A machine learning system developed by researchers at Flinders University and Khalifa University suggests gallium-based semiconductor materials with defined energy gaps. The goal: to shorten the long path from chemical discovery to candidates for new chips and electronic components
The search for the next generation of chip materials is getting a new tool: an artificial intelligence system that can sift through a vast space of chemical combinations and suggest semiconductor materials with desirable electronic properties. An international research team, led by Flinders University in Australia and in collaboration with Khalifa University in the United Arab Emirates, has developed a machine learning platform that serves as a “smart materials discovery engine.”
The study, published in the journal ACS Materials Letters, focused on materials containing gallium. Gallium is well-known in the electronics industry for compounds such as gallium arsenide, which are used in microwave circuits, fast switching circuits and infrared components, among other things. In recent years, it has also received renewed attention due to its importance for advanced chip technologies and electronic components that require speed, efficiency and durability.
The main problem is that the chemical possibilities are almost limitless. To find a suitable material for a chip, an optical component, or a power system, many combinations of elements, chemical relationships, and possible structures must be tested. Such testing in the laboratory is slow and expensive, and even accurate computational simulations require a lot of computing resources. Instead of going through the options one by one, the new system learns from existing examples and directs the search to more promising areas.
The researchers trained the system on thousands of known semiconductor materials from international material repositories. The system then used Bayesian optimization—a statistical method that combines prediction with intelligent selection of the next experiment—to suggest new materials containing gallium. The goal was to find compounds with an energy gap, or band gap, within a predefined range.
The energy gap
The energy gap is one of the most important properties of a semiconductor. It determines how the material responds to electricity and light, and therefore affects its suitability for various uses. Relatively small gaps can be suitable for solar energy applications, medium gaps are important for optical components and LEDs, and larger gaps are needed in power components, radiation-resistant systems, and applications that require operation in harsh conditions.
One of the important points of the study is that the system does not “invent” chemical formulas at random. According to the researchers, it checks whether the candidates it suggests are chemically plausible and physically stable before recommending them for further testing. This reduces the number of unrealistic directions and increases the chance that the selected candidates will be able to move on to the stage of advanced computational verification or laboratory experimentation.
The researchers report that the system was able to suggest several completely new candidates for gallium-based semiconductors that did not appear in the databases on which it was based. This is still an early stage: a computational suggestion of a material is not equivalent to its actual production, and it remains to be seen whether the materials can be synthesized, whether they are stable under working conditions, and whether they actually exhibit the predicted properties. However, the ability to quickly narrow the search space could be very significant for the chip and electronic materials industry.
The broader importance of the research goes beyond gallium itself. The chip industry relies not only on miniaturizing transistors, but also on new materials: materials for power components, sensors, photonics, solar cells, fast communications, and more durable components. As applications proliferate, the need for faster methods to discover materials with precise properties increases.
The approach the researchers present illustrates how artificial intelligence can be integrated into materials research in a practical way. It does not replace chemists, physicists, and materials engineers, but rather helps them choose what to focus on. Instead of spending a lot of time testing thousands of weak options, researchers can start from a smaller list of promising candidates. If the approach continues to be successful in experimental validation, it could shorten the path between a new chemical idea and a useful material in the next generation of chips and electronic components.
For the scientific article: DOI: 10.1021/acsmaterialslett.5c01482
Short FAQ:
What did the researchers discover?
The researchers developed a machine learning system that suggests gallium-based semiconductor materials with predefined energy gaps.
Have the materials already been produced in a laboratory?
Not necessarily. The study suggests computational candidates, and the next step is further validation through precise calculations and experiments.
Why is gallium important for chips?
Gallium compounds are used in fast electronics, microwave components, infrared, and components where high performance is required.
What is an energy gap?
Energy gap is a property that determines how a semiconductor responds to electricity and light, and therefore affects its suitability for uses such as solar cells, LEDs, and power components.
What is the advantage of artificial intelligence in materials discovery?
It can quickly narrow down the number of options to test, directing researchers to more promising candidates.
More of the topic in Hayadan:
