Conventional image sensors achieve color imaging using absorptive organic dye filters. These face considerable challenges however in the trend toward ever higher pixel densities and advanced imaging such as multispectral imaging and polarization-resolved imaging. In this talk, we will present a new paradigm for color imaging based on all-silicon nanowire devices without any color filters. We fabricate pixels consisting of vertical silicon nanowires with integrated photodetectors, demonstrate that t heir spectral sensitivities are governed by nanowire radius, and perform color imaging. We also demonstrate polarization-resolving photodetectors consisting of silicon nanowires with elliptical cross sections. Our approach is conceptually different from filter-based methods, as absorbed light is converted to photocurrent, ultimately presenting the opportunity for very high photon efficiency. This work was performed at Harvard University under the supervision of Prof. Kenneth B. Crozier.

Quantum dot
lasers epitaxially grown on Si are promising for an efficient light source for
silicon photonics. In the talk, I will present record-high performance and
excellent reliability of 1.3 μm quantum dot lasers on Si enabled by a low
threading dislocation density GaAs buffer layer. Continuous-wave threshold
currents as low as 4.8 mA and output powers of 185 mw have been achieved from
narrow ridge-waveguide lasers at 20 °C. P-modulation doped lasers operated up
to 107 °C. 1800-hour reliability tests at 35 °C showed an extrapolated
mean-time-to-failure of more than a million hours. Micro-ring QD lasers on Si
were also fabricated and showed an ultralow threshold current of 0.5 mA. I will
also present a preliminary result on monolithically integrated III/V laser,
waveguide, and photodetectors on Si. This represents a significant stride
toward efficient, scalable, and reliable III-V lasers on on-axis Si substrates for
photonic integrate circuits that are fully compatible with CMOS foundries.

종이접기(Origami) 기반 소프트 로보틱스 연구 

This lecture gives an introduction to neurorobotics and highlights future developments in neurorobotic technology. Participants will learn about the huge diversity of neurorobotic research which ranges from early simple models of robot control which are based on only a few neurons to the state-of-the-art closed-loop integration of large-scale brain simulations with detailed models of physical bodies. Based on this overview, the lecture highlights the open challenges in neurorobotics research how they will be addressed in the future development of the Neurorobotics Platform of the Human Brain Project.

III-nitride laser diodes (LDs) have been developed for many applications such as optical storage, medical, and solid-state lighting. However, the performance of LDs grown on conventional polar (0001) c-plane GaN has been hindered due to polarization-induced electric fields, leading to a reduction of wave-function overlap. Recent progress on semipolar and nonpolar planes has shown significant improvement in high-power performance of LEDs and LDs by reducing the polarization-induced electric field. Conventional III-nitride LEDs have typical bandwidth limits in the MHz range due to comparatively long carrier lifetime (on the order of ns) and significant RC parasitic effect. Even worse, for white lighting configuration, the response is significantly limited by slow relaxation lifetime (on the order of μs) of phosphor conversion. However, recent studies on LDs for light fidelity (LiFi) applications have demonstrated more than GHz bandwidth and Gbit/s data transmission rate with inherent advantage of short photon lifetime. As an alternative to LED technology, the use of laser diodes (LDs) as a transmitter presents a viable approach to overcome the limitations of modulation bandwidth as well as the lighting efficiency. Despite of these merits, only limited studies have been reported for LD-based VLC and LiFi demonstration without notable improvement in modulation capabilities. Since semipolar and nonpolar LDs are expected to have higher differential gain and lower threshold than c-plane LDs as mentioned above, they also have high potential for improvements in modulation bandwidth. Thus, nonpolar and semipolar LD is potential candidate for future LiFi applications requiring high-capacity for high-speed data transmission as well as high-efficiency for lighting.

Human brain functions have long served as the targets of development of artificial intelligent systems. Findings in neuroscience have provided guidance at multiple levels in the designs of machine learning architectures and algorithms. Today’s neuroscience, in turn, necessitates applications of artificial intelligence and machine learning algorithms for making sense of huge datasets.

In this talk, I review examples of co-evolution of AI and brain science and consider how they can help each other for further progress.

This lecture will be given by team-teaching (Prof. Hak-Rin Kim, Prof. Kyu-Ik Sohng, Prof. Jae won Jang). Sequentially, 'I.TFT Theory and Display Fabrication' will be lectured by Prof. Jae won Jang, 'II. Color & Display Image Theory' will be lectured by Prof. Kyu-Ik Sohng, and 'III. Operational Principles of Displays : LCD vs OLED'. 

 In the recent years we are witnessing a dominance of deep neural network learning approach in the machine learning field. Eventhough the neural network concept is more than 50 years old, only the recent developments enabled its wide use. Namely, availability of processing power, especially graphical processing units, availability of large databases, and the refined knowledge about the approach itself. Convolutional neural network as a special case of deep learning approach is widely used in computer vision domain, also in biometry. A lot of image material is obtained through surveillance scenarios, where we can use modalities like face, gait, and ears for recognition, normally within a multibiometrics system. Each modality has to be first detected and then recognized. The following examples are discussed in this context: 1) network for ear detection in the wild, where, different from competing techniques from the literature, our approach does not simply return a bounding box around the detected ear, but provides accurate and detailed, pixel-wise information about the location of the ears in the image; 2) training network with limited training data for ear recognition in the wild, where we explore different strategies towards model training with limited amounts of training data and show that by selecting an appropriate model architecture, using aggressive data augmentation, and selective learning on existing (pre-trained) models, we are able to learn an effective model; 3) face deidentification (anonymization) with generative network that provides privacy guaranties and at the same time retains certain important characteristics of the data even after deidentification.