Design Ghost! Google uses AI to design AI chips
Google is using AI to design next-generation AI chips that can be completed in less than 6 hours
Letting machines design and build other machines is the root cause of the disaster. At least, all science fiction movies tell us that. Computers will develop to the point where they pose a threat to mankind. This will not happen suddenly, but after a few years, artificial intelligence becomes conscious and hostile to the species that created it.
We have not fully reached the point where smart machines can build more complex versions, but Google is already using artificial intelligence to design better artificial intelligence chips. This is an achievement that has been documented in a research paper, and it clearly shows that artificial intelligence can improve its own designs.
Google published a paper in the journal Nature, detailing the technology it uses to teach artificial intelligence to design new artificial intelligence chips that will be faster and more energy-efficient than their predecessors.
What so special about it?
The software developed by Google allows computers to create designs for the next generation of artificial intelligence chips in just six hours which a team of human engineers will take months to design. The company explained that Google uses this technology to manufacture its latest tensor processing unit chip.
How does it work?
What the artificial intelligence algorithm does for Google is to draw the floor plan of the chip. This involves drawing the position of the various components of the system chip to maximize processing speed and reduce power consumption. Components including the CPU, GPU, and memory chips are placed on the silicon chip and connected.
Google provided the algorithm with 10,000 chip floor plans, so it can understand and learn what is most effective in chip design to improve speed and efficiency.
Artificial intelligence manages to surprise researchers by creating designs that humans might not expect. Human chip designers use neat lines to layout components, but Google’s artificial intelligence uses a decentralized approach to design better SoCs.
As CNBC pointed out, this is not the first time that Google’s artificial intelligence “rebel” and proposed unexpected solutions. A few years ago, Google used different artificial intelligence algorithms, so its computer can beat the best Go players in the world. At that time, the human chess player was stunned by the artificial intelligence winning tricks that other human opponents could not think of.
Leaving aside the scenarios of artificial intelligence apocalypse, it is important to note that these complex algorithms are not working or appearing by themselves. Machines are taught to perform a specific task, and over time, they will get better. But artificial intelligence cannot create better artificial intelligence chips on its own. Starting from scratch, no humans have taught it how to propose improvements in speed and efficiency. The same goes for defeating humans in any competition.
According to Google engineers, the new chip design pioneered by Google may have a “significant impact” on the semiconductor field.
Yann LeChun, Facebook’s chief artificial intelligence scientist, expressed admiration for this breakthrough on Twitter. Another editorial in “Nature” magazine also praised Google’s achievements, saying that it may be “a huge help to speed up the supply chain.” The magazine also pointed out that “technical expertise must be widely shared to ensure that the company’s “ecosystem” becomes truly global,” and warned that such technological achievements should not “drive away people with the necessary core skills.”