Aionics uses machine-learning aided discovery and simulations to accelerate the discovery and validation of new formulations for sectors such as advanced manufacturing, clean energy storage, autonomous aviation and aerospace and technological infrastructure.
What are the gains and accelerations enabled by AI in a field like this?
I started working in this field of AI for Science in 2014. At that time, people asked, “What does AI and chemicals and molecules and materials have to do with each other?” Today, the question is, “What can they not do?”
The timeline to commercializing a new material is usually 15 to 20 years from invention in the lab to scaled manufacturing and product deployment. A large portion of this involves finding the right market opportunity and then validating and qualifying the material for that market. This process is long and arduous, with constant back and forth between the materials developer and the end-product maker. AI has fundamentally changed the landscape in material science by making a significant advance in validating and qualifying materials for commercialization by shrinking the time it takes.
Can you share an example of how Aionics is using AI to accelerate this process?
We had a customer program with a large automotive Original Equipment Manufacturer (OEM) to design new electrolyte formulations. The electrolyte sits between the anode and cathode of a battery and controls battery life, charging speed, safety and everything other than the amount of energy in the cell. We were designing for a use case in which the battery had to remain thermally safe while charging quickly. The project’s process was AI-guided. Typically, you might go through tens of iteration cycles over multiple years before finding a formulation that works–we completed four iteration cycles in three months. The goal was to improve a thermal property by 50%–we ended up improving it by 100%. Even as an expert in the field, I could look at molecules and think they might be interesting but could not pinpoint and say this is a molecule we should focus on. AI guides this optimization in just a few months, revealing formulations that fundamentally change the performance landscape for aerospace and electric aviation.
Where do humans still enter the process?
You can accelerate portions of the discovery cycle with AI but once you find a promising material, you still have to build the supply chain to scale it up and validate it in a real device through qualification phases. This experimentation itself is still laborious and requires expert humans to complete that final validation. We have built a robotic test stand called Clio, a microfluidic setup that can mix and test chemicals automatically, but there is still a great deal that requires human expertise and intuition.
Our approach is to discover materials with a customer that can take materials through the final stages of validation. For example, if you design a new electrolyte for an automotive OEM, the electrolyte still has to be tested inside a battery under automotive duty cycles. AI companies are not going to manufacture the car, so the OEM and cell maker must. Human judgment is also essential in deciding which materials can scale and how quickly they can scale, and how sustainability, PFAs or supply chain considerations factor in.
What are the material requirements for electric aviation and what does Aionics bring into the picture?
In 1884, Charles Renard flew a big, blimp-like dirigible for about eight kilometers on a zinc-chlorine battery that weighed 435 kilograms. He said the era of electric aviation was imminent. Here we are 142 years later still waiting for things to take off in a second century of aerospace that will be electric, connected, and autonomous. The public-private M-Air initiative I direct at the University of Michigan focuses on autonomous electric aviation and drones. The 30-acre test facility is complete with roads and traffic lights and fully gated with no human traffic. I joke that I am the mayor of this fake city where companies validate new use cases for electric aircraft and unmanned aerial systems. Broadly, M-Air is a preeminent effort in electric aviation, including passenger aircraft, cargo and other aerial applications.
A modern aircraft turbine delivers roughly 6,000 watt-hours per kilogram, while today’s best batteries are around 200 watt-hours per kilogram. That 30-times energy gap has limited electric aviation to shorter distances so far. Batteries are also still much heavier than jet fuel. You have to change the electrodes–the anode or the cathode, not just the electrolytes–to achieve the necessary energy density. Aionics has been working on discovering new materials that can provide those capacities. Like electric vehicles, the technology will improve over time and there are many projects in electric aviation using today’s battery technology to improve power delivery.
Where do you see the company going in the next few years?
Aionics is a venture-backed startup with investors including UP Partners, Trousdale and the University of Michigan Endowment. Partnerships with commercial companies are equally critical because our projects are structured as joint development agreements. Our projects are joint development agreements that allow us to enter into these ripe spaces.
The incredible transformational capabilities enabled by these AI methods are just getting better–what we can do now is already different from what we could do six months ago. Historically, we focused on energy storage, but the foundation models we built can now reason across many domains. Just this year, we have expanded into geothermal applications and fluids for data center cooling. We have also developed one of the most performant models for predicting how molecules smell, which has application to liquid mixture formulations of lubricants, paints and thermal fluids. We believe AI-driven materials science creates the opportunity for the emergence of a new Dow or DuPont. The goal of Aionics is to go further into other molecular formulation markets that will make big impacts on large parts of the economy.
