Due to their analog nature, system errors can decrease the accuracy of large DNN models processed on this hardware. Analog weighting of optical inputs can be implemented with masks, holography or optical interference using acousto-optic modulation 15, 16, 17, 18, spatial light modulation 19, electro-optic or thermo-optic modulation 20, 21, 22, 23, phase-change materials 24 or printed diffractive elements 25. To overcome these electronic limitations, optical systems have previously been proposed to perform linear algebra and data transmission. 1) due to electronic communication 11, clocking 12, thermal management 13 and power delivery 14. However, despite these advances, a central challenge in the field is scaling hardware to keep up with exponentially-growing DNN models 10 (see Fig. To this end, emerging DNN-specific hardware 5, 6, 7, 8 optimizes data access, reuse and communication for mathematical operations: most importantly, general matrix–matrix multiplication (GEMM) and convolution 9. An important step toward unlocking the full potential of DNNs is improving the energy consumption and speed of DNN tasks. A prominent subset of machine learning is the artificial deep neural network (DNN), which has revolutionized many fields, including classification 1, translation 2 and prediction 3, 4.
Machine learning has become ubiquitous in modern data analysis, decision-making, and optimization. We also analyze the energy consumption of the DONN and find that digital optical data transfer is beneficial over electronics when the spacing of computational units is on the order of \(>10\,\upmu \)m. In a proof-of-concept experiment, we demonstrate optical multicast in the classification of 500 MNIST images with a 3-layer, fully-connected network. The path-length-independence of optical energy consumption enables information locality between a transmitter and a large number of arbitrarily arranged receivers, which allows greater flexibility in architecture design to circumvent scaling limitations. To improve scalability, we propose a digital optical neural network (DONN) with intralayer optical interconnects and reconfigurable input values. This trend has been enabled by an increase in available compute power however, efforts to continue to scale electronic processors are impeded by the costs of communication, thermal management, power delivery and clocking. As deep neural network (DNN) models grow ever-larger, they can achieve higher accuracy and solve more complex problems.
0 kommentar(er)
