Ultra-high density and nonvolatile nanophotonic convolution processing unit
Photonic convolution processing units are the core components of photonic neural networks (PNNs). Traditionally, for an N × N photonic convolution processing unit, it typically requires O(N2) or O(Nlog2N) cascaded photonic devices, which leads to larger sizes and lower density. Therefore, nanophoton...
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2023-12-01
|
| Series: | Results in Physics |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2211379723009919 |
| _version_ | 1797430090061053952 |
|---|---|
| author | Zhicheng Wang Junbo Feng Zheng Peng Yuqing Zhang Yilu Wu Yuqi Hu Jiagui Wu Junbo Yang |
| author_facet | Zhicheng Wang Junbo Feng Zheng Peng Yuqing Zhang Yilu Wu Yuqi Hu Jiagui Wu Junbo Yang |
| author_sort | Zhicheng Wang |
| collection | DOAJ |
| description | Photonic convolution processing units are the core components of photonic neural networks (PNNs). Traditionally, for an N × N photonic convolution processing unit, it typically requires O(N2) or O(Nlog2N) cascaded photonic devices, which leads to larger sizes and lower density. Therefore, nanophotonic neural networks (N-PNN) with high density are highly attractive. In this work, we propose an ultra-high density and nonvolatile N-PNN scheme, requiring only a compact photonic device to represent all weight values in the N × N convolution kernel, which results in a computational density of more than 5 petaflop operations per second per square millimeter (POPS/mm2) under ideal conditions. The footprints of the proposed 2 × 2 and 3 × 3 nanophotonic convolution processing units are only 4 × 4 μm2 and 6 × 6 μm2, respectively. Moreover, based on these two types of units, the N-PNN achieves image classification and recognition capabilities comparable to traditional computers. In addition, Owing to its lower insertion loss, our approach holds great significance for the scalability and on-chip integration of large-scale N-PNNs. This facilitated the integration of N-PNNs with existing electronic systems. |
| first_indexed | 2024-03-09T09:22:36Z |
| format | Article |
| id | doaj.art-e5ad74e071c6499c944d19af5b80c4ed |
| institution | Directory Open Access Journal |
| issn | 2211-3797 |
| language | English |
| last_indexed | 2024-03-09T09:22:36Z |
| publishDate | 2023-12-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Results in Physics |
| spelling | doaj.art-e5ad74e071c6499c944d19af5b80c4ed2023-12-02T06:59:41ZengElsevierResults in Physics2211-37972023-12-0155107198Ultra-high density and nonvolatile nanophotonic convolution processing unitZhicheng Wang0Junbo Feng1Zheng Peng2Yuqing Zhang3Yilu Wu4Yuqi Hu5Jiagui Wu6Junbo Yang7College of Artificial Intelligence, Southwest University, Chongqing 400715, China; Center of Material Science, National University of Defense Technology, Changsha 410073, ChinaUnited Microelectronics Center Co., Ltd, Chongqing 401332, ChinaCollege of Artificial Intelligence, Southwest University, Chongqing 400715, China; Center of Material Science, National University of Defense Technology, Changsha 410073, ChinaCollege of Artificial Intelligence, Southwest University, Chongqing 400715, China; Center of Material Science, National University of Defense Technology, Changsha 410073, ChinaCollege of Artificial Intelligence, Southwest University, Chongqing 400715, ChinaCollege of Artificial Intelligence, Southwest University, Chongqing 400715, ChinaSchool of Physical Science and Technology, Southwest University, Chongqing 400715, China; Corresponding author.Center of Material Science, National University of Defense Technology, Changsha 410073, China; Corresponding author.Photonic convolution processing units are the core components of photonic neural networks (PNNs). Traditionally, for an N × N photonic convolution processing unit, it typically requires O(N2) or O(Nlog2N) cascaded photonic devices, which leads to larger sizes and lower density. Therefore, nanophotonic neural networks (N-PNN) with high density are highly attractive. In this work, we propose an ultra-high density and nonvolatile N-PNN scheme, requiring only a compact photonic device to represent all weight values in the N × N convolution kernel, which results in a computational density of more than 5 petaflop operations per second per square millimeter (POPS/mm2) under ideal conditions. The footprints of the proposed 2 × 2 and 3 × 3 nanophotonic convolution processing units are only 4 × 4 μm2 and 6 × 6 μm2, respectively. Moreover, based on these two types of units, the N-PNN achieves image classification and recognition capabilities comparable to traditional computers. In addition, Owing to its lower insertion loss, our approach holds great significance for the scalability and on-chip integration of large-scale N-PNNs. This facilitated the integration of N-PNNs with existing electronic systems.http://www.sciencedirect.com/science/article/pii/S2211379723009919 |
| spellingShingle | Zhicheng Wang Junbo Feng Zheng Peng Yuqing Zhang Yilu Wu Yuqi Hu Jiagui Wu Junbo Yang Ultra-high density and nonvolatile nanophotonic convolution processing unit Results in Physics |
| title | Ultra-high density and nonvolatile nanophotonic convolution processing unit |
| title_full | Ultra-high density and nonvolatile nanophotonic convolution processing unit |
| title_fullStr | Ultra-high density and nonvolatile nanophotonic convolution processing unit |
| title_full_unstemmed | Ultra-high density and nonvolatile nanophotonic convolution processing unit |
| title_short | Ultra-high density and nonvolatile nanophotonic convolution processing unit |
| title_sort | ultra high density and nonvolatile nanophotonic convolution processing unit |
| url | http://www.sciencedirect.com/science/article/pii/S2211379723009919 |
| work_keys_str_mv | AT zhichengwang ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT junbofeng ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT zhengpeng ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT yuqingzhang ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT yiluwu ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT yuqihu ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT jiaguiwu ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit AT junboyang ultrahighdensityandnonvolatilenanophotonicconvolutionprocessingunit |