A highly scalable and reconfigurable optical convolution paradigm based on wavelength routing is proposed, which leverages the unique sliding property of an arrayed waveguide grating router (AWGR) to execute the sliding window operation of convolution in the wavelength-space domains. By directly loading two input vectors onto two modulator arrays, the convolution result is instantaneously generated at a photodetector array at the speed of light propagation. This enables the entire convolution computation to be executed within one clock cycle, eliminating the necessity for preprocessing or decomposition into elementary MAC operations. The proposed optical convolution unit (OCU) has striking advantages of high scalability, high speed, and processing simplicity compared to those based on optical matrix-vector multipliers (MVM). A proof-of-concept experiment employing standalone optical components is devised to validate optical convolution computing principles with one-bit accuracy. The classification of ten handwritten digit classes sourced from the MNIST database is experimentally demonstrated, achieving a precision of 4-bit. New algorithms for data splitting and reorganization were concurrently introduced to facilitate the convolution calculation of two-dimensional image data. Notably, through Field-Programmable Gate Array (FPGA) across varying data transmission speeds of 1MHz, 5MHz, and 10MHz, inference accuracy rates of 97.32%, 96.25%, and 94.50% were respectively achieved, demonstrating the robustness and versatility of the proposed paradigm.
Optical neural networks (ONNs) have the potential for accelerating the inference of AI models, since ONNs have the advantage of high compute speed, and high parallelism. Integrated diffractive optical network for implementing parallel Fourier transforms has been proved efficient and is promising for large scale ONNs. We propose a novel on-chip Fourier transform implementation based on etched concave mirrors, enabling the construction of a photonic integrated 4F system to perform the convolution computation in the Convolutional Neural Networks (CNNs). One of the input vectors is encoded in a modulator array at the object plane, and the Fourier transform of the other vector is encoded in another modulator array at the spectrum plane. We simulated the computing process by the diffractive propagation of the optical field from the object plane to the image plane according to the Kirchhoff's diffraction formula. Finally, we used our simulation system to replace the traditional convolution layers in the electronic system to implement CNNs on three different datasets, Iris- Flower, MNIST and Fashion-MNIST, and obtained 96.67%, 95.6% and 89.4% classification accuracies, respectively, demonstrating comparable performance with the electronic counterpart.
In this paper, we propose a method to realize complex amplitude modulation with a 4-level phase plate and an amplitude-modulated SLM. We implement this complex amplitude modulator system in classical 4f system and use this system to realize highly precise convolution to verify its feasibility. The optical convolution accuracy is higher than the phase-only modulator or the amplitude-only modulator method. We then incorporate this system with a one-layer CNN as its convolutional layer to accomplish a handwriting recognition inference, the image classification accuracy is also comparable to all-electrical result with the same framework.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.