Ex-TSMC executive: Samsung has the technology — but must spin off foundry to succeed

"Foundry is a service business — not a technology-intensive industry. Whether Samsung has good technology and whether Samsung can run a good foundry are two completely separate questions. Technology is a necessary condition for doing foundry, not a sufficient one. If Samsung Electronics wants its foundry business to succeed, it needs to spin off into an independent company and run a real foundry business."

Yang Guang-lei, director of industry-academia innovation at National Taiwan University of Science and Technology and adjunct professor at National Taiwan University, made the remarks June 3 at the Haoyang Laboratory building on the NTUST campus in Taipei. Memory chips and foundry are fundamentally different businesses, he said, and Samsung Electronics cannot afford to let its foundry division languish in the shadow of its memory operation. Samsung currently houses its memory, foundry and System LSI divisions together under its DS (Device Solutions) semiconductor unit.

Yang studied electrical engineering at National Taiwan University before earning a doctorate in computer science from UC Berkeley, then joined TSMC as director of R&D infrastructure engineering — one of the so-called "Six Knights" credited with building the foundation for TSMC's growth. He later served as an independent director at SMIC and a technology adviser to Intel Foundry before joining National Taiwan University as an adjunct professor in 2023.

With more than 30 years in the semiconductor industry, Yang described the current chip supercycle as an "artificial spike" driven by the massive capital of Big Tech — a surge unlike anything in history and, he said, the steepest ever recorded. That demand shock, he explained, has triggered shortages across the board, including in DRAM.

"At some point it will reach equilibrium and gradually saturate, but it is hard to predict how long that will take or how fast it will happen," he said. He warned, however, that oversupply could emerge once the boom slows, just as chipmakers are scrambling to expand capacity. "Until now, capacity has been expanded gradually in line with demand, keeping supply just slightly below demand. Right now, capacity is far below demand — that is a very unconventional situation," he said.

When oversupply does hit, Yang said, the pain will center on DRAM — as it did in the two previous semiconductor "chicken game" price wars. That dynamic, he argued, is precisely why foundry matters as a business.

"In memory, once a new technology generation arrives, the old equipment becomes worthless. But in foundry, even legacy equipment keeps printing money. The financial stability is almost unimaginably strong," he said. "Invest in foundry and you can keep recovering that investment for 20 to 30 years. Memory equipment, at best, gives you five to seven years — the return on investment simply cannot match foundry."

A foundry operator also has no stake in who wins the CPU or GPU wars between Nvidia, AMD or anyone else — they are all customers. "If you look at historical data, Intel and Samsung's sales swing wildly with the market cycle, but TSMC has always grown steadily," Yang said. "That is because it does not depend on any single product."

Yang said Samsung Electronics' foundry operation has sufficient technology but will struggle to become a true rival to TSMC. The core of the foundry model is the customer, he said, which means a service ethos — what he called, in the extreme, a "servant mindset" — is the most critical ingredient. Samsung, however, has grown accustomed to a product-centric business model built around televisions, smartphones and memory chips, leaving it with what he described as a "great power mindset." Taiwan's chipmakers, by contrast, developed an OEM culture precisely because their domestic market was never large enough to absorb finished products. TSMC and Samsung, he said, have fundamentally different cultural DNA.

Yang also said the memory supercycle that has run since last year is, paradoxically, becoming an anchor around Samsung foundry's neck. "There is a Chinese saying: 'Success is the mother of failure.' You need failure to find success — but because memory is doing so well, there is absolutely no pressure to change or pivot," he said. "The memory boom means there is not enough bandwidth to pay attention to the foundry side, and there won't be going forward. Inside Samsung, the people in the foundry division already feel like second-class citizens being pushed aside by the memory division."

"It is very hard for a small new-business unit to survive inside a giant corporation — that is exactly why Samsung needs to spin off foundry as a separate company," he said. "Unless both memory and foundry are world-class and can be offered together as one seamless system, having the two divisions under the same roof is either unattractive or unlikely to generate meaningful synergy." The remarks stand in direct contrast to Samsung Electronics' own stated strategy of competing as an integrated device manufacturer, or IDM, by linking its memory, foundry and packaging capabilities.

For Samsung foundry to succeed, Yang said, a spinoff must be accompanied by an infusion of outside talent from leadership down to the execution level to build a new culture. The company must also abandon the ambition of excelling at every process node and concentrate 100 percent of its energy on the specific processes and nodes where it is most confident. Without the backing of the US government, he added, Intel is no match for Samsung. Because Samsung's technology is already solid, changing its execution and organizational culture is what could make it a foundry powerhouse, he said.

On whether sub-1-nanometer manufacturing is achievable, Yang was unequivocal. "Processes at 7 angstroms, 5 angstroms and 3 angstroms will all appear — that is technically inevitable," he said. "But I can guarantee that the numbers attached to those nodes will have nothing to do with the actual physical geometry inside the chip. Single-chip planar shrinkage has hit its limits, so advanced packaging — integrating different chips into one — has become the decisive battleground, and it is already the dominant trend." That, he said, is why the industry must focus on packaging technology.

“한국서 할 일 많아…곧 다시 올 것” 거침 없었던 젠슨 황의 4박 5일…제조업·SW 역량 다 가진 韓서 ‘AI 유니버스’ 야심

https://biz.heraldcorp.com/article/10767019?sec=027