Photonic AI processor; optoelectronic physical reservoir computing; optical programmable logic array.
Researchers from Massachusetts Institute of Technology (MIT), Enosemi, and Periplous developed a fully integrated photonic processor that can perform all the key computations of a deep neural network optically on the chip. The chip is fabricated using commercial foundry processes and uses three layers of devices that perform linear and nonlinear operations.
A particular challenge was implementing nonlinear operations on the chip. To overcome this, the team designed nonlinear optical function units (NOFUs), which combine electronics and optics.
First, the system encodes the parameters of a deep neural network into light. Then, an array of programmable beamsplitters performs matrix multiplication on those inputs. The data then passes to the programmable NOFUs, which implement nonlinear functions by siphoning off a small amount of light to photodiodes that convert optical signals to electric current, a process that consumes little energy and eliminates the need for an external amplifier.
“We stay in the optical domain the whole time, until the end when we want to read out the answer. This enables us to achieve ultra-low latency,” said Saumil Bandyopadhyay, a visiting scientist in the Quantum Photonics and AI Group within the Research Laboratory of Electronics (RLE) at MIT and a postdoc at NTT Research, in a statement. The low latency enabled a deep neural network to be trained on the chip. “This is especially useful for systems where you are doing in-domain processing of optical signals, like navigation or telecommunications, but also in systems that you want to learn in real time.”
The optical device was able to complete the key computations for a machine-learning classification task in less than half a nanosecond while achieving more than 92% accuracy. [1]
Researchers from the Tokyo University of Science fabricated a self-powered dye-sensitized solar cell-based physical reservoir computing (PRC) optoelectronic device for efficient edge AI processing of time-series data.
“In order to process time-series input optical data with various time scales, it is essential to fabricate devices according to the desired time scale. Inspired by the afterimage phenomenon of the eye, we came up with a novel optoelectronic human synaptic device that can serve as a computational framework for power-saving edge AI optical sensors,” said Takashi Ikuno, an associate professor in the Department of Applied Electronics at TUS.
The solar cell-based device utilizes squarylium derivative-based dyes and incorporates optical input, AI computation, analog output, and power supply functions in the device itself at the material level. It can be controlled by the input light intensity. It has ultra-low power consumption and was able to classify human movements such as bending, jumping, running, and walking with more than 90% accuracy. [2]
Researchers from Huazhong University of Science and Technology and Wuhan National Laboratory for Optoelectronics developed a large-scale optical programmable logic array (PLA) that uses parallel spectrum modulation to achieve an 8-input system, making it capable of handling more complex logic operations.
The optical PLA showed the ability to handle advanced logic functions such as decoders, comparators, adders, and multipliers. This enabled it to successfully run Conway’s Game of Life and other cellular automata models, including the Sierpinski triangle, without relying on electronic components for nonlinear computing. [3]
[1] Bandyopadhyay, S., Sludds, A., Krastanov, S. et al. Single-chip photonic deep neural network with forward-only training. Nat. Photon. 18, 1335–1343 (2024). https://doi.org/10.1038/s41566-024-01567-z
[2] Komatsu, H., Hosoda, N., Ikuno, T. Self-Powered Dye-Sensitized Solar-Cell-Based Synaptic Devices for Multi-Scale Time-Series Data Processing in Physical Reservoir Computing. ACS Applied Materials & Interfaces (2024). https://doi.org/10.1021/acsami.4c11061
[3] Zhang, W., Wu, B., Gu, W. et al. Large-scale optical programmable logic array for two-dimensional cellular automaton. Adv. Photon. 6(5) 056007 (17 October 2024) https://doi.org/10.1117/1.AP.6.5.056007
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply