The recent breakthrough by Google DeepMind in the field of material science is a testament to the power of artificial intelligence in revolutionizing our understanding of the natural world. In a remarkable feat, the AI has uncovered a treasure trove of 2.2 million new materials, specifically crystals, that were hitherto unknown to science. This discovery is not just a numerical milestone but represents a significant leap in our exploration of the material universe.
Crystals, often perceived as mere objects of beauty or mystical significance, are in reality pivotal in numerous technological applications. They are integral to the function of solar panels, acting as the light-harvesting layer. In industrial settings, crystals catalyze reactions for producing essential chemicals like ammonia and nitric acid. Moreover, the semiconductor industry relies heavily on crystalline silicon for microchip fabrication. This vast array of applications stems from the unique atomic structure of crystals, characterized by a repeating pattern, akin to 3D tessellating tiles.
Google’s AI subsidiary, DeepMind, has been at the forefront of this groundbreaking discovery. The team developed an advanced machine-learning tool, GNOME (Graph Networks for Materials Exploration), which utilized existing chemical structure databases to predict entirely new crystal structures. Their findings, published in the prestigious journal Nature, reveal the discovery of 2.2 million new crystal structures, a figure that dwarfs the previously known 48,000 types of crystals.
To validate these AI-generated predictions, Google DeepMind collaborated with researchers from the University of California, Berkeley. In a joint study, also featured in Nature, they synthesized 58 of the predicted compounds, successfully creating 41 of them in a span of just over two weeks. This successful synthesis further bolsters the credibility of Google DeepMind’s AI predictions. In addition, more than 700 other crystals from this AI-generated list have been synthesized by various research groups worldwide, indicating the broad scientific interest and applicability of these findings.
In a generous move, DeepMind has made a significant portion of its discovery available to the public. This includes what they believe to be the 381,000 most stable crystal structures from their findings. This subset is particularly notable for containing thousands of crystals that could potentially exhibit superconductivity – a phenomenon wherein electrical currents flow with zero resistance. Additionally, the subset includes several hundred potential conductors of lithium ions, which could revolutionize battery technology.
The impact of DeepMind’s work is substantial, yet as noted by Aron Walsh, a materials scientist at Imperial College London who was not involved in the research, it marks just the beginning of a broader exploration. Walsh’s own research, focusing on stable crystals incorporating four chemical elements, known as quaternaries, estimates a potential 32 trillion manufacturable crystals. Google DeepMind’s GNOME, however, restricted its search to crystals forming under relatively modest temperatures and pressures. This limitation indicates that crystals are just one part of a broader spectrum of materials, which includes everything from amorphous solids like glass to gases, gels, and liquids.
An interesting revelation from DeepMind’s research, as highlighted by Ekin Dogus Cubuk, is the discovery of approximately 3,200 senary crystals, composed of six elements. Previously, such crystals were believed to be extremely rare. This insight challenges existing understanding and opens new avenues for exploration in crystal formation and variety.
The ultimate utility of these 2.2 million newly discovered crystals is yet to be fully understood. However, the methodologies and insights gained from this endeavor are valuable in themselves. They not only offer new potential materials for various applications but also illuminate previously unknown rules governing crystal formation. This knowledge could significantly expedite the process of synthesizing new materials, alleviating the need for labor-intensive manual synthesis.