Characterisation & modelling of perovskite-based synaptic memristor device
Introduction
In the modern era, parallel computing architectures play a significant role in high performance and high-speed processing of data. This parallel computing paradigm is used in artificial intelligence to undertake various operations, complex analysis, and high-speed processing of data with a high density of integration. Architecture based on complementary metal-oxide-semiconductor (CMOS) based devices is an optimal choice for a circuit designer up to 90 nm technology node [1]. By following Moore's law, technology is shrinking down, reaching up to the 7 nm, where transistor-based devices suffer from very high leakage current and consequent power consumption issues which hinder high scale integration of neuromorphic devices [2]. Memristor is an excellent choice due to its numerous advantages over transistor-based technology. The Memristor device has a high density of integration [3,4], and a high operating speed [5] with a low power dissipation [6] over transistor-based devices. Memristor is also a memory device, having at least two stable states of the resistance value. The two stable states of the resistance are represented by 0 and 1 and are associated with the high resistance and the low resistance states respectively [7]. The resistance of the memristor device depends on the applied input voltage with regard to the set and reset voltage i.e. Vset and Vreset respectively. This device shows a pinched hysteresis loop in the I-V characteristic along with the switching between two stable states [8]. The changed behaviour of the states can be seen by applying the Vreset or negative voltage for the transition from low resistance state (LRS) to high resistance state (HRS) and Vset or positive voltage for the transition from high resistance state (HRS) to low resistance state (LRS) [9]. This state change behaviour of memristor device in I-V characteristic is shown in Fig. 1. The reproductive behaviour of conductance/resistance of the memristor device depicts the synaptic plasticity, which is the basic remembering and learning process for the neural network.
This memristive behaviour has been analysed in various materials such as TiO2 based devices, various perovskite-based devices including inorganic oxide material, and hybrid organic-inorganic perovskite materials. Perovskite-based solar devices have reached outstanding power conversion efficiencies [10], and have the advantages of low cost and easy low temperature processability. Lead halide hybrid organic-inorganic perovskite semiconductors demonstrate excellent optical and electrical properties, a high absorption coefficient, long electron/hole diffusion length in the visible spectrum, and low excitation binding energy [[11], [12], [13]]. Due to the above-mentioned properties, lead halide hybrid organic-inorganic perovskite materials are extensively used to fabricate devices for various applications such as solar cells, photodiodes [14], light-emitting diodes [15], and lasers [16].
Parallel computing-based architecture is in great demand in the neuromorphic application as it has many advantages over conventional computational architectures. Progress in the development of artificial neuromorphic systems remain still faces the challenges of defining efficient electronic synaptic devices. However, recent reports show the promising use of hybrid organic-inorganic perovskite material for implementing synaptic devices under sun illumination [17]. Perovskite-based synaptic devices have the potential to mimic all the biological functions with low energy consumption of the order of fJ/100 nm2 per event, which is close to the energy consumption of biological synapse. The human brain has a complex structure of billions of neurons in which every neuron has a direct synaptic connection with approximately 10,000 neurons [17].
The artificial synaptic device performs all the functions of a biological synapse. These artificial devices are made up of a memristor device i.e. analogues to the biological synapse. Here, the top and bottom electrodes of the memristor device represent the pre- and post-synaptic neurons, respectively. In an artificial synapse, the conductivity/resistivity of the memristor device symbolises the synaptic weight and variation in conductivity/resistivity represents the potentiation and depression similar to the biological synapse where synaptic weight is the connection strength between neurons. Fig. 2 displays a biological synapse composed of a pre-synaptic neuron, a post-synaptic neuron, and a synapse. The variation (increment/decrement) in conductivity defines potentiation/depression behaviour of synaptic weight in response to the applied spikes [18].
Previously, it has been reported that ion-conducting organometal trihalide perovskites (OTPs) are promising candidates for memristors and synaptic devices. Zhengguo Xiao, et al. [18] proposed a device based on methylammonium lead iodide (MAPbI3) with ITO/CH3NH3PbI3/Au structure which was able to mimic the behaviour of a biological synapse, moreover showing several advantages, such as low cost fabrication, low operating energy, and low energy consumption as compared to other materials, i.e. oxides and Si-based materials. Other hybrid materials, such as MAPbI3, FAPbI3 (formamidinium lead iodide), MAPbI3Cl3-x, and inorganic lead halide perovskites, like CsPbI3, also show the mechanism based on ion conduction and can be therefore used as memristors. Perovskite-based devices show the typical hysteresis in the I-V characteristics under positive and negative bias scanning under illumination [18,19] and dark conditions [20], in all these cases the scanning rate was 0.1 V/s.
This article builds on our previous work (reported in [20]) providing an in-depth analysis for the characterisation of a perovskite (CH3NH3PbI3) based device as applicable to memristor and synaptic behaviours in dark conditions. In this paper, we also present a new physics-based circuit-level SPICE model for CH3NH3PbI3 based perovskite device. This proposed SPICE model is validated through experimental results reported in [20]. The simulation results for the memristive and synaptic behaviours are closely related to the experimental data. This SPICE model is essential for the behavioural analysis of complex and large-scale network simulations in an artificial neural network. The perovskite-based device with Glass/indium tin oxide (ITO)/tin oxide (SnO2)/CH3NH3PbI3/gold (Au) architecture reported in [20] operates at low energy due to the addition of SnO2 layer and also shows stable and uniform behaviour after the 15th day of fabrication.
The rest of the paper is organised as follows: Section 2 describes the fabrication of perovskite-based (Glass/ITO/SnO2/CH3NH3PbI3/Au) device [20]. Device characterisation with experimental results [20] is discussed in Section 3. Modelling and simulation of the Glass/ITO/SnO2/CH3NH3PbI3/Au architecture as a memristor device are explained in Section 4. Simulation results illustrating the synaptic behaviour of perovskite (CH3NH3PbI3) based device are presented in Section 5. Energy-efficient device for memory and neuromorphic application is discussed in Section 6. Finally, Section 7 concludes the manuscript.
Section snippets
Device fabrication
For the purpose of our analysis, we fabricated a planar structure device using CH3NH3PbI3 film, as reported in [20]. The CH3NH3PbI3 layer is sandwiched between ITO/SnO2 and a gold electrode [21,22]. We introduce a SnO2 electron transporting layer (ETL) as compared to the commonly used TiO2 [[23], [24], [25]], since SnO2 has higher electron mobility and a wider energy band gap. Fig. 3(a) [20] exhibits the cross-section view of the Glass/ITO/SnO2/CH3NH3PbI3/Au device that was fabricated and used
Experimental results
Typical memristive behaviour has been observed in the Glass/ITO/SnO2/CH3NH3PbI3/Au device as presented in [20], and shown in Fig. 4. To characterise the memristive behaviour of the device, the voltage was applied to the Au electrode while the ITO was left grounded. The applied dc sweep voltage was 0 V → +1.5 V → 0 V → -1.5 V → 0 V. All the parameters used in the characterisation are shown in Table 1 [20]. All the I-V characterisations were performed using an Autolab 302N Modular Potentiostat
Modelling & simulation
Memristor behaviour can be described through the following equations [40]:and
Here, Eq. (1) represents a simple I-V relation for the resistive device, where the current through the memristor (I) depends on the voltage (V) and the internal state variable of the device is represented by W. The state variable W of the memristor depends upon the present input state and the previous internal state of the device, which is distinct from the other normal resistive devices. Eq. (2)
Synaptic behaviour
It has been formerly reported that halide containing perovskite is a good ion conductor. Conduction in the perovskite layer is due to the ion migration, resulting in the formation of the p-i-n diode. When consecutive positive pulses (0 to +1.5 V) were applied on the device with a scan speed of 0.1 V/s, the amplitude of the dark current increased continuously. This shows that the positive bias scanning progressively changes the device into the p-i-n polarity due to the positive train of pulses,
Energy consumption
Resistive switching behaviour has been distinguished in hybrid organic-inorganic perovskite synaptic device, due to p-i-n type structure caused by the ion migration-induced doping. Ion-migration based mechanism is different from the previously reported mechanism such as phase change, ferroelectricity, ion-motion-induced resistive change and the atomic conduction bridge, etc., which underlie various types of synaptic devices. Such devices require a very low level of doping concentration, only in
Conclusion
A hybrid organic-inorganic perovskite (CH3NH3PbI3) based memristor device (Glass/ITO/SnO2/CH3NH3PbI3/Au) has been fabricated and exemplified in [20] showing a reliable synaptic function under dark condition. Perovskite materials are ideal for memristor and synaptic devices due to their solution-processability, low cost fabrication, low activation energy for ion migration and formation of reversible p-i-n structure. Based on the potential of this device, this paper proposed an in-depth analysis
CRediT authorship contribution statement
Vishal Gupta:Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Visualization.Giulia Lucarelli:Methodology, Validation, Formal analysis, Investigation, Writing - review & editing.Sergio Castro-Hermosa:Formal analysis, Investigation, Writing - review & editing, Funding acquisition.Thomas Brown:Conceptualization, Formal analysis, Resources, Data curation, Writing - review & editing, Supervision, Project
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
We thank the funding agencies, University of Rome Tor Vergata's “Mission: Sustainability” “BiCVision” project, LazioInnova Gruppi di Ricerca project no. 85-2017-15373 SIROH, and Departamento del Huila's Scholarship Program from Huila, Colombia, under grant 677 for financial support. We would also like to thank the anonymous reviewers whose insightful comments helped to improve this manuscript.
References (58)
- et al.
Recent progress in resistive random access memories: materials, switching mechanisms, and performance
Material Science & Engineering R
(2014) - et al.
Highly efficient perovskite solar cells for light harvesting under indoor illumination via solution processed SnO2/MgO composite electron transport layers
Nano Energy
(2018) - et al.
Activity-dependent synaptic plasticity of a chalcogenide electronic synapse for neuromorphic systems
Sci. Rep.
(2014) - et al.
Neural synaptic weighting with a pulse-based memristor circuit
IEEE Transactions on Circuit and System I: Regular Papers
(2011) - et al.
Photoelectric plasticity in oxide thin film transistors with tunable synaptic functions
Advanced Electronic Materials
(2018) - et al.
Implementation of complete Boolean logic functions in single complementary resistive switch
Sci. Rep.
(2015) - et al.
Process variability aware low leakage reliable nano scale double-gate-FinFET SRAM cell design technique
J. Nanoelectron. Optoelectron.
(2015) - et al.
Reliable Ge2Sb2Te5 integrated high density nanoscale conductive bridge random access memory using facile nitrogen doping strategy
Advanced Electronic Materials
(2018) - et al.
Memristive devices for computing
Nat. Nanotechnol.
(2013) - et al.
Flexible all-inorganic perovskite CsPbBr3 non-volatile memory device
ACS Appl. Mater. Interfaces
(2017)