An amazing race is on. A race to battle many threats facing humanity, including global warming and pandemics such as COVID-19. These challenges push science and technology experts to develop long-term solutions for protecting our planet and our people against known and unknown threats. And since we cannot guess what surprises may be lurking around the next corner, having adaptable and easy-to-use methods in place has never been more important.
The National Research Council of Canada (NRC) has worked closely over the years with partners from industry, academia and other research institutes to successfully respond to existing and foreseeable risks. Through the AI for Design Challenge program launched in 2020, the NRC is pushing the envelope by harnessing artificial intelligence (AI) to boost scientific discovery and engineering design to solve the most complex problems in a range of fields. In the end, the results will benefit our long-term well-being.
According to Chief Digital Research Officer Dr. Joel Martin, the NRC is well positioned to not only develop but also test numerous disruptive technologies and solutions. "The core theme in the AI for Design Challenge program devises methods for speeding up simulations, improving design-space searches and making easy-to-understand models that clearly correlate to the real world," he says. "And at the end of the day, we will share with the world AI algorithms that can be used by anyone—including non-AI experts—across a broad set of domains."
Martin adds that additional Challenge programs provide an interconnected testing ground for the solutions developed in the AI for Design Challenge program. Some have already demonstrated tangible results; others are being launched. These include using AI to power research into medical therapies, accelerate materials discovery and improve photonics designs.
Powering medical therapies
AI-powered design for stem cell therapy promises to speed research into treating degenerative muscle diseases. While these cells can grow and repair muscles after an injury, the signals that direct them to act are poorly understood. Additional research into these signals can uncover new ways to promote growth in aging muscles and muscular dystrophies.
"Clinical application of this type of therapy poses immense challenges because it is hampered by the complexity of stem cells and the different ways in which they make decisions to generate functional tissues," says Dr. Xiaojian Shao, NRC Project Lead, AI-powered design of stem cell therapy for degenerative muscle diseases. Stem cells can divide to replace lost (undifferentiated) cells or they can be directed (differentiated) to form new tissues and organs.
"The focus of our research is to improve muscle stem cell function in different specific disease scenarios," adds Dr. Shao. But testing such scenarios requires thousands of trials and experiments so it is extremely time-consuming to do them in a lab. "With the machine-learning / AI methodologies we are developing, we believe we can accelerate experiment design and achieve ideal scenarios much faster."
Underlining the need for AI and machine learning are technological developments that are producing a flood of molecular data. For example, researchers now take into account the latest in genomic research: single-cell genomics technology, which can more accurately identify different populations of cells within a tissue and their inner workings. This technique simultaneously measures thousands of genes in thousands of single cells from a single specimen.
"We and our partner, the Ottawa Hospital Research Institute, will develop a variety of methods to stimulate, manipulate and connect muscle stem cells," says Dr. Shao. "Once we have performed tests and experiments and validated the findings, we will have the right information to design treatments that generate functional muscle cells."
Accelerating materials discovery
High-entropy alloys (HEAs) are revolutionizing materials science. These stronger, more malleable, wear- and heat-resistant alloys contain 5 or more metals in roughly equal concentrations. By mixing and rearranging the components, scientists can create completely new HEAs, with the number of possible configurations running into the millions. Potential applications include state-of-the-art spacecraft, submarines and nuclear reactors. HEA materials are also efficient electrocatalysts for a variety of reactions and exhibit reversible hydrogen storage capacity.
In this vast new space, predicting new HEAs requires more computer modelling resources than conventional alloys, which use fewer elements. To find the shortest way to identify the right elements, NRC scientists from the Energy, Mining and Environment Research Centre and the Security and Disruptive Technologies Research Centre are collaborating with the University of Toronto Computational Materials Engineering lab to exploit the power of AI and machine learning.
"With the AI model we developed, we generated a system to prototype the properties that was larger than any used in HEA studies before," says Dr. Conrard Giresse Tetsassi Feugmo, NRC Research Officer, Machine Learning. "We generated structures containing more than 40,000 atoms in a matter of hours—roughly 1,000 times faster than quasi-random approaches." These structures will be used in large-scale simulations to predict more realistic properties. A paper on this topic by Tetsassi Feugmo et al. was featured on the cover of the American Institute of Physics (AIP) Publishing's Journal of Chemical Physics: "Neural evolution structure generation: High Entropy Alloys."
"The HEAs we develop will be particularly useful in aerospace and clean fuels—and contribute to Canada's goal of net-zero carbon emissions by 2050," adds Dr. Feugmo. "The metallurgy industry needs inexpensive, lightweight and temperature-resistant materials as well as a successful catalyst for water electrolysis in fuel cell processes." The next phase is to test the new system at the NRC's advanced materials research facility in Mississauga.
Enhancing design of photonic power converters
An exciting new alternative to sending electricity over copper wire has seen the light. Using a photonic power converter, power can be transmitted in the form of light and then converted into electricity. These systems are not affected by electromagnetic interference and show improved operation in hazardous conditions. Light can be beamed almost anywhere—the distance travelled depends on its wavelength. By designing photonic power converters to work with a new wavelength, efficient photonic power transmission will be possible over many kilometres, even through the atmosphere.
Photonic power converter chips and related applications are in the early stages of development. Improved efficiency and a new operational wavelength will make them valuable in myriad electric systems—from autonomous cars to satellites.
The NRC's AI for Design Challenge program has recently launched a multi-partner collaboration to make that happen quickly. Team Lead Dr. Karin Hinzer, University of Ottawa Research Chair in Photonic Devices for Energy, explains that these chips have many layers and countless variables to consider. "As we make more complex chips, we need AI techniques for advancements in materials and device optimization." These methods, especially dimensionality-reduction software, will help design a more efficient device that can "make light go farther."
All partners are world leaders in their fields. The NRC, a pioneer in photonic device fabrication and a leader in AI, is helping the University of Ottawa adapt innovative models to AI. The designs will be manufactured by long-standing NRC partners Fraunhofer Institute for Solar Energy Systems ISE and AIXTRON SE. Canadian industry leaders Optiwave Systems Inc. and Broadcom Canada are creating software that will predict how the design works within a system and will scale up the manufacturing.
And at the end of the project, the partners will present a chip unlike anything the world has seen before.
Building on success
These are just some examples of how the AI for Design Challenge program is gaining momentum in Canada. The program has also proved that with more data being generated than humans can manage, the world can no longer progress without AI.
"Since the AI for Design Challenge program was launched we have made great strides in a number of areas," says Dr. Martin. These include using AI and machine learning to find novel solutions for COVID-19 health concerns, speed up drug discovery and characterize the multi-parameter design space of nanophotonic components.