Imagine a computer chip that works like a human brain, where memory and processing happen in the same spot instead of moving data back and forth. It uses light instead of just electricity to handle the heaviest parts of AI tasks almost instantly. This allows devices to make complex decisions in a blink of an eye without draining the battery.
Current AI hardware is too slow and power-hungry for real-time edge use because moving data between memory and processors creates a bottleneck. This prevents devices from reacting in sub-millisecond timeframes within strict power limits.
A hybrid architecture combining photonic convolutional processors using phase-change materials with electronic memristive crossbar arrays and dopant network processing units.
If you are an autonomous vehicle manufacturer dealing with the need for split-second obstacle avoidance — this project developed a hybrid photonic-electronic architecture that enables inference at ~1 microsecond latency. This ensures the car responds in real time to avoid unintended consequences in unpredictable environments.
If you are a robotics systems integrator dealing with high power consumption in edge AI — this project developed a brain-inspired chip that operates within a ~1 watt power budget. This allows high-precision AI inference to happen locally on the robot without needing heavy cooling or massive power supplies.
If you are a smart device OEM dealing with lag caused by cloud data traffic — this project developed a physical-substrate computing system that processes massive heterogeneous data locally. It achieves sub-ms latency, removing the delay associated with sending data to distant servers.
Based on available project data, there is no specific pricing or cost information provided for the hardware.
The project focuses on realizing a pathway for a new technology; however, the consortium includes 2 industry partners to help bridge the gap toward industrial application.
Based on available project data, specific licensing terms are not listed, but the project involves a consortium of 6 partners across 5 countries.
The architecture is designed to run convolution neural networks (CNNs), specifically targeting benchmarks like ImageNet classification.
The project period runs from 2022-05-01 to 2026-10-31, suggesting the technology is currently in the development and validation phase.
The consortium is research-heavy but strategically balanced for a deep-tech project, consisting of 4 universities and 2 industry partners (33% industry ratio). Spanning 5 countries (CH, DE, IT, NL, UK), it combines academic expertise in unconventional computing with industrial application capabilities, including one SME.
Contact Universiteit Twente (NL) regarding the HYBRAIN project coordination.
Contact us to explore licensing opportunities for hybrid photonic-electronic AI hardware.
