other's are still highly profitable.

Are the physics and manufacturing of analog vs digital components really so different that there's no applicable knowledge transfer? I would be surprised to learn that.

Well, there's more transfer than nothing, but the hard parts of digital (sophisticated logic, absurd per-gate reliability, low feature sizes) just don't overlap much with the hard parts of analog, which is all about device physics and materials.

Sure, the people who design logic transistors would probably adapt quickly to designing power transistors, but they are a very small part of a very large machine. For example: a digital organization may have learned through harsh and repeated lessons 100 subtle ways to be more conservative about device changes. FinFETs took 15 years to go from R&D to prod, for example. While this level of risk aversion might be critical for shipping billion-transistor devices, it may be maladaptive for shipping single transistors, where a 15 year deployment leaves you 10 years behind your competition.

Likewise, the processes (patterning, etching, implantation, etc) are largely similar, but the digital side has made colossal investments optimizing for things that just don't matter as much for analog design. The famed EUV machines? What does a power transistor gain from tiny features? I'd be surprised if they were useful, let alone useful enough to justify the cost, let alone useful enough to justify the cost and the risk of repurposing a machine for process innovation. Like using a F1 car to commute to work -- even if cost were no object, extreme optimization for one purpose makes it worse at others.

Power electronics gain surface area from small feature sizes, allowing the electrons to diffuse more efficiently seen as lower loss and higher switching frequencies.

Same with gate design.

It's not that there is zero applicable knowledge transfer, but there are definitely some fundamental differences between digital components and analog power devices, in particular.

For digital, the transistor are CMOS, optimized to be as tiny as possible, running at < 1V, and with individual devices mostly driving microamps, going up to milliamps in some particular areas of the chip.

For power, we're talking about IGBTs, optimized to handle as much power as possible with maximum efficiency, at hundreds of volts.

They are both semiconductor devices, but they are about as similar as the engine for a sports car vs. for a panamax cargo ship.

The feature size of power electronics is fundamentally limited by the dielectric breakdown voltage of materials. Under a high enough voltage potential, even the best insulator [1] will "break down" and become a conductor

The dielectric breakdown voltage of SiO used on wafers is about 2.7x10^9 V/m (2.7 V/nm) so a 600V part that relies on SiO for insulation has to have a minimum gap of at least 222 nm [2] between any conductors that carry that voltage. Cutting edges nodes manufacturing CPUs and stuff generally deal with single digit voltages so while a lot of knowledge can carry over, the meat of the designs are very different.

[1] Even a perfect vacuum would breakdown due to the Schwinger effect, but at an astronomical voltage potential like 10^18 V/m.

[2] It's actually a lot more complicated at that scale, but it's a useful ballpark value.

Audi/Porsche do use 800v systems with fast charging:

https://www.caranddriver.com/news/a28903284/porsche-taycan-e...

>> Power semiconductors aren’t the most bleeding edge. SiC technology is a recent game changer leveling all market players to the same tech level.

IGBTs are not quite a commodity at highest specs, and power modules are yet another years-long development. Infineon has been building modules with high power density for decades and that experience will also carry over to SiC chips. Where I'm at we are building our own SiC "modules" from discrete components for now and these are much larger than Infineon power modules - but we are air cooled ;-)

It seems like there should be a standard form factor for 3-phase power modules by now. That would make them easy to replace even if you have different specs.

>German premium manufacturers also don’t use 800V

Taycan uses cree sic, bleeding edge. GM signed up for those too.

> good luck even coming close to specification of Infineon parts.

You don't have to meet or beat the vendor's expectations when you have the ability to build them in any manner you deem fit. Vertical integration shouldn't be about a direct 1:1 replacement of your vendors. It should be a 2-way street wherein your requirements adjust relative to your revised ambitions. In-house design cycles can allow cheaper and more rapid development so a lot of other levers become available to pull.

Also, power ICs aren't exactly requiring EUV tools, multi-patterning and other bleeding-edge manufacturing knowledge.

Oower ICs require a mostly seperate form of bleeding-edge manufacturing knowledge. Power electronics has been improving immensely over the last few decades, in large part due to such advances.

> you mean they used ocean freight on tiny light components (maybe hundred kg per 10K cars) to cut costs? :)

No… they used ocean freight because that’s how you ship assembled shit from China. This whole chip shortage thing is bullshit, it has nothing to do with the chips themselves, and everything to do with the cheap labor assembling hundreds of parts - many of them chips and electronics, but also not - into a single “part” to dodge import restrictions around how much of a car is made “foreign.” It’s insane how hard this is to communicate to people and how hard they’ve bought into the absolute bullshit.

Reply on news.ycombinator.com