OT: NYT article "How To Construct a Chip Factory
The Huge Endeavor to Produce a Tiny Microchip https://www.nytimes.com/2022/04/08/technology/intel-chip-shortage.html?unloc...
| From: Ivan Avery Frey via talk <talk@gtalug.org> | | The Huge Endeavor to Produce a Tiny Microchip | https://www.nytimes.com/2022/04/08/technology/intel-chip-shortage.html?unloc... Interesting. As Moore's law progressed, a different law also applied: exponentially fewer cutting edge fabs. Simply put, each generation of fab got exponentially more expensive (I don't have actual cost data). It used to be that every chip vendor had its own fab. But they got exponetially more complicated and expensive and a new model developed: fabless chip vendors and merchant fabs. Example: most RISC vendors were fabless Example: AMD spun off its fab as "Global Foundaries". Example: fabs TSMC, SMIC. I don't follow this closely but Intel's fabs may be the last that are mostly producing their owners own chips. Intel has tried to be a merchant fab but I don't think that they have gotten much business. I don't know much about linear chips. Maybe the picture is different there. Intel's vertical integration used to be a plus. They would develop their next "node" (step in feature size or performance) ahead of everyone else. This seemed to be because they were willing to put more money into manufacturing each chip (because they could get it back) and also because each node was developed in parallel with the design of the chips that would use it. Perhaps six years ago their node progression stumbled and now their manufacturing is only on a par with TSMC in performance and it might be much more expensive (usually due to higher defect rates). It is very bad for the world that TSMC is almost the only game in town for cutting edge processors. Samsung is OK too. Taiwan may be the next Ukraine. There are a lot of suppliers of high-tech things needed for fabs. Many with no competitor. Some are in Europe and some are in the US. The way US governments have been throwing around sanctions, you can bet that PRC is trying to create local replacements. Our world is partitioning before our eyes.
Why do cars need 2nm chips? I could understand desktop, laptop, smartphones needing such. But, even Arduino would be overkill in cars, no? (Apparently not) -- On 4/8/22 9:03 AM, Ivan Avery Frey via talk wrote:
The Huge Endeavor to Produce a Tiny Microchip https://www.nytimes.com/2022/04/08/technology/intel-chip-shortage.html?unloc...
--- Post to this mailing list talk@gtalug.org Unsubscribe from this mailing list https://gtalug.org/mailman/listinfo/talk
On Fri, Apr 08, 2022 at 07:46:08PM -0400, William Park via talk wrote:
Why do cars need 2nm chips? I could understand desktop, laptop, smartphones needing such. But, even Arduino would be overkill in cars, no? (Apparently not)
My understanding is that cars don't want the newest chips. They would rather have older designs that are more reliable in different temperatures. Unfortunately those are not as profitable to the chip factories as the newer stuff, so the car makers are not nearly as interesting to them as makers of cell phones and such it seems. -- Len Sorensen
On 2022-04-08 19:46, William Park via talk wrote:
Why do cars need 2nm chips?
There's a surprising amount of processing power required in a modern car. Android Auto uses a tablet-class CPU, and Android Automotive (the lower level OS that does more than infotainment) is similar. Electric cars can have even more complex requirements. cheers, Stewart
4| From: Stewart C. Russell via talk <talk@gtalug.org> | On 2022-04-08 19:46, William Park via talk wrote: | > Why do cars need 2nm chips? | | There's a surprising amount of processing power required in a modern car. | Android Auto uses a tablet-class CPU, and Android Automotive (the lower level | OS that does more than infotainment) is similar. Electric cars can have even | more complex requirements. There is definitely a push for AI in cars. The closer to self-driving, the more processing power is needed. Also: electronic features are a lot cheaper than mechanical features. So there is a tendency to add as many of those as can be imagined. Features seem to be what sell cars. Also: processors are cheaper than wires. So sensors need to talk to shared buses, not simple wires. So each sensor needs something approaching a processor. I understood most things about my early cars. My recent car is another matter. I'm pretty sure it has a lot of features that I haven't even imagined. And it has a tablet-like control/display unit in the middle of the dashboard. Much of the interface seems too complicated to operate while one is driving. I guess I'll learn about the controls of my first self-driving car because I won't be distracted by driving it myself. Simplicity is sophisticated. It's hard to sell. Apple tries sometimes. Some AI that I'd like in cars: When I'm driving, I can often tell something is wrong by the noise or vibration or smell. AI could monitor all these signals and try to infer problems that might require attention.
participants (5)
-
D. Hugh Redelmeier -
Ivan Avery Frey -
Lennart Sorensen -
Stewart C. Russell -
William Park