Intel's sunset: Can it make a comeback by manufacturing chips for others? (Down)

For Intel and the United States, the adoption of the foundry model represents a major change

Shenyi Bureau is a compilation team under 36 Krypton, focusing on technology, business, workplace, life, etc. fields, focusing on introducing new foreign technologies, new ideas, and new trends.

Editor's note: Once upon a time, Intel was the banner of the semiconductor industry, with leading technology, huge profits, and an extremely comfortable life. It has never considered doing OEM work for others. After all, the investment in doing this is too high and the construction period is too long. By the time you finish building it, the technology may fall behind again before the cost is recovered. But things have changed recently. Chip shortages caused by various reasons have made foundries led by TSMC a lot of money, and undertaking different customer needs has also allowed foundries to continuously improve their technology. Not only is Intel not making as much money now, but its technology is lagging behind. So this semiconductor giant plans to attack for the second time and engage in foundry business. This time, can it succeed? The article is from a compilation. Due to length constraints, we have published it in two parts. This is the second part.

Image source: Intel

Key points:

Rather than taking a few small steps, taking too big a step at once is more likely to slip

The real challenge in the foundry market is to ensure that every customer's needs are met

Making customer trust as important as manufacturing capabilities

For Intel and the United States, Adopting the foundry model represents a major change

Intel is in the sunset: Can it make a comeback by manufacturing chips for others?

It’s easy to take too big a step

At the same time, TSMC is on the rise. Throughout the 2000s, TSMC was flooded with financial resources, allowing them to invest in larger and more advanced wafer fabs. The company proved that the foundry business was not only viable, but also highly profitable.

Soon, TSMC began to extend an olive branch to Apple, a move that ultimately pushed the company into the lead. At the time, Apple had begun designing its own chip for the iPhone, and although early versions of it were manufactured by Samsung, a patent dispute led Apple to look for alternative manufacturers.

Apple has been testing TSMC chips for several years. In 2015, they began to directly compete between TSMC chips and Samsung chips on the iPhone 6s. Samsung also has a nominal advantage - the transistors it makes are smaller and should be more efficient than TSMC's. But in most cases, this is not the case. TSMC's chips maintained their position, and by the time the iPhone 7 was launched, Samsung was out. Analysts praised the phone as "extremely thin and light, with a powerful processor, and TSMC has been in iPhones ever since.

TSMC has mastered the art of low-power processors, in part because It has improved its technology in the process of foundry chips for hundreds of different customers, said Hu Chenming, a professor at the University of California, Berkeley and former chief technology officer of TSMC: "Once you have a large number of foundry customers, the product cycles of these customers are not the same. Maybe it's all synchronous. Almost any time there is a new technology, there will be some customers willing to pay for it.

“Compared with taking a few more small steps, it is easier to slip if you take too big a step at once.

——Hu Chenming, a professor at the University of California, Berkeley

Foundries that can juggle multiple customers and technologies can quickly leapfrog competitors. TSMC's scale, for example, allows the company to master extreme ultraviolet lithography faster than anyone else, reducing the time it takes to make advanced chips. Hu Chenming said that large foundries generally have dozens or hundreds of customers, which encourages them to take smaller steps because there will always be. There are interested customers. He said: "It is easier to slip if you take too big a step than to take a few small steps. Companies also recover more easily from missteps.

Historically, Intel has generally adopted a Great Leap Forward approach that attempts to reflect Moore's Law. According to Moore's Law, the density of transistors will double every 18 to 24 months. For most of the company's history, Intel has been successful, rolling out impressive updates that kept the company multiple steps ahead of its competitors. But then in 2015, Intel slipped. The company announced that the launch of its 10-nanometer chip will be postponed. In 2017, it announced another delay. Soon, the industry giant not only failed to match its rivals, but also fell behind.

At the same time, other companies began to follow Apple's lead and design their own chips instead of buying off-the-shelf components like those sold by Intel. As TSMC takes the lead, more and more companies are turning over their designs to TSMC for manufacturing, giving the company more opportunities to improve its processes. Today, about 90% of cutting-edge chips are made by TSMC, while the rest are made by Samsung.

Hu Chenming said: "It is extremely difficult to develop a new generation of technology. In terms of extremely leading technologies, Intel is now lagging behind TSMC and Samsung, which can all be traced back to the fact that Intel is not involved in foundry.

Trust issue

Paul Triolo, head of EurasiaGroup’s geotechnical practice, said: “The real question is, will Intel’s entry into the foundry business help them return to the leading position?

Success in the semiconductor industry depends largely on scale. Selling more means more opportunities to perfect the process, and perfecting the process will help get to the next process node, Triolo said. That's the case. At some point, they will reach a critical mass of capabilities and customer relationships. Then a virtuous cycle is created—that’s what you need.

Hu Chenming said: "Can the United States maintain its technological leadership without engaging in foundry? This is indeed the problem Gelsinger wants to solve. If it does not get involved in the foundry business, Intel cannot maintain Technology leadership - I certainly agree with this conclusion

For Intel and the United States, the adoption of a foundry model represents a significant shift. Historically, leading American companies have either designed and manufactured chips. IDMs, either play the role of fabless designers that outsource production to another company, in part because computer chips make most of their profits from design and sales, not manufacturing. Jennifer Kwan, associate director of research, said TSMC’s success as a pure-play foundry “is almost an anomaly historically. Not only that, she added: "TSMC has proven that this is actually a profitable business. Many people did not expect the result to be like this.

Now, Intel seems to be putting its full enthusiasm into So far, Intel has announced that Qualcomm and AWS have become its customers, and Klaus Shugraf, vice president of foundry strategy and planning, said that there are 100 other companies that have shown signs of this. Interest. “We have a long order list and customers come from all walks of life. Currently, Intel seems to be focusing on the high-performance market. Shugraf said: "We find that the growth in the next five to ten years will mainly come from the leading edge.

However, the real challenge in the foundry market is not whether it can attract customers or even whether it can develop better products. Good technology—it’s about ensuring that every customer’s needs are met, said Hunter, a researcher at Georgetown University. One of the things that makes TSMC unique is that it “can meet a very wide range of customer needs, especially. It is the demand of chip design companies. "Intel doesn't have that much experience in this area.

As a pure-play foundry, TSMC's customers don't have to worry about unique elements of their designs being managed by manufacturers to steal them and use them in their own chips. Stanford University professor H.-S. Philip Wong, who led TSMC’s R&D from 2018 to 2020, said: “When communicating with TSMC, one of the basic principles they mentioned is customer trust. This is as important to them as manufacturing capabilities.

Intel appears to have learned from its last failed attempt to launch a foundry, which died down five years later in 2018. Unlike last time, the new Intel Foundry Services is an independent business unit reporting directly to Gelsinger. Sugarraf said the company has been building key relationships with suppliers over the years so that fabless companies don't have to shoehorn their designs into Intel's way of doing things. Moreover, the process can be set up so that customers do not have to worry about their secrets appearing in other chips.

Employee Margaret Henschel is walking through Fab32, Intel's large-scale manufacturing facility in Chandler, Arizona. Tim Herman / Intel Only those who know can know. He added that the company is also committed to "being fair and transparent with customers in how capacity is allocated. This looks a lot like dealing with other foundries.

The acquisition may help Intel quickly gain access to these Earlier this year, there were rumors that Intel was negotiating with GlobalFoundries, but so far there has been no progress: “Intel’s previous attempts to do foundry failed mainly because it didn’t do it. To be customer-centric, it is not focused on manufacturing a large number of parts for a large number of different customers. But GlobalFoundries did it.

Sugraff acknowledged that adapting to a wide range of customer requirements can be a test for the company, which has historically relied on close relationships between its own in-house designers and its manufacturing operations. But he thinks this is a challenge Intel can overcome. "It's going to make us a little nervous. But fundamentally, we're ready for this.

America at the Crossroads

It's going to take a while for anyone to Observers believe it will take at least three years to determine whether Intel's foundry ambitions are a success, and more likely five years to build a new chip. Designs also take months or years to test and produce.

Shih, a professor at Harvard University, said: "These things always take a certain amount of time.

In the long run, Intel's success also depends on U.S. industrial policy. TSMC's costs are lower than Intel's, This is partly due to strong support. The Semiconductor Industry Association estimates that over 10 years, the cost of building and operating a fab in the United States is 30% higher than in Asia, with about half of the cost difference attributed to the government. Subsidies.

The U.S. Congress is considering injecting about $50 billion into the semiconductor industry to stimulate domestic research and development and construction of wafer fabs. This can promote fair competition to a certain extent, but it will also bring about new developments. Challenges. Jin Zhijie, dean of the School of Engineering at the University of California, Berkeley, said that if this bill is passed, the semiconductor industry will create tens of thousands of new jobs every year. She said: "This means that we need 5,000 to 10,000 fresh graduates every year. born. No single university, not even a university system like the University of California, can meet this workforce development need.

Jin Zhijie, dean of the School of Engineering at the University of California, Berkeley, talks with freshmen taking management, entrepreneurship and technology courses. David Paul Morris/Bloomberg

In the short term, highly skilled immigrants will help fill the gap, Hunter said. But in the long term, the U.S. needs to find "ways to increase the amount of talent," he said. "Talent is needed at every level, from people with professional degrees to people with Ph.D.s." Along these lines, Jin Zhijie and her colleagues have been developing a program that spans industry, universities, and community colleges to meet the need. She said: "We hope to operate the best practices in semiconductor education and promote these practices across the United States.

Whether it is the United States or Intel, there is still a period of time to regain the leadership in semiconductor manufacturing. It's a steep hill to climb, but none of those interviewed had any intention of giving up on their opportunity: "I think there's an argument to be made that having the 'whole company' focus on manufacturing improves TSMC's chances of overtaking Intel," Hunter said. possibility. But I don't think anything was inevitable about what happened. The United States has talent from all over the world, we have strong intellectual property protections, and we have an ecosystem of design companies eager to work with American foundries.

Berkeley professor Hu Chenming agreed. He said: "We have the leader, which is Intel, and we have the technology base that is basically on par with the best universities in the world. And we still have the best universities in the field. Maybe what they lack is an urgency. Feeling. Hu Chenming added that success “really depends on whether you think you have to do it.

For now, many believe the United States — and Intel in particular — has no choice. Maire said: "We have seen this crisis exacerbate a relatively small shortage of semiconductors and send the global economy into chaos. The success of the United States depends in large part on this.

Translator :boxi.