Samsung Foundry (SF) and TSMC have been gaining foundry market share in recent years as a number of their rivals stopped developing leading-edge process technologies. As a result, Taiwan-based TSMC evidently gained more clients. Still, Samsung Foundry believes that as tensions between China, Taiwan, and the U.S. rise, more customers will start using SF's services to mitigate geopolitical risks.
"When I meet customers these days, they think the current geopolitical risk is serious; they need their second source," said Sim Sang-pil, a vice president of Samsung Semiconductor responsible for foundry business planning, reports Bloomberg. "Samsung Foundry has many opportunities with customers who want to have that second source."
In a bid to meet the demand of its parent company and an increasing number of foundry customers, Samsung's contract semiconductor business aims to expand its production capacity by three times by 2027, the company revealed in early October. By that time, the contract chipmaker plans to adopt 2nm (in 2025) and 1.4nm (in 2027) fabrication processes and remain a little bit ahead of TSMC in the nanometer race (which no longer has to do anything with actual physical measurements).
While being a second source for something has historically been a more or less sustainable business model, it may not necessarily work in the future as more advanced process technologies emerge. For example, according to International Business Strategies, an advanced chip design implemented using a 5nm-class fabrication process costs around $540 million, up from around $300 million for implementing a 7nm-class design.
The implementation cost will increase even further with smaller nodes, so implementing one design for two different nodes might be too expensive for the vast majority of fabless chip developers. Furthermore, it remains to be seen whether they are eager to implement one design using both SF and TSMC nodes, as chips made using different technologies tend to offer different performance, power consumption, and yields.
For Samsung, there is another difficulty. At times, the company introduces new process technologies ahead of TSMC. For example, Samsung Foundry's 3GAE (3nm-class, gate-all-around early) uses gate-all-around transistors, whereas TSMC's N3 (3nm-class) still relies on FinFET transistors. As a result, Samsung Foundry's and TSMC's production nodes are vastly different, so it will be particularly hard for fabless chipmakers to source the same chip from two manufacturers as these chips will offer different performance, power consumption, different transistor density, and different costs. It's unclear how Samsung plans to address this issue.
Meanwhile, Samsung has chances to steal clients from TSMC as the company plans to significantly expand its production capacity in the U.S. with its all-new fab near Taylor, Texas, that is set to come online later this decade. Of course, Samsung will also have to fight Intel, which wants to become the second-largest foundry by the decade's end. Intel will also position its fabs in the USA, Ireland, and Germany as its strategic advantages over TSMC. Furthermore, TSMC is building its new fab in Arizona, attracting customers who want to reduce the potential risk of producing their chips in Taiwan.
