Need to put billions of transistors on a chip? Let AI do it



Artificial intelligence is it now helps design computer chips – including those needed to run the most powerful AI code.

Sketching a computer chip is both complex and intricate, requiring designers to assemble billions of components on a surface smaller than a fingernail. Decisions at every step can affect the eventual performance and reliability of a chip, so the best chip designers rely on years of experience and hard-earned expertise to set up circuits that squeeze the best performance and power efficiency out of nanoscopic devices. Previous efforts to automate chip design over several decades have been limited.

But recent advances in AI have allowed algorithms to learn some of the dark arts involved in chip design. This should help companies make more powerful and efficient drafts in much less time. Importantly, the approach can also help engineers design intelligence software together, experimenting with different code settings, along with different circuit layouts, to find the optimal configuration for both.

At the same time, the rise of AI has sparked new interest in all sorts of new chip designs. Top-of-the-line chips are increasingly important for almost every part of the economy, from cars to medical devices to scientific research.

Chip manufacturers, including Nvidia,, Google, i IBM, all test AI tools that help arrange components and wiring on complex chips. The approach could shake the chip industry, but it could also introduce new engineering complexities, as the type of algorithms used can sometimes behave in unpredictable ways.

In Nvidia, the chief scientific researcher Haoxing “Mark” Ren tests how the AI ​​concept known as learning reinforcement it can help you arrange the components on the chip and how to connect them. The approach, which allows the machine to learn from experience and experimentation, has been key to some of the major advances in AI.

The AI ​​tools that Ren is testing are exploring different chip designs in the simulation, training a large artificial one neural network identify which decisions ultimately produce a high-performance chip. Rehn says the approach should halve the engineering effort required to produce the chip, while producing a chip that matches or exceeds the performance of human design.

“You can design chips more efficiently,” says Rehn. “It also gives you the ability to explore more design space, which means you can make better chips.”

Nvidia started making graphics cards for gamers, but quickly saw the potential of the same chips for powerful running machine learning algorithms and is now a leading manufacturer of premium AI chips. Ren says Nvidia plans to bring AI-made chips to market, but declined to say how soon. In the distant future, he says, “you’ll probably see most of the chips designed with AI.”

Reinforcement learning is best known for training computers to play complex games, including the social game Go, with superhuman skills, without any explicit instructions regarding the rules of the game or the principles of good play. It shows promise for various practical applications, including training robots to capture new objects,, flying fighter jets, i algorithmic stock trading.

Song Han, an assistant professor of electrical engineering and computing at MIT, says consolidation shows significant potential for improving chip design because, as with games like Goa, it can be difficult to predict good decisions without years of experience and practice.

His research group recently developed a tool which uses gain learning to identify the optimal size for different transistors on a computer chip, exploring different chip designs in the simulation. Most importantly, it can also transfer what it has learned from one type of chip to another, which promises to reduce process automation costs. In the experiments, the AI ​​tool produced circuit designs that were 2.3 times more energy efficient, while creating a fifth less interference than those designed by human engineers. Researchers at MIT are working on AI algorithms at the same time as the new chip design to make the most of both.

Other players in the industry – especially those who have invested heavily in the development and use of artificial intelligence – also want to adopt artificial intelligence as a chip design tool.


Source link


Please enter your comment!
Please enter your name here