Generative AI for materials science

Prof. Ramamoorthy Ramesh is the executive vice president for research at Rice University, Houston. In his current role, he oversees internal and external research ventures across the university. He is an internationally recognized expert in material sciences and physics, focused on energy efficient computing.

Prof. Ramamoorthy Ramesh shared fascinating insights from his journey, during a recent visit to Infosys Bangalore campus. The interaction was at the intersection of innovation in three critical areas — materials science, sustainable energy, and computing or artificial intelligence in particular.

Moonshot, SunShot to Earthshot

The four pillars of research at Rice University today are sustainability or energy/environment/climate change, health, responsible artificial intelligence (AI), and addressing disparities in urban environments. Way back in the 1960s, the Moonshot or Apollo program was inaugurated following US President Kennedy’s historic speech at Rice University in 1962. More recently in 2011, the US Department of Energy’s SunShot Initiative was modeled after the Moonshot, with Prof. Ramesh as the Founding Director. The initiative had a goal of cutting the per-kWh cost of utility-scale photovoltaic power from $0.28 to $0.05 by 2020. The program was tremendously successful and was renewed in 2016 with the goal of cutting the cost an additional 50% to $0.03 by 2030.

This led to the formulation of the Earthshots, which comprised of a set of five reports that the Biden-Harris transition team for energy wrote for the administration. The first four reports were about hydrogen, CO₂ removal, energy consumption in buildings, and energy storage. If we get the cost of energy storage down to about $30 a kilowatt hour, the entire world will change. By 2040, we could have 70% to 80% of power from renewables and we may not need grids anymore. Local or micro grids are being developed, and energy storage could be their only showstopper. The fifth one was absolutely relevant for computing. Back then in 2014, the energy consumption in electronics was about 5% of the total energy produced; due to the emergence of AI/ML and IoT, this is growing exponentially!

Prof. Ramesh and his team did some calculations, which suggested that because of this exponential growth in electronics and software, the demand for energy will grow significantly. Their forecast was that because of the deployment of AI, machine learning, and IoT, 25% of global energy would go into electronics by 2030. While this was not universally agreed-upon at that time, the Semiconductor Research Corporation ‘s decadal report in late 2020 confirmed these calculations.

Generative AI in Rice University

At a basic level, generative AI is used in the university to find collaborators and to write a proposal for example. At a more fundamental level, it is used for material discovery, which has been very Edisonian in nature in the past. It was like making a dish – add a little more salt, a little more pepper and you get a new dish. But now we can use the power of ab initio calculations using first principles at an atomic level. Knowing the electronic structure, we can put together even 100,000 atoms together, shake the bottle, and the software will tell us what the material properties will be and what we can do with it.

Prof. Ramesh works on correlated materials, where the electrons interact with each other. Suddenly, something new happens. Prof. Ramesh’s colleagues at the University of California, Berkeley are looking at the power of AI for data mining. They are looking at a hundred years of data to look for optimizing material structures and processes to get certain properties. Many properties, whether it's electronic, structural, or magnetic, are very dependent on how the atoms are orbiting and we should be able to predict that.

Energy efficient computing using the laws of physics is another area of research. Semiconductor chips may have 20 to 30 billion lines of codes (or even more) today. They have to be efficient. We need to use the power of mathematics, physics, and chemistry to solve these big problems. We're going to see huge innovations in how software is written. It won't be the same way of writing programs in the future.

India outreach and Infosys collaboration

Rice Global India was recently launched by the university for global collaboration, with a focus on research, innovation, and educational partnerships with top Indian institutions. With Infosys, the plan is to have a joint program focused on each other’s strengths in computing. Two broad domains of application are healthcare and energy transition. Countries like US and India have networks of labs with researchers working on big problems for our society. These problems cannot be solved by a few individuals. These labs were setup with the ethos of doing research in fundamental science to change humanity. Infosys will be a building block for such a network with Rice University.