The increasing complexity of neural networks poses enormous challenges for conventional computer technologies. Electronic hardware is increasingly reaching its limits. This is where MIT’s revolutionary development comes in. A photonic processor that uses light instead of electrons for calculations and heralds a new era of high-speed processing.
At the heart of the system is a fully integrated optical neural network that combines linear and non-linear operations. While linear calculations are performed using light-based matrix multiplications, the new technology overcomes the previous hurdle of non-linear operations. This was solved by special non-linear optical functional units that efficiently combine optics and electronics.
Speed and efficiency in the nanosecond range
The photonic processor processes classification tasks in less than half a nanosecond. A speed that sets new standards. The technology significantly reduces latency times and is extremely energy-efficient. The optical signals are not converted into electrical currents until the final output. This combination enables impressive accuracy values of over 96 per cent during training and more than 92 per cent during inference.
Scalable integration for practical applications
A key advantage of photonic chips is their scalability. They can be integrated into existing systems using commercial manufacturing processes. MIT is thus not only opening up new perspectives for AI applications, but also for energy-intensive areas such as big data and data centres. MIT’s photonic AI chips mark a technological breakthrough in the world of computing. Using light as a computing medium enables ultra-fast, energy-efficient and scalable AI calculations – a step into a new future of data processing.
