Search

This paper presents key building blocks of a monolithic GaN half-bridge solution: 1) a binary-weighted digitally-controlled segmented gate-driver, offering slew-rate control; 2) a high-voltage floating level-shifter with glitch prevention; and 3) a monolithic half-bridge with integrated fixed-strength gate-drivers. Together, they facilitate on-chip active gate-driving, improving the reliability of GaN power ICs. The blocks were fabricated on separate dies using imec’s 200 V GaN-on-SOI process. A controllable dv DS /dt from 68 V/ns to 112 V/ns at room temperature is achieved using the segmented gate-driver. An oscillation-free v DS switching of the half-bridge at 200 V and 5 MHz is demonstrated through experimental results.

In this study, the transient threshold voltage behavior of p-GaN capped AlGaN/GaN high-electron-mobility transistors (HEMTs) is evaluated, by monitoring the threshold voltage from $10 \mu \mathrm{s}$ up to 100 s during positive gate bias stress, as well as during recovery. It is found that the threshold voltage transients are highly dependent on the current distribution in the metal/p-GaN/AlGaN/GaN gate stack. Two different gate processes are evaluated in terms of ON-state gate current, gate conduction mechanism and threshold voltage stability. Devices with perimeter-dependent gate current show large threshold voltage shifts; devices in which the gate current is due to uniform thermionic (field) emission over the Schottky interface into the bulk p-GaN demonstrate stable behavior for $V_{g}\, \lt $5V.

This work presents an integrated GaN senseHEMT structure and segmented gate-driver implemented using ON Semiconductor’s 650-V mixed-signal GaN-on-Si process. The senseHEMT element with 2500:1 sensing ratio is embedded within the 650-V power device for near-lossless current sensing, and the totem-pole driver utilizing low-voltage (LV) e-HEMTs is monolithically integrated with the device. The proposed design is verified in a bridgeless totem-pole power-factor-correction converter operating in boundary conduction mode. The converter is controlled by an FPGA-based hysteretic current-mode controller to achieve sinusoidal current shaping. The controller uses the senseHEMT in conjunction with a current DAC and a high-speed comparator to sense the peak and valley currents through the power HEMT. A set of minimum-sized high-voltage (HV) GaN devices protect the LV current-loop circuitry from the high voltages in the powerstage. Experimental results show sinusoidally-shaped input current synchronized with the ac input voltage, demonstrating dynamic functionality of the senseHEMTbased current-control loop to achieve power-factor correction.

This paper proposes a charge-to-breakdown (Q BD ) approach for SiC/SiO 2 dielectric lifetime extraction. The current through the dielectric is shown to be a combination of Fowler Nordheim (FN) and thermally assisted tunneling (TAT). The former leads to positive charge trapping, the latter to negative charge storage in border traps near the SiC/SiO 2 interface. Both degradation mechanisms have a distinctly different failure distribution function. Time-to-fail extracted from constant field TDDB (standard industry technique) is unable to capture the effects of these different charge trapping mechanisms on the failure distribution function, in contrast to constant current Q BD stress. Hence, time-to-fail from Q BD stress leads to a more accurate and physics based lifetime prediction.

This paper reports the performance of GaN -based building blocks for monolithic integration. An integrated gate driver, low-voltage analog and synchronous digital circuits are fabricated in a 650-V GaN-on-Si process and measured. Calibrated GaN models are compared against simulated silicon designs to identify optimal integration criteria.

Smartphones can collect millions of data points from each of its users daily, contributing to a significant change in how the healthcare community approaches health monitoring. This paper provides a framework for how smartphone sensor data can be collected, cleaned, stored, and modeled to effectively predict human states as a step towards health monitoring. To develop robust contextual models, a three-week study was conducted to collect data through a mobile crowdsensing application named Sensus. In this study, participants used multiple sensing strategies, ranging from infrequent sampling to continuous sampling, to determine the effect each has on data integrity and battery life. For a future study, a dynamic data collection strategy was developed that uses a machine learning model trained on existing data collected from 220 participants to forecast when a smartphone will be active and trigger sensor sampling accordingly. Results of this study include 1) extraction of model features that deliver maximized data quality with minimized battery consumption as compared to pre-existing baseline models, 2) implementation of context-driven modeling of user smartphone data on user's contextual environment, and 3) customization of a time-series database for optimized data queries used in metadata visualizations. The adaptive sensing models produced could be used in future large population studies that efficiently examine patterns of behavior in multiple individuals over extended periods to identify disease indicators present in an average user’s daily life.

This paper reports a detailed analysis of the performance and stability of E-mode GaN HEMTs under soft and hard switching conditions. We developed a novel on-wafer setup that controls the overlapping between the gate and drain pulses and, simultaneously, senses the current at the source. This allowed us to plot the instantaneous power, the switching I-V locus and evaluate the switching losses on wafer level. The dynamic on-resistance RDS,ON is evaluated during hard and soft switching at room temperature and high temperature. The results demonstrate that (i) the properties of the buffer impact on dynamic-Ron, in soft- and hard-switching; devices with non-optimized buffer have stronger dynamic-Ron under hard switching; (ii) Ron-increase under hard switching is not strongly influenced by switching (power) losses; (iii) results indicate that a higher dynamic-Ron in hard switching is correlated to a wider switching locus. The proposed testing approach can be used as a screening tool to evaluate – in one single measurement – the switching losses, the dynamic-Ron and the impact of hard-switching on wafer level.

This paper concerns a machine learning approach for the inverse quantification of set-theoretical uncertainty. Inverse uncertainty quantification (e.g., using Bayesian or interval methodologies) is usually obtained following a process where a distance metric between a set of predicted and measured model responses is iteratively minimized. Consequently, the corresponding computational effort is large and usually unpredictable, leading to an intractable situation for real-time applications (e.g., as is commonly encountered in process control problems). To achieve a real-time solution to this inverse problem, machine learning is applied to train a deep neural network, consisting of multilayer auto-encoders and a shallow neural network, by means of a numerically generated data set that captures typical uncertainty in the model parameters. The method is applied to the challenging DLR AIRMOD problem and it is shown that the obtained accuracy is comparable to existing methods in literature, albeit at a fraction of their computational cost.

We propose an electrical evaluation technique named differential charge pumping ( $\Delta-\mathbf{CP}$ ) to extract information on the interface states (D it ) at the SiC/SiO 2 interface. This technique allows to obtain information on (1) the physical location of the D it ; (2) the energetic spread of the D it in the bandgap; (3) if the D it are donor or acceptor type. First results indicate contributions from both the “S/D” region and the channel regions to the overall CP signal. Surprisingly a substantial contribution of the “S/D” regions of the transistors to the overall CP signal is observed. In the channel region, both donors and acceptor type states are identified, the majority being acceptor type.

This paper proposes a physical-statistical approach to AIGaN/GaN HEMT reliability applied to forward gate bias stress. The Ig- V g characteristics are fitted to a physical conduction model as a function of temperature and field. From the model, the acceleration factors for field and temperature are extracted, and used to predict the lifetime for given use condition. All time-to-fail data are used in the lifetime extrapolation in a self consistent manner. The data follow a LogNormal rather than a Weibull distribution.

We propose a technique to evaluate the time-dependence of the threshold voltage instabilities in GaN-based normally-off transistors under positive gate bias. More specifically: (i) for the first time we experimentally evaluate the V th shift in a wide time window (from $10\ \ \boldsymbol{\mu}\mathbf{s}$ to 100 s), as a function of temperature. (ii) We study the existence of two dominant trapping processes, having different time-kinetics. (iii) a first process, occurring in the initial $100\ \ \boldsymbol{\mu}\mathbf{s}$ , leading to a positive threshold voltage shift, and ascribed to the injection of electrons from the 2DEG towards the AlGaN barrier; (iv) a second process, occurring only for $\mathbf{V}_{\mathbf{GS}}\!>\!5$ V, leading to a negative shift in threshold voltage. This latter process is not thermally activated and strongly dependent on gate leakage.

Probabilistic Topic Models are widely applied in many NLP-related tasks due to their effective use of unlabeled data to capture variable dependencies. Analytical solutions for Bayesian inference of such models, however, are usually intractable, hindering the proposition of highly expressive text models. In this scenario, Variational Auto-Encoders (VAEs), where an inference network (the encoder) is used to approximate the posterior distribution, became a promising alternative for inferring latent topic distributions of text documents. These models, however, also pose new challenges such as the requirement of continuous and reparameterizable distributions which may not fit so well the true latent topic distributions. Moreover, inference networks are prone to component collapsing, impairing the collection of coherent topics. To overcome these problems, we propose two new text topic models based on the categorical distribution Gumbel-Softmax (GSDTM) and on mixtures of Logistic-Normal distributions (LMDTM). We also provide a study on the impact of different modeling choices on the generated topics, observing a trade-off between topic coherence and document reconstruction. Through experiments using two reference datasets, we show that GSDTM largely outperforms previous state-of-the-art baselines when considering three different evaluation metrics.

Enhancement mode 650V rated p-GaN gate HEMTs are fabricated on 200 mm p+ Si substrates by using an industrial, Au-free process. The devices show true e-mode performance, with a high V t of 2.8 V, low off-state leakage current and are dynamic R DS-ON free over the complete V DS and temperature range. High temperature reverse bias (HTRB) testing is done on-wafer and after packaging. For the first time, 650 V e-mode power HEMTs realized on 200 mm Si substrates, show industry ready device performance and pass 1008 hour reliability testing, at V GS =0 V, V DS =650 V.

This paper concerns the comparison of two inverse methods for the quantification of uncertain model parameters, based on experimentally obtained measurement data of the model's responses. Specifically, Bayesian inference is compared to a novel method for the quantification of multivariate interval uncertainty. The comparison is made by applying both methods to the AIRMOD measurement data set, and comparing their results critically in terms of obtained information and computational expense. Since computational cost of the application of both methods to high-dimensional problems and realistic numerical models can become intractable, an Artificial Neural Network surrogate is used for both methods. The application of this ANN proves to limit the computational cost to a large extent, even taking the generation of the training dataset into account. Concerning the comparison of both methods, it is found that the results of the Bayesian identification provide less over-conservative bounds on the uncertainty in the responses of the AIRMOD model.

Through optimization of the GaN buffer structure, 650V rated AlGaN/GaN power devices with negative dynamic R on are obtained. By judicious tuning of the resistivity of the [C]-doped and unintentionally-doped (UID) layers (i.e. the charge transport properties in these layers), positive rather than the deleterious negative charged trapping can be achieved. Long term reliability testing shows that the positive charge can be retained for days.

This paper reports on the trapping mechanism of GaN-based metal-insulator-semiconductor (MIS) high electron mobility transistors (HEMTs) designed to work in a cascode configuration. We defined a novel stress protocol (High Temperature, Source Current, HTSC) to investigate the degradation processes induced by semi-on state operation. We compare the results of HTSC with those obtained by the standard HTRB (high temperature reverse bias), with the aim of identifying different impact on the RON variation. While HTRB stress results in a strong negative bias/temperature instability (NBTI), under HTSC conditions no significant Vth shift is observed. This result is ascribed to the fact that under HTSC conditions the gate-source voltage difference is significantly smaller than under HTRB, thus having less impact on Vth stability. The technique described in this paper is useful to test the Vth stability of normally-on devices used in cascode configuration.