ABSTRACT
A large bottleneck for energy-efficiency has long been the information storage units, which typically rely on charge-storage for programming and erasing. HfO2-based ferroelectric tunnel junctions (FTJs) store data as a change in polarization state, which is read as a modification of resistance state, and represent a unique opportunity as a next-generation digital non-volatile memory, while also being complementary metal-oxide-semiconductor-compatible. FTJs are particularly promising as synaptic devices for neuromorphic computing applications. Neuromorphic, or brain-inspired, computing is an emerging paradigm that mimics the operation of the human brain, which applies very well to solving complex multi-dimensional or temporal classification and regression tasks including video, image-recognition, audio-processing, and deep learning. Neuromorphic computing is only one of many applications for which memristive implementations can improve energy efficiency. The FTJ, by nature, responds desirably to the overlapping waveforms, simplifying the design and improving energy-efficiency.
