In a research reported as the world's first, Australian researchers and their partners successfully verified a quantum machine learning model that is specially used for semiconductor manufacturing and tested based on experimental data, a breakth...
In a research reported as the world's first, Australian researchers and their partners successfully verified a quantum machine learning model that is specially used for semiconductor manufacturing and tested based on experimental data, a breakthrough that could reshape future chip design.
Researcher from CSIRO, a national scientific organization in Australia, said: "We have proven that quantum machine learning has surpassed traditional artificial intelligence in modeling Ohmic Contact resistance, a process step that is important but difficult to model in modern semiconductor equipment manufacturing."
This research topic is "Quantum Kernel Learning for Small Dataset Modeling in Semiconductor Fabrication: Application to Ohmic Contact". Researchers point out that modeling complex semiconductor manufacturing processes (such as OEM contact formation) face challenges with high-data space and limited experimental data. Although the traditional machine learning (CML) approach has been successful in many domains, its performance declines in small-sample and nonlinear situations.
The research team used only 159 experimental nitride high-electronic mobility crystal tube (GaN HEMT) samples to develop a quantum nuclear-coordinated reciprocating device (QKAR) that combines a clean Pauli-Z feature map with a trainable quantum nuclear-coordinated layer. All models, including seven basic CML regressors, are evaluated under a unified pre-processing pipeline based on principal component analysis (PCA) to ensure fair comparison. QKAR continues to surpass the traditional base on multiple indicators (such as mean absolute error, mean square error, root mean square error) and achieves an average absolute error of 0.338Ω·mm in the experimental data verification.
▲ GaN HEMT OEM contact formation modeling process diagram based on quantum machine learning. (Source:Advance Science)
Researchers stress that although traditional machine learning methods have been widely explored to enhance manufacturing process modeling, the limitations of these methods are still a major challenge for relying on large data sets to effectively generalize CML models. Semiconductor manufacturing involves complex nonlinear relationships between process parameters such as annealing temperature, time and atmosphere conditions, which further makes modeling using traditional machine learning techniques even more difficult.
This study was conducted by international teams from Beijing University, Songshan Lake Materials Experimental Office and City University of Hong Kong. These institutions provide a set of manufacturing data for CSIRO to train quantum machine learning models. With the continuous improvement of quantum processors in accuracy and scale, the application of quantum machine learning models in actual semiconductor workflows will become increasingly feasible.
The results of the research have been published in the journal Advance Science.
Quantum Machine Learning Shines in Semiconductor Chip Design
