Experimentally realized in situ backpropagation for deep learning in photonic neural networks

Abstract

Integrated photonic neural networks provide a promising platform for energy-efficient, high-throughput machine learning with extensive scientific and commercial applications. Photonic neural networks efficiently transform optically encoded inputs using Mach-Zehnder interferometer mesh networks interleaved with nonlinearities. We experimentally trained a three-layer, four-port silicon photonic neural network with programmable phase shifters and optical power monitoring to solve classification tasks using “in situ backpropagation,” a photonic analog of the most popular method to train conventional neural networks. We measured backpropagated gradients for phase-shifter voltages by interfering forward- and backward-propagating light and simulated in situ backpropagation for 64-port photonic neural networks trained on MNIST image recognition given errors. All experiments performed comparably to digital simulations ( > 94% test accuracy), and energy scaling analysis indicated a route to scalable machine learning.

Seven Bit Nonvolatile Electrically Programmable Photonics Based on Phase-Change Materials for Image Recognition

“…3−6 To date, many ONNs have been demonstrated for artificial intelligence, including pattern recognition, 7 image classification, 8,9 and perceptron. 10 However, in traditional on-chip ONNs, synaptic weights are determined by basic units of PICs by altering their optical phase or intensity. These basic units typically employ the volatile thermo-optic effect or free carrier dispersion effect, 11,12 suffering from severe heat accumulation, high static power consumption, or/and large footprint, which hampers the scalability of programmable photonic networks.…”

“…With the explosive growth of information in the era of artificial intelligence, the demand for the high efficiency of computing systems is extremely growing. Optical neural networks (ONNs) can achieve efficient computing with more accurate information extraction and fewer network parameters and are considered as a promising candidate for the next generation of neural morphology hardware processors. , That is because ONNs based on photonic integrated circuits (PICs) have unparalleled advantages in complex neuromorphic computing due to the inherent high parallelism, low interconnect loss, and ultrahigh computational bandwidth of up to 10 THz of photonics devices. − To date, many ONNs have been demonstrated for artificial intelligence, including pattern recognition, image classification, , and perceptron . However, in traditional on-chip ONNs, synaptic weights are determined by basic units of PICs by altering their optical phase or intensity.…”

Electrically programmable phase-change photonic memory for optical neural networks with nanoseconds in situ training capability

“…3 Optical neural networks (ONNs) based on photonic integrated circuits (PICs) [4][5][6][7][8][9] have the potential to meet this demand as a consequence of their low latency, high parallel (e.g., wavelength/ spatial division multiplexing), and strong anti-electromagnetic interference capability of PICs, as well as the low cost and high yield provided by a complementary metal-oxide-semiconductor (CMOS) fabrication process. [10][11][12][13] Recently, a series of ONNs have been demonstrated for artificial intelligence, including vowel recognition, 14 perceptron, 15,16 pattern recognition, 17 and image classification. 18,19 However, for real-world applications, more efforts are needed to improve the energy efficiency, scalability, and algorithm accuracy of ONNs.…”

Finding best orthogonal channels through arbitrary optical systems using integrated photonic processors

“…We demonstrate that this approach and extensions of it can in practice find such orthogonal channels in a wide range of optical situations. MZI mesh architectures have been extensively used in previous works to "synthetize" arbitrary non-unitary matrices by implementing on-chip SVD [15,16], with powerful demonstrations in computing systems based on photonic vector-matrix multiplication [17,18], photonic accelerators [19], cryptography [20], photonic neural networks [21][22][23][24], photonic analog processors and equation solvers [25], and quantum photonic processors [26].…”