Epitaxial lateral overgrowth (ELO) is an established technique for the reduction of the high threading dislocation density of GaN grown on sapphire substrates. ELO is typically carried out via a selective area growth process with stripes of SiO2 as a mask. While the benefits of ELO were instrumental in the development of early GaN optoelectronics, the phenomenon known as “wing tilt” is a longstanding drawback. So-named due to the crystallographic tilt induced in the “wing” regions above the SiO2 mask, the effect is associated with a tilt boundary at the coalescence boundary. While various approaches have been suggested to mitigate the impact of wing tilt, there is not currently a scheme to eliminate the effect. In this study, we demonstrate that wing tilt is a direct result of the use of the SiO2 mask, and that the tilt occurs even before coalescence. We also demonstrate GaN layers free of wing tilt using a mask-free facet-controlled ELO (FACELO) scheme. For this investigation, ammonothermally grown bulk GaN substrates were used. One template was prepared for traditional ELO with SiO2 stripes of 2 μm separated by windows of 2 μm for a period of 4 μm; another template was prepared with a mask-free pattern by etching trenches of 2 μm width and depth, leaving behind 2 μm stripes. Both templates had stripes oriented along the m-direction of GaN. As the first step in the ELO growth scheme, these templates were grown at a low temperature (950 °C) to promote the evolution of semipolar facets, resulting in triangular cross-sections over the stripes. The second growth step to achieve c-plane coalescence was then performed at a higher temperature (1100 °C). The samples were characterized at both steps via SEM and XRD, as well as TEM after coalescence. The (002) ω-rocking curve of the SiO2-grown triangles shows symmetric satellite peaks about the main peak characteristic of wing tilt. Additionally, the FWHM of the peak has increased from 40 arcsec for the substrate to 150 arcsec for the regrown structure. This result shows that even before coalescence, the growth of GaN over SiO2 results in a change in crystalline quality. After coalescence, the FWHM of the peak increases to 180 arcsec, a value suggesting a threading dislocation density in the high 108 cm-2 range. By contrast, the maskless sample shows no satellite peaks in the (002) ω-rocking curve, and at both steps the FWHM of the original template is maintained. The findings of the XRD are confirmed via TEM. In epitaxial layers above the SiO2 patterned sample, a significant density of dislocations can be seen. By contrast, the maskless sample showed no visible dislocations. We therefore conclude that wing tilt and its associated dislocations are caused by the presence of the SiO2 mask, and that by use of a maskless etched template, these deformations and defects can be avoided.

Wing Tilt in ELO-Grown GaN

The mid-infrared wavelength range (MIR: 3-30 µm) is of significant technological importance. Many of its applications are well-served by leveraging techniques capable of enhancing light-matter interaction. Phononic systems, which leverage excitations of the crystal lattice coupled to electromagnetic modes, offer one potential approach. The negative real permittivity of a phononic material, in the material’s Reststrahlen Band, results in an optical response similar to that of plasmonic materials, though with lower loss (small imaginary permittivity). In the MIR, strongly polar AlN, with a high-energy longitudinal optical (LO) phonon at a free-space wavelength of 11.2 µm, is an intriguing phononic system, with negative permittivity extending from LO phonon wavelength to the transverse optical (TO) phonon wavelength of 15.9 µm[1]. In this work, we investigate periodic arrays of AlN-capped nanowires supporting a variety of unique phononic modes. We demonstrate localized surface phonon resonances (LSPhR) in the AlN nanostructures, in the frequency range bounded by AlN’s LO and TO phonon energies. We also demonstrate that the nano-pillar growth regime allows for normal incidence coupling to the Berreman mode which is observed in thin films at epsilon-near-zero (ENZ) permittivity (at LO phonon energy) for light incident at large grazing angles. The nanowires are grown by plasma-assisted MBE[2] on a patterned silicon wafer, which serves as the template for GaN nanowire growth, upon which AlN caps are grown. On a silicon wafer, after a series of buffer layer growths, a 70 nm Si3N4 growth mask is deposited on the top GaN buffer layer, and the growth mask is then patterned (by e-beam lithography) and etched to provide apertures to the underlying GaN with 1 µm period and varying diameters. GaN nanowires 1.8 μm in height are grown in the apertures, followed by an AlN cap of 210 nm, above the GaN nanowires. Periodic arrays of AlN-capped GaN nanowires were grown with varying aperture dimensions, resulting in nanowires with diameters of ∼ 210, 300 and 390 nm, During the growth of the AlN cap, a thin cylindrical shell of AlN with thickness in few 10s of nms (decreasing thickness from top to bottom) is deposited on the sidewalls of the GaN nanopillar. The result of the patterned growth are subwavelength AlN phononic resonators that effectively ‘float’ above the substrate, supported by the dielectric GaN nanowire bases. We measure near normal incidence reflectance (±3.5°) of the nanowire arrays using a nitrogen-purged Bruker vertex 70 FTIR with an output beam collimated and reduced to ~ 6 mm through an external double-aperture. The reflectance data is compared to the reflection simulated with a corresponding TM-polarized 2-D model in COMSOL assuming a 20 nm thick uniform AlN shell around GaN nanopillar. In this work, we study representative data from the nanowires with diameter of 390 nm. The reflectance spectra show three prominent resonant dips. The longest wavelength features at 13.3 μm wavelength can be thought as a dipole mode at the top and bottom corner GaN/AlN interface. The next longest wavelength features at wavelengths λ = 12.3 μm corresponds to the LSPhR of the AlN caps that are confirmed via simulation of the nanowire structures. The LSPhR spectrally red-shifts with increasing nanowire diameters. Even more interesting is a sharp reflectance dip in AlN nanowires at wavelength, λ = 11.2 μm (AlN LO phonon energy). The fields at λ = 11.2 μm are calculated with 2-D COMSOL model, which indicate a strong field confinement in the AlN shell surrounding the GaN nanowire. While such a resonance would not typically be seen in a planar AlN film (or even a patterned AlN film) at normal incidence, the vertical orientation of the thin (~20 nm) AlN shell offers strong coupling to the Berreman mode. 1. L. Nordin et al. ... D. Wasserman. Applied Physics Letters, 111(9):091105, 2017. 2. M. Brubaker et al. … K. Bertness, Nanotechnology 30, 234001, 2019.

Localized Phonon Polariton Modes in GaN/AlN Nanowiresx

We present Q-functionals, an alternative architecture for continuous control deep reinforcement learning. Instead of returning a single value for a state-action pair, our network transforms a state into a function that can be rapidly evaluated in parallel for many actions, allowing us to efficiently choose high-value actions through sampling. This contrasts with the typical architecture of off-policy continuous control, where a policy network is trained for the sole purpose of selecting actions from the Q-function. We represent our action-dependent Q-function as a weighted sum of basis functions (Fourier, Polynomial, etc) over the action space, where the weights are state-dependent and output by the Q-functional network. Fast sampling makes practical a variety of techniques that require Monte-Carlo integration over Q-functions, and enables action-selection strategies besides simple value-maximization. We characterize our framework, describe various implementations of Q-functionals, and demonstrate strong performance on a suite of continuous control tasks.

Q-Functionals for Value-Based Continuous Control

Multi-agent reinforcement learning (MARL) suffers from the non-stationarity problem, which is the ever-changing targets at every iteration when multiple agents update their policies at the same time. Starting from first principle, in this paper, we manage to solve the non-stationarity problem by proposing bidirectional action-dependent Q-learning (ACE). Central to the development of ACE is the sequential decision making process wherein only one agent is allowed to take action at one time. Within this process, each agent maximizes its value function given the actions taken by the preceding agents at the inference stage. In the learning phase, each agent minimizes the TD error that is dependent on how the subsequent agents have reacted to their chosen action. Given the design of bidirectional dependency, ACE effectively turns a multiagent MDP into a single-agent MDP. We implement the ACE framework by identifying the proper network representation to formulate the action dependency, so that the sequential decision process is computed implicitly in one forward pass. To validate ACE, we compare it with strong baselines on two MARL benchmarks. Empirical experiments demonstrate that ACE outperforms the state-of-the-art algorithms on Google Research Football and StarCraft Multi-Agent Challenge by a large margin. In particular, on SMAC tasks, ACE achieves 100% success rate on almost all the hard and super hard maps. We further study extensive research problems regarding ACE, including extension, generalization and practicability. Code is made available to facilitate further research.

ACE: Cooperative Multi-agent Q-learning with Bidirectional Action-Dependency

Without an assigned task, a suitable intrinsic objective for an agent is to explore the environment efficiently. However, the pursuit of exploration will inevitably bring more safety risks.
An under-explored aspect of reinforcement learning is how to achieve safe efficient exploration when the task is unknown.
In this paper, we propose a practical Constrained Entropy Maximization (CEM) algorithm to solve task-agnostic safe exploration problems, which naturally require a finite horizon and undiscounted constraints on safety costs.
The CEM algorithm aims to learn a policy that maximizes the state entropy under the premise of safety.
To avoid approximating the state density in complex domains, CEM leverages a $k$-nearest neighbor entropy estimator to evaluate the efficiency of exploration.
In terms of safety, CEM minimizes the safety costs, and adaptively trades off safety and exploration based on the current constraint satisfaction. We empirically show that CEM allows learning a safe exploration policy in complex continuous-control domains, and the learned policy benefits downstream tasks in safety and sample efficiency.

CEM: Constrained Entropy Maximization for Task-Agnostic Safe Exploration

Conventional reinforcement learning (RL) needs an environment to collect fresh data, which is impractical when online interactions are costly. Offline RL provides an alternative solution by directly learning from the previously collected dataset. However, it will yield unsatisfactory performance if the quality of the offline datasets is poor. In this paper, we consider an offline-to-online setting where the agent is first learned from the offline dataset and then trained online, and propose a framework called Adaptive Policy Learning for effectively taking advantage of offline and online data. Specifically, we explicitly consider the difference between the online and offline data and apply an adaptive update scheme accordingly, that is, a pessimistic update strategy for the offline dataset and an optimistic/greedy update scheme for the online dataset. Such a simple and effective method provides a way to mix the offline and online RL and achieve the best of both worlds. We further provide two detailed algorithms for implementing the framework through embedding value or policy-based RL algorithms into it. Finally, we conduct extensive experiments on popular continuous control tasks, and results show that our algorithm can learn the expert policy with high sample efficiency even when the quality of offline dataset is poor, e.g., random dataset.

Adaptive Policy Learning for Offline-to-Online Reinforcement Learning

Social media is a breeding ground for threat narratives and related conspiracy theories. In these, an outside group threatens the integrity of an inside group, leading to the emergence of sharply defined group identities: Insiders -- agents with whom the authors identify and Outsiders -- agents who threaten the insiders. Inferring the members of these groups constitutes a challenging new NLP task: (i) Information is distributed over many poorly-constructed posts; (ii) Threats and threat agents are highly contextual, with the same post potentially having multiple agents assigned to membership in either group; (iii) An agent's identity is often implicit and transitive; and (iv) Phrases used to imply Outsider status often do not follow common negative sentiment patterns. 

To address these challenges, in this talk, we define a novel Insider-Outsider classification task. Because we are not aware of any appropriate existing datasets or attendant models, we introduce a labeled dataset (CT5K) and design a model (NP2IO) to address this task. NP2IO leverages pretrained language modeling to classify Insiders and Outsiders. We show that NP2IO is robust, generalizing to noun phrases not seen during training, and exceeding the performance of non-trivial baseline models by 20%.

Which side are you on? Insider-Outsider classification in conspiracy-theoretic social media

In this paper we report about the relationships between a multi-preferential semantics for defeasible description logics and a deep neural network model. Weighted knowledge bases for description logics are considered under a “concept-wise" preferential semantics, which is further extended to fuzzy interpretations and exploited to provide a preferential interpretation of Multilayer Perceptrons.


Multilayer Perceptrons as Weighted Conditional Knowledge Bases: an Overview

In this work we evaluate the You Only Look Once (YOLOv3) architecture for real-time detection of insect-pests in soybean. Soybean crop images were collected on different days, locations and weather conditions, between the phenological stages R1 to R6, considered of high occurrence of soybean pests. For training and testing the neural network, we used a 5 fold cross-validation considering four metrics to evaluate the classification results: precision, recall, F-score and accuracy; and three metrics to evaluate the detection results: mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination R². The experimental results showed that the YOLOv3 architecture trained with batch size 32 leads to higher classification and detection rates compared to batch sizes 4 and 16. The results indicate that the evaluated architecture can support experts and farmers in monitoring pest control action levels in soybean fields.

Performance analysis of YOLOv3 for real-time detection of pests in soybeans

This paper studies switching behavior and dynamic on-state characteristics of lateral 10 mm depletion-mode β-Ga2O3 met-al-oxide-semiconductor field-effect transistors (MOSFETs). Improvements in device technology and fabrication enable hard-switching operation with an increase of the maximum blocking voltage to 400 V at 0.26 A load current in a double-pulse test configuration. Compared to static measurements, dispersion effects lead to an increased on-resistance in hard-switched operation, which becomes especially pronounced at blocking voltages above 200 V. The extend of dynamic Ron-increase further depends on several operation conditions such as switching speed, on-state duration and load current.

Switching Behavior and Dynamic On-Resistance of Lateral β-Ga2O3 MOSFETs Up to 400 V

Higher cost, lack of availability, and higher on-resistance (RDSon) are just a few drawbacks of semiconductor devices at higher voltages. This applies to the wide-bandgap semiconductors like Silicon Carbide (SiC) MOSFETs as well, where individual devices are not commercially available at voltages above 3.3-kV at the time of preparing this paper. Series connection of individual semiconductor devices is an effective way to achieve higher-voltage switches. However, the inherent unequal static and dynamic voltage sharing problems between the devices in the stack need to be addressed. The majority of the existing voltage balancing schemes are developed for slow-switching Si-based semiconductors, and bring about a significant amount of additional losses in the control circuit or on the switches. Some other techniques are not practical on SiC MOSFETs, given the high switching speeds featured by these devices. This paper tries to survey and compare the voltage balancing techniques proposed by researchers for being used on the faster-switching wide-bandgap devices. The voltage balancing techniques proposed in literature are classified into 5 different categories of “passive solutions”, “self-powered and pulse-powered gate drivers”, “active gate control techniques”, “gate timing control methods” and “voltage clamping solutions”. Then, a comprehensive and unbiased comparison is conducted between different voltage balancing techniques from “cost”, “modularity”, “complexity”, “startup balancing”, “losses” and “voltage balancing” points of view. Finally, conclusions from the conducted comparison are discussed.

Study of Voltage Balancing Techniques for Series-Connected Wide-Bandgap Semiconductors Devices

**Abstract:**

The world seabird community is declining globally, and seabird-fishery interactions have been identified as a significant cause of this general trend. It is therefore important and urgent to better understand the spatio-temporal determinants of such interactions. Historically, those have been extensively studied using at-sea observations and dietary studies. More recently, bird-borne devices allowed tracking seabird behaviour over time, but it is challenging to identify seabird-fisheries interactions exclusively from those movement data. Fortunately, bird-borne videos have brought invaluable information on the subject, enabling to directly observe the surroundings of ocean wanderers. Yet, the analysis of video data is extremely time-consuming. Using combined deployments of GPS and video cameras, we aimed at identifying specific movement patterns (GPS data) associated with seabird-fisheries interactions (as observed from video). If such patterns can be automatically identified, the routine can then be widely applied to similar datasets comprising only movement data. We deployed GPS recorders and cameras on 37 northern gannets (Morus bassanus) foraging in the English Channel from 2011 to 2015. In addition, 185 gannets from the same breeding colony were equipped with GPS only across 2005-2019. Using statistical and modelling procedures, we analysed the fine scale foraging movements of gannets interacting with fisheries to identify specific movement patterns associated with seabird-fisheries interactions. The classification model was then applied to all GPS-tracked Northern gannets to detect and map the spatio-temporal dynamics of gannet-fishery interactions in the English Channel across 2005-2019. Overall, our analysis provides a long-term perspective on seabird-fishery interactions in one of the most anthropized seas of the planet, and we propose a general methodology to help reach a better understanding of seabirds/fisheries interactions worldwide.

**Authors:**

David Grémillet¹, Amélie Lescroël¹, Andréa Thiebault², Jérôme Fort¹, Pascal Provost³, Akinori Takahashi⁴, Clara Péron⁵

¹CNRS, ²NMMU, ³LPO, ⁴National Institute of Polar Research, ⁵MNHN

Underline digital video library

Wing Tilt in ELO-Grown GaN

Localized Phonon Polariton Modes in GaN/AlN Nanowiresx

Q-Functionals for Value-Based Continuous Control

ACE: Cooperative Multi-agent Q-learning with Bidirectional Action-Dependency

CEM: Constrained Entropy Maximization for Task-Agnostic Safe Exploration

Adaptive Policy Learning for Offline-to-Online Reinforcement Learning

Which side are you on? Insider-Outsider classification in conspiracy-theoretic social media

Multilayer Perceptrons as Weighted Conditional Knowledge Bases: an Overview

Performance analysis of YOLOv3 for real-time detection of pests in soybeans

Switching Behavior and Dynamic On-Resistance of Lateral β-Ga2O3 MOSFETs Up to 400 V

Study of Voltage Balancing Techniques for Series-Connected Wide-Bandgap Semiconductors Devices

Identifying seabird interactions with fisheries from GPS tracks: Automatic classification validated using bird-borne video recordings

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES