Organic Memristive Devices for Neuromorphic Applications
- PDF / 525,752 Bytes
- 5 Pages / 595.276 x 790.866 pts Page_size
- 88 Downloads / 222 Views
Organic Memristive Devices for Neuromorphic Applications Silvia Battistoni 1 Accepted: 4 November 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract The emulation of the hierarchical organization of the brain is nowadays considered a very promising approach for the realization of an efficient brain-machine interface and neuronal prosthesis. This bottom-up approach is possible only starting from non-classical electronic elements able to emulate synaptic functionalities such as long-term plasticity and short-term plasticity (STP). These elements then must be interfaced with technology able to mimic fundamental network properties (summation, transfer, and threshold). In this mini review, recent advances in the emulation of this hierarchical approach, obtained using a 3-terminal electronic device (organic memristive device) whose functioning is based on the redox activity of an organic thin film, are reported. Keywords Synaptic activity-dependent brain activities . Organic memristor . Artificial neuronal networks . Perceptron . Bioinspired circuits
1 Introduction The field of neuromorphic engineering is a relatively young field of a research that aims at the hardware implementation of neural processing, emulating biophysical processes that underlie neural computation in real neural systems [1]. Key features of biological neural processing are a lowpower consumption and the ability to carry out robust and efficient computation even in the case of limited precision, highly variable and unreliable components [1, 2]. Moreover, the communication between neurons has to deal with the synaptic modulation of the connections (synaptic weights) that follows the Hebbian learning principle: those that fire together, wire together [3]. This adaptation combined with the highly interconnected network of neurons that produces a massively parallel computing, are believed to be the basis of the synaptic plasticity that is responsible for both memorization and learning in the brain [3–5]. Clearly, the hypothesis of realizing a system with highly efficient neuromorphic properties has to pass the bottleneck of using a bioinspired architecture with energy efficiency and interconnectivity comparable to the brain. Moreover, every element constituent the network should provide features typical of synapses (i.e., synaptic plasticity) to mimic exactly the natural network organization. * Silvia Battistoni [email protected] 1
Istituto dei Materiali per l’ Elettronica ed il Magnetismo (IMEM-CNR), Parco Area delle Scienze 37/A, Parma, Italy
Recent results suggest that memristive devices could be considered key elements for modeling important features of biological synapses [6–10] at the hardware level thanks to their multiple programmable resistance states [2] and the possibility of scaling their dimensions to the nanoscale [1]. Implementations of typical synapses functions, such as spike-timing-dependent plasticity (STDP) [11, 12] and shortand long-term potentiation (LTP) [8, 13, 14], have been reporte
Data Loading...
