Energy harvesting is a technique which converts ambient wasted energy into usable electrical energy. Generated electrical energy can be utilized to drive wireless, wearable and implant device which cannot directly supplied by line. With the remarkable growth of wireless sensor network, the energy harvesting becomes more important. Among various energy harvesting techniques, triboelectric nanogenerator have attracted great deal of attentions based on its cost-effectiveness, design flexibility and high output power. Our group has outstanding know-hows in this fabrication and characterization of these nanogenerators.

By utilizing energy harvesting devices as a sole power source, various self-powered systems can be implemented. Because these system does not require any other power supplier or battery, the systems have superior effectiveness and lifetime (Ex. self-powered water treatment system based on wind energy harvesting device).

Apart from power source application, energy harvesting devices can be utilized as a self-powered sensor unit because those devices can generate electrical signal by itself when external excitation is applied to the devices. With this sense, various novel self-powered sensor systems like fall detection system and wearable fabric touch pad have been developed.

With the recent growth of artificial intelligent, hardware realization of Deep Neural network is becoming important, DNN is the one of the most popular computing algorithm for many pattern classifications. It consists of multiple layer and synapses which have individual weight. With this network, the answer is inferred by matrix operation of input value and weights then the error values are calculated by comparing the inferred answer to the desired value. Then the error values are propagated in backwards and each weight is corrected based on the error and layer values. After numerous iterations, the DNN becomes to accurately classify the patterns. If we can make each weight by utilizing single device, the overall complexity and power consumption can be highly reduced. Simple devices such as RRAM and FeRAM can have multiple conductance levels over 5 bits according to the number of input pulse then it can be a powerful candidate for synapse device of DNN. In this sense, various novel synapse devices have been developed and system level validation was also conducted via own simulation tool.