TSMC: Officially announced 2nm!

On June 17, TSMC introduced information about future advanced processes at the 2022 Technology Seminar. The N3 process will be mass-produced in 2022, followed by N3E, N3P, N3X, etc. The N2 (2nm) process will be mass-produced in 2025.

TSMC first introduced the FINFLEX of N3, including 3-2 FIN, 2-2 FIN and 2-1 FIN configurations with the following features:

• 3-2 FIN – Fastest clock rates and highest performance for the most demanding computing needs

• 2-2 FIN – Efficient Performance, a good balance between performance, power efficiency and density

• 2-1 FIN – Ultra-high efficiency, lowest power, lowest leakage, and highest density

TSMC said FINFLEX extends the product performance, power efficiency and density range of the 3nm series semiconductor technology, allowing chip designers to use the same design toolset to select the best options for each key functional block on the same chip.

As for N2, TSMC said that this is its first node to use gate-all-around transistors (GAAFET) instead of the current FinFET (fin field-effect transistor). The new manufacturing process will provide comprehensive performance and power advantages. At the same power consumption, N2 is 10~15% faster than N3E; at the same speed, power consumption is reduced by 25~30%. However, compared with N3E, N2 only increases chip density by about 10%.

The N2 process brings two important innovations: nanosheet transistors (TSMC calls them GAAFETs) and backside power rails. The channels of GAA nanosheet transistors are surrounded by gates on all four sides, which reduces leakage; in addition, their channels can be widened to increase drive current and improve performance, or narrowed to minimize power consumption and cost. In order to provide enough power to these nanosheet transistors, TSMC's N2 uses backside power rails, which TSMC believes is one of the best solutions to combat resistance in the back-end-of-line (BEOL).

IT Home learned that TSMC has positioned the N2 process for various mobile SoCs, high-performance CPUs and GPUs. The specific performance will have to wait until the subsequent tests are released to know.