#Knowing #material丨Outsourcing #TSMC #overtaken #AMD #Intel #king #chips #slips #throne
In Silicon Valley where technology companies are shining, Intel’s light has been shining for many years.On July 24, the second-quarter financial report showed that Intel’s revenue for the quarter was US$19.7 billion and net profit was US$5.1 billion.
With such a large business plate, Intel’s Q2 revenue and profit can also record 20% and 22% growth respectively, which is very rare.
However, the beautiful report in the second quarter did not return the stock price as expected.Today, Intel’s 7nm process chips again bounced due to yield issues, and Intel’s key personnelThing, chief engineerVenkataAs a result, Intel’s stock price fell 10%, and its market value evaporated by $25 billion.
In sharp contrast, AMD, a competitor that released its earnings report, had less than one-tenth of Intel’s revenue, but its intraday trading rose by 10%, and its market value even surpassed Intel for the first time.
Intel’s stock price trend in the past month, the decline on July 24 has not yet recovered
From Moore’s Law, which created the golden rule of the chip field, to the “Pentium Core”, Bill Gates has been called the “King of Chips” Intel has been outstanding for decades.TSMC and AMD used to be surrounded by Intel.And now, both have already moved towards 7nm and higher process.
Intel’s chip process technology skipping ticket is not embarrassing, this kind of thing has happened before. More importantly, Intel announced to the outside world that, in order to be “more pragmatic”, it has acknowledged the rumor that it is considering outsourcing part of its chip production to a third party.
Intel and TSMC have always been in a cooperative and competitive relationship. When TSMC was just established, orders from some of Intel’s backward manufacturing processes allowed TSMC to gradually come out of a loss-making state, and it was subsequently recognized by major manufacturers.
However, the cooperation between the two parties is different this time. According to media reports in Taiwan, Intel will begin mass production of 180,000 processors or graphics chips using TSMC’s 6nm process next year. The quantity is not low and the technology used is also one of TSMC’s most advanced technologies.
Between TSMC and Intel, from the previous low-end foundry to today’s high-end cooperation. In this vigorous battle over the chip manufacturing process, the latter has been surpassed in literal manufacturing process numbers.
This era of semiconductors dominated by Intel is slowly turning over.
TSMC was inseparable from Intel at the beginning
In business, there are no absolute enemies or absolute friends. Speaking of it, TSMC’s start cannot be separated from Intel’s initial support.
The smaller the width of the circuit carried on the chip, the faster the processor will run, but the chip area is limited, if more circuits are packed in the square, the diameter of the circuit will be smaller and the required process will be more complicated.
The process node of the chip has gradually developed from the micron (μm) level to the current 32/22/16/10/7/5/3nm (nanometer), and the half-node 28/20/14nm is also common.
The picture of the chip process node comes from CITIC Securities
Regarding the evolution of chip circuits,One of the founders of Fairchild Semiconductor and later the founder of IntelMoore, as early as 1965, already proposed the “Moore’s Law” that is still in use today-“The number of transistors integrated on a semiconductor chip will double every 18 months.”
Intel’s first processorThe i4004 processor is basically an actual test of Moore’s Law. During this period, Texas Instruments and IBM also launched related products.
It was not until 1993 that Intel’s iconic “Pentium Chip” appeared, which helped Intel to become the world’s largest semiconductor company in the following decades, and this process war really began.It is also the year when TSMC turned over.
The historical cycle of the rise of great powers is always repeated.
In the late 1980s, under the background of the U.S.-Japan trade war, the US semiconductor industry began to shift, and under the US attack on Japan, the semiconductor industry in Taiwan was given an opportunity to start.
TSMC also got its start under this background. However, in the first few years, TSMC’s production capacity was insufficient, the yield was too low, and it had been at a loss.
Zhang Zhongmou, the founder of TSMC, who is now known as the “chip king”, took a lot of effort to win some orders from Intel. At that time, TSMC, which was many generations behind Intel, was not a problem. Under the United States’ blow to Japan’s support of industries in Taiwan, Intel also helped TSMC produce 0.8μm technology.With the endorsement of Intel’s major manufacturers, TSMC also won orders from STMicroelectronics, Broadcom and Nvidia in the same year, becoming a true chip foundry.
Intel is not idle either.The “Pentium” CPU and Microsoft’s windows95 realize the binding of software and hardware, and the win-tel alliance built is the most powerful combination in the personal PC era. After that, Intel took the lead again and developed new processor brands “Xeon” and “Celeron”, respectively entering the professional processor/low-end processor field, eroding the personal computer market.With the support of Microsoft, Intel’s chips swept the world and became the largest chip manufacturer.
The battle for semiconductor routes: the rise of foundry, the ebb of IDM
There are three production modes in the semiconductor industry. One is the Fabless mode represented by HiSilicon/MediaTek/Qualcomm, that is, a chip supplier without a factory, which is only responsible for the design and sales of the chip, and the production/testing/package links required for the chip OutsourcingThe second is the Foundry foundry model represented by TSMC, which echoes the former. Companies in this model undertake manufacturing tasks from multiple design companies.
The third is the IDM model represented by Texas Instruments, Intel, and Samsung, which undertakes all the links from chip design to manufacturing to packaging.In the early stages of the development of the integrated circuit industry, companies generally used this large-scale IDM model.
However, it is too difficult and too expensive to make cores. Although the IDM model can grasp the overall situation, the rate of return on production lines is too low.In recent years, in addition to Intel’s insistence on self-developed and self-produced chips, other companies would not be able to play without sufficient funds.The semiconductor industry is bound to move towards a fine social division of labor.
In the initial cooperation between Intel and TSMC, Intel undoubtedly had an overwhelming voice.
Whether it is TSMC or Samsung, they have strong technical dependence on American semiconductor companies headed by Intel/IBM.Similar to the “Varner Agreement”, IDM manufacturers cannot disclose the latest generation of technology to the foundry in order to ensure their technological leadership.Therefore, TSMC initially only had access to some of Intel’s relatively backward foundry tasks.
In the game, the foundry companies all want to become collaterals rather than vassals of IDM manufacturers.In order to achieve faster technological catch-up, TSMC is alsoThrough in-depth cooperation with other major IDM manufacturers, we can get rid of technical dependence on Intel.
TSMC factory pictures from the official website
In 1998, TSMC put forward the “Qunshan Plan” to give the actual order demand of five semiconductor giants including Texas Instruments and Motorola, adjust its own technology upgrades and increase production capacity.This has also helped TSMC form a positive cycle-it has a deeper understanding of upstream demand, can better produce orders, and has the capital to continue to polish the follow-up technology, improve the yield rate of the production line, and then get more orders. , Continue to share the high technical costs.
The aggressive TSMC has become a backup force for major IDM manufacturers in production. After many efforts, TSMC also aligned with Intel on the 0.18μm process for the first time in 1999. This is of course still on paper. TSMC at this time Because of the yield problem, it is actually still in the catch-up stage.
Later, TSMC’s ascent was inseparable from the decision of Jiang Shangyi, the person in charge at the time, and the support of ASML (ASML), a lithography equipment manufacturer born out of Philips.
45nm in 2002 and 5nm today are two thresholds in the development of semiconductors. One of the breakthroughs requires the use of an immersion lithography machine and an extreme violet lithography machine (EUV) produced by ASML.
ASML’s first immersion lithography machine was jointly developed with TSMC. The research and development of the extremely purple lithography machine was also jointly funded by Samsung/TSMC/Intel-the former helped all manufacturers including Intel to collectively cross the 45nm mass production test Under the background that most manufacturers choose to abandon the research and development of advanced processes above 7nm, TSMC is also the manufacturer with the most EUV resources, and ASML is also the first to help TSMC make rapid breakthroughs in advanced processes.
ASML lithography machine pictures from the Internet
At present, CPU is mainly the vertical division of labor mode of Fabless+Foundry. The competition between Intel and Nvidia and AMD is essentially the competition between Intel as IDM and TSMC. andIn recent years, Intel has adopted more and more third-party foundries, not only because of capacity issues, but also technical factors, which also indicate that Intel’s IDM model is being torn apart and reformed.
Other IDM manufacturers have successively reformedFor example, AMD spun off its CPU production business in 2009 to establish GlobalFoundries (GF).
But for Intel, a chip company that has dominated for decades, the adjustment of the model has never been an easy change.
Intel slows down, TSMC is chasing
Intel’s lagging behind in advanced manufacturing processes is not an overnight result.
A very interesting example is that at the 30th anniversary celebration of TSMC in 2017, executives from Apple, Qualcomm, Broadcom, Nvidia, Amou, ASML and other companies all came to Taiwan, and one of them was singled out and it is still an industry. Heavy players on the chain.
Allies and competitors are there, but Intel is missing.
TSMC’s model is destined to need to work together with industry opponents and “fight alone”Intel is slowly slowing down.
The mass production time of each fab process node. The picture comes from CITIC Securities
This can be clearly seen from its development strategy. Intel first summarized the “tick-tock” model in 2007. This name is derived from the sound of the clock’s second hand. Each tick represents an updated process technology, and tock represents an architectural improvement.A tick-tock is a cycle, which takes two years, which basically corresponds to the speed of Moore’s Law iterating the next generation process in 18 months.
2014 was the year when Intel and TSMC truly opened the gap. ThenIntel entered the 14nm era, but decided to “squeeze toothpaste”, adjusted the two-year cycle to three years, and inserted a one-year optimization period on the basis of the original “process technology-architecture”.
But in actual operation,Intel’s advanced cycle in chip manufacturing has been lengthened again and again——The 10nm process that was originally planned to be launched in 2016 was finally delayed until 2019. The 7nm process technology that should be launched in 2019 has also been delayed.
At the same time, including Samsung, TSMC’s 10nm process was mass-produced in 2017, and the latter was first used on the Apple A11 chip (the model is equipped with iphone8/8plus/X), and 7nm was mass-produced in 2019 , 5nm is in 2020, and 3nm has also made preliminary progress.
Intel is also aware of this, so it has done a lot of work on increasing the density of transistors. Some scholars also pointed out that the density of transistors on Intel’s 10nm process is actually twice that of TSMC and Samsung, and is approximately equal to the density of TSMC’s 7nm.
Intel needs to maintain an advantage in density,TSMC also needs to use a fast-advancing process to make a return.
The competition has already begun.
When Intel was running for foundry, TSMC’s production capacity, especially the production capacity of the 5nm process, had been occupied by Apple and Huawei HiSilicon in advance.Intel’s problem is how much capacity it can rely on its own factories and third parties to provide Intel with while opening up 7nm technology foundry in the future.
Although the process has fallen behind, users do not need to feel sorry for Intel. Intel has many ways to go. Whether it is in storage, PC traditional business, or new business such as data center, Intel is undoubtedly the overlord, and it also has chess pieces in the Internet of things and autonomous driving (Mobileye).
Many people have also determined that even if Intel regains the established rhythm of Moore’s Law that it invented, it will not be able to catch up with TSMC on 5nm/3nm.But the competition between the former “king of chips” and the “king of foundries” is a speed race for advanced manufacturing processes in the semiconductor industry and a battle for the status of the world’s advanced manufacturing industry.
(I’m Qiu Xiaofen, the author of 36氪, I am concerned about the field of science and technology, and you are welcome to add WeChat: 13750430427 for exchanges and breaking news)