Researchers at the Australian CSIO have achieved the first global demonstration of quantum machine learning in the semiconductor fabric. The quantum enlarged model can improve traditional AI methods and create a microchup designed. The team made an important – but difficult to predict – focused on modeling – called “oh -utmost contact” resistance, which shows that the current flow where the metal semi -conductor meets.
He analyzed 159 experimental patterns of modern gallem nitrate (GAN) transistor (known for high strength/high frequency performance). By combining the quantum processing layer with the final classic reactionary phase, the model pulled out the subtle samples that were missing from the traditional point of view.
Dealing with a difficult design problem
According to StudyCSIRO researchers first encoded many fabric variables (such as gas mixture and enlicing Times) and used the principal component analysis (PCA) to reduce the 37 most important parameters. Professor Mohammad Usman – who guides this study – explained that he did this because “quantum computers that we currently have very limited abilities”.
On the contrary, the classic machine can struggle learning when data is lacking or relationships are non -regional. By focusing on these important variables, the team made the problem for today’s quantum hardware.
A quantum kernel view
To model the data, the team created a QKar architecture with a custom quantum kernel. The five key parameters of each sample were mapped at five cobit quantum estate (using the polyz feature map), which enabled the quantum kernel layer to capture complex communication.
The output of the quantum layer was then fed in a standard learning algorithm, which indicated which manufacturing parameters are the most important. As Usman says, it indicates a joint quantum -class model that takes fabricated steps to keep the device’s performance in view of the maximum device.
In the tests, the QKAR model defeated seven top classical algorithms on the same task. It only needed five pills, which is possible on today’s quantum machines. CSIRO’s Dr. Zheng Wang notes that samples of quantum procedures can reduce classic models high -dimensional, small data problems.
To verify the approach, the team created new GAN devices using the model guidance. These chips performed better. It confirmed that the design created by the quantum has been beyond its training data.