Advanced Semiconductor Materials and Devices Group

Computing devices built on von Neumann architecture suffer from “memory wall” problem, often known as the von Neumann bottleneck limitation which leads to temporal latency, high power consumption, and a high risk of data loss while handling enormous amounts of data resulting in reduced data bandwidth. Brain inspired neuromorphic computing is an emerging technological application which requires handling of massive amount of data along with high clocking speed, making devices based on von Neumann architecture unsuitable. To circumvent this challenge, modern day computing devices follow processing-in-memory (PIM) architecture which fuses the memory module with CPU to reduce or eliminate the frequent data transmission. Memristors enable true PIM as a fundamental device which supports both storage and data computation.

Memristor is a two terminal device non-volatile memory device whose resistance can be precisely modulated depending upon the history of electrical current passed through it. Human brain is composed of 10¹¹ neurons which are interconnected through 10¹⁵ synapses. Biological neurons play a crucial role in processing signals related to the reception, integration, and transmission of sensory and perceptual information. These neurons encode information through action potentials, which are electrical signals in the form of spikes. Neurons collect input signals from pre-neurons through their dendrites. Subsequently, the integrated signals in the soma trigger output signals, which are transmitted along the axons to post-neurons when the sum of input signals surpasses a certain threshold. The biological synapse, found between a neuron's axon and dendrite, facilitates the transmission of signals. The metal electrodes in memristors can be visualized as the pre- and post-synaptic neuron, while the channel separating them as the synaptic cleft. Therefore, like synapse, electric signal transferred through the device can be controlled through its resistance.

Memristors have been classified under four categories depending upon their operation mechanism:

1.     Ionic Migration, where a material is sandwiched between two metal electrodes and the resistive switching occurs due the formation of a conducting filament between the active metal electrodes. Their operation can be explained through electrochemical metallization (ECM)or through valence change mechanism (VCM).

2.     Phase Change, where phase transition of the switching material governs the switching mechanism, and the device is composed of a phase change material sandwiched between two metal electrodes.

3.     Spin, where a non-magnetic layer is sandwiched between two ferromagnetic layers. Here, the relative direction of magnetization of these ferromagnetic layers governs switching mechanism.

4.     Ferroelectric, which follows metal/ferroelectric-film/metal device architecture. Here, polarization of charges in the ferroelectric film governs the switching behaviour.

Memory-intensive applications also demand high density device integration, which necessitates a nano-scale downscaling of the devices' dimensions. To address this issue, 2D materials are being investigated for low power memory device applications.  Rhenium Disulphide (ReS₂) is a layered semiconducting group VII TMDC exhibiting distorted 1T’ crystal structure. Weak interlayer bonding, soft Re-S bonds, and high probability of sulfur vacancies due to their low forming energy makes ReS₂ a potential material for non-volatile resistive switching (NVRS) device application.

Our research group specializes in manufacturing memristors using chemical vapor deposition grown 2D thin films. In our lab, we conducted experiments on a CVD grown ReS2 film to assess its suitability for resistive switching device applications. We investigated the impact of various metal electrodes (Pt/Au and Ag/Au) as well as different channel widths (200, 100, and 50 μm). The observed resistive switching behavior in devices with Ag/Au electrodes was attributed to the Electrochemical Metallization (ECM) mechanism, whereas in devices with Pt/Au electrodes, it was ascribed to defect-mediated charge transport. (https://doi.org/10.1039/D3NR02566G)

Figure: (a) Schematic depicting the formation of conducting filament due to movement of metal ion and defect mediated charge carrier transport. (b) Comparative RS cycle of memristor with Ag/Au and Pt/Au electrode. Ag/Au showed 10⁵ times higher I_ON/I_OFF ratio and reason is attributed to electrochemical active nature of Ag electrodes. (c) I-V curve depicting the effect of channel width (200, 100, and 50 µm) on memristive behaviour. Highest switching ratio for 50 µm is attributed to the ease of formation of CF due to reduced distance between the metal electrodes.