Periodic Reporting for period 1 - Next Gen XRM (NEXT GENENRATION OF X-RAY METROLOGY)
Reporting period: 2023-06-01 to 2024-02-29
The goal of this project was to develop a next generation software solution based on machine learning to facilitate and speed up the data analysis of X-ray metrology data.
X-ray Metrology will be a critical technology for the next generation of semiconductor device production.
The devices are getting smaller, and the material stacks are becoming more complex, therefore the need of more sophisticated metrologies to better understand the material properties and the manufacturing.
Differently from optical metrology, X-rays probe the material at the atomic scale and can penetrate from a few nanometres to millimetres, according to the wavelength used , which matches the low dimensions of semiconductor devices.
One of the greatest challenges for the larger adoption of X-ray metrology into semiconductor wafer fabs has been the relatively slow measurements time (minutes) in comparison to optical metrology (seconds), and the complexity of the data analysis.
A lot of progress has been achieved to speed up the measurements via new X-ray sources and detectors, but the data analysis is still lacking.
We have elaborated a software based on pattern recognition and machine learning applied to X-ray data which will has speeded up the analysis to less than 1 minute.
We have used X-ray 2 dimensional data, ie reciprocal space maps (RSM) which are a detailed representation of the crystalline structure of the area probed, and extract the material parameters relevant to different processes.
This work is very important for new process development and new materials characterization.
X-ray Metrology will be a critical technology for the next generation of semiconductor device production.
The devices are getting smaller, and the material stacks are becoming more complex, therefore the need of more sophisticated metrologies to better understand the material properties and the manufacturing.
Differently from optical metrology, X-rays probe the material at the atomic scale and can penetrate from a few nanometres to millimetres, according to the wavelength used , which matches the low dimensions of semiconductor devices.
One of the greatest challenges for the larger adoption of X-ray metrology into semiconductor wafer fabs has been the relatively slow measurements time (minutes) in comparison to optical metrology (seconds), and the complexity of the data analysis.
A lot of progress has been achieved to speed up the measurements via new X-ray sources and detectors, but the data analysis is still lacking.
We have elaborated a software based on pattern recognition and machine learning applied to X-ray data which will has speeded up the analysis to less than 1 minute.
We have used X-ray 2 dimensional data, ie reciprocal space maps (RSM) which are a detailed representation of the crystalline structure of the area probed, and extract the material parameters relevant to different processes.
This work is very important for new process development and new materials characterization.
The deep tech innovation is the creation of an innovative software that integrates complex theoretical X-ray model simulations, machine learning and pattern recognition into the data analysis software, in order to speed up the data analysis.
The major advantage of this approach is the promise of extracting the parameters of interests directly from images without the use of time-consuming physical model simulations.
We have used convolutional neural network (CNN) to analyze the distribution of diffraction intensity in the reciprocal space.
A reciprocal-space map (RSM) refers to a graphical representation of diffraction data in reciprocal space.
It consists of a set of spots that correspond to the intensity and position of the diffracted beams, providing information about the crystal structure and orientation.
By analyzing the reciprocal space map, researchers can determine the symmetry, crystal lattice parameters, and potential defects in the crystal sample.
Reciprocal-space map patterns can be recognized by using a convolutional neural network (CNN) to analyze the distribution of diffraction intensity in the reciprocal space.
The CNN can be trained on a dataset of known pattern types to learn the features that distinguish different patterns.
By passing the reciprocal-space map images through the CNN, the network can classify the patterns based on their unique features and predict the type of pattern present in the image.
The major advantage of this approach is the promise of extracting the parameters of interests directly from images without the use of time-consuming physical model simulations.
We have used convolutional neural network (CNN) to analyze the distribution of diffraction intensity in the reciprocal space.
A reciprocal-space map (RSM) refers to a graphical representation of diffraction data in reciprocal space.
It consists of a set of spots that correspond to the intensity and position of the diffracted beams, providing information about the crystal structure and orientation.
By analyzing the reciprocal space map, researchers can determine the symmetry, crystal lattice parameters, and potential defects in the crystal sample.
Reciprocal-space map patterns can be recognized by using a convolutional neural network (CNN) to analyze the distribution of diffraction intensity in the reciprocal space.
The CNN can be trained on a dataset of known pattern types to learn the features that distinguish different patterns.
By passing the reciprocal-space map images through the CNN, the network can classify the patterns based on their unique features and predict the type of pattern present in the image.
The important results is that we have used sophisticated X-ray theoretical models combined to to machine learning and image recognition, in order to speed up the data analysis acquired with X-ray instruments.
This is a very important step for the use of X-ray metrology for semiconductor processing and manufacturing. We would like to explore the possibility of applying for a patent for this software before full commercialization.
The goal of this project was to develop a next generation software solution based on machine learning to facilitate and speed up the data analysis of X-ray high resolution measurements.
X-ray Metrology will be a critical technology for the next generation of semiconductor device production.
The devices are getting smaller, and the material stacks are becoming more complex, therefore the need of more sophisticated metrologies to better understand the material properties and the manufacturing.
Differently from optical metrology, X-rays probe the material at the atomic scale and can penetrate from a few nanometers to millimeters according to the wavelength used , which matches the low dimensions of semiconductor devices.
One of the greatest challenges for the larger adoption of X-ray metrology into semiconductor wafer fabs has been the relatively slow measurements time (minutes) in comparison to optical metrology (seconds), and the complexity of the data analysis.
A lot of progress has been achieved to speed up the measurements via new X-ray sources and detectors, but the data analysis is still lacking.
We have elaborated a software based on pattern recognition and machine learning applied to X-ray data which will has speeded up the analysis to less than 1 minute.
We have used X-ray 2-dimensional data, ie reciprocal space maps (RSM) which are a detailed representation of the crystalline structure of the area probed, and extract the material parameters relevant to different processes.
This work is very important for new process development and new materials characterization
This is a very important step for the use of X-ray metrology for semiconductor processing and manufacturing. We would like to explore the possibility of applying for a patent for this software before full commercialization.
The goal of this project was to develop a next generation software solution based on machine learning to facilitate and speed up the data analysis of X-ray high resolution measurements.
X-ray Metrology will be a critical technology for the next generation of semiconductor device production.
The devices are getting smaller, and the material stacks are becoming more complex, therefore the need of more sophisticated metrologies to better understand the material properties and the manufacturing.
Differently from optical metrology, X-rays probe the material at the atomic scale and can penetrate from a few nanometers to millimeters according to the wavelength used , which matches the low dimensions of semiconductor devices.
One of the greatest challenges for the larger adoption of X-ray metrology into semiconductor wafer fabs has been the relatively slow measurements time (minutes) in comparison to optical metrology (seconds), and the complexity of the data analysis.
A lot of progress has been achieved to speed up the measurements via new X-ray sources and detectors, but the data analysis is still lacking.
We have elaborated a software based on pattern recognition and machine learning applied to X-ray data which will has speeded up the analysis to less than 1 minute.
We have used X-ray 2-dimensional data, ie reciprocal space maps (RSM) which are a detailed representation of the crystalline structure of the area probed, and extract the material parameters relevant to different processes.
This work is very important for new process development and new materials characterization