AI plays with Lego
Artificial intelligence can not only control cars or analyse large amounts of data. As of late, it can also deliberately grab individual molecules like Lego bricks and combine them. This opens up new possibilities for nanotechnology.
A tiny metal needle glides over the surface, snags a molecule and deposits it elsewhere. Then the procedure starts all over again. “We work with molecules as if they were Lego bricks. This way, we can comparatively quickly and cheaply build complex structures on the nanoscale,” says Stefan Tautz, Director of the Peter Grünberg Institute (PGI-3), in describing what is possible with the new development by the Jülich and Berlin experts. The scientists use a so-called scanning tunnelling microscope, which is actually used to image atomic details of surfaces. To do this, the extremely fine metal tip measures tiny electrical currents that flow between itself and the surface. The forces that occur in this process can also be used to detach individual molecules from the surface and place them elsewhere.
But: “Until now, a person controlled the tip by hand,” says Tautz. “The process is rather intuitive.” However, since different laws apply in the microcosm than in our everyday environment – gravity plays no role, for example – humans easily make mistakes. This is why scientists are relying on artificial intelligence (AI). AI is not influenced by previous knowledge and stoically learns about the forces that act between the molecule and the metal tip and how molecules can best be moved. “Actions by our AI that are effective are rewarded by the system,” explains Christian Wagner, head of the ERC research group Manipulation of Molecules at Forschungszentrum Jülich. In this way, the AI becomes better and better. “Even so, it needs a great many repetitions under the same constant conditions,” says Wagner. On the nanoscale, however, this is hardly possible because conditions are constantly changing. As soon as the atoms and molecules change their arrangement, other forces take effect. “In a way, it’s as if gravity is constantly fluctuating while you’re building with Legos,” Wagner compares. For this reason, the researchers had their AI also learn something about the environment in which a molecule is being moved. With this knowledge, AI was able to train in the virtual world, which accelerated the learning process enormously.
The researchers see the possibility to apply their invention in the production of prototypes, as it is known from 3D printing – only with many more possibilities: “Molecules come in an almost endless variety,” says Tautz enthusiastically. “It is not yet possible to imagine what can be constructed at the nano level with the new method.” Quantum components or innovative materials, for example, are conceivable possibilities.
Artificial intelligence controls the movements of a scanning tunnelling microscope: it is currently training to grab a molecule with the tip of the microscope and lift it off – later, it will also be able to specifically place it elsewhere. The movements are random at first. After each run, AI learns from the experience gained, thus working increasingly as planned.
Artificial intelligence (AI) was given the task of removing individual molecules from a closed molecular layer. First, a connection is set up between the microscope tip (top) and the molecule (middle). Then the AI tries to remove the molecule by moving the tip without breaking the contact. The movements are random at first. After each run, AI learns from the experience gained, thus becoming better and better.
Illustrations: SeitenPlan, Video: Forschungszentrum Jülich / Christian Wagner