Already a year ago, Google published a study in which they describe that they want to revolutionize the production of chips. Especially the creation of the so-called floor plan is one of the most difficult tasks in the process. In weeks to months of work, the interaction between man and machine works out the design of the chip, which enables the highest possible flexibility and performance.
Google now wants to simplify this step with the help of artificial intelligence and reduce it to a few hours.
When creating a floorplan, the problem lies particularly in the fact that the chips consist of many different blocks, which are again differentiated into individual components and logic gates. The combination and arrangement of these has so far been too much for a machine and needed humans.
In a recent article, the Google team, which has now combined decades of research, describes a new process for what it calls Tensor Processing Units (TPUs). This will allow the tech giant to accelerate AI models in their cloud offering in the future and pave the way for more systems to build on top of it.
The new process heralds a new era in chip generation, which is complemented by other recent developments, including in EUV lithography. With the new, faster and cheaper manufacturing methods, this opens up new potential for supercomputers, cloud services and AI-based applications.