We have seen many hands twist over the immediate death of Moore’s Law. This is the observation by Gordon Moore, founder and former CEO of Intel, that his current iteration calls for a doubling of the chip’s transistors every two years. Using Apple’s A-series chipsets as an example, the A13 Bionic launched a power in 2019 iPhone 11
Can artificial intelligence (AI) keep Moore’s Law alive after a 2 nm node?
Built on a 7 nm process node with just under 90 million transistors per square meter, the A13 Bionic contains 8.5 billion transistors. The A14 Bionic chipset can be found iPhone 12
series and iPad Air (2020) and carries 134 million transistors per square meter. The number of transistors on the chip is 11.8 billion; the more transistors inside the chip, the more efficient and energy efficient it is.
Inside the TSMC production facility
The world’s top foundry, TSMC, in the next quarter is expected to test 4nm chips with mass production of a 3nm process node in the second half of next year. Both TSMC and Samsung are working on a 2 nm process node that can be mass-produced immediately in 2024. The future of Moore’s Law is of particular concern after the 2 nm process node.
But artificial intelligence (AI) may be helpful. An article on the use of artificial intelligence to create a “floor plan” on a chip states that with artificial intelligence, the time to build a floor plan, which can take up to several months, is completed in less than 6 hours with artificial intelligence.
Google has already used artificial intelligence to design Tensor Processing Units
Google employees who wrote about technology Nature
are Azalia Mirhoseini and Anna Goldie. Google has used this system in real life to help create its Tensor Processing Unit (TPU) plan, which is used to “speed up neural networks in its search engine, public cloud service, AlphaGo and AlphaZero, and other projects and products.”
In their article, Mirhoseini and Goldie wrote that “Our method was used to design Google’s next-generation artificial accelerators and has the potential to save thousands of hours of human labor for each new generation. Finally, we believe that more efficient artificial intelligence
The neural network evolves better in chip design over time and is “able” to generalize chips – meaning it can learn from experience to become both better and faster in placing new chips – allowing chip designers to be helped by artificial materials with more experience than anyone could ever get. “The paper states that“ We show that our method can produce chip floor plans that are comparable or better than human experts in less than six hours, while humans take months to produce acceptable floor plans for modern accelerators. Our method has been used in production to design the new generation of Google TPU. “
I hope that the use of artificial intelligence in the design of future chips will lead to solutions that lead process nodes below 1 nm.