Comment by totalZero | Hacker News Reader

Comment by totalZero | Hacker News Reader

Parent Full thread

I have been thinking about this a lot recently, and I think Canon and Nikon, both of whom produce Deep UV machinery (ie, equipment that uses a laser wavelength about an order of magnitude wider than EUV's 13.5nm), have an opportunity due to demand for legacy foundry processes. They are a tiny part of the photolithography equipment sub-sector, but the present moment is marked by overwhelming demand for semiconductors for automotive applications, and those chips aren't running on leading-edge nodes like TSMC 5nm or Intel 10nm (there's no reason for them to). If I were the CEO of a global automaker, I'd be buying legacy (22nm and older) foundry equipment in order to make chips in-house because the eventual product mix transition (ICE vs EV) means I'm going to need more custom semiconductors than ever before.

The semiconductor industry is so complex that it depends upon companies in many nations, including Japan. No one nation owns the semiconductor lifecycle, but each part of the process is dominated (but not entirely owned) by one nation. Japan dominates wafer production. One company in the Netherlands dominates photolithography equipment. The United States dominates processor design. China dominates device assembly. Korea dominates NAND. Taiwan dominates foundry. Companies in all of these places are laying out a bunch of CapEx so clearly there's an anticipation of a rising tide that lifts all boats, rather than a rejiggering of market share that hurts one economy (forgetting China for a second) while helping another.

> those chips aren't running on leading-edge nodes like TSMC 5nm or Intel 10nm (there's no reason for them to). If I were the CEO of a global automaker, I'd be buying legacy (22nm and older) foundry equipment

You can, but you would be few years late to the party. There are already dozens of n-th tier fab wannabes riding the trend.

Really ancient fab tools are flying off the shelf like hot cookies. People even scoop eighties era equipment off Ebay, and such.

And yet, several multinational automakers have shut down production lines.

Why can't they keep things going temporarily by spinning up those n-th tier fab wannabes? (your words lol)

In my estimation, there are few trustworthy foundries that can facilitate the kind of scale that an automaker would need. If you're in a hotly competitive industry, you're not likely to contact some random fab freelancer to do a few runs of some chip you need, even if you have the design and everything ready to go. They're unlikely to do it at the quantity you need, on the timeline you need, to the degree of quality that you need, and with the secrecy that you need.

I mean, if you're Denso or Bosch, shouldn't you be able to make your own chips? Auto parts have a long life cycle, so you can amortize the cost of the machinery over many years.

Bosch has a semiconductor division which operates a few fabs in Germany.

They operate three, and a fourth in Dresden is expected to begin production this year. Despite this, they haven't been able to make delivery of orders to VW Group, who are reportedly considering seeking damages in the eight- to nine-figure range from Bosch for production downtime due to shortages.

https://europe.autonews.com/automakers/vw-may-seek-damages-b...

> Why can't they keep things going temporarily by spinning up those n-th tier fab wannabes?

It takes time to tape out over to another node, and by then they expect to not have the supply issue anymore with their regular suppliers.

> Why can't they keep things going temporarily by spinning up those n-th tier fab wannabes?

Years long lead times. Every 200mm fab around is booked for at least 1y+ despite biggest increases in capacity in decades.

A labs scale manufacturing can be setup within months, but it takes 3-4 years to setup AMHS fit fabs on a commercially profitable scale invariably of the technology level involved.

> Every 200mm fab

I know this is a typo, but that would explain the dimensional trend in smartphones...

Not a typo. 200 mm refers to wafer size.

Ah, that makes a lot more sense than a 200nm chip in 2021.

Still, I'm not going to get the image of a city-sized IC out of my head.

It's size of the wafer.

http://audilab.bmed.mcgill.ca/~funnell/InforMed/Bacon/HW/pri...

It was a rhetorical question answered by the succeeding paragraph.

> Years long lead times. Every 200mm fab around is booked for at least 1y+ despite biggest increases in capacity in decades.

This is precisely my point. There is a need for more capacity.

Auto OEM's have strict rules regarding procurement. Therefore it's impossible to make business with them as a unknown startup. Tier 1 suppliers can introduce un certified sub vendors and such.

Renesas has own fabs in Japan and car manufacturers and Denso tend to order and use their chips at cheap.

Hah! I wonder if I too, could, for a couple hundred grand, make my own little fab using old processes. I wouldn't, but I like to know I could.

Those chips are also low margin and the shortage is only because of a failure of forecasting. It's like building mask factories, now there is a glut.

Very smart comment, and that's the question: is the present spike in semiconductor demand a temporary situation, or a new reality?

If this demand environment is a new reality, then prices will go up and it will take several years of CapEx to get them back down. If it's just a spike, then the pure-play foundries may still invest in those legacy production lines, but there will be less of a long-term presence.

I am personally of the opinion that this is a permanent market shift; WFH is here to stay, so there has been real and lasting demand creation for cars. Plus, the transition to EV is undeniable, and will perhaps see an inflection point in this self-reflective moment where automakers have to shut down production lines and reassess their plans anyway.

> is the present spike in semiconductor demand a temporary situation, or a new reality?

I have access to quite pricey market research from Digitimes, SEMI and such.

Biggest contributors to growth are things very, very remote from both cars, and trendy gadgets.

You just keep getting more ICs put into just about everything, and at the same time very simple devices get bigger global penetration.

For example, a giant growth centre been rooftop solar equipment for developing countries — completely missed, and invisible in the West.

Electric mopeds, scooters, and lsevs — again, things mostly invisible, and unknown in the West.

LED lighting — growth immeasurable, and hard to count

Home appliances

Power ICs, and motor drivers

Security devices: locks, locators, CCTV, alarms — 100x+ growth over a decade

Vending machines, signage — 60x growth over a decade

I like how the Olympic committee has tried to take into account that the buildout of facilities every 4 years follows a similar boom/bust cycle, and they're at least trying to think about how we can satisfy both short-term and long-term goals at the same time.

For instance, a stadium where 30% of the seats are built to last, and the remaining 70% are bolted on, meant to be taken back off and recycled after the closing ceremonies.

I don't know how you build a mask factory so you can use it for something else. I do know of textile factories that have pivoted temporarily. Can you build up a short-lived production line for an integrated circuit and then tear it down but leave all of the infrastructure in place to pivot to something else? What I hear out of Intel is that setting up such a fab is a long and expensive process.

What I also hear is a lot of people challenging Intel's wisdom. So now I don't know what to think.

> Can you build up a short-lived production line for an integrated circuit and then tear it down but leave all of the infrastructure in place to pivot to something else?

Not at those margins.

Most fabs build will be running 24/7 until they crumble to dust.

TSMC still operates fabs built in eighties.

> Most fabs build will be running 24/7 until they crumble to dust.

This is false. Fabs get decommissioned from time to time.

> TSMC still operates fabs built in eighties.

I mean, kind of. TSMC's oldest operating fab is Fab 2; it was built in 1989 and it started production in 1990.

https://investor.tsmc.com/static/annualReports/2016/english/...

Fab 1 was decommissioned twenty years ago.

Well, it has been done in the past - automobile (and many other!) companies built tanks and other war material during WW2.

> Those chips are also low margin and the shortage is only because of a failure of forecasting

Not at all, at all. People have foreseen that at least half a decade ago, it's just the reality beaten even most brave forecasts.

Semiconductor economics is one of the most studied fields after agriculture, and industrial commodities economics.

Is there a mask glut seems the opposite? The N95s I get are a little bit cheaper but still $4 /per for the nice ones.

And they aren't US market masks. I'm 95% sure they are real. Not the kind with the fancy foil seal but if they're fake I'm sure they have basically the same effect. Lately I wear 3M 1870+ and it has the best fit for a tight seal.

In Austria they made N95(FFP2) masks mandatory in closed public spaces so supermarkets sell them at 0.59$ a piece or give them for free with each purchase. At that price, they're not the best quality ones, but better than cloth or surgical ones and their shape makes it difficult for idiots to wear them underneath their noses.

Yeah they don't have to be super fancy! I would think fit is the #1 thing material sounds easy enough (maybe it's not I'm just intuiting).

I am so depressed with my country (US). this should have been so obvious and we should have used Defense Production Act to scale it up 10 months ago.

It sounds like Biden admin is advocating N95s now hopefully they use legal means to push for it.

There is a glut for crappy mask. Decent ones are still not quite in balance.

>because of a failure of forecasting

It depends on which side of forecasting, was it demand or supply? It is always easy for outsider to say they should have built 10 Fabs 4 years ago and they would have been riding the tide. But even older ones are still CapEx intensive and relatively long ROI. Which means from a business prospective it is better safe than sorry.

And that is especially true with TSMC who are extremely conservative.

There was another comment about Auto Maker should built their own Fab. I really dont see how this would make any sense. All Auto Maker has to do to guarantee supply was to sign a long term agreement with TSMC, or even ask TSMC to built dedicated Fabs for them. But not a single client ( apart from Apple ) was willing to bet. No one was willing to bet their game for 3-4 years in advance. There is also the conservative nature of traditional business in demand side. If they sign it now at a cost that is 40% higher than their previous median, will they end up like Airliners hedging on Oil Price and ultimately cost them the billions.

So it really is the problem from both side.

And the same is true in ALL other market. Anyone who has ever worked in buy and sell, supply chain should see how similar the conversations are.

Out of all these, processor design (what you say is USA-dominated) seems like the one that involves the smallest amount of capital investment. And is therefore the easiest crown to lose, right?

In a world where you don't have protections for IP, yes. That's a major part of the reason that Chinese companies like Huawei and SMIC find themselves on the naughty list -- it's not just about national security and slave labor.

However, companies like Qualcomm and ARM wouldn't be able to exist without a strong global IP law infrastructure, but the former is a juggernaut and the latter is an object of desire.

Think it's more accurate to say the US dominates software ecosystems that keep the processor designs relevant. China and Korea also have competitive processors.

My point in the comment to which you responded is not that entirety of each constituent stage of the semiconductor production cycle belongs entirely to a different country. Rather, I contend that the lion's share of each constituent stage is maintained by a different country.

I agree that Korea has competitive processor designs. Of course, even Samsung uses many Qualcomm SoCs and licenses many Qualcomm patents. If you remove US-developed design IP entirely from the Samsung portfolio, how many Samsung products would be affected? I think many.

> However, companies like Qualcomm and ARM wouldn't be able to exist without a strong global IP law infrastructure

They could've easily did that. I myself saw rather new ARM cores being all around Chinese FTPs.

The thing is, just macros, or verilog code alone is near useless without tons of documentation, spoonfeeding by ARM FAEs, and tooling support.

There is absolutely zero chance that Qualcomm could exist without being able to protect its IP in court. The model of their licensing business is to do years of R&D, then spend years selling rights to the work product to pay for the next cycle of R&D. Cost of revenues for the licensing business (R&D) are out of sync with the sales that they later generate. Licensing revenue was 40% of QCOM's total revenue last quarter. How exactly do you suggest they make money by selling IP without a legal environment that protects their contracts and patents?

Same goes for ARM. Their entire business is making and licensing designs internationally. That's what they do. That's how they make money.

> The thing is, just macros, or verilog code alone is near useless without tons of documentation, spoonfeeding by ARM FAEs, and tooling support.

Looking at past cases, I think this is an extremely poor characterization of how infringement actually happens. More likely,

(A) customers (or former customers) use licensed technology for which they underpay (or refuse to pay); or

(B) competitors hiring engineers from customers or former customers to compete with the company.

Huawei did actually have the IP for 5G?

Two crowns are owned by the US: design and verification/design software and hardware tooling (my career is in verification). That's why China is non-export list for all of chip-design tooling and why China is now investing into their own tooling. It will take them time (maybe 2 decades ?) to catch up on tooling.

> That's why China is non-export list for all of chip-design tooling

And that's, in part, why every IC shop I checked in China ran off decade old warezed Cadence EDAs.

Who could've knew that warez market exists even for such rare, pretty much custom made software for which there are less than <100 licensors in the world.

When you hear hoofbeats in Africa and are puzzled, think zebras, not horses: espionage, not warez

I did a six month contract for ASML years ago. Their Asian (and especially Chinese) sales people were notorious for giving away gigabytes and terabytes worth of “free” software or DVD images to any customer or potential customer who asked for them.

This caused major problems for the WAN between ASML headquarters in Eindhoven and the remote offices in those Asian countries.

True, and I know that a famous verilog simulator was sold to a chinese company as a _single_ license... lol ! But without support, especially as you go down the rabbit hole of prototyping/emulation, they won't compete. They know it and that's why China is building a complete set of verification and design tool of their own. It will take a while to catch up though...

Every company that uses Cadence, etc. has somebody on staff who can "handle" the license server "issues", whether in the US or overseas.

The difference is that in the US enough licenses are used to pay for the "average" number of users, while overseas it's zero.

And frankly, if the big EDA tools companies actually fixed and improved their software instead of milking it for decades, legit US clients would care more than they do now.

For those who haven't had the pleasure, most EDA products are like buying a boxed Windows in 2021 and they turn out to actually be Win 3.1 - no exaggeration.

Reply on news.ycombinator.com