Google Relies on TSMC for Future Tensor Chips: New Strategies Ahead | Gpu vs cpu for gaming | Hardware catalog | Types of hardware | Turtles AI

Google Relies on TSMC for Future Tensor Chips: New Strategies Ahead
With Samsung’s departure, the upcoming Tensor G5 and G6 chips will be produced by TSMC, focusing on 3nm and 2nm technologies to improve performance and efficiency
Isabella V16 September 2024

 


Google appears ready to switch foundry partners for its Tensor chips, moving from Samsung to TSMC, mainly because of the low yields of Samsung’s 3-nanometer process. Currently, the Tensor G4 chip, used in the Pixel 9, is still produced with Samsung’s 4-nm technology. However, the Tensor G5, intended for the Pixel 10, will be produced by TSMC using the 3-nm process, a technology considered more advanced and reliable. Looking even further ahead, the Tensor G6, which could power Pixel 11 devices, will likely be manufactured using TSMC’s 2-nm process.

Key points:

  •  Google is switching from Samsung to TSMC for the production of its Tensor chips, starting with the Tensor G5.
  •  The Tensor G5 will be produced with TSMC’s 3-nm process, while the Tensor G6 will use the 2-nm process.
  •  The decision is influenced by Samsung’s low production yields in its 3 nm process.
  •  Collaboration with TSMC is seen as a long-term partnership


Google appears to be planning a significant strategic shift for its future Tensor chips, abandoning its collaboration with Samsung Foundry in favor of TSMC (Taiwan Semiconductor Manufacturing Company) for the next generations of Tensor G5 and G6 chips. Google’s choice is influenced by Samsung’s disappointing performance in the 3nm manufacturing process, which has shown low yield rates. This prompted Google to seek a more reliable partner, with TSMC expected to produce the 3-nm Tensor G5 and future 2-nm Tensor G6 chips.

The transition is important because the Tensor G5 will be Google’s first fully customized chip and will be mass produced in 2025, after a delay in development time. TSMC, an industry leader, offers greater power efficiency and smaller size through its advanced technology. This change also reflects a broader trend by Google to gradually replace Samsung’s components with its own IP, despite logistical and coordination difficulties between development teams in the United States and India.

In addition, the choice of TSMC could help improve the overall performance of Pixel smartphones, even as Google continues to sell fewer units than giants such as Apple and Samsung. This poses economic challenges for Google, as ordering small volumes of chips with advanced manufacturing processes such as 2nm can be particularly expensive. However, the Mountain View giant may decide to temporarily opt for TSMC’s N3P technology, an improved version of the 3nm process, before moving to 2nm for the Tensor G6.

In the end, this transition is an important step for Google in its technological evolution, aiming to improve the efficiency and performance of its devices.

This transition from Samsung to TSMC is mainly due to the low yields of Samsung’s 3nm GAA process, which has only achieved 20 percent success rate, making it less competitive with TSMC. Mass production of the Tensor G5 is already underway at TSMC, while the Tensor G6 is expected to use the new 2nm technology, which can offer significant improvements in power efficiency and performance.

TSMC has consolidated its position as an industry leader through more reliable manufacturing capacity, as well as having already attracted other major customers such as Apple. The collaboration with Google is expected to be long-term, with the possibility of Google using 2nm technology for future Tensor chips as well. However, given the limited market share of Pixel devices, Google may find it more cost-effective to leverage the 3-nm N3P process, which is less expensive than the new 2-nm technology, for relatively small production volumes.

This transition represents a significant change for Google and could strengthen the competitive positioning of future Pixel devices.