IBM’s recent 2nm breakthrough might be perceived as a challenge for TSMC, the current industry leader in chip manufacturing.
While TSMC and its arch competitor, Samsung, are both currently manufacturing at the 5nm process, TSMC has already undertaken intensive R&D at 2nm, which is scheduled to enter risk production in 2023. Meanwhile, Samsung is still developing its 3nm process. It entered risk production in early 2020, and was originally scheduled to enter mass production early this year. Unfortunately, the pandemic might have delayed it to 2022. At the 2nm node, however, few are known regarding Samsung’s progress.
Considering IBM’s strategic partnership with Samsung and its recent R&D collaboration with Intel on next-generation logic technologies, IBM’s 2nm breakthrough is considered by some to be a significant advantage for Intel and Samsung – both TSMC’s main competitors.
IBM claims that its 2nm development can fit up to 50 billion transistors on a chip with the dimension of 150 square millimeters (MTr/mm2). In other words, it represents a transistor density at 333 MTr/mm2. In comparison, TSMC’s most advanced chips at the 5nm node feature 173 MTr/mm2 and Samsung’s 5nm chips feature about 127 MTr/mm2.
Samsung aims to leapfrog TSMC
Most significantly, IBM’s application of Gate-All-Around (GAA) transistors on the 2nm process is perceived as a milestone, and can potentially benefit its partner Samsung. Samsung is introducing GAA at 3nm, while TSMC won’t use it till 2nm. Intel, currently lagging behind the two, plans to use GAA on its 5nm process.
With FinFET (Fin-Shaped Field Effect Transistors) giving way to GAA below the 3nm process, it has been the latest battleground contended among the world’s leading foundries. By being the first to adopt GAA at 3nm, Samsung already hoped to leapfrog TSMC. Ever since TSMC ended Samsung’s exclusive partnership with Apple in 2016, with the latter opting for TSMC’s 16nm process, Samsung has been fighting to beat TSMC.
Now, IBM’s success with GAA at 2nm would only further empower Samsung and potentially Intel.
A GAA-based ecosystem needed
Despite IBM’s success, it remains too early to declare a winner as foundries race to continue Moore’s law. It is worth noting that IBM also pioneered the development of 7nm in 2015 and 5nm in 2017, but IBM’s first 7nm chips had not entered mass production until 2020. IBM’s announcement could, however, hasten the formation of a GAA-based ecosystem by validating GAA’s compatibility with 2nm nodes.
Previously, the chip industry had been unsure of GAA’s feasibility, and it hindered the creation of a GAA-based ecosystem. That, in turn, partly explained why TSMC opted for FinFET as it ventured into its 3nm process: it was a commercially-driven decision. TSMC consumers were not receptive toward adopting GAA, especially when a new transistor design would inevitably change the designing processes of ICs, thereby slowing their production and raising their costs.
A similar issue could possibly challenge Samsung too if it succeeded at its GAA-based 3nm node. Currently, Samsung’s fab capacity mainly serves its own needs, while Nvidia and Qualcomm are among its customers. The lack of capacity and the uncertainty associated with a new transistor design could probably diminish Samsung’s chance to win back Apple’s contract.
A supply chain without Taiwan?
Finally, IBM’s announcement might come with geopolitical significance as well. With the US’ long-time IC champion Intel lagging behind, US industrial policy has made it a top priority to regain its leadership in chip manufacturing. The US over-dependence on TSMC has recently become a national security concern as well, given Taiwan’s rising tension with China.
As the US pursues chip autonomy and seeks to de-risk its chip supply chains, some policy advisers have begun to flirt with the idea of building a semiconductor supply chain that excludes Taiwan. If the partnership between IBM, Samsung and Intel succeeded at commercializing the 2nm process, the US would have come one step closer to build such a value chain.