With the growth in demand for artificial intelligence (AI) technology and its applications, and the supporting infrastructure — particularly for computing, storage, and energy needs — the search intensifies for new hardware technologies based on new materials, and the architecture needed for the transition.
- Also read:Sustainable carbon nanotube
Researchers at IIT-Madras have come up with ‘memristor-based devices’ for designing low-power, high-performance hardware for AI and internet of things (IoT). A memristor is an electrical component that can “remember” the amount of charge that has passed through it and alter its resistance accordingly. Moreover, memristors are capable of ‘non-volatile memory’, namely they can “remember” their resistance state even after the power is turned off.
So how does a memristor-based device store data or perform logical functions?
To understand this, let’s look at the memristor-based device’s structure. To create one, we put a layer of insulator or semiconductor between two metal electrodes. By sending electricity through these electrodes, you can alter the layer’s ability to conduct electricity. When you send electricity in one direction, it makes a pathway between the electrodes and there is flow of electricity. When sent in the opposite direction the pathway breaks, impeding the flow of electricity. We can think of these two outcomes as “on” and “off”, which we can use to store data or make decisions in a computer.
Dr Abhishek Misra, faculty at IIT-M’s department of physics, says, “Memristor-based devices are at the forefront of the development of low-power, high-performance electronic hardware required to implement the emerging concepts of AI and internet of things (IoT). These AI- and IoT-based technologies can serve humanity in various ways such as providing better healthcare, security, and education, to name a few.”
Unidimensional factor
Like many technologies popular today that were born in the 1950s and beyond, the memristor, too, has a long history. In 1971, Leon Chua theoretically proposed the memristor as the fourth fundamental circuit element (after resistors, capacitors, and inductors). In 2008, researchers at Hewlett-Packard Laboratories, led by R Stanley Williams, announced the practical realisation of the memristor using certain metal oxides, making it the first experimental demonstration of memristive behaviour. In the 2020s, the exploration of memristors continues to advance, with research focusing on scaling up memristor-based technologies for commercial applications.
Researchers at the IoE Centre for 2D Materials Research and Innovation (C2DMRI) at IIT-Madras have developed an innovative memristor device architecture using one-dimensional core-shell heterostructures of molybdenum dioxide-molybdenum disulphide.
It offers significant advantages including ultra-low power operation, volatile and non-volatile resistive switching, a smaller footprint, and speed.
Renu Yadav, PhD scholar at IIT-M’s physics department, says “the developed memristors can be as intelligent as a human brain. These devices can store as well as process information — thereby providing smarter solutions to the traditional von Neumann architecture, where the processing and storage units are separated”.
These structures, grown via chemical vapour deposition (CVD), feature a distinctive coaxial electrical contact between the inner core (metal molybdenum dioxide) and the outer shells (atomically thin-layered molybdenum disulphide) with a unidimensional geometry, marking a substantial leap from memristor design involving two-dimensional layered materials.
The devices are categorised as non-volatile (displaying a strong conducting path even without power supply and hence needing a voltage to break it), and volatile (weak conducting path that breaks on its own).
Nano advantage
Conventionally, metal oxides such as titanium dioxide, aluminium oxide and hafnium oxide are used as the sandwiched layer. However, after the discovery of atomically thin, two-dimensional layered materials such as molybdenum disulphide and tungsten disulphide, these are being explored for use as a resistance switching layer. Such memristors based on layered materials have advantages such as ultimate vertical scaling, reduced operating power, and energy efficiency.
The IIT-M team has fabricated the memristors on nanowire to make it unidimensional.
Nanowire consists of a core shell heterostructure, with the metal core molybdenum dioxide serving as one electrode and the layered semiconducting shells of molybdenum disulphide serving as the switching layer.
Further, a silver electrode is deposited on the axial direction of the nanowire to complete the metal-semiconductor-metal memristive structure. Depending on the thickness of the molybdenum disulphide shell in nanowire, both volatile and non-volatile resistive switching are achieved.
- Also read: A new device for cooling data centres
The unidimensional core shell heterostructure has a 10-nm diameter metallic core wrapped by a few layers of molybdenum disulphide, leading to smaller memristors.
Misra says the memristor device architecture has the potential to serve as a basic building block for future integrated circuits.