Closed access
Author
Date
2022Type
- Habilitation Thesis
ETH Bibliography
yes
Altmetrics
Abstract
The increasing rates of data generation, collection, and transfer over collective networks - the Internet of Things (IoT) – are shaping an interconnected society progressively depends on the availability of efficient computing and communication systems. For the last 50 years, the evolution of computational architectures has been largely driven by Moore’s scaling law and the breakthroughs of the CMOS technology. However, the continuous exchange of data between physically sperated memory and computing units leads to limited transfer rates, known as “von Neumann bottleneck”, which is becoming a severly limiting factor in today’s hardware architectures. As a consequence, novel energy efficient ‘beyond von Neumann’ configurations are being explored to tackle complex computational tasks, with particular attention to the processing of large amounts of data.
Novel nano-devices called memristors appear as a promising option to enable ‘beyond von Neumann’ hardware paradigms at the core of future circuit implementations. Memristors are two-terminal solid-state devices with a tunable (non-volatile) electrical conductance. Their functionality is based on electrochemical effects such as faradaic redox and/or local ionic/atomic transport localized in nanoscale domains that can be modified under the influence of external stimuli: optical, electrical, or mechanical. Memristors exhibit a unique combination of attributes including (i) potential atomic-size scalability, (ii) intrinsic non-volatility, and (iii) low switching energy consumption (~fJ/bit). Through atomic displacements, the resistance of these devices can be incrementally tuned by successive programming pulses. These properties unlock the potential of memristors not only for efficient communication and computational kernels, but also to emulate biological synapses [8] (in terms of analog conductance modulation), thus enabling the co-location of memory and computing units. These are key ingredients for brain-inspired (neuromorphic) computational architectures.
In this habilitatation thesis, we present our results on electro-optical memristors that could potentially lay the basis for a new disruptive neuromorphing platform for communication and computation. More precisely, electro-optical memristors have been integrated within a plasmonic/photonic circuit to perform memristive modulation, photodetection, and light-induced efficient computation. Show more
Publication status
publishedPublisher
ETH ZurichOrganisational unit
ETH Zürich
More
Show all metadata
ETH Bibliography
yes
Altmetrics