The aim of this research project is to investigate and implement novel concepts of optical and electro-optical neural networks (deep, recurrent, neuromorphic, etc.) for optically assisted edge computing.Research challenges
Proposing a series of innovations in the area of hardware accelerators, design stack and middleware software that revolutionize the ability of edge computing platforms to operate federatedly, leveraging sparse resources that are coordinated to create a powerful swarm of resources is the aim of a newly granted Horizon KDT-RIA CLEVER. Focus Topic is on Processing solutions for AI at the edge, addressing design of accelerators and processing engines, of layer stack and middleware.
Within this project, TU/e aims to collaborate with international and national companies to contribute to ultra-low power non-von Neumann compute architectures for sensor-edge intelligence. This task will explore the practicalities of applying non-von Neumann (i.e. neuromorphic computing) analog-mixed signal computing elements in the form of a programmable hardware platform to sensing applications.
Photonics can play an important role here by designing and developing compact, energy-efficient, low latency and high-capacity computing engines. Several on-chip computing paradigms have already been proposed based on the use of light, but never designed for a specific application in mind nor validated for real applications, if just toy example. A different approach needs to be explored and realized in this program, which starts from application specifications, and proceed with the aimed metrics for the design and mapping of artificial neural networks for computation in photonics. The exploitation of brain-inspired architectures for the design, realization and integration of on-chip processing architecture, as well as for the control plane optimization is foreseen to provide a step-change in the edge AI processing engines and their overall impact on edge computing performance.
In this research program, the student will understand the edge computing requirements, explore novel optical techniques for neuromorphic computing using different encoding techniques and architectures (all connected and sparse), map these models on chip, realize the chip and envision the chip embedding and interface within state-of-the-art engines. He/she should design and realize integrated optical circuitry which can perform these functions efficiently. The work comprises system understanding, design, simulation, fabrication and experimental activities. The final goal is to demonstrate the feasibility of optical AI chips for edge computing.The team
The PhD position is within the Electro-Optical Communication Systems (ECO) group, part of the Institute for Photonic Integration (IPI), in the Department of Electrical Engineering of Eindhoven University of Technology, The Netherlands. ECO has about 30 members, 20 of which are PhD students. The Institute for Photonic Integration (IPI) has five dynamic and ambitious research groups, which are closely cooperating: a systems group, a photonic integration technology group and three materials research groups. Moreover, the newly launched Eindhoven Hendrik Casimir Institute (EHCI), with a predominant focus on neuromorphic and quantum computing, is the ideal eco-system where to develop this research. The PhD student will participate in this international KDT research program and will collaborate intensely with a Post-Doc from TU/e, also involved within the project, as well as with the involved Dutch companies, Innatera (https://www.innatera.com/) and Synopsis (https://www.synopsys.com/