In materia reservoir computing with a fully memristive architecture based on self-organizing nanowire networks

Neuromorphic behaviour in discontinuous metal films

“…Recent demonstrations of RC in networks of nanowires 29–32 as well as in other systems such as all optical networks 56,57 confirm the potential of networks of brain-like components for information processing, and a next step in the field is to more carefully consider practical considerations such as power consumption and ease of fabrication. In the present (unoptimised) devices power consumption is in the range 10–100 mW, which is comparable to that in ref.…”

“…14 These complex all-metal self-assembled networks exhibit interesting memristor-like switching behaviour [17][18][19][20][21][22] and correlations that are promising [23][24][25] for realizing brain-inspired computational approaches such as reservoir computing (RC). [26][27][28] RC is just one of many possible approaches, but it is a useful example because of its conceptual simplicity and because recent successful demonstrations in nanoscale networks [29][30][31][32] mean that the approach is becoming well-established. The basic concept is that a reservoir of highly interconnected non-linear nodes (which evolve dynamically in response to input signals) can be used to map input signals onto outputs of higher dimensions, which are then examined by an external readout function in order to perform a variety of tasks.…”

Grid-graph modeling of emergent neuromorphic dynamics and heterosynaptic plasticity in memristive nanonetworks

“…Short-term dynamics and heterosynaptic plasticity effects of NW networks endow the system the capability of processing spatio-temporal inputs in multiterminal configuration for the implementation of reservoir computing, where the reservoir state is represented by the network conductivity map [16]. In this framework, the model that is able to reproduce both temporal dynamics and heterosynaptic behavior of the reservoir state for arbitrary input voltages (including pulses) can be exploited for simulating different strategies of computing implementation to solve both static and time-dependent tasks.…”

“…These kinds of self-organized systems including NW networks, percolating nanostructured networks and atomic switch networks, are characterized by a structural topology that is more similar to biological neural networks and have been shown to exhibit structural plasticity, homo-and hetero-synaptic plasticity, paired-pulse facilitation (PPF) and collective neural-like dynamics [5][6][7][8][9][10][11][12][13][14][15]. These dynamics have been exploited for the implementation of unconventional computing paradigms such as reservoir computing [16][17][18][19][20][21]. While top-down realized crossbar arrays several models have been developed, the development of models able to describe the emergent behaviour of selforganized memristive networks at the macroscale still represent a challenge.…”

In Materia Should Be Used Instead of In Materio

“…Soon after, in materio started being associated especially with physical reservoir computing implemented on designless nanonetworks (van Damme et al, 2016;Dale et al, 2017aDale et al, , 2017bPrzyczyna et al, 2020;Banerjee et al, 2021;Boon et al, 2021;Kotooka et al, 2021;Lilak et al, 2021;Usami et al, 2021). Very recently, the same meaning was attributed to a slightly different Latin expression: in materia (Milano et al, 2021). Which one is correct?…”