Samsung, IBM, Nvidia, Google: artificial intelligence in the design process of electronic components

0
Samsung, IBM, Nvidia, Google: artificial intelligence in the design process of electronic components

While the shortage of electronic components is raging all over the world, this does not prevent some major technology groups from innovating and exploring new ways of designing. This is the case of Samsung, which announced that it has integrated the DSO.ai artificial intelligence module to speed up the design process of its Exynos processors. According to many experts, AI is an ideal tool for designing ever more powerful chips.

From Design Space Exploration to Design Space Optimization

Chip design is a painstaking optimization process, as each millimeter of chip contains millions of transistors and complex connections between components. Traditionally, the techniques employed by chip design experts can take several months to complete. The traditional approach used by engineers is based on Design Space Exploration (DSE), which requires them to manually scan very large spaces and fields of possibilities to design efficient chips.

The more recent approach of Design Space Optimization (DSO) used by Synopsys consists in exploiting reinforcement learning to automatically search for design spaces, and thus, find optimal solutions.

schéma DSO conception performante puces
Synthetic diagram of a DSO agent

Many experts believe that the use of artificial intelligence and more specifically that of reinforcement learning is an interesting tool if not the right tool to speed up the process of designing chips, but also to propose new designs for chips that can be increasingly small as evidenced byIBM ‘s announcement on its future chips engraved on 2nm.

Samsung welcomes AI in its production lines with the DSO.ai tool

Samsung announced that it has already integrated Synopsys’ DSO.ai AI module into the specialized software that helps them design Exynos processors. Note that the use of AI remains something relatively expensive, but would allow to obtain in a few weeks results comparable to those obtained in several months of research by teams of engineers.

The Korean multinational did not specify whether the chips designed by the AI have already entered production. Other companies are interested in reinforcement learning to speed up the design of semiconductors, such as IBM, but also Nvidia. Google has also been working on developing a relatively similar AI lately, the Mountain View firm states:

“Our AI automatically generates chip designs that are superior or comparable to those produced by humans in all key metrics, including power consumption, performance and chip area.”

Translated from Samsung, IBM, Nvidia, Google : l’intelligence artificielle dans le processus de conception des composants électroniques