Electro-photonic Integrated Deep Learning Processor using Si Photonic Integrated Circuits
Dr. Mitsuru Takenaka
Professor
School of Engineering, Department of Electrical Engineering and Information Systems
The University of Tokyo
*The organization and the title are those when awarded
Research summary
Prof. Takenaka has conducted pioneering research on the application of devices that integrate heterogeneous materials such as compound semiconductors, phase-change materials, and two-dimensional materials into silicon photonic devices for deep learning processors.
Deep learning processors utilizing reconfigurable silicon photonic circuits (programmable photonic circuits) are expected to be capable of performing high-speed, low-power, and low-latency summation and multiplication operations, thereby improving the performance of artificial intelligence (AI) regardless of semiconductor miniaturization. Research on this next-generation computing technology is being conducted worldwide. However, in practical-scale programmable photonic circuits, precise measurement and control of optical phase within the circuit and measurement techniques that can convert optical operation results into low-power and high-speed photodetection are of utmost importance.
Prof. Takenaka has been challenging the precise measurement and control of optical phase and intensity within photonic circuits by integrating compound semiconductors and phase-change materials into silicon photonic circuits. He is also engaged in research to achieve a new programmable photonic circuit that allows learning acceleration through error backpropagation on the optical circuit. These achievements are expected to greatly contribute to the early realization of deep learning processors using silicon photonic circuits.
A semiconductor material composed of multiple elements, exhibiting a wide range of physical and electronic characteristics and offering advantages such as high electron mobility and improved optical properties.
Materials that exhibit the property of changing their phase (state) in response to variations in temperature or pressure.
Materials that are extremely thin, with a surface structured in a two-dimensional manner. They are composed of atomic or molecular monolayers or a few layers.
An integrated circuit designed to perform specialized digital information processing by combining photonic and electronic circuits. It is used to execute the machine learning technique called deep learning.
A silicon-based circuit that can control the flow of light through programming.
Information related to the position and direction of propagation of light waves.
A technique that converts computational results performed within an optical circuit into electrical signals and reads the data.
A learning method in neural networks where errors are propagated in the reverse direction to adjust the weights and biases, enabling accurate output generation.
A semiconductor material composed of multiple elements, exhibiting a wide range of physical and electronic characteristics and offering advantages such as high electron mobility and improved optical properties.
Materials that exhibit the property of changing their phase (state) in response to variations in temperature or pressure.
Materials that are extremely thin, with a surface structured in a two-dimensional manner. They are composed of atomic or molecular monolayers or a few layers.
An integrated circuit designed to perform specialized digital information processing by combining photonic and electronic circuits. It is used to execute the machine learning technique called deep learning.
A silicon-based circuit that can control the flow of light through programming.
Information related to the position and direction of propagation of light waves.
A technique that converts computational results performed within an optical circuit into electrical signals and reads the data.
A learning method in neural networks where errors are propagated in the reverse direction to adjust the weights and biases, enabling accurate output generation.
Introduction to Research
