In this paper, we demonstrated the possibility of predicting full transistor current-voltage (IV) and capacitance-voltage (CV) curves using machines trained by Technology Computer-Aided Design (TCAD) generated data. 3D FinFET IDVG and CGVG predictions are used as examples. The machine is constructed through manifold learning using Autoencoder (AE) to extract the latent variables which are then correlated to physical parameters through 3rd-order polynomial regression. No device physics domain expertise is required in the machine learning process because there is no need to extract device metrics such as transconductance (gm) or Drain-Induced-Barrier-Lowering (DIBL) from the TCAD training data. We show that the machine can predict not just the full IV/CV curves but also gm (1st derivative quantity) and DIBL (extracted from two machines trained by different data). Good results can be obtained even with < 50 training data. Our work shows that manifold learning is possible in device IV and CV to capture the complex physics and, thus, it is expected that it is possible to predict the IV/CV of novel devices using limited experimental data before the underlying physics is well-understood.
This work presents the temperature-dependent forward conduction and reverse blocking characteristics of a high-voltage vertical Ga2O3 power rectifier from 300 K to 600 K. Vertical β-Ga2O3 Schottky barrier diodes (SBDs) were fabricated with a bevel-field-plated edge termination, where a beveled sidewall was implemented in both the mesa and the field plate oxide. The Schottky barrier height was found to increase from 1.2 eV to 1.3 eV as the temperature increases from 300 K to 600 K, indicating the existence of barrier height inhomogeneity. The net donor concentration in the drift region shows little dependence on the temperature. The reverse leakage current up to 500 V was found to be limited by both the thermionic-field electron injection at the Schottky contact and the electron hopping via the defect states in the depletion region. At 300–500 K, the leakage is first limited by the electron injection at low voltages and then by the hopping in depleted Ga2O3 at high voltages. At temperatures above 500 K, the thermionic field emission limits the device leakage over the entire voltage range up to 500 V. Compared to the state-of-the-art SiC and GaN SBDs when blocking a similar voltage, our vertical Ga2O3 SBDs are capable of operating at significantly higher temperatures and show a smaller leakage current increase with temperature. This shows the great potential of Ga2O3 SBDs for high-temperature and high-voltage power applications.
18 September 2020 in IEEE Journal of the Electron Devices Society
This work, for the first time, experimentally demonstrates a TCAD-Machine Learning (TCAD-ML) framework to assist the analysis of device-to-device variation and operating (ambient) temperature without the need of physical quantities extraction. The ML algorithm used in this work is the Principal Component Analysis (PCA) followed by third order polynomial regression. After calibrated to limited ’expensive’ experimental data’, low cost’ TCAD simulation is used to generate a large amount of device data to train the ML model. The ML was then used to identify the root cause of device variation and operating temperature from any given experimental current-voltage (I-V) characteristics. We applied this framework to study the ultra-wide-bandgap gallium oxide (Ga2O3) Schottky barrier diode (SBD), an emerging device technology that holds great promise for temperature sensing, RF, and power applications in harsh environments. After calibration, over 150,000 electrothermal TCAD simulations are performed with random variation of physical parameters (anode effective work function, drift layer doping, and drift layer thickness) and operating temperature. An ML model is trained using these TCAD data and we found 1,000-10,000 TCAD data can train an accurate machine. We show that without physical quantities extraction, performing PCA is essential for the TCAD trained ML model to be applicable to analyze experimental characteristics. The physical parameters and temperatures predicted by the ML model show good agreement with experimental analysis. Our TCAD-ML framework shows great promise to accelerate the development of new device technologies with a significantly more efficient process of material and device experimentation.
A machine learning (ML) model by combing two autoencoders and one linear regression model is proposed to avoid overfitting and to improve the accuracy of Technology Computer-Aided Design (TCAD)-augmented ML for semiconductor structural variation identification and inverse design, without using domain expertise. TCAD-augmented ML utilizes TCAD simulations to generate sufficient data for ML model development when experimental data are inadequate. The ML model can then be used to identify semiconductor structural variation for given experimental electrical measurements. In this study, the variation of layer thicknesses in the p-i-n diode is used as a demonstration. An ML model is developed to predict the diode layer thicknesses based on a given Current-Voltage (IV) curve. Although the variations of interest can be incorporated easily in TCAD simulations to generate ML training data, the TCAD-augmented ML model generally is overfitted and cannot predict the variations in experiment well due to hidden variables which also alters the IV curves. We show that by using an autoencoder, this problem can be solved. To verify the effectiveness, another set of TCAD simulation data is generated with hidden variables (dopant concentration variation) to emulate experimental data. Testing on the second set of data shows that the proposed model can avoid overfitting and has up to 15 times improvement in accuracy in thickness prediction. Moreover, this model is used successfully to perform inverse design and can capture an underlying physics that cannot be described by a simple physical parameter.
29 April 2021 in IEEE Transactions on Electron Devices
In this article, using experimental data, we demonstrate that the technology computer-aided design (TCAD) is a very cost-effective tool to generate the data to build machine learning (ML) models for semiconductor device and process characterization. Characterization of the emerging ultra wide bandgap gallium oxide (Ga₂O₃) Schottky barrier diode (SBD) is used as an example. Machines are trained by using only TCAD I-V's and then used to deduce the effective Schottky metal work function (WF) and ambient temperature (T) of an experimental SBD based on its I-V. Besides noise, the experimental device also suffers from relatively large variations in drift layer thickness and doping concentrations. Both ML models with domain expertise (WDE) and without domain expertise (WoDE) are studied and compared. The ML model WDE requires the use of device knowledge to extract relevant features (e.g., subthreshold slope and turn-on voltage) for ML. The ML model WoDE obviates such a requirement and can be extended to cases where domain expertise is difficult to apply. Denoising autoencoder is used in the WoDE case. We showed that with only 500 TCAD I-V's, we can train machines WDE and WoDE that can deduce the experimental device WF and T reasonably well. In particular, the ML model WoDE has an acceptable prediction accuracy despite the noise and additional variations in the experimental device.
After learning basic quantum computing concepts, it is desirable to reinforce the learning using an important and relatively complex algorithm through which students can observe and appreciate how qubits evolve and interact with each other. Harrow-Hassidim-Lloyd (HHL) quantum algorithm, which can solve linear system problems with exponential speed-up over the classical method and is the basis of many important quantum computing algorithms, is used to serve this purpose. The HHL algorithm is explained analytically followed by a 4-qubit numerical example in bra-ket notation. Matlab code corresponding to the numerical example is available for students to gain a deeper understanding of the HHL algorithm from a pure matrix point of view. A quantum circuit programmed using qiskit is also provided for real hardware execution in IBM quantum computers. After going through the material, students are expected to have a better appreciation of the concepts such as basis transformation, bra-ket and matrix representations, superposition, entanglement, controlled operations, measurement, quantum Fourier transformation, quantum phase estimation, and quantum programming. To help readers review these basic concepts, brief explanations augmented by the HHL numerical examples in the main text are provided in the Appendix.