Taiwan Semiconductor Manufacturing Co. at present formally launched its N2 (2nm class) manufacturing expertise, its first node that can use gate-all-around field-effect transistors (GAAFETs), at its 2022 TSMC Know-how Symposium. The brand new fabrication course of will supply a full-now efficiency and energy advantages, however on the subject of transistor density, it is going to barely impress in 2025 when it comes on-line.
Being an all-new course of expertise platform, TSMC’s N2 brings in two important improvements: nanosheet transistors (which is what TSMC calls its GAAFETs) and bottom energy rail that each serve the identical purpose of accelerating performance-per-watt traits of the node. GAA nanosheet transistors function channels surrounded by gates on all 4 sides, which reduces leakage; moreover, their channels could be widened to extend drive present and enhance efficiency or shrunken to attenuate energy consumption and value. To feed these nanosheet transistors with sufficient energy and now waste any of it, TSMC’s N2 makes use of bottom energy supply, which the foundry considers to be among the many greatest options to combat resistances within the back-end-of-line (BEOL).
Certainly, on the subject of efficiency and energy consumption, TSMC’s nanosheet-based N2 node can boast of a ten% to fifteen% larger efficiency on the similar energy and complexity in addition to a 25% to 30% decrease energy consumption on the similar frequency and transistor rely when in comparison with TSMC’s N3E. Nonetheless, the brand new node will increase chip density by solely round 1.1X in comparison with N3E.
| N2 vs N3E | N3E vs N5 | N3 vs N5 | N5 vs N7 | |
|---|---|---|---|---|
| Pace Enchancment @ Identical Energy | 10% ~ 15% | +18% | +10% ~ 15% | +15% |
| Energy Discount @ Identical Pace | -23% ~ -30% | -34% | -25% ~ -30% | -30% |
| Chip Density | >1.1X | 1.3X | ? | ? |
| HVM Begin | H2 2025 | Q2/Q3 2023 | H2 2022 | Q2 2022 |
Generally, TSMC’s N3 does supply full-node efficiency will increase and energy consumption reductions. However density-wise, the brand new expertise can hardly impress. For instance, TSMC’s N3E node provides a 1.3X chip density enhance over N5, which is a considerable enhance. For justice, we have to word that TSMC makes use of considerably convoluted ‘chip density’ metrics to explain transistor density on N3E and N2 in its supplies printed at its 2022 Know-how Symposium. Chip density primarily describes a hypothetic chip consisting of fifty% logic, 30% SRAM, and 20% analog circuits. Fashionable designs are very SRAM intensive, however SRAM barely scales, identical to analog circuits; therefore an N2 chip that includes 50% of circuits that don’t scale will reveal mediocre scalability in comparison with an N3E IC. If in comparison with N3S, a transistor-density optimized model of N3, the outcome is likely to be even much less spectacular.
TSMC positions its N2 for varied functions, together with cellular SoCs, high-performance CPUs, and GPUs. Among the many options of the N2 platform (along with GAA nanosheet transistors and bottom energy rail), the world’s No.1 foundry mentions ‘chiplet integration,’ which most likely implies that many functions that can use N2 can even use multi-chiplet packages to optimize efficiency and prices.
As reported, TSMC will start high-volume manufacturing of chips utilizing its N2 node within the second half of 2025, so making an allowance for how lengthy modern semiconductor manufacturing cycles are, count on business 2nm chips to emerge available on the market solely in late 2025 or as a substitute in 2026. In fact, prior to now and in 2026, TSMC will supply a wide range of N3 (3nm class) nodes, however it is a totally different story.