Soohyun Kwon

POSEIDON: A framework for application-specific Network-on-Chip synthesis for heterogeneous chip multiprocessors

In recent years, the rise in the number of cores being integrated on a single chip has led to a greater emphasis on scalable communication fabrics that can overcome data transfer bottlenecks. Network-on-Chip (NoC) architectures have been gaining widespread acceptance as communication backbones for multi-core systems, due to their high scalability, predictability, and performance. However, NoCs are also power hungry, and synthesizing a NoC fabric for a particular application requires solving a multitude of non-trivial design problems. Due to the large design space associated with various possible NoC configurations and design constraints, it is critical to automate the exploration process and arrive at a customized NoC that meets performance goals, while minimizing power and peak temperature. In this paper, we present a novel application specific NoC synthesis framework (POSEIDON) that combines multiple algorithms and heuristics to efficiently explore the solution space. Our results indicate that the proposed framework provides a reduction of up to 15.7% in power consumption, 21.08% in average latency, 27.05% in total energy, and 42.7% in energy-delay product compared to state-of-the-art approaches, as well as a 4.2% reduction in peak temperature when the framework is customized for thermal-aware synthesis.

Classification of precipitation types using fall velocity-diameter relationships from 2D-video distrometer measurements

Fall velocity-diameter relationships for four different snowflake types (dendrite, plate, needle, and graupel) were investigated in northeastern South Korea, and a new algorithm for classifying hydrometeors is proposed for distrometric measurements based on the new relationships. Falling ice crystals (approximately 40 000 particles) were measured with a two-dimensional video disdrometer (2DVD) during a winter experiment from 15 January to 9 April 2010. The fall velocity-diameter relationships were derived for the four types of snowflakes based on manual classification by experts using snow photos and 2DVD measurements: the coefficients (exponents) for different snowflake types were 0.82 (0.24) for dendrite, 0.74 (0.35) for plate, 1.03 (0.71) for needle, and 1.30 (0.94) for graupel, respectively. These new relationships established in the present study (PS) were compared with those from two previous studies. Hydrometeor types were classified with the derived fall velocity-diameter relationships, and the classification algorithm was evaluated using 3× 3 contingency tables for one rain-snow transition event and three snowfall events. The algorithm showed good performance for the transition event: the critical success indices (CSIs) were 0.89, 0.61 and 0.71 for snow, wet-snow and rain, respectively. For snow events, the algorithm performance for dendrite and plate (CSIs = 1.0 and 1.0, respectively) was better than for needle and graupel (CSIs = 0.67 and 0.50, respectively).

Incorporation of parameter uncertainty into spatial interpolation using Bayesian trans-Gaussian kriging

Quantitative precipitation estimation (QPE) plays an important role in meteorological and hydrological applications. Ground-based telemetered rain gauges are widely used to collect precipitation measurements. Spatial interpolation methods are commonly employed to estimate precipitation fields covering non-observed locations. Kriging is a simple and popular geostatistical interpolation method, but it has two known problems: uncertainty underestimation and violation of assumptions. This paper tackles these problems and seeks an optimal spatial interpolation for QPE in order to enhance spatial interpolation through appropriately assessing prediction uncertainty and fulfilling the required assumptions. To this end, several methods are tested: transformation, detrending, multiple spatial correlation functions, and Bayesian kriging. In particular, we focus on a short-term and time-specific rather than a long-term and event-specific analysis. This paper analyzes a stratiform rain event with an embedded convection linked to the passing monsoon front on the 23 August 2012. Data from a total of 100 automatic weather stations are used, and the rainfall intensities are calculated from the difference of 15 minute accumulated rainfall observed every 1 minute. The one-hour average rainfall intensity is then calculated to minimize the measurement random error. Cross-validation is carried out for evaluating the interpolation methods at regional and local levels. As a result, transformation is found to play an important role in improving spatial interpolation and uncertainty assessment, and Bayesian methods generally outperform traditional ones in terms of the criteria.

Improvement of Radar Rainfall Estimation Using Radar Reflectivity Data from the Hybrid Lowest Elevation Angles

A novel approach, hybrid surface rainfall (KNU-HSR) technique developed by Kyungpook Natinal University, was utilized for improving the radar rainfall estimation. The KNU-HSR technique estimates radar rainfall at a 2D hybrid surface consistings of the lowest radar bins that is immune to ground clutter contaminations and significant beam blockage. Two HSR techniques, static and dynamic HSRs, were compared and evaluated in this study. Static HSR technique utilizes beam blockage map and ground clutter map to yield the hybrid surface whereas dynamic HSR technique additionally applies quality index map that are derived from the fuzzy logic algorithm for a quality control in real time. The performances of two HSRs were evaluated by correlation coefficient (CORR), total ratio (RATIO), mean bias (BIAS), normalized standard deviation (NSD), and mean relative error (MRE) for ten rain cases. Dynamic HSR (CORR

A flexible and efficient spatial interpolator for radar rainfall estimation

ABSTRACT A key challenge in rainfall estimation is spatio-temporal variablility. Weather radars are used to estimate precipitation with high spatial and temporal resolution. Due to the inherent errors in radar estimates, spatial interpolation has been often employed to calibrate the estimates. Kriging is a simple and popular spatial interpolation method, but the method has several shortcomings. In particular, the prediction is quite unstable and often fails to be performed when sample size is small. In this paper, we proposed a flexible and efficient spatial interpolator for radar rainfall estimation, with several advantages over kriging. The method is illustrated using a real-world data set.

Dual Run-time Environments for Dual Data Memory Bank Architecture

Most vendors of digital signal processors (DSPs) support a Harvard architecture, which has two or more memory buses, one for program and one or more for data and allow the processor to access multiple words of data from memory in a single instruction cycle. We already addressed how to efficiently assign data to multi-memory banks in our previous work. This paper reports on our recent attempt to manipulate dual run-time environment. The run-time environment for dual data memory banks requires two run-time stacks to control activation records located in two memory banks corresponding to calling procedures. Unfortunately, several existing compilers use only single stack or fully static dual run-time stack. The former cannot utilize dual data memory banks, and the latter waste run-time memory. Therefore, we provide dual run-time environment based on stacks in this paper. The experimental results have revealed that our run-time environment utilize dual data memory banks efficiently and diminished usage of run-time memory