FEI：自主红外智能视觉系统的传感能量与信息融合处理.pdf

1、Tsinghua UniversityFEI:Fusion Processing of SensingEnergy and Information for Self-Sustainable Infrared Smart Vision SystemAsia South Pacific Design Automation Conference 2025(ASP-DAC2025)Innovative Techniques for Energy-Efficient and Reliable Hardware Systems SessionHaijin Su,Xin Hong,Maimaiti Nazh

2、amaiti,Ce Zhang,Li Luo,Qi Wei,Zheyu Liu,Wenjie Deng,Yongzhe Zhang,Fei QiaoOutlineBackgroundFEI:Fusion processing of sensing Energy and InformationThe IPC2SystemIPCp:Information Power CouplerIPCtrl:Intelligent Power ControllerSimulation ResultsConclusion2BackgroundInfrared smart vision systems are in

3、creasingly prevalent in the era of the Internet of Things.Widely deployed on the edgeAlways-on workingSense the ambient environmentPerform low-complexity but frequent tasksUnmanned AerialVehicle Navigation2Security Monitoring1Human BehaviorTarget Tracking3Significant power consumption.Power-limited

4、batteries bring huge cost to device maintenance.34BackgroundAdvanced low power processing architecturesEfficient energy harvesting methods to eliminate batteriesEnergy and Information are processed SeparatelyLower power consumptionHigher energy efficiencyDeeper integrationUltra-low-power always-on f

5、ace detector4Energy-Autonomous Sensor System5Fusion5BackgroundEnergy and information are deeply entwined,mutually constraining and complementing each other.FEI:Fusion processing of sensing Energy and InformationExploit the natural merit.Energy-efficient and self-sustainable.Enables energy harvesting

6、 and information processing on the same focal plane.6FEI:Fusion processing of sensing Energy and InformationSimultaneous energy harvesting and low power in-pixel-computing.Schedule the harvested energy to complete low power NN inference.Software-hardware co-design strategy to exploit the layer-wise