Highly Flexible Artificial Synapses from SiO₂-Based Conductive Bridge Memristors and Pt Nanoparticles through a Crack Suppression Technique

Abstract

The growth of flexible memory devices with intrinsic neuromorphic properties is attracting tremendous attention due to the wide spread of wearable electronics. The fabrication of robust artificial synaptic structures that will have the ability to remain operational after various strain loads and perform artificial synaptic weight modulation procedures is considered the cornerstone for the development of wearable neuromorphic computing systems. Along these lines, a low-temperature Ag/SiO 2 /TiN memristive device is demonstrated on a polyethylene naphthalate substrate decorated with Pt nanoparticles (NPs) with a 5 nm average diameter. The memory devices are completely forming-free, operate at a low voltage of 500 mV, and exhibit great stability after bending numerous times under a record high mechanical strain of 4.16% without any severe degradation of the pulse endurance and retention capabilities. Concrete pieces of evidence regarding the crack suppression effect induced by the layer of Pt NPs on the robust memory performance are also provided. Moreover, a thorough analysis of the neuromorphic properties of our prototypes is presented, including the implementation of short-term plasticity effects, such as paired-pulsed facilitation and paired-pulse depression, which are of great importance for the rehearsal procedures of neurons during the learning function as well as for the linear distribution of the conductance pattern during synaptic potentiation and depression tasks. Additionally, Hebbian spike-time-dependent-plasticity responses were also recorded.

Highly sensitive stretchable sensor combined with low-power memristor for demonstration of artificial mechanoreceptor properties

“…In parallel, our sensor exhibits similar mechanical properties to physical skin and operates in the same range of external stimuli as physical skin [50]. This step-like behavior of the sensor gives rise to another interesting feature when combined with a highly flexible memristor with intrinsic synaptic properties that have been previously reported [51]. Consequently, an artificial mechanoreceptor is demonstrated here for the conversion of mechanical stretch loads into electrical outputs that are consequently employed by synaptic elements for executing various neuromorphic functionalities.…”

“…With this in mind, a conductive bridge memory device was fabricated using a sputtered silicon oxide layer and a Ag electrochemically active top electrode fabricated at room temperature on a plastic substrate (polyethylene naphthalate). It is presented in figure 7 along with I-V hysteresis curves for various external mechanical strains of the device, while more details on its operation are given in [51]. We then show that the circuit emulates the functionalities of specified receptors beneath the human skin that have the ability to detect mechanical deformation stimuli [62].…”

Material design strategies for emulating neuromorphic functionalities with resistive switching memories

“…More specifically, the high surface roughness of Pt NPs (∼5 nm) could lead to the formation of sharp tips that can significantly enhance the local electric field distribution. 65,67) On the other hand, a different effect takes place for the VO x -based sample. Due to the bigger diffusivity value for Ag ion migration within this layer, 68) the operating current values are reduced below 1 μA.…”

“…The influence of the dense layer of Pt NPs, which was used as a bottom electrode, was examined for mimicking a wide range of artificial synaptic functionalities. 67,69) By taking into account that the manifestation of various neuromorphic properties is directly correlated with the growth mechanism of the percolating CFs, a reconfigurable synaptic pattern can be achieved by adjusting the effective diameter of the CF. Along these lines, the presence of NPs can impact the thermal accelerated migration of the ionic species, due to the local Joule heating distribution.…”

Memristor-based neural networks: a bridge from device to artificial intelligence

“…13 The composition of metal oxides is simple, easy to prepare, and compatible with complementary metal oxide semiconductor (CMOS) devices. 14 In addition, there are solid electrolyte materials with lattice defects and rapid ion migration, such as Ag 2 S, 15 SiO 2 , [16][17][18] and Cu 2 S. 19 Besides, organic materials with good flexibility and low cost, such as PVP, 20 PMMA, 21 and PFO, 22 have also been used to prepare memristive devices. Nowadays, the preparation of memristive devices can be achieved using many physical and chemical methods, such as atomic layer deposition, spin-coating method, magnetron sputtering, sol-gel method, pulsed laser deposition, and so on.…”