2019-02-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/685848摘要：本計畫以業界現有的產品為基礎，研究適用於視覺應用之高能效及高速的軟體及電路技術。其中子計畫一與業界的合作部分，採用目前最新的非機器學習演算法優化既有的深度圖，另外為了適用於行動裝置，功率消耗的問題也會在演算法及架構設計中納入考慮。而子計畫二以目前發展的高速傳輸介面系統 (如DRAM) 為基礎，解決其晶片、電路板與封裝結構的訊號及電源完整度的問題。另外子計畫三主要著重於設計讓固態硬碟聰明的分流資料存取負載，並將應用需求埋進固態硬碟之儲存子系統的 FTL層設計。最後子計畫四則是以USB3.1的規範設計訊號重整電路，並考慮未來商品量產的可行性，以符合業界標準。<br> Abstract: This project aims at the exploring of the power-efficient/high-speed software and circuit technologies for target vision-based application/system based on the existing products provided by the company. The first sub-project will apply state-of-the-art non machine-learning algorithm to increase the quality of the disparity map. To make it suitable for mobile devices, the power issue would be concerned in algorithm and architecture design. Then, the second sub-project provides solutions for signal/power integrity problems in chip-package-board for high-speed transmission, such as double-data-rate synchronous dynamic RAMs (DDR SDRAMs). In addition, the third sub-project will also propose hardware-adaptive FTL design so that it can efficiently exploit the limited hardware resource to achieve higher performance and parallelism degree. Finally, the fourth sub-project will develop a self-adaptive and regenerate signal re-driver based on USB 3.1 specification to fulfill the industry standard.深度圖優化機器學習消弭光照陰影影響高速介面傳輸訊號及電源完整度快閃記憶體智慧調適性儲存系統動態隨應用調適設計自我調適訊號失真驅動器能量獵取技術depth map refinementmachine-learningillumination invarianthigh-speed interfacesignal and power integrityflash memoryintelligent adaptive-storage systemapplication-adaptive designself-adaptivere-driverpower-harvesting technique