Need to install billions of transistors on a chip?Let artificial intelligence do it


Artificial intelligence is Now help design computer chips-including the chips needed to run the most powerful chips artificial intelligence Code.

Drawing computer chips is complex and intricate, requiring designers to lay out billions of components on a surface smaller than a nail. Every step of the decision will affect the final performance and reliability of the chip, so the best chip designers rely on years of experience and hard-won professional knowledge to lay out circuits to obtain the best performance and energy efficiency from nanoscale devices. Over the past few decades, efforts to automate chip design have had little effect.

But the latest advances in artificial intelligence make it possible for algorithms to learn some of the dark arts involved in chip design. This should help the company develop a stronger and more efficient blueprint in a shorter period of time. Importantly, this method can also help engineers design AI software together, try to make different adjustments to the code and different circuit layouts to find the best configuration of the two.

At the same time, the rise of artificial intelligence has sparked new interest in various novel chip designs. Cutting-edge chips are becoming more and more important to almost all areas of the economy, from automobiles to medical equipment to scientific research.

Chip manufacturers, including Nvidia, Google, with IBM, Are all AI tools that test to help arrange components and wiring on complex chips. This approach may shake up the chip industry, but it may also bring new engineering complexity, because the types of algorithms deployed sometimes operate in unpredictable ways.

At Nvidia, the chief research scientist Haoxing “Mark” Ren Is testing how the AI ​​concept is called Reinforcement learning Can help arrange components on the chip and how to connect them together. This method of letting machines learn from experience and experimentation is the key to some major advances in artificial intelligence.

The AI ​​tool Ren is testing explores different chip designs in simulations and trains a large human Neural Networks Identify which decisions will ultimately result in high-performance chips. Ren said that this method should cut the engineering work required to produce chips by half, while the performance of the chips produced can reach or exceed those of artificially designed chips.

“You can design chips more efficiently,” Ren said. “In addition, it gives you the opportunity to explore more design space, which means you can make better chips.”

Nvidia started manufacturing graphics cards for gamers, but soon saw the potential for the same chips to run powerfully Machine learning Algorithm is now a leading manufacturer of high-end AI chips. Ren Zhengfei said that Nvidia plans to bring chips made using artificial intelligence to the market, but declined to say how quickly. He said that in the more distant future, “you may see most of the chips designed with AI.”

The most famous use of reinforcement learning is to train computers to play complex games, including the board game Go with superhuman skills, without any clear instructions about the rules of the game or the principles of good games.It shows the right Various practical applications, Including Train robots to grasp new objects, Flying fighter, with Algorithmic stock trading.

Song HanThe MIT assistant professor of electrical engineering and computer science said that reinforcement learning shows great potential for improving chip design, because like games like Go, it is difficult to predict good decisions without years of experience and practice.

His research team recently Developed a tool By exploring different chip designs in simulations, reinforcement learning is used to determine the optimal size of different transistors on a computer chip. Importantly, it can also transfer the knowledge it has learned from one chip to another, which is expected to reduce the cost of automated processes. In the experiment, the energy efficiency of the circuit design produced by artificial intelligence tools is 2.3 times that of the circuit design designed by a human engineer, and the interference generated is one-fifth of the circuit design designed by a human engineer. Researchers at MIT are studying artificial intelligence algorithms and novel chip designs to make the most of both.

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

Leave a Reply

Your email address will not be published. Required fields are marked *