Ryoichi Ishihara

Associate Professor, Group leader

Qutech, Dep. Quantum and Computer Engineering, Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology

Ishihara-lab focuses on the integration technologies for unconventional electronic systems; quantum computers, quantum sensors, neuromorphic computers, and biodegradable sensors. Our work involves new materials, scalable fabrication of electronic and photonic devices, and 3D heterogeneous integration, aiming to realize unconventional electronic systems.

Enhancing the Selection of Single Color Center Defects in Diamond Using Machine Learning

Background:

Single color center (CC) defects in diamond, such as nitrogen-vacancy (NV) centers, are crucial for the advancement of next-generation quantum technologies, including quantum computing, quantum communication, and high-precision sensing. NV centers, in particular, have garnered significant interest due to their unique quantum properties, which can be harnessed for applications in quantum bits (qubits), quantum repeaters, and magnetometry.

However, naturally occurring NV centers are randomly distributed within the diamond lattice and display a wide range of optical and spin properties. This variability poses a significant challenge in identifying NV centers that meet the stringent criteria required for specific quantum applications. The current process of selecting suitable NV centers from experimental data, such as 2D photoluminescence (PL) scans, is both time-consuming and labor-intensive, often involving extensive manual analysis and trial-and-error.

Objective and Key tasks:

The goal of this thesis project is to develop machine learning models that efficiently and accurately identify high-potential NV centers from large experimental (say, PL) datasets. By automating the candidate selection process, the project aims to significantly reduce the time and effort required to identify suitable NV centers for quantum applications. Additionally, the project seeks to uncover new correlations within the data, contributing to a deeper understanding of CC defects in diamond.

1. Data Preprocessing and Feature Engineering:
   • Organize and preprocess the extensive experimental data, which includes 2D photoluminescence scans and performance metrics of NV centers.
   • Perform feature extraction to identify relevant parameters that may influence the quality and performance of NV centers, such as PL intensity, emission wavelength, and spatial distribution.
2. Development of Machine Learning Models:
   • Implement supervised learning models (e.g., Random Forest, Support Vector Machines, Neural Networks) to classify NV centers based on their likelihood of meeting specific quantum technology criteria.
   • Explore unsupervised learning techniques (e.g., clustering, anomaly detection) to identify novel patterns or subcategories of NV centers that may have been overlooked in manual analysis.
3. Model Training and Validation:
   • Split the dataset into training, validation, and test sets to evaluate the performance of the developed models.
   • Use cross-validation techniques to fine-tune hyperparameters and prevent overfitting.
4. Model Interpretability and Correlation Analysis:
   • Utilize explainable AI methods (e.g., SHAP values, feature importance) to interpret the decision-making process of the models.
   • Investigate correlations discovered by the machine learning models to gain insights into the physical properties of NV centers that contribute to their performance.
5. Application and Optimization:
   • Apply the trained models to new datasets to streamline the selection of high-potential NV centers.
   • Continuously refine the models based on feedback from experimental results to improve accuracy and reliability.

Expected Outcomes:

• A robust machine learning pipeline capable of rapidly identifying candidate NV centers with high potential for quantum technology applications.
• Reduced manual effort and time required for the selection of NV centers, enabling faster experimental workflows.
• New insights into the characterization of NV centers that are most relevant for specific applications, potentially leading to better-targeted characterization experiments and efficient candidate selection

Interested? Please contact Ryoichi Ishihara r.ishihara@tudelft.nl or Salahuddin Nur <S.Nur@tudelft.nl>