Samsung's success in the 2nm process has finally begun to bear fruit. Not only is its Exynos 2600 processor up to par, but now Tesla is also expanding its procurement and doubling its 2nm order production capacity. Korean media broke the news that Tesla executives will go to Samsung this week to discuss additional orders for 2nm chip AI6. In the previously signed agreement, the AI6 order has a production capacity of 16,000 wafers per month.Tesla hopes to add an additional 24,000 wafers to its production capacity, which will eventually reach a monthly production level of 40,000 wafers.
2nm production capacity orders have more than doubled, which is great news for Samsung, especially since the chip factories built in the United States, which have invested heavily, can increase production capacity utilization, which means that costs can be reduced and profitability can be improved.
Samsung signed an OEM cooperation order with Tesla last year to produce advanced process chips such as AI5 and AI6 for Tesla by the end of 2033. The total contract value is as high as 22.8 trillion won, approximately US$17 billion.
Tesla's AI6 chip will be used not only for its own self-driving system, but also for the highly sought-after Optimus humanoid robot, as well as Tesla's AI data center, especially after Tesla abandoned its originally planned Dojo AI chip.
Although Samsung has manufactured a variety of chips for Tesla before, it has lost significantly to TSMC in the 7nm to 4nm foundry market. It has bet on the 2nm process this year, and it seems that the bet is right so far.
According to Samsung, its first-generation 2nm process SF2 improved performance by 12%, or reduced power consumption by 25%, and reduced area by 5% compared to SF3's 3nm process.
However, there will be an enhanced version of SF2P technology in the future.Based on SF2, the performance will be improved by 12%, or the power consumption will be reduced by 25%, and the area will be reduced by 8%.——In fact, this is the 2nm ontology. The first generation SF2 is essentially the renamed optimization of SF3.
In the future, there will be SF2X and SF2Z processes optimized for HPC and AI. Among them, SF2Z will be powered by BSPDP backside, and SF2A is a process for automotive applications.
