事件驅動能源控制之高能源效率氣體辨識系統

Abstract

在以感測器為主的應用中，針對應用所設計的系統可以有效地實現一個低耗能系統。特別是在電子鼻的應用上，低耗能氣體辨識系統是非常重要的議題。本論文提出一應用於金屬氧化物感測器陣列的事件驅動氣體辨識系統，並結合了高能源效率的概念，進行演算法暨架構共同設計。所提出的氣體辨識系統可以偵測四種不同氣體，並對其氣體進行濃度分析；在氣體類型辨別上可達到100%的準確率，而在氣體濃度分析中，準確率亦高達89.4%。 為更進一步降低能量消耗，本論文採用了電源管理技術，並將一可操作在接近閾值電壓的靜態隨機存取存儲器(SRAM)與一可在低電壓運作的嵌入式可變電阻式記憶體(ReRAM)分別整合進所提出的氣體辨識系統。在65奈米低功耗製程下，一個氣體感測期間所消耗的總能量分別為8.62微焦耳及2.04微焦耳。透過非揮發性記憶體ReRAM，本論文實現了一個極微量漏電電流超低耗能的氣體辨識系統。本論文亦將此系統實現於Lattice MachXO2-1200ZE的現場可規劃邏輯陣列(FPGA)晶片上，搭配時脈閘控技術，可以減少43%的動態功率消耗。

For sensor-based applications, an application-specific architecture is necessary to realize an energy-efficient system. Particularly in the application of electronic nose, a low-energy gas recognition system is essential. For metal oxide sensor array, the energy-efficient algorithm-architecture co-design of gas recognition system with event-driven pseudo-zero-leakage structure is proposed. The proposed low energy gas recognition system can recognize four different gases with concentration analysis, achieving 100% recognition accuracy for gas type and 89.4% accuracy for concentration analysis. To achieve energy efficiency further, a near-threshold SRAM, and a low-voltage embedded ReRAM are integrated into the proposed system separately. We analyze the system in TSMC 65nm LP CMOS process, and the total energy in a sensing period is 8.62uJ, and 2.04uJ, respectively. Especially, we can utilize a ReRAM to realize a pseudo-zero-leakage recognition system, an ultra-low-energy architecture.