Mimicking the Brain – Resistive switching for memory storage

With Artificial Intelligence gaining popularity, there is a need to store more data, causing data centres to pop-up all over the place. Data centres require a large amount of power, with the current version of handling data being very energy inefficient for AI applications. Computing that is inspired by neural networks in the brain, could pave the way towards more efficient storage.

Neurons are the fundamental cells of the brain that allow information signals to be passed between them. Their connections (synapses) are what determines our memory. This makes our brains special in the sense that both information processing and storage takes place in one organ. This is unlike our current computer systems, where their separation causes slow processing and energy loss. Our aim is for these processes to occur in the same place. So called, memristors, can make this happen.

Memristors can simulate neurons and their synapses by storing information through changes in resistance. The internal structure of the device can be changed by applying a current or voltage. When the current or voltage is removed, the change in device structure is retained. Therefore, ‘remembering’ input information. It is also possible to input different information, i.e. overwriting previously stored data, by changing the structure through applying other currents or voltages. Thus, allowing for data to be saved and changed within the memristor.

Their ability to physically change the structure, allows memristors to store more diverse information than current computers. In traditional resistor-based computers, information is stored in ones and zeros. One represents an ‘on’ state (where charges can flow) and zero represents an ‘off’ state (no flow of charges). With memristors, it is possible to have various different resistive states, therefore not limiting it to two states (one and zero). This opens up the possibility of storing more information in one device.

Some struggles with the production of memristors, is making them uniform across the whole device and producing devices with identical properties. Therefore, studying the mechanism in which resistive switching occurs in memristors is important. This can be analysed by changing the deposition conditions of the materials involved and evaluating the device performance. In this project the material Yttria Doped Zirconia (YSZ) is used, which is commonly applied in the CMOS (ie memory chip) industry. Therefore, making it easier to adapt memristors for industry. To make them widely available to the public, memristors have to overcome the named challenges in production, as well as fulfil a range of physical requirements. They need to be able to switch fast between the different resistive states. The device must also be able to switch repeatedly, to store various information. Furthermore, memristors also need to be able to remain in their changed state for an extensive period, otherwise the stored information will be lost over time.

Summarising, this project is focused on making the fabrication process as industry friendly as possible, while maintaining and optimising the resistive switching abilities of the memristors.

Isabella Laetitia Teck

NanoDTC PhD Student, c2023