• Technical Name
  • AI-Semi : A platform for Intelligent Optimization in Semiconductor Device Designs, Manufacturing, and Performance
  • Operator
  • National Chiao Tung University
  • Booth
  • Online display only
  • Contact
  • 吳添立
  • Email
  • tlwu@nctu.edu.tw
Technical Description As the semiconductor world is moving fast to keep the pace up with the technological demand, new innovations, including the geometrical approaches, physical designs, new materials, are coming up in the semiconductor world. However, device modeling and optimization processes in the advanced semiconductor technologies become challenging and time-consuming due to the complex device designs, complicated processes and material properties.
In order to address the challenges mentioned above, machine learning-based neural network could provide the advantages in terms of the fast proceeding and high accuracy.
In our platform, we have developed the machine learning engines that can accelerate the semiconductor device designs and optimizations. These machine learning engines can consider the varied input, such as materials, device designs and process, and varied output, such as VTH, subthreshold slope, current, breakdown voltage, and reliability performance.
Scientific Breakthrough “AI-Semi”can accelerate the optimization of the semiconductor devices toward the targeted performance.
1. Machine learning engine to accelerate the device modeling and designs
Our platform can model the electrical characteristics and reliability in advanced technologies, which are the challenges for the physics-based models. In addition, based on the developed machine learning engines, our platform can predict the electrical characteristics.
2. Autoencoder to reconstruct the device parameters based on the target performance
Considering the following targeted performance: VTH(V)=.014V, SS(mV/dec)=98, ID_Max(mA/mm)=0.025 and ID_OFF(mA/mm)=3.6×10-12, the required device designs, which are (tAlGaN, Lg, Recessed depth)=(10.3nm, 1.47μm, 4.9nm), can be obtained by using the autoencoder.
Industrial Applicability The conventional method of optimizing the performance is a time-consuming process that requires a lot of resources. For the devices based on the novel materials and designs, the lack of accurate physics-based models results in a challenge to predict the device characteristics, leading to the uncertainties of product performance. Our platform can be used in a variety of complicated semiconductor technologies to achieve an accurate analysis of the key designs and processing parameters, further efficiently optimizing device characteristics and reliability. This platform can significantly reduce the cost, time and resources in the post-Moore era, providing the useful assistance in More Moore (such as sub-10nm beyond CMOS) and More than Moore (such as power semiconductor devices) technologies.