AmoebaNet : an efficient and scalable SDN-enabled network service for extreme-scale distributed science

Shah, Syed Asif Raza University of Science and Technology 2019 국내박사

The extreme-scale distributed science workflows play an essential function for scientific discoveries. Today’s large scientific experimental facilities are generating tremendous amount of data. In recent years, the significant growth of scientific data analysis has been observed across scientific centers. The scientific experimental facilities are producing unprecedented amount of data and scientific community encounters new challenges to data intensive computing. The performance of extreme-scale distributed science is highly depends on high-performance, adaptive, and robust network service infrastructures. To support data transfer for extreme-scale distributed science, there is the need of high performance, scalable, end-to-end, and programmable networks that enable scientific applications to use network efficiently. The existing network paradigm that support extreme-scale distributed science workflows consists of three major components: terabit networks that provide high network bandwidths, Data Transfer Nodes (DTNs) and Science DMZ architecture that bypasses the performance hotspots in typical campus networks, and on-demand secure circuits/paths reservation systems, such as ESNet OSCARS and Internet2 AL2S, which provides automated, guaranteed bandwidth service in WAN. This network paradigm has proven to be very successful. However, to reach its full potentials, we claim that existing network paradigm for extreme-scale distributed science must address three major problems: last mile problem; scalability problem; and the agility, automation and programmability problem. The recently emerged concept in network world is called Software-Defined Networking (SDN). This latest technology introduced the new methods of configuration and management of networking. In SDN, the underlying network devices are simply considered as packets forwarding elements and control logic of network is managed centrally by using a software program that dictates the entire network behavior. To address above mentioned problems, this thesis proposed a solution called AmoebaNet. AmoebaNet applies SDN technology to provide “QoS-guaranteed” network services in campus or local area networks. AmoebaNet complements existing network paradigm for extreme-scale distributed science: it allows application to program networks at run-time for optimum performance; and, in conjunction with WAN circuits/paths reservation system such as ESNet OSCARS and Internet2 AL2S; in result, it solves the problems of last mile, scalability, and the agility, automation and programmability. In this thesis, we also presented Congestion Aware Multipath Optimal Routing (CAMOR) solution which can be an additional service for AmoebaNet.

Controllability of Non-thermal Effect in Low Current Arc Plasma and Its Applications in Fuel Conversion : 저전류 아크 플라즈마에서의 저온 플라즈마 효과 제어 가능성 및 이를 이용한 연료 전환 공정

DINH DUY KHOE University of Science and Technology (UST) 2019 국내박사

Scientific and Technological Knowledge Production Trends and Collaboration Patterns Based on an Ecosystem Perspective: Case of Educational Robotics

RICHA KUMARI University of Science and Technology 2020 국내박사

The new and emerging technologies such as artificial intelligence, machine learning, and robotics have changed the current industrial structure and recreated the new industries that are majorly based on the value chain of the global network environment. In a recent setting, the emergence of technological development and innovation is much dependent on a collaborative structure that can facilitate the recombination of existing knowledge and technologies to generate innovation. To enhance the collaborative structure and technological recombination, it is important to establish an environment that is less bounded and blurry and supports an open ecosystem environment. The innovation and knowledge ecosystem is characterized by a community of actors that assist evolving characters of knowledge structure and performance to co-produce innovation. The changing dynamics of interactions in an ecosystem provide a better understanding of the development and competitive strategy of emerging technology leading to value creation. Hence, this study utilizes the perspective of an ecosystem to analyze the development trends of scientific and technological knowledge and knowledge flow structure of a specific emerging technology. For this, the study uses the case of educational robotics technology and developed a framework to examine the comprehensive knowledge structure, evolutionary trends, and collaborative patterns in this technological area. In the first part of the study, the ecosystem framework is evaluated and the theory of the knowledge ecosystem is updated in the context of this study. The theoretical study evaluates the knowledge production pattern and type of knowledge produced within the structure. In the second part, the importance of educational robots has been highlighted to understand its role and future potentials. A special focus is given to highlight the role of educational robots in the current scenario of the COVID-19 pandemic. Moreover, the paper analyzed the scientific knowledge structure by applying the bibliometric and scientometric based evaluation methods to examine the productivity and performance of major countries and players in educational robotics domain. Also, the different principles of social network analysis like hubs, authorities, and broker analysis are used to identify the key countries and institutions working in the educational robotics area. The co-citation analysis at the country and institutional level is done to quantify and evaluate the connections among these players. Finally, an interaction among the players has been visualized by using a network map. The findings of the analysis showed that educational robotics research is more prominent in developed countries like the US, UK, Japan, and East Asian, and countries of other developing regions still lack scientific research in this area. USA, UK, Belgium, and the Netherlands are the most significant hubs and authorities acting as an important point in knowledge transfer. Netherland, Japan, and the USA play an important role as gatekeeper functionalities by acting as important bridge agents in knowledge transfer activities. Similarly, the most important institutions found were also mostly from advanced nations like Australia, the US, Sweden, and Canada. The competitive analysis is helpful to evaluate the country’s position and performance and the result can support R&D investment and policy-related decisions. In the next part of the study, the paper identified the important concepts and representative research areas from the scientific knowledge data by using keyword co-occurrence analysis. Further, representative research areas are selected by using centrality based measures that were used to find important and influential keywords. Also, topic modeling based on latent Dirichlet allocation (LDA) algorithms is applied to technological knowledge (patents) data to identify the latent knowledge structure and valuable topics. The model offers emerging technology areas and trends and contribute to the understanding of the emergence and development of technology over time and in forecasting the technology for the near future. At the final step of the analysis, the views and expectations of users on educational robotics technology have been analyzed by using the hype curve, and sentiment analysis. The analysis is conducted on twitter data to provide a better understanding of the response and sentiments of users. Social media, as a source of knowledge exchange, has an impact on the innovation ecosystem and support open innovation models. Understanding the polarity and sentiments by using social media helpful in analyzing the market expectation on technology. This result of the analysis can be useful to understand the educational technology adoption process in the market and can assist in other market-related decisions. 인공지능, 기계학습 및 로봇공학과 같은 신흥기술은 현재의 산업 구조를 변화시켰으며, 글로벌 네트워크 환경의 가치사슬을 기반으로 하는 신산업을 재창조하였다. 이러한 최근 환경에서 기술 개발과 혁신의 등장은 기술혁신을 창출하기 위해 기존 지식과 기술의 재조합을 촉진할 수 있는 협력구조에 크게 의존하고 있다. 협력구조 및 기술의 재조합을 강화하기 위해서는 경계가 보다 명확하고 개방된 생태계 환경을 구축하는 것이 중요하다. 혁신과 지식 생태계는 진화하는 지식 구조와 성과를 통해 혁신을 공동 창출하는 행위자들의 커뮤니티로 특징지어진다. 생태계 내에서 변화하는 상호작용의 역학관계는 가치창출로 이어지는 신흥 기술의 개발 및 경쟁 전략에 대해 보다 깊은 이해를 제공한다. 따라서 동 연구는 생태계 관점을 활용하여 특정 신기술의 과학기술 지식 및 지식 흐름 구조의 개발 동향을 분석하고자 한다. 이를 위해 동 연구에서는 교육용 로봇의 사례를 이용하며, 해당 기술분야의 포괄적인 지식 구조, 발전 추세 및 협력 패턴을 검토하기 위한 프레임 워크를 개발하였다. 제1장에서는 생태계 프레임워크를 평가하고, 동 연구의 맥락에서 지식 생태계 이론을 업데이트하였다. 이론적 연구에서는 지식 생산 패턴과, 그 체계 내에서 생성된 지식의 유형을 평가한다. 제2장에서는 교육용 로봇의 역할과 미래 잠재력에 대한 언급을 통해 교육용 로봇의 중요성을 강조하였고, 현재 COVID-19 시나리오 하에서 교육용 로봇의 역할에 특별히 중점을 두었다. 또한 교육용 로봇 분야에서 중심적인 역할을 수행하는 국가 및 기관의 성과를 조사하기 위해 계량서지 및 사이언토매트릭 기반의 평가 방법을 적용하여 과학적 지식 구조를 분석하였다. 또한 교육용 로봇 분야에서 협력하고 있는 주요 국가 및 기관을 식별하기 위해 허브, 권위 및 브로커 지수 등 다양한 소셜 네트워크 분석 원리를 활용하였다. 국가 및 기관 수준에서의 공동 열거 분석은 이러한 참여자 간의 연관성을 수량화 및 평가하기 위해 수행되었다. 마지막으로 네트워크 지도를 활용하여 참여자 간의 상호작용을 시각화 하였다. 분석 결과에 따르면 미국, 영국, 일본 및 동아시아와 같은 선진국에서 교육용 로봇 연구가 더욱 두드러지고 있으며, 다른 개발도상국에서는 여전히 동 분야에 대한 과학적 연구가 부족한 것으로 나타났다. 미국, 영국, 벨기에 및 네덜란드는 지식 이전의 중요한 요충지로서 작용하는 가장 중요한 허브 역할을 하고 있다. 네덜란드, 일본 및 미국은 지식 이전 활동에서 중요한 교량 역할을 수행하고 있다. 마찬가지로, 핵심 기관들은 호주, 미국, 스웨덴 및 캐나다와 같은 선진국에서 나타났다. 경쟁분석은 국가의 연구단계 및 성과를 평가하는 데 도움이 되며, 그 결과는 R&D 투자 및 정책 관련 의사 결정에 활용될 수 있다. 제3장에서는 키워드 동시 발생 분석을 통해 과학 지식 데이터에서 중요한 개념과 대표 연구 영역을 파악하였다. 또한 중요하고 영향력 있는 키워드를 찾는 데 사용된 중심성 지표를 사용하여 대표 연구 영역을 선별하였다. 또한 잠재 디리클레 할당(LDA) 알고리즘을 기반으로 하는 토픽 모델링은 기술 지식(특허) 데이터에 적용되어 잠재적 지식 구조 및 중요 주제를 식별하였다. 동 모델은 신흥 기술 분야 및 트렌드를 제공하고, 기술의 출현 및 발전을 이해하는 데 도움을 주며, 가까운 미래의 기술 예측에 기여한다. 또한 정책 결정자와 기업의 향후 의사 결정에 도움이 될 수 있다. 분석의 마지막 단계인 제4장에서는 하이프 사이클과 감정분석을 활용하여 교육용 로봇 기술에 대한 사용자 견해 및 요구를 분석하였다. 트위터 데이터에 대한 분석은 사용자 반응 및 정서를 보다 잘 이해할 수 있도록 제공되었다. 지식 교환의 원천인 소셜 미디어는 혁신 생태계에 영향을 미치며 개방형 혁신 모델을 지원한다. 소셜 미디어를 활용하여 극성 및 정서를 이해하는 것은 기술에 대한 시장의 기대치를 분석하는 데 도움이 된다. 동 분석 결과는 시장의 교육 기술 채택 과정을 이해하고 다른 시장 관련 의사 결정에 도움이 될 수 있다.

Development of human ear-mimetic construct based on three dimensional cell printing technology

이정섭 Pohang University of Science and Technology 2016 국내박사

Tissue engineering is an interdisciplinary field integrating biotechnology, materials engineering and mechanical engineering that focuses on restoring and regenerating various tissues and organs, such as the bladder, airways, and myocardium. In particular, cell printing is a promising technology for effectively regenerating tissues and organs whereby a construct is fabricated based on a layer-by-layer process using appropriate cells at a high cell density, effective hydrogels, and growth factors. Such cell-printing technology allows three dimensional (3D) living tissues and organs to have anatomical cell arrangements and geometrical shapes similar to native tissues and organs by directly printing cells. Despite the outstanding potential of cell-printing technology, most studies remain in the early stages of regenerating tissues, with relatively simple shapes and functions. Furthermore, there is a limitation with regard to regenerating composite tissues of similar shape and size to human tissues and organs because the technology needed to fabricate complex-shaped constructs of large volumetric size has not yet been developed. In this research, a 3D human ear-mimetic cell-printing technology was developed and applied to ear regeneration. Here, the 3D human ear-mimetic cell-printing technology was validated through fabricating a human ear-mimetic cell-printed construct and evaluating the cartilage and adipose tissue formation. Human tissues and organs are composite tissues, comprising two or more types of cells and tissues. With respect to fabricating the cell-printed construct, a multi-head tissue/organ building system (MtoBS) with six independent dispensing heads was developed to enable the dispensing of widely varying biomaterials with high- and low-viscosity properties. Additionally, in fabricating the human ear-mimetic cell-printed construct, the prolonged 3D printing time can cause low cell viability and inadequate performance of the construct because the cells can be exposed to a harsh environment over a long printing time. With this in mind, the MtoBS was improved, with a clean-air workstation, a humidifier, and a Peltier system, providing a more suitable printing environment for a large-volume construct to maintain high cell viability. With the advanced MtoBS, it was confirmed that better control of the printing temperature enabled the enhanced printability of hydrogels and higher cell viability for the construct, despite a prolonged printing time. The human ear has a complex shape and an anatomically complex composition of tissues. A bottom-up fabrication method has the limitation of not being able to stack constructs with overhanging, curved, or hollowed shapes, given the cell-printing technology. To fabricate a cell-printed construct with a complex shape, a sacrificial layer process and computer-aided design and manufacturing (CAD/CAM) technologies were developed. In the sacrificial layer process, the main part was printed with poly-caprolactone (PCL) and the cell-laden hydrogel, and polyethylene glycol (PEG) was then deposited as a sacrificial layer to support the main structure. PEG can be removed readily in aqueous solutions, and the procedure for removing PEG does not affect cell viability. CAD/CAM software was developed to enable a cell-printed construct to be fabricated with two polymers and two cell-laden hydrogels by independent control of the dispensing heads. Fabrication conditions were established for creating a construct for ear regeneration. Appropriate line widths and pore sizes were determined for fabricating the construct with an ear-like shape and similar mechanical properties to those of the human ear through the evaluation of mechanical strength. Once the fabrication conditions were established, the sacrificial layer process, and cell-printing technologies allowed an ear-shaped cellular construct to be manufactured. According to these results, the advanced MtoBS enabled a cell-printed construct with complex shapes to be fabricated while maintaining high cell viability using the sacrificial layer process and CAD/CAM technology. For the effective regeneration of composite tissue in the ear, porcine auricular cartilage and human adipose tissue-derived decellularized extracellular (ear-cdECM and adECM) hydrogels were developed for printing cells and inducing target tissue formation. Human adipose-derived stem cells (ASC) were encapsulated with 2% ear-cdECM and 3% adECM hydrogels, and the cell-laden hydrogels were used to fabricate the cell-printed construct, which regenerated cartilage and adipose tissue in both in vitro and in vivo tests. Compared with control groups, which consisted of cell-printed constructs with 4% alginate hydrogel and transforming growth factor-beta (TGF‒β) for cartilage tissue formation, and with 3% collagen hydrogel and basic fibroblast growth factor (bFGF) for adipose tissue formation, it was confirmed that the two kinds of dECM hydrogels induced cartilage and adipose tissue formation at the same level of tissue regeneration as with specific growth factors. Based on the powerful dECM effect, the cell-printed construct in the shape of an ear was fabricated successfully with ASC-laden ear cdECM and adECM hydrogels, and was implanted subcutaneously in a nude mouse model for in vivo testing. For 4 and 8 weeks, the human ear-mimetic cell-printed constructs maintained their initial shapes, and cartilage and adipose tissue were formed in the parts with ear-cdECM and adECM hydrogels. These results demonstrated that dECM hydrogel can induce target tissue formation without specific growth factors and allow the cell-printed construct to form composite tissues. Thus, this validated the 3D cell-printing technology developed for fabricating human ear-mimetic cell printed constructs and regenerating composite tissues.

Design and experimental characterization of low and high viscosity inkjet printheads

Shah, Muhammad Ali University of Science and Technology 2022 국내박사

잉크젯 프린팅 기술은 저가의 직접 적층 제조 기술을 이용하여 프린팅을 하는 기술로 다양한 분야에 적용이 가능하다. 이 논문은 산업 현장에서 사용하는 잉크젯 프린팅 시스템에서 핵심 기술인 잉크젯 프린팅 헤드 기술을 연구하였으며, 세부적으로 압전 구동 기반 잉크젯 프린팅 헤드(저점도 잉크 프린팅) 및 Fabry-Pérot 공진기 기반 잉크젯 프린팅 헤드(고점도 잉크 프린팅)에 대한 연구를 수행하였다. 압전 잉크젯 프린팅에서 원하는 잉크젯 성능은 설계 매개변수와 전압 파형을 최적화하여 얻을 수 있다. 유한요소법(FEM)을 사용하여 잉크젯 프린팅 헤드의 잉크젯팅 성능을 설계하는 방법은 매우 오랜 시간이 소요된다. 따라서 잉크젯 프린트 헤드의 설계 시간을 줄이기 위해서, 기계적 및 전기적 등가해석 모델인 LEM(Lumped Element Modeling) 방법을 제안하였다. LEM방법은 FEM 방법에 비해 시뮬레이션 시간이 매우 짧다. 제안된 LEM을 사용하여 압력 챔버의 압력과 노즐 출구의 속도를 예측하고 원하지 않는 진동을 억제하기 위한 최적화된 구동 전압 파형을 설계하였다. 또한, LEM을 이용하여 젯팅이 되는 단일 노즐이 젯팅이 되지 않는 인접 노즐에 미치는 영향과 멀티 젯팅 노즐이 단일 중앙 노즐에 미치는 영향을 예측하는 간섭(Cross-talk)에 대한 연구를 수행하였다. 압전 잉크젯 프린팅 헤드에 대한 LEM 설계 결과의 타당성을 확인하기 위해서 잉크젯 모니터링 장비를 이용하여 액적의 속도 및 크기를 측정하였고, 이를 LEM 설계 결과와 비교하였다. 압전 잉크젯 프린트 헤드는 상용화되어 다양한 용도로 사용되고 있지만 고점도 잉크를 분사할 수 없기 때문에 매우 높은 점도의 잉크를 젯팅하기 위해서는 고점도 잉크 젯팅 방식을 적용해야 한다. 따라서 이 연구에서는 Fabry-Pérot 공진기를 기반으로 한 고점도 잉크젯 프린팅 헤드 기술을 연구하였다. FP resonator 기반 잉크젯 프린팅 헤드 연구는 음향 캐비티 내부의 음향 압력에 대한 해석이 수행되고, 최대 음향 압력을 얻기 위한 고점도 잉크젯 프린팅 헤드가 설계되었고, 실험실 수준의 고점도 잉크젯 프린팅 장치가 제작되어 기본적인 성능에 대한 실험 결과를 얻었다. Inkjet printing technology uses the low-cost direct deposition manufacturing technique for printing and is applicable in various fields. This thesis deals with the needs and issues in inkjet printing technology. Piezoelectric inkjet printing (low viscosity ink printing) and Fabry–Pérot resonator based printing (high viscosity ink printing) methods are presented. In a piezoelectric inkjet printing, the desired droplet properties and high dots per inch (DPI) can be obtained by the optimization of design parameters and voltage waveform. In the literature, the desired properties are either achieved by finite element method (FEM) simulations or experimentally using trial and error method. These both methods take very long time to get the desired properties and high DPI. In this study, lumped element modeling (LEM) of a piezoelectric inkjet printhead is done to target the needs. LEM takes very less time compared to FEM and trial and error method. Using proposed LEM, the pressure at the pressure chamber and velocity at the nozzle exit are predicted and optimized driving voltage waveform for the suppression of undesired vibrations is obtained. Furthermore, LEM for crosstalk in done in which the effect of a single activated nozzle on the non-activated neighboring nozzles, as well as the effect of multi-activated nozzles on a single central nozzle is investigated. Experimental analysis is performed to measure the droplet velocity, trajectory, and size. The LEM and experimental results are compared to validate the modeling. Although piezoelectric inkjet printhead is commercialized and used in various applications, however it is unable to jet inks with high viscosity, therefore, one must change the printing method to jet very high viscosity inks. In this study, a high viscosity printing technology based on Fabry–Pérot resonator is simulated, fabricated and tested. The extensive pressure analysis inside the acoustic cavities of FP resonator’s based printing technology is done and based on these analysis an optimized high viscosity inkjet printing device is presented for 18 kHz.

Development of active and durable electrocatalysts for oxygen evolution reaction and oxygen reduction reaction

Jahowa Islam University of Science & Technology 2022 국내박사

Fossil fuels are the world's most widely used energy source, but this source is limited and will be depleted over time. Attention is therefore being paid to renewable energy sources to reduce the dependence on fossil fuels. Energy storage and energy conversion technology play an essential role in the renewable energy sector. In particular, proton exchange membrane water electrolyzers (PEMWEs) and proton exchange membrane fuel cells (PEMFCs) are examples of energy storage and energy conversion technologies. Although these technologies offer several advantages over other technologies of renewable energy sectors, there are still some challenges that prevent the commercialization of PEMWE and PEMFC technologies. For PEMWEs, the main roadblock is the anodic oxygen evolution reaction (OER) catalyst, whereas for PEMFCs, it is the cathodic oxygen reduction reaction (ORR) catalyst. Electrocatalysts for the OER/ORR are an important area where many breakthroughs are needed to improve the slow kinetics of the OER/ORR and reduce the amount of precious metal loading. In this light, the aim of the present research is to develop electrocatalysts for the OER/ORR with promising activity and durability. In terms of the OER catalyst in a PEMWE, a series of boron carbide-supported iridium catalysts were prepared via the wet impregnation method using NaBH4 as a reducing agent. Boron carbide has good electrical conductivity and corrosion resistivity. Physical and electrochemical properties of the catalysts were controlled by changing the synthetic reduction temperature (30 °C–100 °C) and iridium content (10 wt%-60 wt%) on the boron carbide support. The prepared Ir/B4C catalyst is the most promising catalyst for the OER and was synthesized at 100°C reduction temperature. In addition, at 40% loading, Ir/B4C showed maximum OER catalytic performance. The 40%-Ir/B4C catalyst outperformed all synthesized catalysts as well as two commercial catalysts in both activity and durability. The improved performance of 40%-Ir/B4C can be correlated to three key factors: i) high electrochemical surface area, (ii) better electrical conductivity, and (iii) high concentration of Ir(III) on the surface. Controlling the synthetic reduction temperature and iridium content on the B4C support was found to help develop the interaction between iridium and B4C. This metal-support interaction prevents the oxidative dissolution and aggregation of iridium species. In a single cell test, Ir/B4C-40% showed outstanding cell performance of 1.61 V at 1.0 A/cm2 at 0.5 mg/cm2 loading. Using this catalyst in the anode of a PEMWE, the precious catalyst metal loading can be reduced by more than six orders. The Ir/B4C-40% catalyst also showed outstanding durability, e.g. only a small voltage increase of 11 mV during the durability test in MEA performance after operation for 48 hours at a constant current density of 2.0 A/cm2. In terms of the ORR catalyst in a PEMFC, a durable carbon-based electrocatalyst support was synthesized. Platelet-type carbon nanofiber (PCNF) and Vulcan carbon black (CB) were coated in a uniform and discrete manner with silica, followed by platinum deposition on it. Accelerated degradation testing of the silica-coated catalysts showed higher durability than non-silica-coated catalysts under potential cycling. The silica coating can reduce carbon corrosion during the potential cycling between 1.0 and 1.5 V by i) blocking the carbon support from direct contact with the oxygen source and ii) preventing the effect of oxygen spillover from the platinum to carbon. The results suggest the silica coating on a carbon support is an effective strategy to improve the durability of Pt-based electrocatalysts under potential cycling.

Stem Cell and Gene Editing Technology-Based Approaches for Hypoimmunogenic Regenerative Medicine

안주현 University of Science and Technology 2023 국내박사

Efficiency Enhancement in Ultrathin Cu(In,Ga)Se2 Solar Cells Prepared on Transparent Conducting Oxide Back Contacts

MuhammadSaifUllah University of Science and Technology (Campus: Kore 2018 국내박사

Analog and Digital Designs for On-Chip Learning in Neuromorphic Systems : 온칩 학습용 뉴로모픽 시스템을 위한 아날로그/디지털 디자인

Vladimir Kornijcuk University of Science and Technology 2018 국내박사

현대 디지털 컴퓨터는 인간의 두뇌에 비해 훨씬 더 빠르고 정확하게 논리 및 산술 연산을 수행할 수 있다. 그러나 주변의 환경을 실시간으로 인식하고 학습하는 능력에서 디지털 컴퓨터를 비롯한 인공 시스템은 아직 포유동물의 두뇌 수준엔 미치지 못한다. 지난 수십 년 동안 하드웨어 인공 시스템으로 포유동물 신경망의 기능을 실현하기 위한 연구가 이루어져 왔으며, 이는 ‘뉴로몰픽 (neuromorphic) 공학’ 이란 이름으로 알려져 있다. 본 논문은 확장가능성과 전력 효율성을 갖추면서 온칩(on-chip) 학습이 가능한 뉴로몰픽 시스템용 회로와 시스템을 제시함으로써 뉴로몰픽 공학에 기여하였다. 첫 번째로, leaky integrate and fire(LIF)에 기반한 새로운 형식의 인공 스파이킹(spiking) 뉴런을 제시하였다. 제시된 디자인은 통상적으로 축전기 기반 integrator를 활용하는 것과 달리 플로팅 게이트(floating gate, FG) integrator를 활용하고 있다. FG에 저장된 전하의 방전시간은 터널의 넓이보다는 높이, 두께에 대한 의존성이 더 크며, 이는 뉴런의 집적도 향상 가능성이 높음을 시사한다. 회로 동작은 BSIM 4.6.0 상보성 금속 산화막 반도체(CMOS) 모델을 활용한 시뮬레이션을 통해 진행되었다. 다음으로, spike timing dependent plasticity(STDP) 모델을 구현한 시냅스 회로를 제안하였다. FG-LIF 뉴런과 마찬가지로, 이 회로 역시 FG leaky integrator를 기반으로 구성되었다. randomly spiking neuron을 적용한 비지도 학습과 지도학습을 시키는 회로 시뮬레이션 결과, 두 경우 모두 시냅스간의 경쟁이 나타남을 확인하였다. 또한 몬테-카를로 시뮬레이션을 통해 CMOS의 오차가 존재할 경우에 대한 평가도 진행하였다. 세 번째로, 온칩 학습을 위해 random access memory, content addressable memory, partitioned RAM, pointer(PTR)의 네 종류의 순람표(look-up table, LUT) 기반 스파이크 라우팅 시스템을 제안하였다. 각각의 시스템에 대하여 라우팅의 지연이 발생하지 않는 최대의 네트워크 크기를 측정하기 위한 이론적인 근거를 제시하였고 스파이크 라우팅 속도, 가능한 신경 네트워크의 크기, 회로의 과부하 등에 관련된 장, 단점을 분석하였다. 마지막으로, a Xilinx Virtex-7 field programmable gate array (FPGA)를 활용한 완전 디지털화 뉴로몰픽 시스템의 프로토타입을 제시하였다. 이 프로토타입은 1024개의 LIF 뉴런들과 199,680개의 STDP 시냅스들로 구성되어있다. 비지도학습을 통해 시각적인 자극(막대)의 방향을 정확히 인식하도록 학습시킬 수 있었고, 이를 통해 이 시스템에서 온칩 학습이 가능함을 확인하였다. Achieving human-level cognitive performance using artificial systems has long been a motivating challenge that inspires researchers across different research fields. To date, one of the most widely used platforms to this end is general-purpose hardware, which partially owes to its rapid development and availability. The astonishing computational precision and speed make this platform very powerful in simulating behavioral models of biological neurons, synapses, and spiking neural networks (SNNs). Simulating large-scale SNNs in real time, however, is a daunting challenge due to the complexity, which makes their behavior description immensely complex at large scales. An alternative approach, neuromorphic system engineering, attempts to overcome this issue by using very large-scale integrated circuits to synthesize SNNs on a silicon wafer, instead of simulating their behaviors. This dissertation makes a contribution to this approach by proposing a range of circuits and systems for realizing scalable and power-efficient neuromorphic systems capable of on-chip learning. First of all, a new type of artificial spiking neuron based on leaky integrate-and-fire (LIF) behavior is proposed. A distinctive feature of the proposed design is the use of a floating gate (FG) integrator rather than a capacitor-based one. The relaxation time of the charge on the FG relies mainly on the tunnel barrier profile, e.g., barrier height and thickness (rather than the area). This opens up the possibility of large-scale integration of neurons. The circuit was designed by using 65 nm complementary metal oxide semiconductor (CMOS) technology and its feasible operation was examined by performing circuit simulation using the BSIM 4.6.0 CMOS model. The circuit simulation results offered biologically plausible spiking activity (<100 Hz) with a capacitor of merely 6 fF, which is hosted in an FG metal-oxide-semiconductor field-effect transistor. The FG-LIF neuron also has the advantage of low operation power (<30 pW/spike). Additionally, the proposed circuit was subject to possible types of noise, e.g., thermal noise and burst noise. In particular, thermal noise is likely prominent with regard to the use of such low capacitance. The simulation results indicated remarkable distributional features of interspike intervals that are fitted by Gamma distribution functions, similar to biological neurons in the neocortex. Second, a scalable synaptic circuit realizing spike timing dependent plasticity (STDP) model is presented. Like the FG-LIF neuron, this circuit is based on FG leaky integrators and is designed by using 65 nm CMOS technology. The circuit simulations feature (i) weight-dependent STDP that spontaneously limits the synaptic weight growth, (ii) competitive synaptic adaptation within both unsupervised and supervised frameworks with randomly spiking neurons. The estimated power consumption is merely 34 pW, perhaps meeting one of the most crucial principles (power-efficiency) of neuromorphic engineering. Additionally, the robustness of the proposed circuit in light of CMOS process variability effects (line edge roughness and random dopant fluctuations) was evaluated by performing Monte Carlo simulations. Despite notable parameter variability, the STDP behavior of the circuit could be validated as a whole, other than few exceptions. Third, four look-up table (LUT)-based spike routing schemes aimed for on-chip learning are presented. These are the random access memory, content addressable memory, partitioned RAM, and pointer (PTR)-based routing schemes. First of all, theoretical means are provided for evaluating the maximum network size for each scheme without routing congestion—experimentally justified using field-programmable gate array (FPGA) implementations. Given that they vary in spike routing rate, allowable neural network size, and circuit overhead, the pros and cons of each scheme are analyzed with regard to them. The results indicate that the PTR-based scheme supports a neuromorphic core consisting of 50,000 neurons (simultaneously firing at 50 Hz) and 10 million synapses at 1 GHz clock speed with minimum circuit overhead. The PTR-based scheme was further applied to multiple cores in a large-scale neuromorphic cluster, revealing that the cluster can theoretically hold 3.63 million neurons and 3.63 billion synapses at 200 MHz global clock speed when all cores operate at 1 GHz local clock speed. Finally, a fully-digital neuromorphic system prototype implemented in a Xilinx Virtex-7 FPGA is presented. The prototype comprised arrays of 1,024 LIF neurons and 199,680 STDP synapses and employed partitioned RAM routing scheme. The on-chip learning in the system was verified by performing a real-time orientation selectivity development experiment, during which it successfully “learned” to recognize the orientation of a visual stimulus (a bar) in the absence of teaching supervisor (unsupervised learning).

A Study on the Wastewater Treatment System by Advanced Phytoreactors

Oktavia Ratnasari University of Science and Technology 2012 국내석사

The increase of human activities in industrial or agricultural sectors to supply global consumptions causes deterioration in environment. In aquatic environment, water eutrophication is known as notorious for primary environment destruction. In atmospheric environment, anthropogenic global warming is continuously becoming one of the most important problems, despite many efforts have been done. It has been recognized that global warming is caused by CO2 emission mainly from fossil fuels. Various methods in capturing and sequestering CO2 have been intensively researched throughout the world. Phytoremediation is one of promising new technologies in addition to conventional treatment systems. This technology works by utilizing plants to remove contaminants in various media such as water, air, and soil. Through phytoremediation process, it may stimulate biological processes or physico-chemical characteristics of plants to aid the process. It also has been applied for treating numerous contaminants including heavy metals, pesticides, petroleum hydrocarbons, chlorinated solvents, explosives, radioactive substances and landfill leachates. These technologies have been broadly used in various fields because of its advantages in economic aspect, sustainability and operation as well as providing aesthetical value. In water environment, aquatic plants play a major contribution for sustaining nitrogen cycle and returning N2 gas back to the atmosphere through bacteria-mediated reactions of nitrification and denitrification. Several studies have demonstrated the effectiveness of these plants to remove nutrients, especially nitrogen and phosphorus, from polluted water. In the aspect of air pollution, studies have proven that phytoremediation has an important function in removing excessive amount of CO2 in the atmosphere by photosynthesis process of plants in terrestrial and aquatic ecosystems. Photosynthesis can produce plant biomass which can be utilized to produce renewable material for bioenergy. Since 1990s, light-emitting diodes (LEDs) have been developed as advanced light sources for indoor plant cultivation, replacing traditional fluorescent and incandescent lamps. LEDs have a very superior life span compared to conventional lamps and excellent features including energy efficiency, specific wavelength, easily modified light intensity and quality, small size, low thermal output, and high photoconversion efficiency. LEDs have been widely developed in indoor cultivation of various plant species, plant tissue culture, space agriculture, algaculture and plant disease reduction. In the aspect of light spectra, blue LED (400-470 nm) and red LED (630-665 nm) match with the absorption spectra of chlorophylls and carotenoids. As for phytochrome for plant growth, their peak absorbance of 660 nm and 730 nm can also be matched with red LED (630-665 nm) and far-red LED (730 nm). Since the discovery of LED technology, research related to plant growth and photosynthesis has been increased. This study was conducted to investigate nutrient removals from wastewater using effective aquatic plants in various systems incorporated into advanced phytoremediation of wastewater. Three types of phytoreactors were studied for their performances in removing nitrogen, phosphorus and organic contaminants. Nutrient removals were determined by kinetics, removal rates, and removal efficiencies. In the experiments using advanced plant growth acceleration technology, the growth rates of studied plants were also measured and compared to plant’s growth rates under controlled condition. The effects of light wavelengths in ABPs, ASBPs and ASCPs were also analyzed. In addition, the effect of CO2 enrichment in ASCPs was also considered. Lastly, prediction of nutrient removal using kinetic models was presented. Surface area occupied by plants and plant’s weights were employed to evaluate the performance of plants (P.stratiotes, E.crassipes and A.gramineus) to remove TN, TP and COD by batch reactors under normal and accelerated conditions. The removal of TN, TP and COD by plants differed between the treatment of light sources (p < 0.05). Using first order kinetic equation, based on unit area, TN removal rates by batch reactors ranged from 1,760 mg N to 3,720 mg N m-2 d-1, while based on unit weight, removals ranged from 862 mg N to 1,990 mg N kg-1 d-1. Area-based TP removals were between 141 mg P and 508 mg P m-2 d-1, and weight-based removals varied from 101 mg P to 283 mg P kg-1 d-1. For COD, area-based removals ranged from 528 to 1,330 mg COD m-2 d-1, and weight-based resulted in the range of 281 and 703 mg COD kg-1 d-1. Water lettuce (Pistia stratiotes) was selected for experiments with advanced plant growth acceleration (PGA) technology. In advanced sequential batch phytoreactor (ASBP), statistical analysis by ANOVA declared that the difference of area-based and weight-based nutrient removal rates by FL and PGA treatments was significant (p < 0.05 for TN, TP and COD removals. TN and COD removals were best achieved by phytoreactors treated using red light (=630+660 nm), while highest TP removal was reached using blue light (=400+440 nm). Further, in experiments using advanced sequential continuous phytoreactor (ASCP), in which different concentrations of CO2 were employed, p < 0.01 validated that kinetic coefficients for TN and TP removals was significantly different between normal and advanced phytoreactors. COD removal showed a p value of 0.08, suggesting that the application of LED didn’t improve the removal of COD in the system. Plants cultivated under advanced reactors showed superior results of growth rates compared to controls. In batch experiment under red light (=660 nm), P.stratiotes showed growth rate of 1.86 times higher than those grown under FL lamps, while for E.crassipes and A.gramineus, their growth rates were 2.14 times and 2.29 times compared to those grown under FL lamps. In ex