Researchers at Tohoku University have built an open-source laboratory platform that carried out repeated syntheses of ZIF-8 while recording the conditions of each run. ZIF-8 is a porous crystal studied for gas separation and chemical sensing. In a preprint posted on September 14, the team reports that changing how quickly one solution was dispensed produced a repeatable change in the particle sizes measured after synthesis. The experiment tests whether routine laboratory work can be automated without losing process details needed to interpret the result.
Materials experiments are sensitive to actions easily omitted from a written recipe. Liquid-addition rate, intervals between steps and container position can affect a reaction. Manual records can miss those details; custom robotic laboratories are expensive and hard to adapt. The Tohoku group instead used commercial robot arms and electric pipettes, a webcam, a balance, connected sensors and 3D-printed fixtures.
The equipment follows a structured sequence and records its actions and measurements. An optional AI assistant, built with Gemini 3.5 Flash, can turn a plain-language request into a JSON procedure. A researcher reviews it in a graphical interface, and the software checks configured limits before sending commands to the equipment. The project's repository notes that the reported ZIF-8 batches used a spreadsheet-defined workflow rather than the AI-generated procedure. The experiment therefore tests automated execution and logging, not autonomous planning.
For the demonstration, the system mixed two 5-millilitre solutions and varied the dispensing speed of the second one. The three settings corresponded to about 1, 3.5 and 11 millilitres per second, or roughly five seconds, 1.5 seconds and half a second to deliver the full volume. It completed nine syntheses, three at each setting, with the conditions alternated to make drift over time easier to detect. Additional batches were prepared for X-ray diffraction and dynamic light-scattering measurements.
X-ray diffraction confirmed that ZIF-8 was the main crystalline phase. Apparent crystallite sizes ranged from 25 to 36 nanometres, with no systematic dependence on dispensing speed. Dynamic light scattering, which estimates the size of particles in liquid, showed a different pattern: measured particles became larger as dispensing slowed. The trend appeared in two independent syntheses at each setting. Its cause remains unresolved because the method cannot separate crystal growth from particle aggregation, and the study included no direct comparison with manual synthesis.
The contribution is a traceable experiment linking a controllable process variable to the material measured at the end. Such records are useful to researchers and machine-learning systems that need consistent experimental data. The result does not establish a general rule for ZIF-8: earlier studies using other concentrations and addition orders reported different trends. Independent laboratories, manual baselines and other reactions will be needed to test whether the platform can preserve this repeatability beyond its first demonstration.
Related coverage
Sources
- Yusuke Hashimoto and colleagues, “Quantitative control and recording of materials-synthesis processes using an automated experimentation platform,” arXiv, September 14, 2026. Platform design, ZIF-8 experiment, measurements and limitations.
- SAIGEN project repository. Software, hardware files, operating documentation and clarification of the workflow used for the reported experiment.
- Lead image: Figure 5 from the paper's HTML edition, reused unmodified under CC BY 4.0. It shows the reported automated solution-mixing system.