Energy-efficient hardware implementation of memristor-based in-memory computing

Neuromorphic engineering is an emerging bio-inspired discipline that morphs the biological brain on custom silicon.

Although memristors rose as a potential synapse to solve the density challenge in a memristive crossbar, the scalability of the crossbar is limited by its power dissipation and chip area.

To contribute to low power dissipation, I focus on improving the energy efficiency of the synapses (non-filamentary category) and neurons, which are the fundamental constituents of the neural network-enabled IOTs.

The energy efficiency of the synapses will be improved by applying fast switching pulses on the bulk-based synapses that result in low switching currents.

The energy efficiency of the neurons will be improved by taking advantage of the FDSOI28nm technology node by which the neurons are designed.

A dedicated PCB will be designed, assembled and mounted that house both the synapses, and the neurons, which will be energy-efficiently used to recognize digits or letters by unsupervised learning rule.