Google’s AI designed its own chip in only 6 hours

Recently, Google said it has developed a way to use Deep Reinforcement Learning (RL) to create computer chip design diagrams in just six hours – Google says this is inherently a complex task. Complexity often requires people months to achieve.

According to the team published in a paper published in the journal Nature on Wednesday, June 9, the chips developed by Google’s AI are equal to or even superior to what humans can make.

More than just research, this project will apply to one of Google’s commercial products for the first time, research being used for the company’s upcoming tensor processing unit (TPU) chips, optimized for AI computing.

A Google Development Diagram

As such, Google uses the AI ​​approach to chip design, and ultimately AI-designed chips can be used to improve and accelerate the future development of AI.

“Our approach is used to design the next generation of Google’s artificial intelligence (AI) accelerators and has the potential to save thousands of human hours per century,” said the team. new system”.

The big breakthrough is that Google’s AI method can be used for the floor-planning design of chips, as the article says ” Despite 5 decades of research, chip mapping has challenges automation, requiring months of effort on the part of physical design engineers to create manufacturable layouts.”

A TPU chip from Google launched in 2017

The team trained its AI system by giving it 10,000 chip diagrams so it could learn what was “right” and what was “wrong”.

“As a result, our approach uses past experience to become better and faster at solving new cases of the problem, allowing chip design to be realized,” the team said. rendered by more experienced man-made agents than any other designer.”

However, the AI’s diagram looks quite different from the human-made diagram. Instead of neat rows of components arranged on a die, they are almost randomly scattered across the silicon. An illustration from Nature shows the difference, with a human design on the left and a machine learning design on the right.

The team notes that the positive impact on semiconductor chips is very high.

Recently, a lot of interesting things are happening on the semiconductor chip front. For example, IBM just had a “breakthrough in semiconductor design” by creating the world’s first 2-nanometer chip. They are not the only ones with breakthroughs in chip design, as a team from MIT, National Taiwan University and Taiwan Semiconductor Manufacturing Company have found a way to produce semiconductors at scale. 1 nm.

Reference: InterestingEngineering