Posted on 01/16/2024 4:34:01 AM PST by FarCenter
Samsung Electronics is planning to fully automate its semiconductor factories by 2030, with “smart sensors” set to control the manufacturing process, according to South Korean media reports.
The world’s largest maker of memory chips aims to create an “artificial intelligence fab” that operates without human labor, the reports said. The ground-breaking project is reportedly already underway, the same reports said.
Samsung has signaled since last summer it aims to AI to optimize integrated circuit (IC) design, materials development, production, yield improvement and packaging. Identifying the cause of defects in the production process is reportedly a top priority of the AI plan.
Samsung is developing its own sensors and switching procurement from foreign to domestic suppliers to gain control of the technology and develop relevant South Korean expertise. Measuring plasma uniformity in deposition, etching and cleaning is one key application; real-time monitoring of production processes is another.
The technology will be applied both to Samsung’s DRAM and NAND flash memory operations and its contract manufacturing operations. Catching up with Taiwan’s TSMC and staying ahead of America’s Intel are of vital importance to Samsung as die-shrinks progress from 3nm now to 2nm by 2025 and 1nm late in the decade.
The finer the circuit line widths, the greater the risk of microscopic defects dragging down chip production yields. Deploying AI to keep the problem to a minimum is increasingly critical to maintaining competitiveness. A completely automated factory would also eliminate the risk and cost of human contamination.
You want Skynet cuz this is how you get Skynet!
In 1981 I was to be part of the largest leap forward in manufacturing to date. Millions of dollars were spent to produce (pause for effect) a paperless factory! Despite the planning and money spent, people were generating paper trails all over the place. Nothing worked the way it was supposed to. It might have worked if nothing ever went wrong. If every process was perfect. If some of the people weren’t (not to put too fine a point on it) idiots.
Thing is, you can declare whatever you want to the press. Then, you have to deal with reality. And reality is a female dog.
Analysts are pointing out that AI is a power hog. Will the power grid by these new chip factories be able to take the load? Will we see the deployment of SMR power modules to localize power generation for these plants?
This is going to be hilarious! The AI will probably determine they are building coasters and start setting drinks on them.
Really . Self replicating AI. In the real world, not virtual.
That’s what AOL CDs were for.
AI is just doing averaging on large data sets, so I surmised the AI would find the AOL CD’s and turn them into the coasters we all did.
A lot of projects to automate in the ‘80 ran into problems because they still relied on humans to provide inputs of what they saw or they were doing. They also spit out work orders to humans to do things. Neither work very well, which is why sensors and actuators controlled by the system are essential. You have to take ALL the humans out of the system.
Another problem in the ‘80s was that systems weren’t able to handle all cases, so complicated decisions or problem situations where data was incorrect or insufficient were referred to humans at workstations.
There was also a lot of resistance by middle managers responsible to work center. None of them really wanted to shrink the number of people working for them, and they wanted to focus to be on automation to support their existing work center responsibilities, rather than to automate full processes that spanned work centers.
Not surprised...
Most people do not realize that micro chip production has been the most automated industry in the world for at least the last 20 years.
I am still stunned that the USA allowed 80% of its chip business to move to Asia.
I know USA labor and environmental costs are very expensive, but the idea that we waved a white flag and surrendered to Taiwan, Japan, China, and South Korea is incomprehensible.
Actually, this has been the trend since the invention of the semiconductor. I.e., hiring of smokers is out. All people shed particles even when fully suited up. So, as chips continue to be miniaturized, and ever more complex people become a bigger problem.
Your assessment is well informed and accurate.
My view is that the most innovative leader at the front of each technological wave discovers the problems and will probably be bankrupted by the efforts required to resolve them. It’s best to be in the second wave so you can benefit from the resolution others discovered and were bankrupted by. History is full of innovators who, for one reason or another failed. DEC, HP, IBM are examples. Of course, each was destroyed by multiple leadership failures or market changes.
I think a better approach would be to build out a section of your business as AI to run in parallel to the existing processes, so you don’t hit a hard stop if something goes wrong. Building a whole plant to a new model, all while the plant was under contract to perform, was a disaster most of us worker types accurately predicted.
As you indicated, one big roadblock is the present management and employee cadre feeling threatened. Therefore, the parallel process should be more-or-less lowkey and in a separate building. Yet, the existing staff needs to feel less threatened, so they don’t sabotage the inputs or outputs of the new process. Resolving that is a case-by-case issue. I’d say if the process could be divided into A+B+C= product, you first automate C, then B+C, then, A+B+C. Then, you move everyone involved in the product to something else, or you generously buy them out.
>>I am still stunned that the USA allowed 80% of its chip business to move to Asia.
“Allowed” is probably not the right word.
From a geopolitical perspective, it was necessary to develop the economies of Japan, South Korea, Taiwan, etc. so that they could function as barriers to Soviet and Chinese communist expansion in Asia.
A manufacturing sector like consumer electronics and semiconductors made sense to develop there since the products did not cost much to ship to the US and since they could be made highly reliable so that ongoing support/maintenance was manageable.
I think that shipping those industries to Asia was US policy. I do know that in a later decade, companies were “encouraged” by the US government to find software development jobs that could be allocated to India.
“Analysts are pointing out that AI is a power hog.”
I don’t get that. I’d say the increasing amount of electronic gear in use results in significant power consumption, but from a power standpoint AI is just another electronic thing.
There are numerous articles about the AI need for power.
Here’s one
https://www.wsj.com/tech/ai/ai-energy-consumption-fc79d94f
Semiconductor fabs are already power hungry. A single ASML EUV lithography machine draws a million watts of power.
AI is very power intensive to train. Once trained, it can be replicated and used at much lower power levels.
“There are numerous articles about the AI need for power.”
When it comes to power consumption, what is AI? What exactly is AI?. Seems to refer to advances in speech and face recognition and recognition of things like the road in front of a moving car.
Isn’t AI, then, just the broad continuation of the advancement of computer uses, made possible by better and cheaper components and the thousands of hardware and software engineers doing their jobs?
The power going into an electronic device doesn’t know what it is being used for.
Will there be proportional chips of color, and especially, nonbinary chips?
Most fabs are fully automated.
This isn’t about AI but about people are too dirty to be in submicron fabs
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.