Physics Colloquium: Ultra-low switching Energy Memories to artificial neurons
Prof. T. Venky Venkatesan, National University of Singapore
Abstract: Memory devices are responsible for a significant fraction of the energy consumed in electronic systems- typically 25% in a laptop and 50% in a server station. Reducing the energy consumption of memories is an important goal. For the evolving field of artificial intelligence, the compatible devices must simulate a neuron. We are working on three different approaches towards these problems- one involving an organic metal centered azo complex, the other involving oxide based ferroelectric tunnel junctions and the last involving real live neuronal circuits.
In the organic memristors that we have built on oxide surfaces the device performance exceeds the ITRS roadmap specification significantly demonstrating the viability of this system for practical applications. More than that, these organic memories exhibit multiple states arising from interplay of redox states, counter ion location (studied by in-situ Raman and UV-Vis measurements) and molecular self-assembly leading to the possibility of neuronal systems. These molecular devices are extremely stable and reproducible- a significant departure from conventional organic electronics.
On the oxide- front the significant results are that ferroelectric tunnelling is seen even in barriers with single and two atomic layers of BaTiO3 or BiFeO3. Oxygen vacancy motion can also play an important role in changing the device characteristics leading to synaptic characteristics. Last but not the least, oxide surfaces can be utilized to force neurons to grow at specific places on a surface giving the potential for fabricating live neuronal circuits.