Skip to main content

100xfaster synthesis

IBM RoboRXN: Making a new material, without ever going into the lab.

*Comparison of historical synthetic efforts (person-decades) with the typical retrosynthetic prediction (minutes)

Synthesizing and testing a new material is often the biggest bottleneck in the discovery process. On top of requiring whole teams of devoted technicians, it can take years to identify and produce the appropriate chemicals, conditions, and reactions to properly carry out the synthesis.

See how we used IBM RoboRXN to create a new molecule

How it works

Scientists identify the molecule. Our integrated lab figures out the recipe, and makes it.

Synthesizing a new material is not unlike baking. Once you’ve decided on what you want to make, the next step is determining the appropriate ingredients and recipe. It’s a matter of figuring out the right sequence, conditions, and ratios to bring everything together.

AI radically accelerates chemical “recipe development.” Through a process called “retrosynthetic analysis,” a researcher simply uploads the desired material to the IBM cloud, and our system will use an AI model to break the molecule down into a chemical recipe, while also identifying a set of commercially available ingredients to be used in the synthesis process.

The recipe is then automatically executed by a robotic autonomous lab. In literally minutes, the drafted molecule becomes a reality. It once took decades to find out how to synthesize a new material; AI accelerates every step of this process, from conception to execution.

Figure R1.


Reverse-engineering the recipe

Scientists upload the molecule to the IBM RoboRXN cloud.

An AI model trained on over 3 million chemical reactions (derived from patents) quickly breaks the molecule down into a series of reactions and ingredients.

Graphic that shows chemical reactions


Graphic that shows various trees of nodes

Optimizing executability

The AI model determines if the necessary ingredients are commercially available. If not, it will keep breaking the ingredients down until it gets to commonly available components.

The model will also present multiple possible ingredient/sequence combinations for scientists to pick from and refine according to their expertise.


Instructions for robots

Once the recipe is ready, the AI engine translates it into a series of detailed steps (including time, temperature, order, etc.) that are sent via the cloud to our robotic autonomous lab for execution.

Screenshot of AI generated instructions intended for a robotic autonomous lab


Testing remotely

Scientists can remotely monitor the synthesis and verify the results with online automated analytical instruments.

Getting from molecular design to reality at record speed. Our AI-driven autonomous lab combines the brainpower of AI, the speed and precision of robotic automation, and the seamlessness of cloud

How can a
chemist work
in a home
Dr. Teodoro Laino Ph.D.
IBM RoboRXN at work:

Project Photoresist

A new set of environmentally-friendly PAGs, synthesized in minutes.

Through a combination of generative modeling and AI-enriched simulation, our experts had arrived at a shortlist of novel PAG candidates. The final step was synthesis.

We began by running a retrosynthetic analysis on our candidates. Although our AI model had already been trained on over 3 million chemical reactions, we realized that only 30 of the 3 million were involved in the synthesis of a photoresist. To find the best chemical recipes for our PAGs, we’d need to specialize the model’s knowledge base.

By training the AI with an additional 1,000 photoresist-specific chemical reactions (derived from targeted scientific publications), we were able to make the model twenty times more accurate for our use case—a testament to the flexibility and tailorability of AI models.

Once approved, we sent the ingredients and their instructions to our remote autonomous lab in Zurich. Our scientists were able to manufacture these novel PAGs without once setting foot in a lab—a critical cost and time efficiency.

Combined, IBM RoboRXN delivered a 100x faster synthesis than traditional laboratories. In less than a year, we were able to go from studying, to simulating, to testing new and sustainable PAG materials. With AI, IBM was able answer an environmental challenge in record time—getting the world that much closer to more sustainable computing technology.

Figure R2.

Final choices

Our experts determined a shortlist of PAG candidates ready for synthesis. After running a retrosynthetic analysis, they narrowed in on the best chemical recipes to choose from.

Graphic of a molecular structure

Figure R3.

Photo of lab equipment


Our cloud-based AI autonomous lab handled the innumerable logistics, calculations, conversions, and integrations required to successfully synthesize our candidates. And, it did it 100x faster than traditional processes.

Find out more about IBM RoboRXN

Retrosynthetic Analysis AI

Language Models for converting Experimental Procedures, predicting Chemical Reactions or Retrosynthesis Pathways and automating Chemical Synthesis.

AI lab

IBM RoboRXN for Chemistry is a pioneer project combining AI, Automation and Cloud to accelerate material discovery.


Philippe Schwaller, Riccardo Petraglia, Valerio Zullo, Vishnu H. Nair, Rico Andreas Haeuselmann, Riccardo Pisoni, Costas Bekas, Anna Iuliano, and Teodoro Lainoa

Chemical Science (2020)

Alain C. Vaucher, Federico Zipoli, Joppe Geluykens, Vishnu H. Nair, Philippe Schwaller, and Teodoro Laino

Nature Communications (2020)

Philippe Schwaller, Teodoro Laino, Théophile Gaudin, Peter Bolgar, Costas Bekas, and Alpha A Lee

Chemical Science (2019)

Philippe Schwaller, Theophile Gaudin, David Lanyi, Costas Bekas, and Teodoro Laino

Chemical Science (2018)

Discovery Workloads
on the Hybrid Cloud

Emerging discovery workflows are posing new challenges for compute, network, storage, and usability. IBM Research supports these new workflows by bringing together world-class physical infrastructure, a hybrid cloud platform that unifies computing, data, and the user experience, and full-stack intelligence for orchestrating discovery workflows across computing environments.