Multiplexed optical neuromorphic computing
The innovative system is combining signal processing with an advanced neural network architecture, specifically using electromagnetic radiation for communication between nodes within a multi-layered network.
Challenge
Synchronization of signals between the emitting and receiving nodes across multiple substrates can be challenging, especially when dealing with high-speed data and large-scale networks. Ensuring minimal signal loss and reliable communication between nodes in the system can be crucial. Managing the modulation and demodulation of signals, especially across multiple layers with different characteristics, requires precise control and sophisticated algorithms to avoid errors and optimize the signal-to-noise ratio. Proper alignment of the substrates and ensuring coupling of transmission and reception elements can be difficult, particularly when the network scales up. Misalignment can lead to signal degradation and reduced performance. The control unit and the emission elements in the system require substantial power, especially when the network operates with high-frequency modulation and large numbers of nodes. Moreover, efficient power management is crucial for long-term system operation.
Our Solution
The present system consists of multiple logical layers within the neural network, with each layer formed by a group of nodes. The first logical layer is composed of transmitting nodes that are capable of emitting electromagnetic radiation. The second logical layer consists of receiving nodes that are designed to capture the radiation emitted by the transmitting nodes. These two logical layers are connected by a common pass-through range through which the emitted radiation is transmitted from the sending elements in the first layer to the receiving elements in the second layer. A control unit is used to adjust the emitted signal characteristics, based on learned weights, enabling the network to dynamically optimize signal properties for effective transmission and reception. Multiple substrates are stacked to form a multi-layer neural network with interconnected logical layers. The first unit controls the modulation of the emitted signals (frequency modulation, amplitude modulation, etc.), while the second unit handles the demodulation of the received signals. The first assignment unit may be responsible for multiplexing the transmitted signals, while the second assignment unit handles the demultiplexing of received signals. The system can utilize pulse-width modulation or other modulation techniques to control the emission of the nodes, depending on the network's operational requirements.
Schematic representation of the multiplexed neuromorphic computing technology (image generated with Perplexity AI).
Advantages
- High-speed signal processing
- Scalability: stacked substrates and modular components allows for easy scaling
- Reduced electrical interference: system less prone to electrical noise and intereference
- Energy efficiency: use of modulation techniques like pulse-width modulation and optimized power control
- High adaptivity: system’s use of learned weights enables dynamic adaptation to different signal conditions
Applications
- High-speed communication networks
- Artificial intelligence (AI) and neural networks
- Internet of Things (IoT) and telecommunications
- Medical imaging systems
- Quantum computing
Development Status
System functionality has been successfully tested. Prototype available.
Patent Status
European patent application filed.
Patent holders:
Ostfalia University of Applied Sciences (Germany)
TU Braunschweig (Germany)
Contact
Dr. Mirza Mačković
Patent & Innovation Manager Technology
E-Mail: Diese E-Mail-Adresse ist vor Spambots geschützt! Zur Anzeige muss JavaScript eingeschaltet sein!
Tel.: +49 551 30724 153
Reference: MM-2393-FHBW
Published: July 8, 2025.
