JAIST Scientists Break Solar Cell Efficiency Records Using Machine Learning 

  • JAIST researchers utilize machine learning for substantial efficiency improvement in SHJ solar cells.
  • Professor Ohdaira’s team pioneers constrained Bayesian optimization for optimizing Cat-CVD deposition.
  • The discovery of specific deposition conditions leads to a record-setting 25.6% efficiency for SHJ solar cells.
  • Implications extend beyond solar cells, highlighting the broader applicability of machine learning in material processes.
  • JAIST’s advancements promise more affordable and effective solar power solutions.

Main AI News:

In a landmark study featured in ACS Applied Materials and Interfaces on February 21, 2024, researchers from the Japan Advanced Institute of Science and Technology (JAIST) have utilized machine learning techniques to substantially boost the efficiency of silicon heterojunction (SHJ) solar cells. This breakthrough not only signifies a pivotal advancement in solar energy research but also heralds a promising stride toward addressing climate change through more effective renewable energy resources.

Revolutionizing Solar Cell Production Methods 

Professor Keisuke Ohdaira spearheaded the research, supported by a team that included Ryota Ohashi, Huynh Thi Cam Tu, Koichi Higashimine, and Kentaro Kutsukake. Their focus was on refining the deposition parameters for the passivation layer in SHJ solar cells using catalytic chemical vapor deposition (Cat-CVD). Employing a novel approach called constrained Bayesian optimization (BO), a machine learning technique, the team pinpointed the ideal conditions for generating high-quality intrinsic hydrogenated amorphous silicon (i-a-Si:H) films. This methodology minimizes the traditional trial and error associated with Cat-CVD, expediting the journey towards higher efficiency solar cells.

Surpassing Efficiency Thresholds 

The application of constrained BO enabled the researchers to systematically explore and identify deposition conditions that were previously overlooked or considered unachievable. Notably, they discovered a specific combination of substrate temperature and precursor gas flow rate that effectively mitigates carrier recombination, a prevalent issue diminishing solar cell efficiency. With just twenty optimization cycles, the team achieved a groundbreaking efficiency of 25.6% for SHJ solar cells, surpassing the previous record of 22.3%. This advancement holds significant implications for the solar energy industry, paving the way for more cost-effective and potent solar power solutions.

Future Prospects for Solar Energy 

The significance of this study extends beyond the remarkable enhancement in solar cell efficiency; it underscores the potential of machine learning methodologies like constrained BO in streamlining and expediting material and process optimization across diverse domains. The researchers’ capacity to derive new scientific insights through optimization underscores the transformative potential of integrating machine learning into traditional manufacturing processes. As the global community continues its quest for sustainable energy solutions, the strides made at JAIST represent a substantial leap forward in harnessing solar energy more efficiently and economically.

The ramifications of this research transcend solar cells, suggesting broader applications for machine learning in optimizing intricate material processes. This could revolutionize not only the manufacturing of solar cells but also a myriad of electronic devices. As we progress, the endeavors led by Professor Ohdaira and his team at JAIST are poised to ignite further innovation, propelling the advancement of sustainable technologies and bolstering the global fight against climate change.


The breakthrough achieved by JAIST researchers in enhancing solar cell efficiency through machine learning signifies a pivotal shift in the solar energy market. It promises more cost-effective and potent solar power solutions, paving the way for sustainable energy adoption on a larger scale. Additionally, the broader applicability of machine learning in material process optimization hints at transformative changes not only in solar cell manufacturing but across various industries reliant on electronic devices. This development underscores the urgency for market players to embrace innovative technologies to stay competitive and contribute to the global transition towards clean energy solutions.