Machine learning has found novel MXenes – a promising class of 2D inorganic materials – hiding in plain sight among pre-existing materials.1 Using insights from this data mining, the researchers created novel MXenes with properties not found in conventional ones such as magnetism. They also developed a new rapid, low-energy MXene synthesis route to produce them.
MXenes were first developed in 2011 by materials scientist Yury Gogotsi and colleagues at Drexel University in Pennsylvania. They traditionally comprise conductive 2D materials in which the M is a transition metal and the X is either carbon or nitrogen. They have potential applications ranging from supercapacitors to dry lubricants. However, their possibilities are constrained by difficulty synthesising them – harsh etchants such as hydrofluoric acid are often used – the absence of magnetism and the rarity of semiconductivity. ‘One of the most frequent questions I’m asked at conferences when giving talks is whether there are semiconducting MXenes,’ says Gogotsi. This property could open their potential as promising components in, for example, 2D field effect transistors with higher carrier mobility than silicon.
Computational materials physicist Johanna Rosen at Linköping University in Sweden and her colleagues reasoned that many ‘unrecognised’ MXene structures could have been synthesised before 2011. They dredged two huge databases – the Materials Project and the Open Quantum Materials Database – for structures with MXene-like chemical formulas. Crucially, they allowed the M element to include some rare earth metals, as well as transition metals, and also allowed variation at the X site to include other light elements such as oxygen and phosphorus.
Based on the crystal structures, the researchers identified 33 previously unknown MXenes and synthesised seven of them, including exotic materials such as thorium nitride oxide. To do this, they developed a ‘self-propagating high-temperature synthesis’ protocol that produced the two-dimensional MXene directly, eliminating the need for both continuous external heating and etching of a 3D crystal. Ali Saffar Shamshirgar at Linköping University, the work’s first author, says that many of the materials originally required weeks of high-temperature synthesis to obtain small samples. ‘This approach allows us to screen a wide range of chemistries in a very short timespan and expend much less energy,’ he explains.
The researchers used the data they had mined to train a machine learning algorithm to predict promising MXene candidates from materials that had never been synthesised. They thereby produced 11 new rare earth-containing MXenes and characterised them using both density functional theory and experimental techniques such as Raman spectroscopy and x-ray diffraction, obtaining good agreement. Most were semiconducting and they showed a variety of magnetic properties: ytterbium-containing MXenes, for example, were ferromagnetic, whereas holmium-containing ones were antiferromagnetic. These properties could prove useful in applications such as spintronics and magnetic memories.
The researchers now plan to explore even more widely. ‘We want to characterise exactly how the synthesis process works,’ says Rosen. ‘If we learn that, we can more easily make even more materials, use even more chemistries and get even more properties.’
‘I would say this is one of those rare papers in which multiple important advances have been reported in the same work,’ says Gogotsi. He says that ‘direct synthesis is very important: one can go right away from relatively inexpensive precursors to the MXene’. In addition, he says, ‘they also show very attractive properties, particularly quantum properties, semiconducting properties and magnetic materials’. He notes – as do the researchers – that magnetic rare-earth based MXenes were independently reported recently2, but he says that, after this study, ‘there will definitely be further growth, and there will be further expansion of both chemical variety and access to properties which conventional MXenes that we discovered 15 years ago don’t offer’.
