It would be easier to name devices that do not have a microchip than to mention those that have an incorporated one. Currently, almost everything that can be plugged in carries a microchip: from smart watches to radars, from the world of the minimum to the absurdly large. Luckily, we know very well how they work.
And, until now, these microchips were designed by humans. The problem is that, while the rest of the mortals ask an AI to make us a design, translate a text or find a certain location, Scientists have asked him to develop a wireless microchip driven by AI. Nothing less. And the results, published in Nature, are far from what we can expect.
The study, led by Kaushik Senguta of Princeton University, describes how the deep learning technique was used to devise new chips designs, and although the chips seem to work, The Senguta team recognizes that it is not completely sure how. The designs “seem to have random ways and we cannot understand them completely,” says Sengarta in A statement.
In fact, the photos of the chips have a design similar to the video games of alien AI could “be better understood as an alien intelligence than as an imitation of our own cognition.”
In the tests carried out by the Senguta team, the deep learning model generated highly optimized electromagnetic structures that, when they were tested, surpassed their counterparts designed by humans. The authors also discovered that His model was very suitable for a reverse engineering design approach, which basically begins with the desired result and lets the model work back to complete the blank spaces.
The usual method to design microchips is tedious and is based on a combination of expert knowledge, templates proven in battle and the old method of test and error. This process usually has been synthesis, imitation and evidence in real life for days or weeks, and even so, Humans find it difficult to understand the astronomically complex geometry of the chips they produce.
Despite being more efficient, Senguta also points out that It is a tool, not the definitive solution for hardware engineeringespecially because the deep learning algorithm detected defective designs with the same effectiveness with which it produced effective designs.
“There are problems that still require human designers to correct them -concludes Segupta -. The objective is not to replace human designers with tools, but Improve productivity with new tools. The human mind is better used to create or invent new things, and the most mundane and utilitarian work can be delegated to these tools. ”