Brains in the Loop: Neuromorphic Computing in Future Communication Systems
Tuesday, 2 June 2026, 14:00 – 17:30, room M3
Organisers
- Sebastian Karl (Fraunhofer IIS, DE)
- Eike-Manuel Edelmann (KIT, DE)
- Fariborz Derakhshan (Nokia Bell Labs, DE)
- Konstantinos Nikitopoulos (Uni Surrey, UK)
- Alexander von Bank (Karlsruhe Institute of Technology, DE)
- Ankit Gupta (VIAVI Solutions Inc., UK)
- Stephen Wang (VIAVI Solutions Inc., UK)
- Leonidas Richter (Fraunhofer IIS, DE)
- Shi Li (VPI Photonics, DE)
- Andre Richter (VPI Photonics, DE)
- Laurent Schmalen (Karlsruhe Institute of Technology, DE)
Motivation and Background
Future 6G communication systems face key challenges in delivering unprecedented capacity at ultra-low latency and superior energy efficiency. Emerging applications, as well as novel network capabilities such as integrated sensing and communication, require intelligent decisions and network management at latencies well below one millisecond. Simultaneously, sustainability targets necessitate the support of billions of devices with minimal energy consumption. Traditional AI/ML systems have demonstrated their ability to increase network capacity. However, their dense GPU/CPU-based computations struggle to meet strict energy and latency requirements.
Neuromorphic computing promises to achieve exceptional AI/ML performance with lower computational demands through biologically inspired architectures, such as spiking neural networks and analog neural processors. By transmitting information via discrete events, exploiting hardware-level parallelism, and co-locating memory and computation, these systems emulate the mammalian brain’s parallel, event-driven processing, which is unprecedented in terms of energy efficiency. By enabling real-time, scalable, and lowpower AI at the edge, neuromorphic computing stands out for its innovativeness and relevance to 6G’s most pressing challenges. Additionally, recent results show that neuromorphic processing has the potential to go beyond AI-based algorithmic approaches to realize highly efficient PHY processing architectures leveraging new, implementation-efficient, signal representations capable of delivering practical, substantial, and measurable gains.
Current research neuromorphic computing and novel AI-driven approaches aims to enhance the capabilities of future communications. The work presented ranges from highly-efficient, intelligent processing of physical-layer signals with both digital and analog hardware, over hardware designs with innovative electronic and novel photonic components, to unconventional neuromorphic approaches beyond purely data-driven intelligent processing.
These innovations are essential for realizing the full potential of the next generation of communication technologies and invite further research. The session aims to provide an analysis of the current capabilities of neuromorphic computing for communication systems and firsthand insights from current pioneering research on applications in academia and industry.
Structure
Duration: 2 x 90 minutes
Impulse: Neuromorphic Computing in Communication Systems
▪ Speaker: Fariborz Derakhshan (Nokia Bell Labs, DE)
NeuroPHY: Towards Ultra-Efficient, Neuromorphic Inspired PHY Processing Architectures
▪ Speaker: Konstantinos Nikitopoulos (Uni Surrey, UK)
▪ Abstract: In this talk, we will explore how implementation-intensive physical layer (PHY) tasks can be reformulated into isomorphic mathematical representations that are naturally suited to spiking neuronal architectures and event-driven computation. Moving beyond conventional approaches that convert existing deep neural networks into spiking equivalents, NeuroPHY exploits the inherent graph-structured nature of PHY algorithms to enable native neuromorphic computation without reliance on computationally intensive, data-driven training. We will demonstrate how neuromorphic processing can realize critical transceiver building blocks and share key insights and system-level lessons learned. We will highlight where hybrid digital-neuromorphic designs introduce performance bottlenecks and how neuronal architectures expose distinct complexity–performance trade-offs.
Neuromorphic Signal Processing for Optical Communications: SNN-based Equalizers and Demappers
▪ Speaker: Alexander von Bank (Karlsruhe Institute of Technology, DE)
▪ Abstract: In this talk, we demonstrate that spiking neural networks (SNNs) can serve as building blocks for potentially energy-efficient receivers in high-speed communication systems. We introduce SNN-based approaches for equalization and demapping in short-reach optical communication systems employing intensity modulation with direct detection, which are widely used in applications such as data centers and metro-area networks. We show that a variety of design choices are possible, differing in neural encoding and decoding schemes, the use of recurrent connections, and the application of regularization during training. Finally, we examine how these design parameters affect performance across multiple metrics, including bit error rate, model complexity, latency, and symbol-wise energy consumption.
NeuromorphicRx: From Neural to Spiking Receiver
▪ Speakers: Ankit Gupta, Stephen Wang (VIAVI Solutions Inc., UK)
▪ Abstract: In this talk, we show how spiking neural networks can be utilized to design energy-efficient 5G-NR/6G OFDM receivers. These receivers replace multiple traditional signal processing blocks, including channel estimation, interpolation, equalization, and symbol de-mapping, within a unified neuromorphic framework. Particular emphasis will be placed on the generalization capability of SNNs and their robustness under quantization. Further, we provide deeper insights into the performance and underlying mechanisms of spiking-based OFDM receivers.
Break
Neuromorphic Edge AI Processors for Sustainable Communication Systems
▪ Speaker: Leonidas Richter (Fraunhofer IIS, DE)
▪ Abstract: We present two analog/mixed-signal ASICs for AI workloads. ADELIA is an analog-crossbar accelerator for ultra-low power inference in the sub-mW regime with a variety of established DNN-architectures. SENNA is a novel neuromorphic SNN-accelerator offering ultra-low latency event-based computations in the range of nanoseconds in its Field-Programmable Spiking Neuron Array architecture. Both analog accelerator cores are embedded in a digital signal flow, enabling seamless integration and cooperation with digital signal processors.
Component and System Modelling for Photonic Neuromorphic Computing Architectures
▪ Speaker: Shi Li (VPI Photonics, DE)
▪ Abstract: We introduce key concepts in photonic neuromorphic computing and discuss the challenges and potential solutions for modeling and simulating the physical characteristics of these architectures. Further, we present latest research results from our work in the SUSTAINET-Advance project addressing novel application cases in datacom and telecoms.
Panel discussion “Neuromorphic computing in communication systems – why isn’t everyone using it?”
▪ Moderator: Sebastian Karl (Fraunhofer IIS, DE)
▪ Topics: What are the current obstacles and future opportunities for neuromorphic computing for wireless and optical communication systems?; How can we show the benefits of neuromorphic computing in communication systems?; How to overcome the gap between SW development, neuromorphic HW and established communication protocols to foster wide-scale adoption?
