“Atomristors” could be denser than human synapses, say researchers
Reporting their results in ACS’ Nano Letters in a paper titled “Atomristor: Nonvolatile Resistance Switching in Atomic Sheets of Transition Metal Dichalcogenides”, the researchers extrapolate that the “atomristors”, as they describe their devices, could yield memristor densities in the range of 1015/mm3, which would translate to a theoretical areal density of 6.4Tbit/in2 for single-bit single-level memory storage.
Although it had been observed that a number of solution-processed multilayer two-dimensional (2D) material morphologies could yield a nonvolatile resistance switching (NVRS) behaviour, where a device’s resistance can be modulated between a high-resistance state (HRS) and a low-resistance state (LRS) and retain those states without power, it is the first time that such a NVRS behaviour is observed in atomically-thin vertical metal–insulator–metal (MIM) devices, claim the researchers.
electrodes (TE and BE) could be gold, while the TMD
could be MoS2.
To build their “atomristors”, the scientists prepared synthetic atomic sheets of transitional metal dichalcogenide (TMD) such as MoS2, MoSe2, WS2 and WSe2, using standard chemical vapor deposition (CVD) and metal–organic CVD processes. They then transferred the atomically-thin sheets to sandwich them between different types of electrodes including silver, gold and even graphene. Nonvolatile resistance switching was observed in all cases. The crossbar devices consisted of TMD atomic sheets between top and bottom electrodes, on top of a Si/SiO2 device substrate.
In a particular implementation, with 2D MoS2, the atomristor had a high-resistance state (low currents measured) until a 1V bias was applied to set the atomic-layer switch to a low-resistance state. The device retained it resistance value until a negative bias was applied to reset it. In this vertical metal–insulator–metal configuration, which lends itself well to 3D integration, the researchers measured an on/off ratio over 104 and truly zero-static power for data retention in ambient conditions.
They also highlight that atomristor devices offer distinct advantages in terms of ultimate vertical scaling, down to an atomic layer with forming-free operation. By replacing the metal electrodes with graphene, the entire memory cell could be scaled below 2nm.
“Moreover, the transparency of graphene and the unique spectroscopic features of 2D materials affords direct optical characterization for in situ studies and in-line manufacturing testing”, they write in the paper.
Data endurance is still something that need to be improved, but memory retention was tested up to a week, which could be already sufficient for certain neuromorphic applications involving short and medium term plasticity. In fact, due to the sub-nanometre thinness of the monolayers, they extrapolate that at a loose pitch of 10nm, an atomristor density of 1015/mm3 could be achieved. Those could be used to mimic human synapses (about 109/mm3) in neural networks.
According to the authors, designed as a single-bit single-level memory device, these stacked atomristors would yield a theoretical areal density of 6.4Tbit/in2.
in monolayer nanomaterials layered into an “atomristors”.
Credit: Cockrell School of Engineering, The University of
Texas at Austin.
Another potential application that the researchers anticipate for their atomristors is zero-static power radio frequency (RF) switching. In fact, in one of their experiments, they demonstrated a monolayer switch operating to 50GHz with acceptable insertion loss of about 1dB and an isolation in excess of 12dB. The authors believe the atomristors could be further optimized through scaling to operate at 100s of THz. They conclude that in frequency switching applications, the high breaking strain and ease of integration of 2D materials on soft substrates would make the atomristors suitable candidates for the manufacture of flexible nonvolatile digital and analog/RF switches.
Related articles:
Plastic-based artificial synapse beats all energy-efficiency benchmarks
Closely mimicking synapses: diffusive memristors
Synaptic transistor learns as it switches
Neural network built in plastic
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
