Photonic Reservoir ComputingNTT Device Technology Laboratories

NTT is conducting research on "optical reservoir computing" technology, which emulates the neural processing in a human cerebellum with optical interference.
By taking advantages of ultrafast optical processing, NTT is exploring applications for future AI system such as ultrafast decision making.

Nowadays, machine learning technique based on neuromorphic information processing is gathering much attention for AI systems. As this operation requires huge computational resources, the reduction of their power consumption and processing speed becomes a social problem.

• Large-scale parallel information processing thanks to the nature of light
• Compact integration using optical circuit technology developed for optical fiber communications

• Rough pre-processing of large amounts of data (video, etc.)
• Classification, prediction, and generation of ultra high-speed time series data, such as communication signals

Nowadays, neuromorphic optical circuits, which mimic the information processing in human brain, is gathering much attention for the ultrafast and energy efficient machine learning. In the ANN, input signals propagate through nonlinear nodes (corresponding to neuron in the brain). Each node response are weighted and summed (corresponds to synaptic weights) to learn the complex feature of the data. Our technology uses light propagation and interference for this operation. Thanks to the nature of light, we can realize much lower power consumption and faster computation speed beyond the conventional digital electrical computation. Although such type of optical processor is attractive candidate for the ultrafast neural information processing, it was difficult to implement the large-scale ANNs into the optical chip because it requires fine tuning of large-scale optical circuits. Here we focuse on reservoir computing. In this framework, the most of connection are randomly generated and only the output layer is trained. We are now challenging to solve various problems by using optical reservoir computing circuits.

NTT Device Technology Laboratories - Photonics-Electronics Convergence Laboratory