Dec 3, 2024
4:00pm - 4:30pm
Sheraton, Second Floor, Republic A
Simanta Lahkar1,Aida Todri-Sanial1
Technische Universiteit Eindhoven1
Simanta Lahkar1,Aida Todri-Sanial1
Technische Universiteit Eindhoven1
Current computing paradigm based on von Neumann architecture faces several critical challenges mainly due to the separation of memory from processing, which causes large power consumption. This challenge is becoming even more critical with the increasing demand for AI workloads that require memory-intensive calculations. Neuromorphic computing provides an alternative approach where processing and memory are merged, similar to biological neural networks. This work investigates the unique properties of advanced materials, like phase change materials (PCMs) and resistive RAM for computational devices to implement a neuromorphic computing paradigm based on coupled oscillatory neural networks.<br/>Certain transition metal oxides (TMOs), such as VO<sub>2</sub>, exhibit an insulator-to-metal transition (IMT) - when subjected to external stimuli, it undergoes a phase change until the entire material converts to its highly conductive metallic state - forming an interesting category of PCMs. We found that, for VO<sub>2</sub>, the external stimuli causing IMT can directly be linked to the temperature inside the material using a physical model giving high experimental accuracy in predicting its behaviour. Furthermore, lowering the temperature can trigger the converse metal-to-insulator transition (MIT) until the entire material reverts to its initial insulator phase upon sufficient reduction of its temperature. This offers an interesting adjustable resistive change that can be triggered using bias voltage due to Joule heating mechanism. A two-terminal VO<sub>2</sub> device exhibits a negative differential resistance (NDR) region that can be exploited to create self-sustaining oscillations. An external resistance and capacitance are used to design such a VO<sub>2</sub>-based relaxation oscillator.<br/>The coupling elements between oscillators play an important role in the dynamics of the neural network. At first, we investigate coupling VO<sub>2</sub> oscillators with discrete components, such as resistance and capacitance, to achieve in-phase and anti-phase relations. Next, we explore the impact of coupling resistance by utilizing a non-volatile memristor, such as bilayer HfO<sub>2</sub> devices. This marks the first-ever investigation of combining phase change VO<sub>2</sub> devices with resistive RAM HfO<sub>2</sub> devices to implement in-memory computing based on a coupled oscillatory neural network paradigm. This project has received funding from the EU’s Horizon program under Projects No. 101092096, PHASTRAC.