What has happened to American semiconductors?
Since the invention of integrated circuits by Fairchild Semiconductor and Texas Instruments, it has only been a year old, and it has almost completely lost its ability to manufacture semiconductors from the cradle of semiconductors back then, and TSMC is needed to continue its life?
Today we're going to talk about itWhy can't semiconductor manufacturing go back to the United States?
December 6, 2022, Arizona, USA.
U.S. ** Biden, Apple CEO Tim, Nvidia CEO Jensen Huang, AMD CEO Su Ma, and 91-year-old TSMC founder Zhang Zhongmou and other leaders of world-renowned semiconductor companies gathered together.
Today, they are all gathered here to celebrate their good buddy, TSMC.
TSMC's planned wafer factory in the United States has finally waited for its first machine.
* Biden is even more excited to announce that the manufacturing industry is back!
Wait, since the word "back" has been uttered.
That means that to a certain extent, semiconductor manufacturing in the United States is "lost".
There is some truth to this, as we can see from the documents released by the White House in 2021.
The U.S. share of global semiconductor manufacturing has also been declining in recent years, from 37% in 1990 to 12% in '21. Even, if no timely action is taken, this data is likely to continue to decline in the coming years.
So, what exactly has the U.S. semiconductor experienced?
Since the invention of integrated circuits by Fairchild Semiconductor and Texas Instruments, it is only a year old now, and it has almost completely lost the ability to manufacture semiconductors from the cradle of semiconductors back then? And TSMC is needed to continue its life.
Hello everyone, today let's talk about semiconductor manufacturing, why can't we go back to the United States?
Unlike the semiconductor factories that we have now that are full of yellow lights and automated equipment, the newly born semiconductors are still quite "manual" in the production processIt is a labor-intensive enterprise in a sense.
At that time, there was no concept of "integrated circuit", and if you wanted to make a circuit, you needed to make basic components such as resistors, capacitors, inductors, diodes, and transistors.
Then use wires to connect these most basic components to make the most basic logic circuit.
The circuit that does this is naturally time-consuming and wasteful, and at the same time, due to the complex connection, once it is moved, the stability performance is not too good.
Later, it wasn't until around 1958 that Texas Instruments' Jack Kilby suddenly patted his head and found that these originals could be produced together.
Is it possible that we can make these components together during production, and the stability will be much better.
After a period of scientific research, they found their own way.
A triode is first fabricated on a germanium wafer, then a small amount of doping in a pure germanium crystal is used to make a resistor, and a reverse diode is used to make a capacitor.
And so, they didThe first integrated circuit in human history: a phaser
However, the IC has not yet solved a key problem – that is, the human power required to connect the wires to these components.
Later, Fairchild Semiconductor's Robert Noyce improved on this by deciding to replace the thermally welded wires with the method of evaporative deposition of metal.
This time, the steps of connecting the wiring have also been solved.
But this also brings some new problems, and that is:Costs are getting a little out of control.
Although Fairchild's process is more elegant, it needs to use the more expensive silicon process at that time, and the output demand of the new technology at that time is small, so it is very difficult to control.
In the sixties of the last century, an integrated circuit could even be sold450 USD, in terms of inflation, is equivalent to the current oneTwo iPhone 15 Pro Max, 1t Bank of China non-Hainan version.
So, is there any way to save the cost of IC manufacturing?
In 1959, Charles Spock, who had worked at General Electric, joined Fairchild Semiconductor, and his job was to find a new place for the company to build a factory, and they had just won a large semiconductor order and needed to significantly increase production.
Spock was looking for a place to build a factory in a place where the trade unions were not so well developed.
Because his last job was ruined by the union in New York.
One of the first places he considered was Portland, in the northeast corner of the United StatesThe trade unions there are not so developed, and the labor cost is relatively cheap.
But his colleague Noyce advised him to open his mind, knowing that there was a good place, not only with cheap human resources, but also with weak trade unions.
Moreover, with a certain amount of Western education, it is not much of a problem to communicate when speaking English. And it's also a free port, which can avoid a lot of imports and tax problems.
This place was Hong Kong, a highly industrialized manufacturing center at the time.
For them at that time, making semiconductors was actually no different from clothes.
According to Spock's recollections,Workers in those years earned about 25 cents an hour, one-tenth of that of American workers.
Not only that, but " local workers are twice as profitable as American workers and are willing to accept harder work ».
Except for being somewhat sensitive to wages.
If wages in the factory next door had risen by 5 percent, they might have quit their jobs at the speed of light and gone to work in a garment factory across the street.
In such an environment, Fairchild's Hong Kong factory was officially put into operation in 1963, and the United States semiconductor manufacturing officially took the first step to go overseas.
However, the semiconductor manufacturing that was transferred to Hong Kong at that time was actually not complete.
Fairchild converted a slipper factory in Hong Kong into a semiconductor factory, where it was only responsible for packaging and testing wafers made in the United States. By the way, part of the sales task is solved, and the chips produced can be sold directly in Hong Kong to all parts of East Asia.
In the end, thanks to the efforts of American engineers + thousands of Hong Kong workers working in three shifts, in 1963 alone, Fairchild produced 1200 million semiconductor chips.
This performance not only made the company very satisfied, but also gave peers a covetousness. In the United States, you can find such cheap labor and such efficient production capacity.
So everyone followed suit, and later other American companies, including Texas Instruments and Motorola, also began to set up factories in Hong Kong.
After tasting the delicious taste of industrial transfer, the follow-up development is out of control. It also set its sights on Singapore and Malaysia, where labor costs are cheaper.
After all, although Hong Kong's hourly wage was only one-tenth of that of the United States, it was already one of the highest in East Asia at that time.
At this point, it is already a clear trend to move semiconductor production away from the United States.
Traditional semiconductor manufacturing can be divided into three major links:Chip design, wafer fabrication, and packaging and testing.
At that time, the United States can be said to have transferred packaging and testing, which is the link with the lowest technology content and the highest labor cost, to overseas. But soon, wafer manufacturing was also targeted. After all, it is not an exaggeration to describe the semiconductor industry as alchemy.
This is river sand, the essence is silica, usually sprinkled on the road may not be picked up, if you want to go to the building materials market to buy it, about 170 yuan can get a ton, may not be as expensive as the handling fee.
This is the better kind of high-purity quartz sand in sand, and its value has increased several times, about 5w US dollars a ton.
And this is Intel's latest release of the 14900k desktop processor, priced at 4999 yuan, weighing 357g。That's about 140 million tonnes in translating.
And the raw material for making him is high-precision quartz sand.
In other words, as long as there is a way to design a chip and make it, you can make a lot of money.
Japan on the coastline is eyeing the chip fat gap.
After all, semiconductor processing is a jobAt that time, there was no need to consume too much energy, and it would not occupy too much space, which was a direct response to XP for Japan, which was relatively short of energy and did not have enough land resources.
It just so happened that the post-war United States also had the idea of using support for Japan to counter the Soviet Union.
Thanks to this, Japan has absorbed a lot of technology from the United States at a low cost.
Many of the Japanese companies we are familiar with today also took advantage of the opportunity to rise at that time.
Whether it is transistors, computers, or integrated circuits, these technologies were quickly studied by Japan after they were studied by the Americans, and then they were launched with slightly weaker performance, but cheaper imitation products.
As a result of these reasons, Japan's production technology has not been pulled down much.
Within a few years, Japan had found a way to nibble on more American meat.
In 1966, Japan, as usual, prepared to build a high-performance computer, the Hitac 8000, modeled after IBM in the United States.
In order to make this System-360, IBM recruited more than 6W new employees, obtained more than 300 patents, and overcame a series of difficulties in operating systems, databases, integrated circuits, etc.
The total cost of the project was about $5.2 billion, which in those days was equivalent to the cost of seven nuclear-powered aircraft carriers.
However, the budget for this imitation project in Japan is less than one percent, ( 0.$3.4 billion) is still a five-year phased plan.
In the end, of course, it failed unsurprisingly.
However, during the development of the HITAC 8000, Japan accumulated a lot of experience in developing new memory due to the high memory requirements of the time.
Finally, in 1968, a 144-bit N-channel MOS memory was developed.
This valuable experience helped Japan accumulate a great deal of experience in the subsequent DRAM era.
When Intel finally launched its full-fledged DRAM product, the C1103, Japan followed suit with the fastest bowl of soup, and NEC launched a similar chip the following year.
Many people may have some doubts about the rise of Japanese semiconductors.
A DRAM that can be seen everywhere now, why can it knock down many large manufacturers in the United States?
But at the time, storing data was a big problem, and older computers were even using core memory, which was not as fast as DRAM, with physical storage and reliability.
For example, if a company releases a battery with twice the energy density of today, all other parameters are all advantages, and it has made a revolutionary breakthrough in battery energy storage.
Next year, all friends may want to use this battery.
You can imagine how big this market is, Intel eats meat, and Japan drinks soup shallowly.
In the process of such research and development,The production process of the semiconductor industry has also transformed from a labor-intensive industry to an asset-intensive industry.
The wire bonding process, which used to require a female worker to hold a microscope, can now be completed with an automatic wire bonder, one factory, 100 machines, or even 10 people.
In the follow-up, an ultra-quiet workshop and a clean room were arranged, which directly improved the yield of the product.
The story that follows is the one we are familiar with.
Years of technology accumulation have made Japanese DRAM far superior to American products in terms of yield and quality.
Various companies are also showing their magic in cooperation, and this series of technologies has been integrated. In one fell swoop, it accounted for 90% of the DRAM market, and those technology companies on the American continent were even more discarded.
Even Intel, which invented DRAM, was beaten out of the DRAM market, and if IBM hadn't pulled it halfway, it might have to go bankrupt or be acquired.
It was only at this time that the United States reacted in hindsight and began to attack Japanese semiconductors.
From the adoption of legislation to determine dumping in the Japanese semiconductor industry, to the promotion of the establishment of a global division of labor in the semiconductor industry to carve up the Japanese market.
We have also talked about this content before, so I won't repeat it.
After the operation of the United States, Japan's semiconductor industry cannot be said to have plummeted, but at least the momentum of force has been suppressed.
But with that comes new problems come the way.
Japan's side has been hammered, and the chip can't be made without man.
It is impossible to build it yourself, it is expensive and troublesome, but you can't let others do the OEM with full authority, so as not to be stuck by others.
At this critical moment, TSMC came out and directly began to shout:
It's okay, I can build it, and I'm only responsible for OEM, you just need to take care of your own design."
If the high labor costs and strict labor unions in the United States were digging a hole for the relocation of the semiconductor manufacturing industry.
Now that TSMC is leading this model, it can be said that it has covered a cup of soil for the manufacture of new local processes in the United States.
is only responsible for wafer foundry and not chip design, which can be said to be perfectly in line with the appetite of American companies.
He has two distinct advantagesOn the one hand, it reduces the internal friction between chip design manufacturers.
In the past, in order to keep the chip scheme made by themselves confidential, everyone had to hide it and not design it for others, so they could only build their own production line for production.
But now it's different, TSMC is one"There is no quarrel with the world"The wafer foundry does not involve chip design, and everyone can hand over the latest chip design to me for foundry without worrying about leaks.
Moreover, TSMC itself has become a practitioner, and each chip is done, so you can summarize the aspects that need to be improved after the process is not done well, and then use it in the next round of optimization design.
On the other hand, it lowers the barrier to entry for chip design.
New players who enter the market in the future do not need to spend a lot of money on research chip manufacturingAs long as it can be designed, manufacturing problems can be handed over to TSMC.
This model can be said to have worked very perfectly and has been running for decades, but now it seems that the only problem may be that TSMC is developing too well.
This part of the "outsourcing" foundry industry chain has been rolled out of their invisible high technical barriers, making it difficult for new entrants to find a way.
In the early years, the United States moved its industry out in pursuit of low labor costs, but now this part of the industry can blossom and bear fruit overseas, and it has lived a perfect life.
It's fine if there are no geopolitical problems, but this time the United States suddenly came to its senses after sanctioning the day and sanctioning the landNow there is only one advanced process factory that can be relied on, TSMC.
Now if you suddenly want to manually transfer TSMC's output, the cost paid is different.
To figure this out, we have to look at what factories TSMC has moved over?
According to TSMC Chairman Liu Deyin, it can be seen that the first phase of the Arizona plant is expected to start mass production of 4nm chips in 2024, and the second phase of the plant will also start construction at the same time, and it is expected to start producing 3nm chips in 2026.
It's all wafer fabs, and there is no plan to build a packaging fab.
Ah yes, the packaging was a low-end production line back then, but now it has long been "the return of the dragon king". That's right, "advanced packaging" is also packaging.
To put it simply, the advanced process can determine how well a chip can perform, and what advanced packaging needs to do is to help multiple chips fit together as much as possibleSo that they can work together to play better, to achieve the effect of 1 + 1 is greater than 2.
For example, Apple's amazing M1 Ultra used the advanced packaging process to connect two chips together to achieve 2With a data traffic volume of 5 TBS, the performance of the glue chip is perfectly exploited.
NVIDIA's computing crown H100 is also an HBM memory that relies on advanced packaging to obtain better performance and lower power consumption on graphics cards.
But none of these awesome technologies can be used in the United States, and even if the Arizona fab starts all the way, the chips must be sent back to Taiwan, China for further packaging.
There are not many advantages to this cost, but let's not talk about this packaging and testing factory that has not yet been skimmed, even the wafer factory under construction is actually full of twists and turns.
At the ** meeting in Q2, Liu Deyin, chairman of TSMC, said that the American workers are not skilled enough.
Due to the shortage of local skilled workers in the United States, companies may have to temporarily transfer experienced technicians from Taiwan, which will delay the start of mass production at the first factory until 2025. ”
The Americans at the scene also began to shake the pot, giving the opposite statement:
Accusing TSMC of chaotic management in the process of building the factory, and did not do a lot of safety measures, resulting in their failure to start work smoothly. ”
Good guys, no one in the world remembers a fab that started on time, right?
Even if you add a layer of buff to the United States and let the wafer factory land smoothly, I believe that there will be a lot of chaos in the follow-up operation.
Because the core of semiconductor production, in fact, the most critical is still in people.
From the invention of integrated circuits in 1960 to the automated manufacturing of semiconductor factories in 2023.
Although from the earliest manual welding of wires, manual pouring of resin, to the current automated process production, it seems that a lot of labor costs can be saved.
However, saving costs does not mean that people are not needed, after all, people revolve around the equipment, not the equipment revolves around people.
Poor friends who often carry buckets and run experience should find that the more production line machines that can run with only a small number of simple operations, the more the phenomenon of three shifts will occur.
This phenomenon is especially evident in semiconductor fabs.
If you are a PE (process engineer), in order to ensure that the production line can not have problems, although you do not have to go to the production line to operate in person, but when it is your turn to be on duty, you still have to ensure that you can be on call, and you have to go to the production line to troubleshoot the problem.
If you are an OP (operator), you need to wear dust-free clothes every day, and you may only be able to go to the toilet two or three times in twelve hours, watching the overhead shuttle crane, listening to the roar of countless machines around you, smelling the pungent smell of photoresist, and living an isolated work life.
As long as human intervention is required in the semiconductor production process, it is necessary to recruit people who can accept such a working environment.
Although these technicians are very ordinary, they are definitely not as simple as screwing on the automobile production line.
A very funny contradiction is that TSMC's biggest attraction to the United States is to create tens of thousands of jobs, but the problem is that there are enough people in the United States to find people who can match these jobs.
According to a study by SIA (Semiconductor Industry Association), the number of jobs available in the U.S. semiconductor industry is expected to increase from 34 by 203050,000 to 460,000, but according to the current talent development plan.
At that time, there may be 6 of these new jobs70,000 jobs are not hired.
Moreover, the report also pointed out that there are only a few students studying STEM (science, technology, engineering, mathematics) at present, and even those who are currently studying these majors will not give priority to getting involved in the semiconductor industry when they are employed.
Speaking of which, I believe you already have the answer to whether American semiconductor manufacturing can recast its glory.
Perhaps the United States hopes to use the "CHIPS Act" to inject a shot in the arm for semiconductor manufacturing.
But as in 2011, Obama did the same question as Jobs asked.
Why can't we have Apple's iPhones, iPads made in the United States, and why can't we bring these jobs home?
These jobs won't come back