The power of mythic materials such as vibranium, unobtainium and kryptonite have captured our minds and powered entire worlds of imagination. This underscores the fact that materials enable technological revolutions. There would be no aeroplane without duralumin, no skyscrapers without steel and no inexpensive internal plumbing without plastics.
Yet, the development of new materials is painfully slow. The number of materials that we use today is an atom of the possible combinations of elements. Our search for new high strength low weight alloys and room temperature superconductors have not yielded results. The promise of materials is heavy and unquestioned, while the delivery takes forever and often seems hopeless.
Thomas Alvin Edison is said to have tested over 2000 materials as a possible filament in his newly discovered electric bulb. He even tested his own hair before stumbling on carbon that later gave way to tungsten. A century later, material scientists have not left behind this ‘Edisonian’ approach — research in new materials zoom around predictable compositional spaces, and new results move one decimal point at a time.
This is where the burgeoning field of Artificial Intelligence (AI) promises to make a daring revolution. Are there clues in data that we have missed? Is there a material that is already out there, a sleeping giant, undiscovered, in some obscure publication?
Perhaps we can extract it from text using Natural Language Processing (NLP). The millions of images published each year – domains, grains, graphs, spectra – can we build enough neural networks to consume them all? What data do we need to feed this monster? What does it take to predict a new material?
Perhaps the new databases that are being assembled all over the world can help. Perhaps, there is not enough data and what we might want are robot graduate students who can carry out a thousand experiments a day. This field of ‘high throughput’ experimentation is already here and we are combing through multidimensional phase spaces, gathering every feature that hides in plain sight.
Maybe it is not data that we need, maybe it is the right technique, an algorithm, a model? New Informatics departments are starting all over the world, training a new batch of scientists, ones forged in traditional fabrication labs, but trained in a new way of thinking – that of the data scientist.
Finally, we ask ourselves – what is knowledge? Can materials be so complex that there are phenomena that exist in million-dimensional spaces that we have not even encountered? But maybe we are approaching this question all wrong! When we create our robot scientist what do we teach him/her/it? We are poised on the brink of a revolution, slowly, one bit at a time, it gathers force, bringing upon us a world uncharted.
In this series of articles, I explore the rising field of AI in materials science, the various paths that lead to it, the obstacles on the way and the possibilities ahead.
A modern dagger from the distant past
In 1922, the English archaeologist Howard Carter entered the tomb of the Egyptian Pharaoh Tutankhamun in one of the most famous archaeological excavations of modern times. Carter catalogued 5398  items from the tomb including such treasures as the famous golden face mask, the gold coffin, thrones, chalices and sandals.
However, amidst all this high karat glory he found an item wrapped in linen hugging the pharaoh’s thigh – a dagger made of iron. Spectacularly, the three millennia old iron dagger showed no signs of rusting, suggesting that it behaves more like modern-day low carbon steel – an observation confirmed through chemical analysis .
Iron was a total mystery in ancient Egypt. Iron deposits were few, the technology to produce iron was not known, nor was any technique to preserve iron. Three millennia later, steel is everywhere. It’s strong, cheap and well understood. We know that the addition of a tiny amount of carbon makes iron stronger. We know that this is because tiny carbon atoms lock the dislocations inside iron from moving under stress. We know that the addition of chromium keeps iron from rusting through the formation of a thin protective layer.
We can control these factors precisely and reproducibly. Steel is unarguably the gift of modern science – a gift that keeps giving as cars, spoons, bridges and skyscrapers.
The development of steel is one of the finest examples of materials engineering. Indeed, the first phase diagram most materials engineers learn is that of iron-carbon. The neat curves and eutectics give the impression of a mature field of knowledge, confident and proud. Yet, the path to modern steel was anything but that.
If you think about it, why would someone throw in a piece of black charcoal to glowing iron melt and hope for wonder metal? Yet, that is what they did – even before we knew the science behind it. A little bit of carbon, a little bit of chromium, a dash of nickel and a spot of manganese. Steelmaking was once an act of cooking – more art than science.
It is not just steel however; the discovery of all new materials is a painfully slow process.
The Augean Stables of elements
Let’s say that we systematically study the combination of every possible element with every other element, in binaries, ternaries and so on. Among these are dozens of new ceramics, room temperature superconductors, strong lightweight alloys, undiscovered shape memory metals, new types of topological insulators and of course, flubber. However, starting with 120 elements we are left to explore a mind-boggling 10100 materials !
Even if we reduce the phase space by fixing on say, just ten elements, we still need to make and test around 4 million materials, each across a range of temperature and pressure.
It is no wonder then that the discoveries of new materials have been rare. Perhaps, as a result, materials have fascinated us since our history began even as their values have fluctuated. Marching through the Stone Age, Bronze Age and Iron Age, we have entered into the age of paper, plastic and silicon. Materials have gifted us everything we value. Iron made agriculture possible. Paper kick-started the Renaissance, bringing Europe out of the dark ages as plastic pipes enabled easy food transport and lightweight duralumin alloys crafted aeroplanes.
In our mythologies, we talk of Thor’s hammer made from the Asgardian metal Uru and of the crystal kryptonite that weakened Superman. The crew of James Cameron’s Avatar ransacked Pandora looking for unobtainium, a room temperature superconductor. The Wakandan tribe in Black Panther used vibranium, a powerful, flexible, kinetic energy reflecting material. Materials are magical, the legends of the ‘stuff’.
Today, materials science is undergoing a silent revolution. The dawn of artificial intelligence in materials science has already begun.
Armed with data, Graphics Processing Unit (GPU) driven computing and powerful algorithms, we are beginning to teach a computer how to look for and discover new materials – without actually making them. For example, given a list of elements – their atomic sizes, atomic weights, ionisation energies, bond strengths etc – a machine performs a thought experiment.
It adds a certain amount of each element into iron, calculates the changes in electronic structure, the behaviour under stresses and reactivity with oxygen, assigning each element a grade. In the next iteration, it varies the amount of each element – say from 0% to 10 % by weight and notes the changes in the property. In yet another iteration, each of these compositions is heated from 0 K to 3500 K and in three atmospheres. Sure, it’s a lot of work, but as long as a computer does it, who cares?
At the end of this calculation, we will hopefully see that carbon is one of the highest rated as is chromium and nickel. What is exciting, however, are other discoveries that we have not yet tried. Does adding something to iron makes steel that is green? Thermally insulating? Flexible steel? Leaving steel aside, can this approach predict a room temperature superconductor? How about new piezoelectric materials? Biodegradable, ultra-strong, non-hydrocarbon polymers? The scope is endless.
This is the promise of the AI revolution in materials science. A thousand wonder materials, each a revolution in itself. It’s not all hyperbole – materials are often the basis or technological revolution as we have seen. The latest approach to this problem – of designing new materials and improving existing materials, processes and performance – is to develop robust artificial intelligence that can do the task for us.
Unlike other attempts at AI, such as in natural language processing or image recognition, we have on our side centuries of experimental intuition and a system of knowledge called science to help us. How is this challenge being attempted by some of the smartest minds today? What are the first problems that are being attacked with AI?
We will find all answers in the next part: “The Age of AI in Material Science – Part Two” (will be published next week – April 9th).
*This article is the work of the guest author shown above. The guest author is solely responsible for the accuracy and the legality of their content. The content of the article and the views expressed therein are solely those of this author and do not reflect the views of Matmatch or of any present or past employers, academic institutions, professional societies, or organizations the author is currently or was previously affiliated with.