CQS Seminar: Neuromorphic Computing: Linking Atomic Changes to Device Operations

An immense amount of data currently processed by AI systems using conventional computers based on von Neumann architecture results in unsustainable energy consumption. An effective solution to this conundrum is provided by neuromorphic computing, which merges logic and memory storage operations. The memory state is determined by the magnitude of current through the cell material; therefore, the key issue is to control charge transport-assisting atomic rearrangements occurring within the time scale (sub-nsec) of circuitry operation frequencies. Accomplishing this task requires combining material modeling, charge transport simulations and electrical measurements data while explicitly considering their functional interconnectivity in real time.

The evaluation approach developed within this framework is used here to assess the characteristics of several classes of promising resistive non-volatile memories. The effectiveness of neuromorphic systems, which are required to achieve several orders of magnitude in energy savings, depends on the consistency of the process modulating structural features of employed memristors, while application conditions introduce additional constraints (energy supply, operation speed, environmental stability, etc.) to device characteristics imitating adaptive synaptic changes. This approach enables tracking the effect of subtle atomic modulations on gradual changes in device characteristics, allowing to predict how well a given technology can perform in specific applications.  

Hope you can make it!