Science Communication in Media
Smart devices and technologies surround us more than we are aware. From face recognition smartphones to consumer preference cookies, we produce data at every technological interaction. To give one example, operating the engines for a two-jet plane on a 24h journey can produce as much data as Facebook does in one day [1].
Our rate of data production is projected to rise, as the demand for automated, wireless and self-improving technological services increases. Some potential future applications of data-intense algorithms (like artificial intelligence or deep learning) include implementation of proactive healthcare, energy-efficient usage monitoring, security surveillance. Data is projected to account for a third of global electricity by 2030 [2], and yet the way in which data consumes energy is non-trivial. Let’s start by considering the hardware level: here, traditional computer architecture physically separates units for data storage and data processing. On the software level, the data is shuttled back and forth between memory and processing to perform a task like train an algorithm [3]. This physical separation is at the root of the problem: the more complex the algorithms, the more ‘expensive’ this shuttling data becomes.
Interestingly, the software of most autonomous systems mentioned earlier is largely inspired by the brain: indeed, by the 1940s, new frontiers in neuroscience inspired computer scientists to mimic biological neural networks for problem-solving applications [4]. One of the ways in which the brain is orders of magnitude more energetically efficient involves co-locating memory and processing units in synapses (Fig. 1, left). There is an extent of incompatibility, therefore, between software and hardware: the former takes inspiration from nature, whereas our current traditional computer architecture is not, resulting in high energetic demand.
Memristors have been researched since the 2010s [5] to address this discrepancy: memristive materials are ‘resistors with a memory’, where changes in resistance are voltage-controlled, and these resistance states are maintained as non-volatile memory. This allows storage and processing to potentially coexists. Memristive materials have been mostly researched for memory applications, but have also been investigated in the field of neuromorphic (brain-inspired) computing.
Whereas the application is neuromorphic computing or not, my PhD will focus on studying memristive materials as novel hardware material for specific (energy intense) applications. My research is aimed towards understanding one fundamental event that leads to memristive behaviour: ion transport through materials. In particular, understanding how to control these mechanisms could facilitate back-engineering of large-scale systems. In fact, one of the biggest challenges in the field is variability across samples, and this is partly due to the random nature of ion motion (Fig. 1, right) – at least at the nanometre scale, where it becomes atomistic. Another difficulty deriving from the nanometre magnitude is the observation of material changes during operation, imperative to understand conduction mechanism and differentiate electronic from ionic contributions to the resistive switching events.
In order to tackle this problem, I will begin by looking at inducing Li-diffusion in 2D atomically thin mica, and studying how properties vary by exchanging different metal cations. Mica – a layered aluminosilicate material (Fig. 2) – serves as an excellent model device thanks to its stability and easy mechanical exfoliation.
Check our recent publication here: https://pubs.acs.org/doi/full/10.1021/acs.chemmater.4c00965
References:
- https://www.industryweek.com/technology-and-iiot/systems-integration/article/22006020/internet-of-aircraft-things-an-industry-set-to-be-transformed
- M. Koot, F. Wijnhoven, Usage impact on data center electricity needs, Appl. Ener., 291 (2021)
- A. Mehonic, A. Sebastian, A. J. Kenyon, “Memristors—From In-Memory Computing, Deep LearningAcceleration, and Spiking Neural Networks to the Futureof Neuromorphic and Bio-Inspired Computing”, Adv. Intell. Syst., 2 (2020)
- S. Kumar, W. D. Lu, “Dynamical memristors for higher complexity neuromorphic computing”, Nat. Rev. Mater. (2022)
- D. Strukov, G. Snider, et al., The missing memristor found, Nature, 453, 80–83 (2008)
NanoDTC PhD Student, c2021