Abstract
Reconfigurable linear optical processors can be used to perform linear transformations and are instrumental in effectively computing matrix–vector multiplications required in each neural network layer. In this paper, we characterize and compare two thermally tuned photonic integrated processors realized in silicon-on-insulator and silicon nitride platforms suited for extracting feature maps in convolutional neural networks. The reduction in bit resolution when crossing the processor is mainly due to optical losses, in the range 2.3–3.3 for the silicon-on-insulator chip and in the range 1.3–2.4 for the silicon nitride chip. However, the lower extinction ratio of Mach–Zehnder elements in the latter platform limits their expressivity (i.e., the capacity to implement any transformation) to 75%, compared to 97% of the former. Finally, the silicon-on-insulator processor outperforms the silicon nitride one in terms of footprint and energy efficiency.
Highlights
In the field of artificial intelligence (AI), deep learning has been tremendously successful over recent years, reaching unprecedented milestones in various fields, such as image recognition, natural language processing, genome analysis, and autonomous driving [1].The use of graphics processing units (GPUs) as accelerators for computation required in deep neural networks (DNNs) was pivotal for their success
Convolutional neural networks (CNNs) are the kind of DNNs used in computer vision, whose convolutional layers are composed of square filters with slow-changing parameters
In this paper we have reported the comparison of two reconfigurable linear optical processors based on thermally tunable Mach–Zehnder interferometer (MZI) meshes
Summary
In the field of artificial intelligence (AI), deep learning has been tremendously successful over recent years, reaching unprecedented milestones in various fields, such as image recognition, natural language processing, genome analysis, and autonomous driving [1]. We present two photonic integrated reconfigurable linear processors based on 2 × 2 interferometric elements [5] These devices are versatile, being exploited over the years for optical processing and filtering [6], programmable true-time optical delay lines [7], and quantum information processing [8,9]. Recent results exploiting reduced-precision numerical formats for neural network computation strongly motivate the use of a few-bit resolution. This opens up the possibility to exploit analog hardware to accelerate DNN with tolerable or even negligible accuracy losses in the DNN model prediction [10,11,12,13].
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.