| 小河 亮 | 1451031: M, 2回目発表 | 井上 美智子，池田 和司，大下 福仁，米田 友和，大和 勇太 |
|
Title: Burn-in Test time reduction by using data mining approach
Abstract: The Burn-In (BI) test process is one of important Very Large Scale Integration (VLSI) test processes to screen out early-life failure VLSI. However, not only it takes too much time and cost to conduct BI test, but also VLSI deteriorates through the BI because of its characteristics. The data mining approach is expected to omit the BI process from VLSI test processes. It has the possibility to predict the result of BI test by using test data (current, voltage frequency, etc.) that are collected before BI. In my presentation, I will explain my research progress. Language of the presentation: Japanese | ||
| YUTTAKONKIT YUTTAKON | 1461024: D, 中間発表 | 中島 康彦，井上 美智子，高前田 伸也，TRAN THI HONG |
|
title:Performance and Energy Efficiency Evaluation and Optimization of Lightfield Application on GPU
abstract: Lightfield application has been largely implemented on many field from mobile device to industrial. Since battery driven devices is one of the most applied device, energy consumption and power efficiency has to be concerned. Therefore, we implement and evaluate performance and power efficiency characteristic of lightfield application mainly on GPU and multicore processor. Power efficiency result indicates that for the mobile device, GPU is nominated both performance and power efficiency. Besides for desktop and workstation, GPU performed better for low data dependency application such as rendering and lightweight depth extraction but not clearly outstanding with general depth extraction. However, GPU shows more versatile for hardware scaling benefited from low system overhead. Furthermore, we analyzed the capability for incoming technology as 144 megapixels data input, which we found that current mobile GPU cannot practically be used for depth extraction. We investigate for device improvement which we found the power efficiency improvement method for rendering. Due to performance is bounded by fix ratio of memory and core clock at 1.6x, we can step tuning the device for better efficiency. To increase the performance, we use information from multiple GPUs to analyzed the configuration's properties correlation. For rendering, which is found as a memory bound application, can take benefit of faster memory by applying the GDDR5 memory, the performance gained is predicted to be high as 8.8x. Also for lightweight depth extraction, we found that increasing the core clock can increase the performance to 1.5x. language of the presentation: English | ||
| 井上 万実 | 1451011: M, 2回目発表 | 笠原 正治，井上 美智子，大岩 寛，Cyrille Artho，川原 純 |
|
Ttitle: Performance of Heavy Chain
Abstract: Bitcoin has attracted considerable attention as a new virtual currency , and the population of Bitcoin users has grown faster than other virtual currencies. However, the processing speed of Bitcoin transactions is much lower than that of conventional payment systems such as VISA. In this research, we show " heavy chain", a way of shorting the wait time of each transaction and processing more transactions. We explain how the transaction-validation time is affected by heavy chain. language of the presentation: Japanese | ||