Why is it harder for China to build chips than rockets: building factories is just the start-up phase of smashing money

The chip shortage has become a headache for all industries around the world. But due to the atomic level manufacturing process, cleaner than the operating room clean space and often worth millions of dollars of manufacturing equipment, making it extremely difficult to increase production of chips. Each chip manufacturing plant to contribute $3 billion in profits to avoid losses, the industry only a total revenue of $ 188 billion Intel, Samsung and TSMC three companies have the ability to build more factories.

The chip shortage has hit car makers and technology giants hard, giving alarm bells ringing in all industries around the world. This chip crisis raises a fundamental question: Why can’t we produce enough chips?

The answer is both simple and complex. In short, it is extremely difficult to manufacture chips, and it is becoming increasingly difficult.

There is a saying in the chip industry is quite flirtatious, “chips are not building rockets, but much more difficult than building rockets.”

To complicate things, the construction of facilities to build chips not only takes several years, and often billions of dollars. For companies in the industry, even if they can afford to spend money and time, if the manufacturing technology can not catch up with competitors, will lose the advantage. Craig Barrett, former president of Intel (Craig Barrett) said that the company’s chips are the most complex devices ever manufactured by mankind.

The world is considering manufacturing its own chips, but success is not a sure thing.

It usually takes more than three months to build a chip, with large factories, completely dust-free clean rooms, multi-million dollar machines, tin fluids and lasers. The ultimate goal of integrating so many elements is to turn a block of silicon into a tiny switching network of billions of transistors. These transistors form the basis for the circuits of various devices, ultimately allowing a telephone, computer, car, washing machine, or satellite to perform their critical functions.

So small, yet so complex

Most chips are groups of circuits that run software, process data and control the functions of electronic devices. The circuits in different chips are arranged differently and serve different purposes. Take NVIDIA’s graphics graphics chip, the GeForce RTX3090, for example, which is currently the best chip for converting computer code into video game graphics.

NVIDIA’s graphics graphics chip GeForce RTX3090

Chip companies always want to package more transistors in a single silicon wafer, thus increasing performance and device power efficiency. The first chip Intel made, the 4004, came out in 1971 and contained only 2,300 transistors of 10 microns in size. For the next few decades, Intel’s leadership in the chip industry remained undisputed. But by 2015 to 2020, Intel’s dominance of this position is not maintained. Competitors TSMC and Samsung Electronics began to make chips with smaller transistor sizes, with individual transistor sizes even as small as 5 nanometers, or 5 billionths of a meter. By comparison, the average width of a human hair is 100,000 nanometers.

Chip design and manufacturing

Cleaner than an operating room

An exceptionally clean clean room is needed before putting silicon wafers into a chip manufacturing machine. A single transistor is many times smaller than a virus, and a tiny speck of dust can cause serious damage and burn millions of dollars for nothing. To reduce this risk, chip manufacturers choose to house their manufacturing machines in clean rooms that are essentially free of any dust.

To maintain this clean environment, the air in the clean room is also constantly filtered, and very few people get in. If there are more than one or two head-to-toe wrapped employees appearing next to the chip production line, it may indicate that something is wrong with the chip production. The talented engineers who develop chip products are actually thousands of miles away.

Even with all these precautions, the silicon wafers used to make the chips cannot be directly touched by workers or exposed to the air. They are placed in boxes and moved back and forth with robots mounted on tracks in the cleanroom ceiling. The wafers are only ejected from the box’s safety device when they enter the manufacturing machine, and only then do they enter the critical process of chip manufacturing.

Atomic-level manufacturing process

There are up to 100 layers of material on a single chip. The material has to be stacked and then some of it removed to create the complex three-dimensional structure that connects all the micro-transistors. Some of these single layers of material are even as thin as an atom. Machines made by companies such as Applied Materials, Panlin Group and Tokyo Electron need to deal with a range of variables such as temperature, pressure, electric fields, and magnetic fields to achieve this goal.

One of the most difficult parts of the manufacturing process is lithography, which is currently done primarily by machines made by Asmac. The lithography machines produced by Asmac use light to etch the material deposited on the silicon wafer. The corresponding etched pattern eventually becomes the transistor. The scale of all this is so small that the etching can currently only be done using extreme ultraviolet light, which only exists naturally in space. To reproduce this process in a controlled environment, Asmac’s lithography machine irradiates the tin solution with laser pulses. When this metal evaporates the required extreme ultraviolet light is produced. But this is not enough; the wavelength of light needs to be made shorter with mirrors.

There are more than 59 types of equipment for the front end of the photolithography machine alone, and eight types of equipment for the back end. The silicon wafers have to go through multiple steps such as oxidation and coating, lithography, development and drying, etching, doping, metal deposition and etching to create the finished wafers.

Huge Costs

Chip manufacturing plants run 24/7. This is done because the cost is too high. It costs $ 15 billion to build an entry-level factory capable of producing 50,000 wafers per month. Most of it is spent on specialized equipment. Sales in this market exceeded $60 billion for the first time in 2020.

Intel, Samsung and TSMC, three companies with combined revenue of $188 billion last year, have invested the most in building chip manufacturing plants. Each factory costs more than $ 20 billion, and the technology used is more advanced. TSMC will invest up to $ 28 billion in new factories and equipment this year. And the United States plans to provide $50 billion in investment in the chip industry in five years.

No matter how much money is spent, chip factories will be obsolete in five years or less. In order to avoid losses, each factory of chip makers must contribute $3 billion in net profits. Currently only the chip industry’s largest companies can afford to build multiple factories.

And the more chips are built, the better they can be built. Chip yield is still the most critical measure, if less than 90% will cause serious problems. But to improve the yield, chip makers can only buy lessons by constantly burning money.

The chip industry’s brutal economic model means that fewer and fewer companies can catch up with the trend. The industry as a whole ships about 1.4 billion smartphone chips a year, but most of them are made by TSMC; Intel has an 80 percent share of the computer chip market; and Samsung dominates in the field of memory chips. For all other companies, trying to enter this market and stand firm is by no means an easy task.