Periodic Reporting for period 1 - MENESIS (Memristor-Enabled NEuromorphic System for Intelligence in Space)
Reporting period: 2021-09-01 to 2023-08-31
Today's technology relies on transistor-based devices for all kinds of computation power. However, the transistor is not a rad-hard element due to its bulky architecture, allowing a huge amount of dose to be deposited into it; this results in electrical failure in computing systems. Hence, up to now, we do not have reliable artificial intelligence (A) accelerators deployed to space due to this reliability reason. This challenge makes us rely a lot on the ground station (cloud server) to compute the data streamed from the satellite. Cloud servers are power-hungry installations with high costs of operation and maintenance; the increased need for AI computation would make us keep relying on building new servers in the future, and this will eventually increase carbon and silicon footprint. The project aims to design rad-hard ultra-thin memristor devices as a novel computing element enabling AI computation in space for an efficient data stream. The architecture of the device is so thin that most of the high-energy radiation passes through the device, limiting the undesired energy deposition.
In the first stage, the architecture of the memristors was designed, and we played out the fabrication platform to produce high throughput and high-yield memristor wafers. The electrical and material characteristics were investigated, and the effects of high-energy radiation were studied. It is important to design memristor cells having very thin switching layers; this is the key to maintaining the performance of the device during and after radiation. Nevertheless, any electrical deviation induced by the radiation can be exploited to assess the radiation damage. We developed a circuit board prototype to measure the radiation dose that was absorbed by the devices; this method could realise adaptive electronic systems. We also developed ultra-thin selectors with robust endurance and high non-linearity by limiting the interfacial reaction at the electrode/oxide junctions; we further exploited this stable junction property to design non-linear memristor cells that could be useful to avoid sneak-path issues in cross-bar array configuration. The results of this project are published in peer-reviewed journals; please see the dissemination section for the list of publications.
The results from this project provide a strong groundwork and benchmark for the follow-up projects. Two follow-up grants were awarded after the end of the action, allowing the Fellow to work on the ideas that are still left undone. The Fellow also applied for bigger grants to pursue the groundwork to higher technological readiness levels, initiating new collaborations and partnerships with various research institutions and industries. The Fellow received tremendous support from the host, and several trainings for technical and professional development were received. Thanks to these skills gained during the action, the Fellow was successful in securing a permanent faculty post at the University of Southampton, effective after the end of the project.