Neuromorphic Computing – AI at the Edge

In this work, we investigate new concepts of transistors or circuit level for high improved performance in terms of materials, devices, components as well as processing technologies. We study metal-insulating transition (MIT) VO2 devices as an oscillating device for emulation neurons as an ultra-low power device. We perform atomistic simulation and TCAD modeling of the electrical and thermal properties of MIT VO2 and 2D memristive devices. Additionally, we develop compact and ultra-low power 2D material memristor devices for emulating synaptic weights.

We explore innovative concepts on neural networks and architectures for energy-efficient computation circuit architectures for neuromorphic computing. We investigate a novel computing paradigm for neuromorphic computing based on oscillatory neural networks (ONN) where the information is encoded on the signal phase rather than time and signal amplitude in contrast to existing neuromorphic computing based on spiking-neural networks (SNN) or convolutional neural networks (CNN).

The von Neumann architecture is the basic building block of almost all computers today. However, the energy-intensive data communication and storage process, known as the “von Neumann bottleneck” presents a severe limitation for data-driven computing that requires significant memory updates. By contrast, neuromorphic computing architectures combine both computation and memory through an array of neuron-like elements with massive synapse-like interconnections that can provide a significant improvement in computational capability for specific types of analyses. However, it is still very challenging to implement DNN algorithms in resource-constrained embedded devices, which hinders their widespread adoption in the vast array of Internet of Things (IoT) applications and markets. In this work, we investigate system implementation for low energy, low latency implementation and architectures for in-sensor and edge computing solutions and applications.