The HYBRAIN demonstrator is an advanced platform integrating various technologies developed within the HYBRAIN project, aiming to achieve groundbreaking neuromorphic computing performance by combining unique hardware architectures to enhance efficiency, speed, and low-energy consumption.

The primary objective of the demonstrator is to establish an architecture capable of ~1 μs (microsecond) inference using ImageNet at 224×224 resolution with 3-channel RGB data. This target outperforms existing approaches that take over 1 ms (millisecond) in state-of-the-art approaches by achieving high-speed processing while maintaining a power budget below 10 W. Ultimately, the project aims to integrate photonic and electronic components into a seamless multi-chip system.

The demonstrator is pivotal to HYBRAIN’s vision of pushing the limits of edge AI with low-latency, energy-efficient processing. It embodies the project’s mission by showcasing the potential of integrating photonic and electronic systems to achieve unprecedented computational speeds and efficiency.

The demonstrator aims to address significant challenges in neuromorphic computing, including:

• Effective Integration of Diverse Technologies: Ensuring compatibility across various technological timescales.
• Minimising Electro-Optical Conversion Losses: Reducing energy losses that occur during electro-optical transitions, which are crucial for maintaining efficiency in data processing.

The demonstrator aims to establish HYBRAIN’s architecture as a new benchmark for ultra-fast, low-power neuromorphic computing. The expected outcomes include significant breakthroughs in latency reduction and energy efficiency, which will have a wide-ranging impact on applications requiring real-time, edge AI processing.

The findings and data generated from the demonstrator will feed directly into HYBRAIN’s broader goals, providing empirical validation for its technologies and laying the foundation for future innovations in neuromorphic systems.

The demonstrator serves as a proof of concept for integrating photonic and electronic systems, pioneering a new direction for neuromorphic computing research. It sets a precedent for performance that could redefine the capabilities of this field.

By integrating photonic and electronic components, HYBRAIN’s demonstrator surpasses existing neuromorphic computing solutions in both latency and energy efficiency. The hybrid architecture removes bottlenecks that limit conventional systems.