Dissertation Defense

Metal Oxide Memristors with Internal Dynamics for Neuromorphic Applications

Chao DuPhD Candidate

Metal oxide memristors, a two-terminal nanoscale semiconductor device whose resistance/conductance can be regulated according to the history of applied stimulations, are initially proposed as a promising candidate for the next generation non-volatile memory. Bearing the similarity to the weight change of synapses in human brain, they are recently being intensively investigated as a critical component in neural network for neuromorphic applications.

The resistive switching mechanism is attributed to the redistribution of oxygen vacancies under electric field and spontaneous diffusion. Based on this understanding, 2nd order switching dynamics is discovered and thoroughly investigated for the first time in both WOx memristor and Ta2O5-TaOx memristor and more comprehensive resistive switching models are proposed to quantitively capture the internal ionic dynamics. The dynamics is utilized to implement important synaptic functions including paired pulse facilitation, spike-timing dependent plasticity, experience dependent plasticity, in single cell and in a bio-realistic fashion. WOx memristor crossbar network is used to implement several important neuromorphic applications including: 1) sparse coding, as the network can easily conduct matrix operation, especially dot product and the resistance of each cell at the crosspoint can be regulated to store information needed for computation, 2) temporal information processing through memristor-based liquid state machine, as WOx memristor has the ability to process temporal information due to its short-term memory which is caused by its spontaneous decay characteristics. Improvement of both single cell performance towards better synaptic behaviors and memristor crossbar network performance for large scale applications are achieved by the optimization of fabrication methods.

Sponsored by


Faculty Host

Wei Lu