Data is all around us in the modern world. Whether it’s from basic processes within our cities (such as traffic control), medical services, industrial processes, or continuous surveillance, the use of data is continuing to grow—and is likely to grow further as we head towards a more data-driven society. The Internet of Things (IoT), artificial intelligence (AI) and big data have all been responsible for this data boom and now every person is responsible for around (on average) 5 terabytes of data.
The collection of such large amounts of data usually involves many analogue signals to be gathered at sensor terminals and uploaded to the cloud server, where huge amounts of data can be stored and processed remotely from a data centre. Even though this storage process is vital from an improved data storage, productivity and information security perspective, the transmission of redundant and invalid data to the cloud often leads to a waste of computational resources and large power consumption to compensate.
Given the energy waste with many cloud systems, energy starvation is seen as one of the most severe industry challenges. A range of new devices manifesting—that are low-power edge computing options—could become a good supplement for cloud computing technology. Overall, low-power edge computing technologies could process and pre-screen real-time data before storing it to the cloud, which could help save on energy by only processing valid data signals. It’s thought that memristors could play a key role in these technologies.
The Need for Memristors
Memristors are seen as a vital component of low power edge-computing systems for several reasons. They have already been used in a range of resistive memory (ReRAM) devices. Memristors are a component which limits and regulated the flow of an electrical current in a circuit. However, one key differentiator between memristors and other resistors is that memristors remember the current that has flowed through it. Another key feature of memristors is that they are a non-volatile component to retain their memory without the need for power.
Memristors have several benefits for low power edge computing applications. The first, is that they are CMOS compatible to be designed to interact and integrate with existing technology architectures. Moreover, they have a fast-switching speed, a low power potential, and the possibility for high density information storage.
Another key aspect of edge-computing applications is that the non-volatile resistance can be reconfigured. Alongside their two terminal structure and 3D integrating capability, it can enable memristor arrays to perform large-scale in-memory computing tasks. These features can eliminate the frequent movement of data between the central processing units and memory architectures in computer systems, leading to lower energy consumption for data-intensive workloads.
While many memristors have been designed for different applications, nano-sized polymers are seen as a suitable material candidate for the switching matrix within memristors in low-power flexible edge computing applications.
Several functional mechanisms occur within polymer memristor devices. These are charge trapping and de-trapping, charge transfer, electrochemical redox reaction, conformation reconfiguration, ion migration, and local resistive switching mechanisms. These mechanisms tend to come from conductive regions within the polymer matrix (often conductive filaments and defect-rich areas), where these electrical phenomena tend to generate easier under a local electric field.
However, in many polymer matrices, the conductive elements (and the defects) tend to be randomly distributed. There is a drive to scale polymer memristors down to the nanoscale to be integrated into more (and smaller) devices. However, there is currently an inherent issue with the scaling down process.
When trying to scale down the elements to the nanoscale, the inhomogeneity of the conductive regions within the polymers can result in devices that have drastically different functionalities. This ranges from devices that show completely different electronic behaviours, to devices that do not possess any resistive switching abilities.
So, there is a lot of interest and potential for nano-memristors made from polymers, but many are not currently reliable enough for the demands of modern-day computing. However, modern-day chemical solutions offer the possibility to fine-tune the geometry and electronic structure of the different polymers (and their monomer building blocks).
A New Memristor for Edge Computing
A new 2D conjugated polymer nano-memristor has emerged, taking advantage of the different chemical manipulation methods available. It was a polymer conjugation strategy. The device created was efficiently minimised, with a high fabrication yield of 90%, and exhibited homogenous resistive switching characteristics and low-power potentials.
The memristor incorporated coplanar macromolecules with 2D conjugated thiophene derivatives. These molecules enhanced the π-π stacking interactions and the crystallinity of the polymer. These interactions helped create a more regular atomic structure, resulting in a homogenous response, rather than a random inhomogeneous response seen with other polymers and other fabrication approaches. The different electron donating and electron accepting layers were regulated and ordered on the molecular level so that the resistive switching phenomenon propagated throughout the device.
Moreover, the polymer network included some triphenylamine chemical groups, and these are redox active materials. Their inclusion enabled the device to be switched between ON and OFF states and only took 32 nanoseconds to perform this switching function. The team also demonstrated the potential for the polymer nano-memristors in the in-memory Boolean logic and arithmetic operations of a general computer and a hardware accelerator for binary neural networks in pattern recognition tasks.
Polymer memristors can be used in different technologies. However, scaling them down has prevented nanosized polymer memristors from manifesting on a wider scale. A plausible chemical solution has emerged where the device’s reliability is much higher and can be used to produce functional devices. Work still needs to be done in this area before they are used in edge-computing applications. However, there is still time to overcome the energy challenges of the industry, and the research to date can be built on using other chemical solutions to find the most effective way of producing polymer nano-memristors on both a small and large scale.