Google Developing Strange New AI

( Claiming a breakthrough in artificial intelligence that could have “major implications” with the semiconductor sector, Google, in a paper published in the journal “Nature,” claims it has developed artificial intelligence software that can design chips in far less time than it would take humans to develop.

Google has already used the artificial intelligence to develop its latest tensor processing unit chips that are used to run AI-related tasks.

In its report, Google says that its method to generate chip floorplans through artificial intelligence is shown to be “comparable or superior to human experts in under six hours.”

It ordinarily takes humans months to produce chip floorplans for modern accelerators.

A chip’s “floorplan” involves plotting where components like CPU processors, graphics processing units, and memory are placed in relation to each other on the silicon die. Precise floorplan positioning is important because it affects both power consumption and processing speed.

Google’s “deep reinforcement learning system” — an algorithm trained to take certain actions that will maximize its chance of earning a reward – can plot floorplans with very little effort.

While similar algorithms currently exist that can defeat humans at complex problems, in Google’s chip scenario, the artificial intelligence is trained to find the best combination and placement of components to make the chip as computationally efficient as possible.

In order to train the AI system, Google fed it 10,000 chip floorplans in order to learn which combinations work and which do not. Where human chip designers typically lay components out in orderly lines, Google’s artificial intelligences takes a more scattered approach to chip design.

In an accompanying editorial in Nature, Google’s artificial intelligence chip design is hailed as an “important achievement” that will speed up the supply chain. However, the journal editorial also urges that this new technical advancement be widely shared in order to ensure the “ecosystem” of companies becomes “genuinely global.” The editorial also cautioned that the creation of these “time-saving techniques” do not “drive away people with the necessary core skills.”