Intel Is Investing $20 Billion Towards a Massive New Semiconductor Plant

Joined
Nov 16, 2002
Messages
36,392
Location
NJ

Intel is set to invest $20 billion into a massive new semiconductor chip manufacturing site near Columbus, Ohio, a report from Reuters reveals.

The news comes as an analyst informs that China, which has invested heavily in semiconductor technology in recent years, is "three or four generations" away from being at the cutting edge of semiconductor production.
 
TSM has been around for quite some time and is a staple in the semiconductor business, but it's clear that further production is needed. This is good news not just for the semiconductor supply problems, but for the U.S. moving forward. The problem is that it doesn't really help the here and now, unfortunately.
 
It takes about 2 years to bring up a wafer fab and $10B or more. A mfg line is designed and built with tools for a certain class of chip. Changeover takes time, months or more, and millions to complete. And since geometries (technology nodes) evolve rapidly, the usefullness of a given chip tends to be short lived.
 
After 20 years of driving local businesses out, the politicans in Utica NY finally managed to get a chip plant in Marcy, NY. We're supposed to be getting one soon in the Syracuse area (clay).

Of course local residents are already complainin about an increase in traffic. This is a pretty depressed area economically. A chip plant with high paying jobs seems like it'd be a great idea.

And this whole area has such a high water table that it's pretty much ideal for chip manufacturing.
 
And then shipped back here as a finished product or to be put in a finished product 😁

Logistics. Expensive logistics.
They have been doing that since, I don't know, 90s?

Those Made in Malaysia, the Philippines, Costa Rica Intel CPUs were from FABs in US, but the backend process like packaging and testing done there. A caddy of CPUs wafers probably is worth 100K or so at today's prices, what's another $200 to fly them around?
 
OK, so domestically produced chips will be shipped to Asia to be placed on circuit boards?

Not really. These are wafer fabs we're talking about. They take bare silicon wafers and then imprint all the features on them. It's then sent to a different packaging plant to cut into individual dies, test, and put it into a package where it's connected to the pins or solder balls using "bond wires" to pads on the silicon. The country of origin that you see on the package is where it was packaged. You'll never see a complete Intel chip that says it was made in the US, even if the silicon (the most expensive part) came from the United States. It is kind of interesting what AMD puts on their packages. The one on the left obviously obviously has silicon from Taiwan and the US, but the whole "diffused" terminology is new to me.

2021-11-07-image-2.jpg


But there are so many different ways to do it now. Some packages hold multiple silicon "die". I remember the first time I heard about a "multi-chip module" but they're routine these days.
 
They have been doing that since, I don't know, 90s?

Those Made in Malaysia, the Philippines, Costa Rica Intel CPUs were from FABs in US, but the backend process like packaging and testing done there. A caddy of CPUs wafers probably is worth 100K or so at today's prices, what's another $200 to fly them around?

Intel does have a few fabs outside the United States, but not many. I remember when they used to ping pong generations of different processors between design in Oregon and Israel, although I'm not sure if that was where they made their silicon.
 
Intel does have a few fabs outside the United States, but not many. I remember when they used to ping pong generations of different processors between design in Oregon and Israel, although I'm not sure if that was where they made their silicon.

My understanding is they were copy exact bug to bug at least back in the days. Today I think it most likely has different process for different market (cheap Desktop use minor older processes, expensive data center CPU use the best, and laptop use middle of the road). Doesn't matter anyways, just whoever gets the most investment in tooling.
 
And then shipped back here as a finished product or to be put in a finished product 😁

Logistics. Expensive logistics.

There's very little packaging still done in the United States. They've crunched the numbers and figured that they could still make money with silicon in the US, but not packaging and boards. I do remember one place my old workplace used that made boards. It was really expensive and only used for our prototypes. And when we needed board level work on it, we spent a lot of money getting specialty repair since it was beyond our capability to rework a board with all those surface mount chips.
 
My understanding is they were copy exact bug to bug at least back in the days. Today I think it most likely has different process for different market (cheap Desktop use minor older processes, expensive data center CPU use the best, and laptop use middle of the road). Doesn't matter anyways, just whoever gets the most investment in tooling.

It gets interesting because the most advanced processes shrink in size (basically Moore's Law), so theoretically the cost of the silicon is cheaper. But of course the big cost is in development and the latest equipment to make all that. Of course we're approaching the size of an atom, so Moore's Law is going to reach a physical limit.

I remember taking classes where the discussion was on how scaling down in semiconductor feature size improves speed. Of course one can maintain the same voltage while scaling down and the performance can theoretically go way, way up. But then the problem is that amount of energy that has to be dissipated just goes way up and it's not a good thing if it just burns up. But the voltage can be cut, the energy used will be lower, and the performance is still better. However, I think there are still some specialty applications where they're willing to do things like liquid cooling. Processors used to come in plastic packaging, but now they come in these fancy packages designed to connect to a heat sink and where there has to be a fan.

 
It gets interesting because the most advanced processes shrink in size (basically Moore's Law), so theoretically the cost of the silicon is cheaper. But of course the big cost is in development and the latest equipment to make all that. Of course we're approaching the size of an atom, so Moore's Law is going to reach a physical limit.

I remember taking classes where the discussion was on how scaling down in semiconductor feature size improves speed. Of course one can maintain the same voltage while scaling down and the performance can theoretically go way, way up. But then the problem is that amount of energy that has to be dissipated just goes way up and it's not a good thing if it just burns up. But the voltage can be cut, the energy used will be lower, and the performance is still better. However, I think there are still some specialty applications where they're willing to do things like liquid cooling. Processors used to come in plastic packaging, but now they come in these fancy packages designed to connect to a heat sink and where there has to be a fan.

I remember hearing data center GPU will reach the limit where liquid cooling is required by about now. CPU though maybe still air cooled. We are seeing even storage processors (not nand or dram) must go down to 14nm by about now and 28nm weren't cutting it anymore.
 
Taiwan already beat china (and us) to the punch.
I'm not sure Taiwan beat us to the punch; it is more like we gave away the business.
When I started in SEMI, in the 80's, about 70% of tool shipments were in the US.
Now TSMC, UMC, etc make about 70% of the chips and about 90% of the dense geometry chips.

If there is a mess in Taiwan, China has the world by the shorthairs.
 
This is good but we also need passives. No fun at the moment, searching for subs that keep going out of stock.
 
OK, so domestically produced chips will be shipped to Asia to be placed on circuit boards?

Most of the time, the packaging is also overseas in Asia, biggest in the Phillipines, Malaysia and Singapore.
This is when the Wafer (full of printed circuit) cut up into Die and then put into the IC packaging that we see mounted on the electronic board.

Then they go to the electronic assembly board.
 
I remember hearing data center GPU will reach the limit where liquid cooling is required by about now. CPU though maybe still air cooled. We are seeing even storage processors (not nand or dram) must go down to 14nm by about now and 28nm weren't cutting it anymore.

I've seen crazy stuff before. My first visit to Comdex in the mid-90s, I saw a company that did cryogenic cooling of Intel CPUs and then they overclocked it by maybe 3-4x. The speed of semiconductors goes way up when they're cooled because conductance increases. But that probably wasn't good for the hard drives, but of course the storage system has to be kept away from the cooling.

Apple was having issues with their Power Mac G5 with the IBM PowerPC processor, where it had to be liquid cooled. I believe the liquid they used was some sort of mineral oil. However, their switch to Intel was supposedly because Motorola/Freescale (which was making the mobile versions) just couldn't get the power consumption down for a mobile G5. But this rebuild of a G5 shows all these tubes and a radiator.



I do remember asking a professor why not just maintain the voltage for better performance when shrinking. I think he stated that it could theoretically be faster if that were done. But the answer was "How long do you think it's going to last with 8 times the energy being dissipated per unit area?"
 
OK, so domestically produced chips will be shipped to Asia to be placed on circuit boards?
Made in China for labor costs ? How about Mexico instead - they already make thousands of car parts
We need to thin out these MGO cargo ships if we want to say EV is for the planet …
 
Top