Tensorized Optical Neural Network Architecture

Abstract

Researchers at the University of California, Davis have developed a large-scale, energy-efficient, high-throughput, and compact tensorized optical neural network (TONN) exploiting the tensor-train decomposition architecture on an integrated III–V-on-silicon metal–oxide–semiconductor capacitor (MOSCAP) platform.

Full Description

The technology provides a solution of using a TONN architecture to address and mitigate challenges of optical neural networks. The TONN architecture is scalable to 1024×1024 synapses and beyond, which is extremely difficult for conventional integrated ONN architectures, by using cascaded multi-wavelength small-radix (e.g., 8 × 8) tensor cores.

Applications

• Computer vision 
• Speech recognition 
• Machine translations 
• Medical diagnoses 
• Advanced gaming 
• Large-volume and cost-effective EPIC manufacturing

Features/Benefits

• Scalable synapses 
• Utilizes fewer Mach–Zehnder interferometers (MZIs) and fewer cascaded stages of MZIs than conventional ONNs 
• Maintains a training accuracy for Modified National Institute of Standards and Technology handwritten digit classification tasks 
• Reduces the footprint-energy compared with digital electronics ANN hardware 
• Steps ahead compared with silicon photonic and phase-change material technologies 
• Overcomes the limited scalability of conventional ONNs 
• Reduces the dependency on the type of task, unlike CNNs 
• Does away with the need for alignment of III–V diode laser chips to Silicon Photonics chips, thereby eliminating related losses and packaging complexity

Patent Status

Patent Pending