An AI-powered revolution in clean energy chemistry
UBC’s Project Ada, the first self-driving lab to fast-track clean energy materials discovery, has come of age
By Geoff Gilliard
With the signing of the Paris Climate Accord, most of the world’s governments agreed that we’d need to slash fossil fuel emissions in half to reach net zero by 2050. Net zero is the point of equilibrium between the greenhouse gas we produce and what’s removed from the atmosphere by forest and ocean carbon sinks.
Clean energy technologies such as solar panels, improved batteries and fuel cells offered hope in reducing emissions, but the Paris Agreement made clear that there was massive gap in getting affordable new products to market. Materials discovery is a major part of the innovation cycle of clean energy technology, but developing commercially viable materials typically takes over a decade.
That technological shortfall sparked Mission Innovation, a global initiative launched in 2017 to find ways to break the research and development bottleneck. Mission Innovation issued a call to action — combine materials science with automation and artificial intelligence to accelerate materials discovery by at least ten-fold. Canada was the first to answer this call with the launch of Project Ada, a University of British Columbia-led initiative that positioned the nation as a world leader in self-driving labs.
In 2018, an $8 million Natural Resources Canada grant enabled UBC professors Dr. Curtis Berlinguette and Dr. Jason Hein, along with Dr. Alán Aspuru-Guzik of the University of Toronto, to assemble “Ada”, the first self-driving lab for clean energy materials. The multidisciplinary team chose the name in honour of Ada Lovelace, the world’s first computer programmer. Self-driving labs like Ada think for and work by themselves, combining automation with machine learning to plan, conduct and analyze experiments much faster than humans can.
“It was a time where people were just starting to look at AI,” Dr. Hein recalls. “There were a lot of promises about the wonders AI would deliver, but it fell short because it was oversold and under delivered.”
The Project Ada team quickly identified the importance of choosing AI-suitable challenges in order to accelerate discovery.
“If you want to make better concrete and your test involves making the concrete and then waiting 20 years to see how it lasts, that’s a really bad loop,” Dr. Hein says. “You have to come up with an experiment we can learn from in a meaningful timeline. We learned not only that a self-driving lab could work, but about the structures that define what a self-driving lab actually is.”
Once the scientist defines the search space and presses “go”, AI plans the experiments for the automated hardware. The resulting data then informs the design of subsequent experiments in a closed loop. Error correction is built in to the process and results are fed back to the algorithm to prevent a disconnect between what the AI says should happen and what the robot actually delivers.
“Often we start empirically from a blank slate because it’s a new problem, and the AI is learning on the fly,” says Dr. Hein. “In the initial days, early algorithms were ridiculously data hungry. Data from millions of experiments was needed before the AI could make a guess. We’ve gotten a lot better in the last couple of years at figuring out how to bound the problem and use different kinds of algorithms so the guessing happens much faster.”
When the autonomous lab community started, self-driving labs were one-off efforts built from scratch. Commercially available platforms could take years to design and build, and were rigidly customized to a very specific workflow. Inevitably the lack of an industry standard led to duplicated efforts. Due to the resources, time and people required to build a self-driving lab, Project Ada focused on flexible automation which could quickly pivot to different materials and applications if — or more likely, when — the plans for an experiment changed.
The first Ada platform consisted of modules linked together to autonomously make and test thin-films for solar cells. With humans at the helm, optimizing the unique and unconventional properties of thin-films was time intensive. A self-driving lab produced a much more efficient search.
“We were able to pivot to explore different materials and applications by reconfiguring and adding new modules to meet the evolving needs of our experiments,” says Dr. Berlinguette. “To date we’ve focused on making and testing thin-film materials, because they’re a key component of basically any clean energy technology.”
New funding secures Canada’s edge
Drs. Hein and Berlinguette lead the Vancouver branch of the Acceleration Consortium which recently received a $200-million grant for self-driving lab development from the Canada First Research Excellence Fund (CFREF). The boost will help researchers at UBC and the University of Toronto to accelerate research translation from early-phase discovery to real-life deployment of new materials and molecules — from life-saving medications and biodegradable plastics to low-carbon cement and renewable energy.
“The CFREF grant was only possible because of that early investment with Ada,” Dr. Hein says. “We’re running full steam and now is the time for Canada to invest in self-driving labs so we can keep our advantage. CFREF is one of very few mechanisms in Canada to really focus on long-term personnel development and capacity building. This will help to train the next generation of scientists to be leaders in the field as it grows out.”
Some of Dr. Hein’s and Dr. Berlinguette’s graduate students and postdoctoral fellows have already moved directly into senior positions in industry. Several members of the Ada team now work at UBC spin-off enterprises.
Dr. Hein’s spin-off company, Telescope Innovations, uses self-driving labs to develop scalable manufacturing processes and tools for the pharmaceutical, biotech, mining and chemical industries.
“We don’t silo ourselves,” says Dr. Hein. “We’ve been at the ground level helping to form that narrative about a new way of doing science. A lot of my time now goes into helping people who want their own self-driving lab. We compress timelines to get a product to market faster. It’s a way of doing business better.”
With a modular toolkit at their disposal, the Hein and Berlinguette labs have a blueprint for building a self-driving lab that can readily be configured to suit different workflows.
“It’s as much teaching people how to use the tool as it is to build it,” Dr. Hein says. “A chemist may need 20 kilos of some molecule. They know how to make a gram, but not how to make a larger amount — and they only have a month. We can figure out very quickly the right path to make that work. Instead of just repetitiously grinding through and trying everything, we can identify optimal quicker.”
Getting a new self-driving lab off the ground starts with a conversation so the researchers can understand a company’s expectations, the problem, and what resources are available. Then they help assemble a suitable team. Developing Ada highlighted the benefits of embedding automation engineers and machine learning scientists alongside chemists so they could accelerate workflows.
“Depending on their field, scientists use the same language to describe very different ideas. Understanding how to broker and manage these conversations across disciplines that don’t often talk to each other is key,” Dr. Berlinguette says. “If you don’t have that communication you run the risk of engineers building fancy tools that scientists don’t need.”
A window on the future
Dr. Berlinguette’s spin-out, Miru Smart Technologies, developed “Adam”, a next generation self-driving lab. Adam employs Ada’s core spray coating infrastructure to accelerate the commercialization of electrochromic — or “smart” — windows. Smart windows feature layers of thin-films sandwiched between glass. The film can be controlled by a smart phone to change their opacity between clear, tinted or dark.
“Smart windows can improve building energy efficiency by 20 per cent,” Dr. Berlinguette says. “About 27 per cent of greenhouse gas emissions are due to the heating and cooling of buildings, which is more than the emissions that come from all of the airplanes and cars in the world. Windows are arguably the most energy inefficient component of buildings today. We’re trying to bring the first innovation to windows in the last 40 or 50 years.”
The challenge to making effective smart windows is filtering through the vast number of parameters that can potentially be used in their manufacture. Adam optimized the formulations required to make windows with a colour-neutral tint and very high optical clarity. Other commercialized smart windows tend to have a blue or yellow tint, which hasn’t gone over well in the marketplace.
“We built out Adam in about six months, and it took less than a month for the robot to optimize the product characteristics of smart windows that otherwise would have taken us years,” Dr. Berlinguette says.
Decarbonization research is also a key priority for the Berlinguette Group.
“We design and build electrochemical reactors to convert CO2 into useful products, to drive nuclear fusion at low temperature, and to electrify the fuels, chemicals, and cement industries,” says Dr. Berlinguette. “We view flexible automation and self-driving labs as an enabler to accelerate discovery and translation of new technologies in these areas from lab to market.”
With support from Natural Resources Canada and the National Research Council of Canada, the Berlinguette Group expanded their first Ada platform to build a self-driving lab for the discovery and optimization of materials used in CO2 electrolyzers. Millions of formulations are available for the catalysts, and several different process parameters need to be optimized such as temperature, pressure and flow rates.
“All of these different parameters are why the development of fuel cells, for example, has taken so long to commercialize,” Dr. Berlinguette says.
“Ada helps us optimize electrochemical reactors much more quickly. We’re just starting to put it into practice now at UBC, so we’re excited because the broader mission is to accelerate decarbonisation technologies. Smart windows are a part of that, but also figuring out how to make clean fuels so we’re not sourcing all our energy from fossil fuels.”
