Optimizing synaptic properties of Gd(1−x)CaxMnO3 memristor devices

Abstract

The gap between memory and processing power in traditional von Neumann architectures is enlarging, and there is a great need for new computing architectures to meet the demand of more and more data centric applications of modern day. Neuromorphic computing is a promising biology inspired computing architecture that mimics the brain and nervous system. It is a necessity to find new components that would allow this structure of synapses and neurons to be built. Memristors have been found to act like synapses, which makes them a promising component for artificial synapsis in neuromorphic computing. In this thesis I studied Gd0.3Ca0.7MnO3 (GCMO) memristive thin films and their synaptic properties. The goal of the thesis was to firstly test if these thin films could be used as a artificial synapse and secondly optimize the pulse shape if they work. I measured the spike-timing-dependent plasticity (STDP) of the thin films using two different pulse shapes and multiple different voltages. I then simulated neural networks that used synapses based on the GCMO thin films. The synapses were benchmarked based on the image recognition accuracy of the neural networks. My measurements and simulations showed that GCMO synapses are very likely to work in real world applications. The neural networks trained in this thesis are using identic synapses that are based on a single measurement result. The results are very promising, but further research is needed to mitigate the device-to-device variation. My results also show the optimal pulse shape and voltage from the ones studied. Relation between STDP parameters and accuracy was found, which will help in future optimization of the system. There is very little research published on the synaptic properties of GCMO devices, unlike other perovskite oxide thin films like Pr(1−x)CaxMnO3. There is no published STDP results for GCMO thin films. The main purpose of my thesis is to deepen the knowledge about the synaptic properties of the GCMO thin films, that could be possibly used in future applications.