Research

MINDnet combines technological know-how in photonics, electronics, a fundamental understanding of physical system dynamics, and expertise in neurotechnologies and neuroscience. Within this project the consortium will create synergies across heretofore separate research communities and will provide holistic knowledge and skills to optimize the computing system all the way from single nonlinear devices to full hardware-software co-design, including also the training methodology and links to application requirements. 

Our part in Ilmenau will be to develop method to augmented memory and adaptability in physical reservoir computers. We will model the nonlinear dynamics of the physical devices, analyze their modulation properties and bifurcation structure and find ways for efficient operation of these devices in analog computing  applications.

The Ilmenau School of Green Electronic (ISGE) is a highly interdisciplinary pool of research projects with the goal to develop an information technology for the future, which is sustainable and climate-neutral not only in operation, but along the entire value chain and in the entire material cycle, i.e. during operation, production, repair, and recycling.

Inside of the general focus on the areas of bio-inspired microelectronics, intelligent materials, devices and technologies as well as on energy-efficient computing, our project P6 deals with the investigation of the performance of an electrooptical reservoir computing system (ORC) in its dependence on the architecture of the reservoir and the experimental implementation as an optical spatio-temporal recurrent neural network. An existing experimental setup for a reservoir computing system based on a spatial light modulator in an electronics feedback loop serves as a starting point. The systems will be adapted according to the requirements using optical systems design and three-dimensional beam shaping. The dependence of the performance on the optimized reservoir structure is investigated using benchmark experiments and evaluated in terms of energy efficiency as an example of green electronics.

Brain-inspired or neuromorphic chips that use biologically inspired spiking neural networks have attracted attention because they promise highly efficient ways to process data. Developing neuromorphic systems using electronic and photonic hardware is part of the new European collaborative project SPIKEPro (Spiking Photonic-Electronic IC for Quick and Efficient Processing) within the framework of the European Innovation Council (EIC). TU Ilmenau is one of the partners in this consortium, which includes TU Eindhoven, the University of Strathclyde, University College London, and HP Enterprise Belgium. SPIKEPro aims to achieve a breakthrough from science to technology by combining low-energy electrical and photonic neurons into a unified spiking neural network on an integrated circuit. SPIKEPro’s chip integration approach is based on a common technology platform that connects ultrafast laser-based optical neurons with efficient electrical spiking diodes via non-volatile synaptic weights. This enables the project to simultaneously capitalize on the advantages of both electronics and photonics to deliver efficient, high-speed spiking neural networks (SNNs) that surpass existing implementations. In addition to reducing the energy consumption per spike in the network, SPIKEPro will also develop novel learning strategies and algorithms capable of functioning with a reduced number of synaptic connections. The results of SPIKEPro will have a lasting economic, societal, and scientific impact. The project will bring ultrafast and efficient neuromorphic hardware to the diverse fields of edge computing, sensor data processing, high-speed control, and computational neuroscience.

Reservoir computing is a machine learning method that can be easily implemented in hardware and is being intensively studied by our research group. In the NeurosensEar project—funded by the Carl Zeiss Foundation and focused on neuromorphic acoustic sensors for high-performance hearing aids of the future—we are investigating micromechanical resonators as in-sensor reservoir computers.

DFG Project (2020–2023)
“Hybrid Photonic Computing in Delay-Coupled Nonlinear Systems with Memory”

Subproject within the Collaborative Research Center SFB910 (2019–2022)
“Collective phenomena in laser networks with nonidentical units”

• Implement all-optical reservoir computing schemes (evaluation via benchmark tasks: chaotic time series prediction, memory capacity, channel equalization, etc.)
• Develop a numerical framework for simulating highly connected, delay-coupled networks with memory,
• analyze the impact of network topology on computing performance
• Explore correlations between performance and bifurcation structure

• Modeling semiconductor quantum-dot lasers with optical feedback, injection, or network coupling
• Emission stability of two-state lasing devices, optical switching applications, neuronal spiking
• Complex emission dynamics of nano- and micro-lasers

• Pulse shaping in passively mode-locked lasers
• Timing jitter calculations and performance tuning via optical feedback
• coupled mode-locked lasers