“Samsung’s 3nm Process: Over-Optimistic Claims and Yield Concerns”

Samsung saw its operating profit plunge by 96%, the company’s biggest financial blow in 14 years. He thus tried to enlighten the financial report with at least positive news on the subject of the 3nm process. However, there are some indications that this may be somewhat over-optimistic.

Samsung’s 3nm process is somewhat unique. This is the first deployment of GAA transistors, namely Samsung’s own MBCFET variant. The deployment of new technology was intended to enable the company to make a greater intergenerational shift in process parameters and, if possible, to overtake the competition. However, this opportunity was not used, and Samsung is actually trying to get closer to TSMC. Still, its 3nm MBCFET process could be interesting.

The company has now mentioned that it has acquired “various customers” including mobile chip manufacturing and highlighted the “stable yield” of the process.

Although, when contrasted with TSMC’s 3nm process news, these statements might give the impression that Samsung’s production will be in very good shape, at least two indications suggest that it will not be rosy in reality. The first follows from the statement that the 3nm process achieves up to 23% higher performance or up to 45% higher energy efficiency than the 5nm process. Samsung is comparing here with a 5nm process, not with the (albeit slightly) more advanced 4nm. If Samsung’s 4nm process is energy-wise comparable to TSMC’s 6nm process, then a 45% improvement in energy over 5nm production (which could therefore be comparable to TSMC’s 7nm process) won’t be a complete miracle, but at best somewhere on the level of TSMC’s 4nm process (5nm the TSMC process is 30% better than 7nm in terms of power, with the 4nm TSMC N4P process doing a further 22% better).

So this is a significant energy improvement, but it will not be comparable to TSMC’s 3nm process. The second fact worth noting is the fact that Samsung is said to be manufacturing its own Exynos 2400 on the 4nm LPP+ / SF4P process. Not on 3nm GAA. This in turn indicates that the yield of 3nm production does not even reach such a level that it is worthwhile for Samsung to use it for its own SoC. The company did talk about “stable yields”, but stable ≠ high.