Memristors, also called RRAMs or resistive switching devices, have attracted tremendous attention as possible candidates for many applications such as neuromorphic computing hardware, next-generation memory cells, logic applications and security applications. The inherent memory effect in the simple two-terminal devices allows efficient data storage and parallel write/read-out system. Other properties such as high density, low power consumption, long cycling endurance and sub-nanosecond switching speed have been also demonstrated in memristor devices. However, conventional memristors suffer from unavoidable spatial-temporal variation due to uncontrollable, stochastic filament formation. The Lab is now developing a new strategy to achieve uniform switching through CMOS compatible materials/fabrication steps.
YTN Science News about new memristors - CLICK HERE!
Artificial Neural Network
In recent years, deep learning and artificial neural networks has achieved unprecedented accuracies in large-scale recognition and classification tasks by utilizing supercomputing resources. While several application-specific integrated circuit (ASIC) solutions utilizing CMOS have been previously proposed, limitations still exist on communication bottlenecks, energy consumptions and online learning capabilities. To address all issues in AI hardware, the community is moving towards utilizing memristor as artificial synapses because they can offer fast parallel neuromorphic computing at extremely small device footprint with low power consumption. The goal of this project is to develop large-scale neural network arrays for artificial intelligence (AI) hardware based on new design of artificial synapses (memristors).
Integrated Systems Development
Another major focus of the ENTIS lab would be the integration of intelligent systems from input sensors to computing units. By utilizing memristor-based computing systems, the team will demonstrate fully integrated systems from artificial neurons (CMOS) to artificial synapses(memristors). The team will also involve in on-body decoding system from neural probing and security applications using biometrics.
Memristor-based deep learning and AI hardware can be utilized to various application where articifial intelligence is needed. Several parameters, such as accuracy, heat dissipition and power consumption, should be analyzed by both simulation and experiment for the certain applications.