The most recent approaches for clustering single-cell RNA-sequencing data rely on deep auto-encoders. However, three major challenges remain unaddressed. First, current models overlook the impact of the cumulative errors induced by the pseudo-supervised embedding clustering task (Feature Randomness). Second, existing methods neglect the effect of the strong competition between embedding clustering and reconstruction (Feature Drift). Third, the previous deep clustering models regularly fail to consider the topological information of the latent data, even though the local and global latent configurations can bring complementary views to the clustering task. To address these challenges, we propose a novel approach that explores the interaction between local and global latent configurations to progressively adjust the reconstruction and embedding clustering tasks. We elaborate a topological and probabilistic filter to mitigate Feature Randomness and a cell-cell graph structure and content correction mechanism to counteract Feature Drift. The Zero-Inflated Negative Binomial model is also integrated to capture the characteristics of gene expression profiles. We conduct detailed experiments on real-world datasets from multiple representative genome sequencing platforms. Our approach outperforms the state-of-the-art clustering methods in various evaluation metrics. 

Exploring the Interaction between Local and Global Latent Configurations for Clustering Single-cell RNA-seq: A Unified Perspective

Current generative transformer-based models have achieved state-of-the-art performance in long document summarization. However, this task witnessed a paradigm shift in developing ever-increasingly computationally-hungry models, focusing on effectiveness while ignoring the economic, environmental, and social costs of producing such results. Consequently, such extensive resources impact climate change and raise barriers to small and medium organizations characterized by low-resource regimes of hardware and data. As a result, this unsustainable trend has lifted many concerns in the community, proposing tools to monitor models' energy costs. Despite their importance, no evaluation measure considering models' eco-sustainability exists yet. In this work, we propose Carburacy, the first carbon-aware accuracy measure that captures both model effectiveness and eco-sustainability. We perform an extensive benchmark for long document summarization, comparing multiple state-of-the-art quadratic and linear transformers on several datasets under eco-sustainable regimes. Finally, thanks to Carburacy, we found optimal combinations of hyperparameters that let models be competitive in effectiveness with significantly lower costs.

Carburacy: Summarization Models Tuning and Comparison in Eco-Sustainable Regimes with a Novel Carbon-Aware Accuracy

Modern power systems will have to face difficult challenges in the years to come: frequent blackouts in urban areas caused by high peaks of electricity demand, grid instability exacerbated by the intermittency of renewable generation, and climate change on a global scale amplified by increasing carbon emissions. While current practices are growingly inadequate, the pathway of artificial intelligence (AI)-based methods to widespread adoption is hindered by missing aspects of trustworthiness. The CityLearn Challenge is an exemplary opportunity for researchers from multi-disciplinary fields to investigate the potential of AI to tackle these pressing issues within the energy domain, collectively modeled as a reinforcement learning (RL) task. Multiple real-world challenges faced by contemporary RL techniques are embodied in the problem formulation. In this paper, we present a novel method using the solution function of optimization as policies to compute the actions for sequential decision-making, while notably adapting the parameters of the optimization model from online observations. Algorithmically, this is achieved by an evolutionary algorithm under a novel trajectory-based guidance scheme. Formally, the global convergence property is established. **Our agent ranked first in the latest 2021 CityLearn Challenge**, being able to achieve superior performance in almost all metrics while maintaining some key aspects of interpretability.  

Winning the CityLearn Challenge: Adaptive Optimization with Evolutionary Search under Trajectory-based Guidance

Adaptive Temporal Planning for Multi-Robot Systems in Operations and Maintenance of Offshore Wind Farms

This paper demonstrates a GaN-based three-level inverter design for an electric vehicle (EV) traction application. The basic power block is a three-level active neutral point clamped (3L-ANPC) converter operating at 800 V DC bus voltage with the use of two paralleled 650 V, 60 A GaN E-HEMTs. The paper focuses on switching mode selection for better thermal performance of high-power 3L-ANPC inverters. The paper presents design details and guidelines for power layout structure and gate driver. The key parasitic inductances of the designed power block are extracted using ANSYS Q3D. A PLECS simulation model using the device loss parameters shows approximate power delivering capability of 41 kW for the designed forced air-cooled three-phase inverter. Experimental double pulse test results are presented with low voltage overshoot to verify the low inductance power loop design. Experimental results of the three-phase 3LANPC inverter operation at 70 kHz switching frequency are presented for delivered output power of 5.1 kW at 550 V DC.

Design Considerations of a GaN-Based Three-Level Traction Inverter for Electric Vehicles

Radio frequency (RF) engineers and power (Power)  electronics engineers are increasingly using gallium nitride (GaN) in their products replacing Si technology.  While there are differences in RF and power GaN technology, such as substrate, GaN epi thickness, etc., the core properties of GaN, that leads to advantages over Si are the same, such as higher breakdown voltage, high electron mobility in 2DEG compared to Si MOSFET. In a typical company, RF and power electronics engineers operate in separate silos, rarely interact with each other except to discuss power supply specs and are literally decoupled by bypass capacitors or capacitor banks.   At Tagore Technology, we have brought together RF and Power engineers under the same roof with the common thread being GaN. Our goal is to explore co-design of GaN power supply and GaN power amplifiers to create a compact solution at the PCB/Module level now and possibly on the same wafer in future. This talk will highlight one such application where pulsed radar power amplifiers are supplied by a high frequency GaN converter leading to 20x reduction in capacitor bank size without impacting noise properties of the RF signal. The challenges of co-design such as mixing of power supply switching noise as a phase noise of the RF signal as well as thermal management challenges will be highlighted.

Bringing RF & Power closer through GaN Technology

The first GaN power devices has been commercialized over a decade ago. These devices have been demonstrated in multiple end applications including chargers for cell phones, motor drives, DC/DC converters for servers or class-D audio amplifiers. GaN devices delivers clear benefits in system performance or cost in those end applications, but true commercial success has been mostly limited to chargers for cell phone or other portable devices. This limited commercial success so far is at least partly due to a restricted supply of GaN device production at a competitive cost point. The various fabless startups are facing the issue of high GaN wafer price at foundries (6-inch today) and limited value adder of discrete device design. More functionality integration on GaN die aims to increase their value adder of these products vs. discrete devices with mixed results and market adoption success. This presentation will argue that true integrated device manufacturer with high volume 8-inch internal manufacturing, fully focused on GaN is the game changer required to bring GaN power devices out of peripheral applications and into mainstream high-volume end-products, including mobile phones. In such ecosystem, roadmap based on discrete devices from 650V down to 30V and system-in-package integration is highly attractive for foreseeable future. This talk will give examples of latest device innovation including 40V bi-directional GaN switch reducing on-state resistance by 50%, chip size by 70% and temperature rise by 40%. First ever GaN product designed directly into mobile phone handset.

GaN device innovation matched with manufacturing capacity and quality propels GaN to new markets

Wide-bandgap Device and related technology to Enable Medium Voltage Distribution Power Grid

To reduce global warming / human health impacts, governments around the globe have been committed to reduce greenhouse gas and pollutant emissions.  To achieve these goals, vehicles with internal combustion engines (ICEs) are being replaced at very fast pace by vehicles with electrified power trains, (e.g., battery-electric vehicles (BEVs), plug-in hybrid electric vehicles (PHEVs), fuel-cell electric vehicles (FCEVs), etc.) To offer electrified functions, many power-electronic (PE) modules are incorporated on these vehicles (e.g., traction inverters, DC/DC converters, on-board / off-board chargers, etc.)  In these PE modules, WBG devices have shown promising potentials replacing Si ones to achieve better performance, smaller footprints, lighter weights, etc.  However, from many view angles, many WBG products are working under higher stresses than the Si ones to deliver better performance, leading to concerns of lower long-term reliability / higher parameter drifts.  To convince automotive manufacturers / suppliers, WBG vendors should disclose the raw-wafer / epi-layer defect info (e.g., densities, failure mechanisms, long-term aging effects, etc.), quality assurance processes and qualification results, EOL screening processes and pass/fail criteria, production yields, test-to-failure data (e.g., Weibull charts, bath-tub curves), and actual field return results.  For comparison, Si experience will be shared in this presentation.

Reliable wide-bandgap (WBG) products for safety-critical automotive applications – what are expected

Insulated metal substrate (IMS) boards have been widely used in industry for their remarkable electrical and thermal performance. However, the IMS boards require a separate driver board that affects the power density. With Gallium Nitride enhancement-mode high electron mobility transistors (GaN E-HEMTs) being adopted in modern power converters, power density is becoming critical more than ever. This paper proposes the next generation of single-board IMS power stages (i.e., IMS4). In this board, the power board and the driver board are embedded into a single board. Furthermore, a CerStrate™ ceramic substrate single-board solution is proposed that maintains the exact electrical specifications as observed in IMS4 and significantly enhances the thermal performance. The superiority of these single-board solutions is verified through finite element analysis (FEA) simulations and experiments.

Next Generation of GaN Single-Board High Power Stages

In this work, Infineon’s 100V 3mΩ CoolGaN™ is evaluated and compared with a 100V similar Rdson MOSFET using OptiMOS™5 silicon technology in a three-phase inverter for battery powered motor drive applications. In order to minimize conduction losses, 30ns dead time is achieved for GaN due to its lower gate and output charge in comparison to 60ns of dead time for silicon. A motor-generator setup is developed to evaluate a 48V drone motor up to ~1kW of inverter power. For a motor speed of 4000RPM, with a bus voltage of 48V and 100 kHz switching frequency, GaN shows higher efficiency across the entire load range. At 1/3rd load (~300W), 98.67% and at full load (~900W) 98.10% power efficiency values are achieved. At light load (~45W), a significant increase of ~5% is also observed in efficiency. This experimentally demonstrates that 100V CoolGaN™ is ideal for developing high frequency switching inverters with high power densities and motor efficiencies in the kilowatt power range.

100V GaN for Highly Efficient 1kW Motor Drive Applications

With the development of SiC technology, the onresistance of discrete devices drops rapidly and can potentially replace the expensive power modules in high power converters. Unlike parallel discrete devices or a power module with built-in parallel dies, using a single device does not need to slow down the gate driver speed hence can achieve lower switching loss. However, there are still many challenges that need meticulous design considerations. This paper uses the 1200V TO-247-4 package device as an example to provide detailed guidance on device selection, loop optimization, insulation, thermal management, and mounting. A DC-DC converter design example is given with a very low loop inductance and excellent thermal performance. The aircooled hardware demonstrates the hardware limitation is around 40kW in an 800V to 470V buck operation. Normalized to the Ron of the two devices used, this demonstrates an extremely high power density figure of merit of PDFOM=87kW/cm2 .

Underline digital video library

Exploring the Interaction between Local and Global Latent Configurations for Clustering Single-cell RNA-seq: A Unified Perspective

Carburacy: Summarization Models Tuning and Comparison in Eco-Sustainable Regimes with a Novel Carbon-Aware Accuracy

Winning the CityLearn Challenge: Adaptive Optimization with Evolutionary Search under Trajectory-based Guidance

Adaptive Temporal Planning for Multi-Robot Systems in Operations and Maintenance of Offshore Wind Farms

Design Considerations of a GaN-Based Three-Level Traction Inverter for Electric Vehicles

Bringing RF & Power closer through GaN Technology

GaN device innovation matched with manufacturing capacity and quality propels GaN to new markets

Wide-bandgap Device and related technology to Enable Medium Voltage Distribution Power Grid

Reliable wide-bandgap (WBG) products for safety-critical automotive applications – what are expected

Next Generation of GaN Single-Board High Power Stages

100V GaN for Highly Efficient 1kW Motor Drive Applications

Design of High Power Converter with Single Low Ron Discrete SiC Device

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES