From Serendipity to Systemic Design

We have the privilege of overabundance of data and big data from a combination of research institutions, company projects, field experiments, etc. The challenge now: how do we process and translate this data into real-world applications to discover new materials? We can’t expect to find the next-generation of materials the way we discovered penicillin—by chance.

In 2022, the total revenue of the global chemicals industry topped $5.72T. This translated to approximately 935 Mt of direct CO₂ emissions, the third largest industry subsector emitter. To meet net-zero emissions, the global chemical industry needs to reduce nearly a fifth of emissions by 2030, despite a forecasted increase in production. But—big surprise—we’re not on track to meet these goals.

What’s more, the chemical sector is the largest industrial energy consumer, particularly in China. And in the U.S. alone, over 70,000 products are produced from fossil fuels daily. We rely so heavily on fossil fuels for our everyday lives, e.g., plastics, that’s it’s difficult to produce new materials and chemicals that compete with the products we’ve become so accustomed to. Traditional materials discovery takes years, generally decades, to progress. For example, batteries have not seen significant progress since the lithium-ion battery was invented in the 1980s.

But with the advent of digital computational systems like artificial intelligence (AI) and machine learning (ML) coupled with hybrid cloud technologies and computers, we are witnessing a paradigm shift in modern materials discovery. Perhaps the most important challenges in our lifetime will be to characterize the key chemistries behind photosynthesis (ammonia synthesis), discover high-performance batteries, and even unlock reliable energy sources like stable tokamaks for fusion reactors.

Data-Driven Discovery

While chemical databases contain billions of identified and characterized compounds, Materials Project has only 150K materials in its known materials database. There may be an excess of 10­¹⁰⁸ potential carbon-based molecules that could be of significant benefit that require advanced analytics to process beyond serendipity.

In 2023, Google DeepMind produced 380K stable materials for everything from batteries to superconductors. But there still exist significant gaps in experimentation, modeling, and physical reproducibility. The integration of digital systems like AI could aid not only in data mining from databases like ChemMine or IBM DeepSearch, but also in providing language models to help us efficiently discover like IBM RXN.

Still, research suggests that in practice generative models are most useful when accompanied by the deep expertise of humans for data cleaning and validation. This is the reason that UK-based Materials Nexus, who I recently chatted with, is reverse-engineering materials with its team of materials scientists. It’s raised $2.7M and uses AI, ML, and computers to co-discover and develop metals and magnetic alloys. The team transfers digital findings into physical validation. It seeks to license or sell its intellectual property (IP) to partners. Forward looking, Materials Nexus will manufacture products or operate similarly to a fabless manufacturer.

UK-based Cusp.AI has raised $30M for its search engine which leverages generative AI, deep learning, and molecular simulation for materials design. Its team is led by Dr. Chad Edwards, former leader at Quantinuum, Google, and BASF. Cusp.AI recently partnered with Meta to further its open science contributions (data), specifically to advance materials for cleantech applications, e.g., the discovery of novel direct air capture sorbent materials.

Faster Time-to-Market

This month, France-based, Altrove, raised $4M for its AI-based predictive tools for physical validation in automated labs. It’s currently focused on discovering substitutes for rare earth materials for use in transition technologies, electric vehicles, and other advanced electronics. Altrove‘s technology browses the latest existing and predicted materials databases, runs predictions on material properties and presents the best candidates for a use case in 2-4 weeks. Its automated lab then tests and validates scalable processes to manufacture materials in just 2-6 months. Materials can be purchased directly from Altrove’s manufacturing partners, or its IP can be integrated into existing processes.

Quantum Leap in Materials

Germany-based Quantistry raised $3.2M earlier this year from investors like Chemovator, the business incubator of BASF, for its SaaS chemical simulation platform. The platform combines the latest expertise in small-scale quantum computing and AI. Just a of couple weeks ago, Quantistry partnered with IQM Quantum Computers to explore hybrid quantum solutions for the chemical and material industry.

While a majority of AI solutions will utilize desktop computers, some solutions also leverage advanced super computers. As we inch closer to quantum computing solutions, we are sure to see the integration of small-scale quantum computers in materials discovery in the next few years or at least by the 2030s. Quantum computers have ultra-fast computing speeds with high precision to process incredibly complex datasets that would take traditional computers lifetimes to process. The likes of IBM, Microsoft, and Google are competing to deliver quantum computing services (for more on quantum computers, I highly recommend Dr. Michio Kaku’s Quantum Supremacy).

Germany-based HQS Quantum Simulations is currently providing quantum computing-based SaaS solutions to predict material properties. HQS offers a full software workflow as well as the development of a quantum-level module that integrates with an existing workflow. It’s raised over $17.3M from notable investors like b2venture and HTGF.

Don’t Be Alarmed, AI Isn’t Taking Jobs—Rather, It’s Enabling Them

As we race against time, we need to quickly and efficiently discover new materials. The challenge lies in harnessing the correct data from an overabundance of sources. Digital solutions are enabling the rapid discovery of materials just as some of the most exciting technological innovations begin to come online, e.g., quantum computers. Still, human expertise remains critical. The future of materials discovery lies in a synergistic collaboration between these innovative technologies and the expertise of scientists and engineers. After all, a computer is only as intelligent as the engineers who build it.

• To efficiently and rapidly discover the next generation of materials, we must deploy digital solutions like AI and ML to analyze big data for rapid data mining, high-throughput computation and testing, and for reverse engineering of materials
• AI-powered materials design can transform decades of slow, incremental progress into discovery in just weeks to months; however, human expertise remains crucial for guidance in steps like data cleaning and validation
• By the 2030s, quantum computing will unlock the most important challenges in our lifetime like the discovery of the biological catalyst to produce ammonia (i.e., photosynthesis), high-performance batteries, etc.