Fraunhofer IIS is developing AI chips based on spiking neural networks to significantly improve signal quality and boost data transmission rates in fibre optic networks.
The project, known as SpikeHERO, aims to enhance Europe’s digital infrastructure and consolidate its technological sovereignty through improved fibre optics.
The Institute is working on the project with partners from the Netherlands, the Czech Republic, and Belgium.
The importance of fibre optic networks in a digital future
Fibre optic networks are a cornerstone of the digital future. As industries compete for key technologies, demand for higher bandwidths and lower latencies continues to grow.
Future 6G networks, in particular, must meet extremely high data transmission requirements.
The challenge is that as the desired data rate increases, the signal quality tends to degrade. To compensate for interference, providers currently rely on digital signal processors.
“However, these come with high costs and are not a sustainable long-term solution due to their enormous power consumption,” explained Michael Rothe, manager of the Embedded AI group at Fraunhofer IIS.
“As a result, further increases in fibre optic data rates are hitting technical limits.”
Overcoming difficulties with neural networks
The SpikeHERO project is working on a solution to technical limits in fibre optics.
The project partners are developing a novel AI processor architecture that combines an optical and an electrical spiking neural network chip.
These neural networks will continuously monitor the communication channel, analyse signals, and correct any interference at the receiver using control parameters. By maintaining signal quality, new possibilities emerge for increasing data rates in fibre optic systems.
Specifically, the project aims to triple the bandwidth from 10 GHz to 30 GHz and reduce latency from 10 microseconds to under 6 nanoseconds. At the same time, energy consumption is expected to drop from 7-10 watts to just 1-2 watts.
Inspired by the brain
Spiking neural networks (SNNs) are considered a promising advancement in artificial intelligence.
Their mode of operation mimics the principles of the human brain: information is processed in the form of pulses and only when a critical relevance threshold is exceeded. This makes SNNs ideal for AI applications that require both real-time responsiveness and energy efficiency.
There are different approaches to developing SNN chip hardware. Optical semiconductors transmit spikes via photons, while electrical counterparts use voltage and current.
Each type has its advantages, and SpikeHERO aims to combine both. For the electrical SNN chip, the project will use the SENNA chip developed by Fraunhofer IIS and Fraunhofer EMFT.
“We’re currently working on the second generation, which promises even higher spike rates with lower energy consumption,” said Rothe.
