Google is using machine learning algorithms that will accelerate the manufacturing process of artificial intelligence chips. The method used to produce these processors is faster than the manual process and you can get the same results.
According to Google in an interview with Nature, this project will be used for the first time in a commercial product. Specifically, in a future version of the TPU chips ( Tensor Processing Unit). A component capable of performing artificial intelligence tasks and that is used in Google's data centers.
The task that machine learning algorithms must perform is faster and more efficient than that of humans for a reason: AI does not have to study or solve, but rather try another combination. Thus, until it is right.
The people in charge of manufacturing artificial intelligence chips must study the processes, check the operation of each component, connect them together, make use of external tools, etc. If there is a failure, the engineering team looks for a solution, something that can take weeks to manufacture the component.
The manufacture of chips for artificial intelligence is “like a game”
Image: Receiver chip. Fraunhofer Institute On the other hand, this process is much easier using AI, which takes this task “like a game” , according to the Google engineers. The mechanics are very simple: there is an integrated circuit to which they must place small components. The goal is to find the right fit. If artificial intelligence fails that combination, try another. Thus, until you get the correct one. The funny thing about AI is that it can perform unimaginable combinations, which in turn help the component work properly.
This process has helped Google train a reinforcement learning algorithm, which has been able to create a set of 10,000 planes of varying quality. Each design has been labeled with a “score” and the algorithm uses this data to generate the correct drawings to make the chips.
While it is true that the results are not surprisingly different from those obtained with a human, the speed at which artificial intelligence can make artificial intelligence chips, seems to be a good reason to use this method.
Google has shared an image with the results obtained through AI. The stamp, which is blurred because the design is confidential, shows the difference between a chip designed by humans (left) and a blueprint designed by artificial intelligence (right).
