A5073278616- Advanced Memory and Neural Computing
- Ferroelectric and Negative Capacitance Devices
- Advanced Neural Network Applications
- Topic Modeling
- COVID-19 and Mental Health
- Magnetic properties of thin films
- Natural Language Processing Techniques
- COVID-19 Pandemic Impacts
- Nanopore and Nanochannel Transport Studies
- Advanced Numerical Analysis Techniques
- Mental Health Treatment and Access
- Machine Learning in Materials Science
- Advanced Data Storage Technologies
- Biochemical Analysis and Sensing Techniques
- Carbon Nanotubes in Composites
- Catalytic Processes in Materials Science
- Machine Learning in Bioinformatics
- Software Engineering Research
- Semiconductor materials and devices
- Photosynthetic Processes and Mechanisms
- TiO2 Photocatalysis and Solar Cells
- Hydraulic flow and structures
- Quantum and electron transport phenomena
- Parallel Computing and Optimization Techniques
- Magnetic Field Sensors Techniques
Korea Advanced Institute of Science and Technology
1996-2024
Yonsei University
2005-2021
Center for Global Health
2020
Seoul National University of Science and Technology
2013
Cornell University
1987
Memorial Sloan Kettering Cancer Center
1987
Whitehead Institute for Biomedical Research
1987
Abstract Summary: This manuscript describes a new simple method to disperse single wall carbon nanotubes (SWNTs) in various organic solvents. The is based on using amphiphilic block copolymer micelles as dispersant. We have found that the stabilization of SWNTs by adhering surface much superior either surfactants or high‐molecular‐weight polymers currently used. Our nondestructive beneficial because property enables us stabilize polar solvents nonpolar ones at same time. also demonstrate...
4F <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> selector-less crossbar array 2Mb ReRAM test chip with 54nm technology has been successfully integrated for high cell efficiency and density memory applications by implementing parts of decoders to row/column lines directly under the area. Read/write specifications operation in a are presented minimizing sneak current through unselected cells. The characteristics (nonlinearity, Kw >;8,...
A transparent PS-b-P4VP copolymer/SWNTs nanocomposite film is used as both source and drain electrode in a pentacene OTFT with transmittance electricresistance values of 85% 6000 Ω/□, respectively. The low electric resistance the achieved by selective doping HAuCl4 P4VP cores self-assembled block copolymer micelles, which simultaneously provides sufficientstability for SWNTs suspension. Supporting information this article available on WWW under...
The processing in-memory (PIM) approach that combines memory and processor appears to solve the wall problem. NAND flash memory, which is widely adopted in edge devices, one of promising platforms for PIM with its high-density property intrinsic ability analog vector-matrix multiplication. Despite potential, domain conversion process, converts an current a digital value, accounts most energy consumption on flash-based accelerator. It restricts usage compared other platforms. In this paper,...
Resistive RAM (ReRAM) is widely regarded as a promising platform for deep neural network (DNN) acceleration. However, the ReRAM device suffers from severe thermal problems that degrade lifetime and inference accuracy of ReRAM-based DNN accelerator. To address issues, we propose <u>t</u>hermal-aware <u>op</u>timization framework <u>a</u>ccelerating on <u>R</u>eRAM (TOPAR). TOPAR includes 3-stage offline optimization online thermal-aware error compensation. Offline consists weight...
Energy-efficient Resistive RAM (ReRAM) based deep neural network (DNN) accelerator suffers from severe Stuck-At-Fault (SAF) problem that drastically degrades the inference accuracy. The SAF gets even worse in realistic ReRAM devices with low cell resolution. To address issue, we propose a fault-resilient DNN on devices. We first analyze device and 3-stage offline compilation lightweight online compensation. proposed work enables reliable execution of only 5% area 0.8% energy overhead ideal...
The growing computational demands of AI inference have led to widespread use hardware accelerators for different platforms, spanning from edge the datacenter/cloud. Certain application areas, such as in high-frequency trading (HFT) [1–2], a hard latency deadline successful execution. We present our new accelerator which achieves high capability with outstanding single-stream responsiveness demanding service-layer objective (SLO)-based services and pipelined applications, including large...
Transformer-based language models have become the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">de-facto</i> standard model for various natural processing (NLP) applications given superior algorithmic performances. Processing a transformer-based on conventional accelerator induces memory wall problem, and ReRAM-based is promising solution to this problem. However, due characteristics of self-attention mechanism accelerator, pipeline hazard...
A soil fulvic acid was extracted from topsoil of the Okchun Basin, Republic Korea, and purified characterized by chemical methods (elemental analysis, number averaged molecular weight) 13C 1H NMR spectroscopy. An "average" structure a has been developed modifying one structural models proposed for Suwannee River acid, IHSS (International Humic Substances Society) reference sample, based on comparative data. The assignment moieties facilitated use set subspectra CHn (n = 0 to 3) groups,...
Deep neural networks (DNNs) are widely used for real-world applications. However, large amount of kernel and intermediate data incur a memory wall problem in resource-limited edge devices. The recent advances binary deep network (BNN) computing in-memory (CIM) have effectively alleviated this bottleneck especially when they combined together. previous CIM-based accelerators BNN highly vulnerable to process/supply voltage/temperature (PVT) variation, resulting severe accuracy degradation...
Recently, multiple transformer models, such as BERT, have been utilized together to support natural language processing (NLP) tasks in a system, also known multi-task BERT. Multi-task BERT with very high weight parameters increases the area requirement of resistive memory (ReRAM) architecture, and several works attempted address this model size issue. Despite reduced parameters, number computations remains same, leading massive energy consumption ReRAM-based deep neural network (DNN)...
Resistive RAM(ReRAM) is gaining attention as a suitable memory platform for accelerating deep neural networks(DNNs) in an energy-efficient way. However, ReRAM-based DNN accelerators suffer from serious Stuck-At-Fault(SAF) issues that significantly degrade the inference accuracy. SAF device-level non-ideality, and problems of worsen realistic ReRAM with low cell resolution. To address problem ReRAM, we present framework mitigating on accelerators(Fault-free). We first analyze impact...
Many recent studies have focused on Processing-in-memory (PIM) architectures for neural networks to resolve the memory bottleneck problem. Especially, an increased interest in Spin Orbit Torque (SOT)-MRAMs has emerged due its low latency, high energy efficiency, and non-volatility. However, previous work added extra computing circuits support complicated computations, which results large overheads. In this work, we propose a new PIM architecture with relatively small peripheral circuit,...
Abstract Background According to the World Health Organization, coronavirus disease 2019 (COVID-19) pandemic has created situations that have a negative effect on people and threaten their mental health. Paraguay announced Estado de Emergencia Sanitaria (Presidential Decree No. 3456) March 16, 2020, which was followed by imposition of 24-h restriction movement order 21. Self-quarantine at home may been most effective method preventing spread infectious diseases; however, with global becoming...
ReRAM-based Processing-In-Memory (PIM) has been widely studied as a promising approach for Deep Neural Networks (DNN) accelerator with its energy-efficient analog operations. However, the domain conversion process operation requires frequent accesses to power-hungry Analog-to-Digital Converter (ADC), hindering overall energy efficiency. Although previous research suggested address this problem, ADC cost not sufficiently reduced because of unsuitable ReRAM. In paper, we propose...
Transformer-based language models have recently gained popularity in numerous natural processing (NLP) applications due to their superior performance compared traditional algorithms. These involve two execution stages: summarization and generation. The generation stage accounts for a significant portion of the total time its auto-regressive property, which necessitates considerable repetitive off-chip accesses. Consequently, our objective is minimize accesses during expedite transformer...
With the superior algorithmic performances, BERT has become de-facto standard model for various NLP tasks. Accordingly, multiple models have been adopted on a single system, which is also called multi-task BERT. Although ReRAM-based accelerator shows sufficient potential to execute by adopting in-memory computation, processing extremely increases overall area due fine-tuned models. In this paper, we propose framework area-efficient execution accelerator. Firstly, decompose of each task...
Titanium dioxide was coated onto buoyant polypropylene granules. granules (TCPG) had high mechanical and attritional stability as well appreciable photocatalytic activity under solar irradiation. Chosen model pollutant, Methylene Blue (MB) could be totally decolorized partially mineralized within 3 h of treatment. TCPG might successfully applied suspended in the column flow-through system. Degradation rates depended strongly on pH, initial dye concentration, optimal media dosage, following...
Thylakoids are the photosynthetic complexes that absorb light energy to produce electrons (PEs) in chloroplast of plant cells. PE production starts from water splitting at photosystem II (PSII) thylakoid membrane and they excited by absorbing photon energy. Over last decade, many researchers have attempted extract PEs enhancing attachment membranes an electrode or using artificial mediators. However, these some limitations including increased ohmic resistance length tethering molecules loss...
Abstract Background: This study investigated the factors influencing depressive feelings in Paraguayan public officials caused by coronavirus disease (COVID-19) pandemic. Methods: used a web-based cross-sectional method to analyze COVID-19-induced officials. The study’s research area was Asuncion and Limpio Paraguay. Results: results of Model 4 indicated high levels among officials, as well concerns about COVID-19 infection female also found that officials’ were related duration...