Potential of vanadium dioxide for artificial sensory neurons
(2024)
Files
phd_thesis_noemie_bidoul_final.pdf
Open Access - Adobe PDF
- 41.49 MB
Details
- Authors
- Supervisors
- Flandre, Denis
- Abstract
- Growing volumes of data are produced by sensors and connected devices, requiring increasing physical resources to process them. A first solution would be to rationalize data production by questioning its relevance in each application. Simultaneously, we can design energy-efficient systems to generate and process information. The field of neuromorphic engineering does so by taking inspiration from the biological brain, which is able to perform complex inference tasks on real-time stimuli at low energy cost. This efficiency is partly attributed to the encoding of sensory information in the timing of electrical spikes. This paradigm leverages the sparsity of sensory events, as neurons are mostly silent in the absence of relevant information. In this thesis, we harness the phase-change properties of vanadium dioxide (VO2) to design devices emulating sensory neurons behavior. Thanks to their temperature-dependent resistive switching properties, these devices encode temperature stimuli into the timing of electrical spikes. More specifically, our work investigates four key limitations that presently hinder VO2 artificial neurons development. First, the inherent stochasticity of the resistive switching, which we minimize through design parameters. Second, the way this stochasticity impacts the neuron resolution, which we unravel and optimize. Third, the limited sparsity of common encoding techniques, to which we propose an alternative. Last, the lack of knowledge on the environmental impact of the device fabrication, which we quantify through Life Cycle Assessment. From the material synthesis to the device operation in a compact electronic circuit, our work provides guidelines on how to design VO2 artificial sensory neurons with minimized energy consumption and reduced direct environmental impacts.
- Affiliations
Citations
Bidoul, N. (2024). Potential of vanadium dioxide for artificial sensory neurons. https://hdl.handle.net/2078.5/234253