A research team at VinUniversity has developed a new artificial intelligence (AI) technology capable of automatically classifying experimental samples in the search for new materials. A task that once demanded 1,000 hours of manual work can now be completed in just 10 hours with AI. The study has been published in Nature Synthesis in October, 2025.

Few people realize that the answers to many issues troubling global leaders lie in something fundamentally familiar: materials. They form the foundation of every breakthrough: from capturing industrial and residential CO₂ emissions, to harvesting water from arid desert air, to building next-generation high-efficiency solar panels.

It is a vision of a brighter future, one where humanity lives in harmony with nature and leverages science to build a sustainable coexistence. But the journey ahead remains steep. Materials science still relies heavily on trial-and-error methods dating back to the 19th century: mixing chemicals, heating them, waiting, and manually inspecting each sample one by one.

 

 

• One scientist spends 1,000 hours on manual experiments
• 90% of trials fail
• Thousands of recipes may be tested before discovering one new material

“It’s like searching for a needle in a haystack, except you don’t even know if the needle exists,” shared Prof. Omar Yaghi, one of the world’s leading materials scientists.

With such manual practices, the likelihood of discovering new intelligent material structures is extremely low, while the process remains laborious, inefficient, and limiting compared to the vast number of possible chemical combinations.

 

 

 

 

 

After more than 20 years as a professor at the University of California, Berkeley, Prof. Laurent El Ghaoui moved to Vietnam in 2020 to serve as Vice Provost of Research and Innovation at VinUniversity. “I saw tremendous potential here, a young university,” he shared.

In 2021, Prof. Omar Yaghi, one of Prof. Laurent’s former colleagues at UC Berkeley, was honored with the VinFuture Prize for pioneering work on Metal–Organic Frameworks (MOFs). During his visit to Vietnam, Prof. Laurent introduced him to VinUni.

“Omar told me about his idea for the project ‘Algorithmic Iterative Reticular Synthesis (AIRES),’ a technology that uses AI to discover new materials. I immediately thought of the VinUni team,” Prof. Laurent recalled.

Dr. Phùng Thị Việt Bắc, VinUni’s Head of Research Administration at that time, instantly recognized the opportunity: “This is exactly what we are pursuing at VinUni: using AI to discover green materials that can replace harmful chemicals. The connection through VinFuture opened a door to world-class collaboration.”

“When Laurent introduced the VinUni research team, I knew these were capable scientists. This wasn’t a collaboration in title, but a genuine scientific partnership,” Prof. Yaghi emphasized.

 

Following this connection, Dr. Phạm Huy Hiệu and Dr. Lê Duy Dũng from VinUni’s College of Engineering and Computer Science traveled to California with nothing but their laptops and determination to learn. At Prof. Yaghi’s lab, they faced the enormous challenge ahead: 20,000 microscope images to be classified into crystalline successes and non-crystalline failures, with more than 85% accuracy required.

“Typically, AI needs a million images to train well. But at VinUni, we only had 650 images. A problem arose: how do we train an effective AI model with so little data?” shared Dr. Dũng.

“To solve the data scarcity problem, we designed an algorithm that leverages powerful computer-vision models, applies data augmentation, and focuses on learning features that distinguish the samples most prone to confusion,” added Dr. Hiệu.

Once back in Vietnam, Dr. Hiệu and Dr. Dũng immediately set to work. The team adopted an inventive approach: using EfficientNetV2-S, leveraging ImageNet pretrained weights, and applying enhancements such as flipping, contrast adjustments, and blurring. The result exceeded expectations: with a recall rate of 88%, the AI could correctly detect 9 out of 10 crystalline samples.

AIRES is a compelling example of the rsearch model VinUni is constructing: faculty have dedicated time for research, laboratories are equipped with modern tools, and computing infrastructure is strong enough to support advanced AI projects.

 

Nhân and Huyền are two young research assistants who played hands-on roles in developing the AI model, processing data, and conducting experiments. Their names appear as co-authors in the Nature Synthesis publication, a testament to their contribution.

And they are not the only ones.

At VinUni, research assistants work full-time on international projects, co-author high-impact publications, and receive mentorship from experienced scientists. Graduate programs provide access to global academic networks, collaborative research, and real-world scientific challenges.

With a goal to enter the top 100 universities in the QS Rankings by 2030, VinUni is intensifying research investments across fields, including health sciences, environmental science, and smart urban technologies. High-impact scientific publications play a crucial role in this evaluation.

“The publication in Nature Synthesis is one of the first clear milestones,” Dr. Bắc shared. “We are building a strong system that supports the consistent production of high-quality research.”

 

 

Before the support of AI trained by VinUni, 1,400 random experiments produced only 10 new material structures while requiring 1,000 hours of continuous work. With AIRES, everything changed. With just 700 experiments, the team achieved the same results in under 10 hours, a 100-fold acceleration.

The impact is unmistakable:

• The Zn-ZIF material library expanded by one-third, the largest dataset collected in a decade.
• Chemical-mixing robots now operate five times faster than humans.
• AI automatically suggests optimal chemical recipes, dramatically shortening the path from idea to discovery.

 

 

 

 

 

 

 

 

At the VinUni lab, Dr. Nguyễn Đăng Tùng, faculty member at CECS and principal investigator of the AI-accelerated green materials project at CEI, has set ambitious goals: discovering smart materials that can absorb toxic gases, replace harmful chemicals, and support clean, sustainable energy.

The promise of AIRES has already been validated. “Now we have proof in Nature Synthesis that VinUni can do world-class science. The next step is scaling this impact for Vietnam,” said Prof. Laurent.

 

This publication carries special meaning for VinUni’s journey toward strong research capacity. Genuine, rigorous collaboration and international academic credibility have reinforced the university’s standing.

“VinUni has shown that a five-year-old university in Vietnam can contribute meaningfully to global science. More than just a prize, VinFuture has served as a bridge that fosters substantive scientific connections,” noted Prof. Laurent El Ghaoui.

Moving forward, the team plans to expand international partnerships, train the next generation of Vietnamese AI scientists, and build an automated laboratory at VinUni, step by step realizing the vision of an advanced research ecosystem.

 

The team is already pursuing the next project, this time entirely in Vietnam.

“From VinFuture to Nature Synthesis, we’ve shown that ‘Made in Vietnam’ is not only about manufacturing, but also scientific innovation,” affirmed Dr. Tùng. “And this is only the beginning.”

 

From just 650 images, VinUni has contributed to a global breakthrough. This story goes beyond AI or materials, it strengthens the belief that Vietnam can create world-class science. Step by step, Vietnamese researchers are building an advanced research ecosystem where AI and materials science propel each other forward to create transformative discoveries.

 

---

[1]. Rong, Z., Chen, Z., Luong, F. et al. Algorithmic iterative reticular synthesis of zeolitic imidazolate framework crystals. Nat. Synth (2025). https://doi.org/10.1038/s44160-025-00939-9