Novel Materials for Optically Stimulated Memristors

Nowadays, our computers and personal devices have incredible capabilities to recognize our faces, understand our voices, and even predict our needs. This is all possible because of advanced machine learning algorithms. They not only impact our personal lives, but also are very important for finance, manufacturing, and healthcare, among others. These advances are incredible; however, they depend on computer systems designed for completely different applications, like graphics processing units. Because of this, their performance and energy efficiency can be very poor. To improve the situation, a new type of computer is necessary. Since many machine learning algorithms are inspired by the human brain, one promising idea is to develop brain-inspired components already at the hardware level. This field is called neuromorphic computing and requires fundamentally rethinking how computers work. To reach this goal in the most energy-efficient way, new building blocks for computer chips are necessary to complement transistors: memristors. These memristors have very similar properties to the synapses existing in the brain. Because of their variable electrical resistance, they can simultaneously store information and perform operations, just like synapses. Also, certain memristors types are very sensitive to light, which can mirror the effect of neuromodulators like dopamine in the brain. This makes them even more attractive as building blocks of neuromorphic computers. However, memristor technology is still very new. Scientists do not fully understand exactly how they change their electrical resistance. For this reason, researchers do not even know the best materials to create them. The design space is huge as different compounds can be combined with each other. Trying all combinations one by one experimentally would be very expensive and time- consuming. Also, these new computing elements should be designed to operate reliably for a long time. In this collaboration between scientists from TU Wien and ETH Zurich, the issue of material choice for brain-inspired computers will be studied. Using state-of-the-art supercomputers and the experience obtained by the principal investigator at the Christian Doppler Laboratory for High Performance TCAD, new simulation techniques will be developed. Then, different materials stacks for memristors will be theoretically investigated. Only the best candidates will be experimentally prototyped at ETH Zurich and electrically tested at TU Wien. With this combined approach it will be possible to validate the simulation methods, to better understand the physics of memristors, and to propose new brain-inspired devices with improved characteristics.

No additional funding sources recorded.