E-mail Alert
RSS
| Citation: |
Wan YJ, Liu XD, Wu GZ et al. Efficient stochastic parallel gradient descent training for on-chip optical processor. Opto-Electron Adv 7, 230182 (2024). doi: 10.29026/oea.2024.230182
|
Efficient stochastic parallel gradient descent training for on-chip optical processor
-
Abstract
In recent years, space-division multiplexing (SDM) technology, which involves transmitting data information on multiple parallel channels for efficient capacity scaling, has been widely used in fiber and free-space optical communication systems. To enable flexible data management and cope with the mixing between different channels, the integrated reconfigurable optical processor is used for optical switching and mitigating the channel crosstalk. However, efficient online training becomes intricate and challenging, particularly when dealing with a significant number of channels. Here we use the stochastic parallel gradient descent (SPGD) algorithm to configure the integrated optical processor, which has less computation than the traditional gradient descent (GD) algorithm. We design and fabricate a 6×6 on-chip optical processor on silicon platform to implement optical switching and descrambling assisted by the online training with the SPDG algorithm. Moreover, we apply the on-chip processor configured by the SPGD algorithm to optical communications for optical switching and efficiently mitigating the channel crosstalk in SDM systems. In comparison with the traditional GD algorithm, it is found that the SPGD algorithm features better performance especially when the scale of matrix is large, which means it has the potential to optimize large-scale optical matrix computation acceleration chips. -
-
References
[1] Richardson DJ, Fini JM, Nelson LE. Space-division multiplexing in optical fibres. Nat Photonics 7, 354–362 (2013). doi: 10.1038/nphoton.2013.94 [2] Berdagué S, Facq P. Mode division multiplexing in optical fibers. Appl Opt 21, 1950–1955 (1982). doi: 10.1364/AO.21.001950 [3] Ryf R, Fontaine NK, Wittek S et al. High-spectral-efficiency mode-multiplexed transmission over graded-index multimode fiber. In 2018 European Conference on Optical Communication (ECOC) 1–3 (IEEE, 2018);http://doi.org/10.1109/ECOC.2018.8535536. [4] van Uden RGH, Correa RA, Lopez EA et al. Ultra-high-density spatial division multiplexing with a few-mode multicore fibre. Nat Photonics 8, 865–870 (2014). doi: 10.1038/nphoton.2014.243 [5] Ding QC, Liu B, Ren JX et al. High-security SCMA-OFDM multi-core fiber transmission system based on a regular hexagon chaotic codebook. Opt Express 30, 36010–36024 (2022). doi: 10.1364/OE.471151 [6] Puttnam BJ, Eriksson TA, Mendinueta JMD et al. Modulation formats for multi-core fiber transmission. Opt Express 22, 32457–32469 (2014). doi: 10.1364/OE.22.032457 [7] Zhu L, Zhu GX, Wang AD et al. 18 km low-crosstalk OAM + WDM transmission with 224 individual channels enabled by a ring-core fiber with large high-order mode group separation. Opt Lett 43, 1890–1893 (2018). doi: 10.1364/OL.43.001890 [8] Zhu GX, Hu ZY, Wu X et al. Scalable mode division multiplexed transmission over a 10-km ring-core fiber using high-order orbital angular momentum modes. Opt Express 26, 594–604 (2018). doi: 10.1364/OE.26.000594 [9] Wang J, Yang JY, Fazal IM et al. Terabit free-space data transmission employing orbital angular momentum multiplexing. Nat Photonics 6, 488–496 (2012). doi: 10.1038/nphoton.2012.138 [10] Zhao YF, Liu J, Du J et al. Experimental demonstration of 260-meter security free-space optical data transmission using 16-QAM carrying orbital angular momentum (OAM) beams multiplexing. In Optical Fiber Communication Conference Th1H. 3 (OSA, 2016); http://doi.org/10.1364/OFC.2016.Th1H.3. [11] Koebele C, Salsi M, Milord L et al. 40km transmission of five mode division multiplexed data streams at 100 Gb/s with low MIMO-DSP complexity. In 37th European Conference and Exposition on Optical Communications Th. 13. C. 3 (OSA, 2011); http://doi.org/10.1364/ECOC.2011.Th.13.C.3. [12] Randel S, Corteselli S, Badini D et al. First real-time coherent MIMO-DSP for six coupled mode transmission. In 2015 IEEE Photonics Conference (IPC) 1–2 (IEEE, 2015);http://doi.org/10.1109/IPCon.2015.7323761. [13] Randel S, Sierra A, Mumtaz S et al. Adaptive MIMO signal processing for mode-division multiplexing. In Optical Fiber Communication Conference OW3D. 5 (OSA, 2012);http://doi.org/10.1364/OFC.2012.OW3D.5. [14] Diamantopoulos NP, Shariati B, Tomkos I. On the power consumption of MIMO processing and its impact on the performance of SDM networks. In 2017 Optical Fiber Communications Conference and Exhibition (OFC) 1–3 (IEEE, 2017). [15] Yang L, Zhou T, Jia H et al. General architectures for on-chip optical space and mode switching. Optica 5, 180–187 (2018). doi: 10.1364/OPTICA.5.000180 [16] Clements WR, Humphreys PC, Metcalf BJ et al. Optimal design for universal multiport interferometers. Optica 3, 1460–1465 (2016). doi: 10.1364/OPTICA.3.001460 [17] Cao XP, Zheng S, Long Y et al. Mesh-structure-enabled programmable multitask photonic signal processor on a silicon chip. ACS Photonics 7, 2658–2675 (2020). doi: 10.1021/acsphotonics.9b01230 [18] Little BE, Chu ST, Haus HA et al. Microring resonator channel dropping filters. J Lightwave Technol 15, 998–1005 (1997). doi: 10.1109/50.588673 [19] Pérez D, Gasulla I, Capmany J. Programmable multifunctional integrated nanophotonics. Nanophotonics 7, 1351–1371 (2018). doi: 10.1515/nanoph-2018-0051 [20] Zhuang LM, Roeloffzen CGH, Hoekman M et al. Programmable photonic signal processor chip for radiofrequency applications. Optica 2, 854–859 (2015). doi: 10.1364/OPTICA.2.000854 [21] Ribeiro A, Ruocco A, Vanacker L et al. Demonstration of a 4 × 4-port universal linear circuit. Optica 3, 1348–1357 (2016). doi: 10.1364/OPTICA.3.001348 [22] Lu LJ, Zhao SY, Zhou LJ et al. 16 × 16 non-blocking silicon optical switch based on electro-optic Mach-Zehnder interferometers. Opt Express 24, 9295–9307 (2016). doi: 10.1364/OE.24.009295 [23] Ferrera M, Park Y, Razzari L et al. On-chip CMOS-compatible all-optical integrator. Nat Commun 1, 29 (2010). doi: 10.1038/ncomms1028 [24] Dong JJ, Zheng AL, Gao DS et al. High-order photonic differentiator employing on-chip cascaded microring resonators. Opt Lett 38, 628–630 (2013). doi: 10.1364/OL.38.000628 [25] Liu WL, Li M, Guzzon RS et al. A fully reconfigurable photonic integrated signal processor. Nat Photonics 10, 190–195 (2016). doi: 10.1038/nphoton.2015.281 [26] Zhou HL, Dong JJ, Cheng JW et al. Photonic matrix multiplication lights up photonic accelerator and beyond. Light Sci Appl 11, 30 (2022). doi: 10.1038/s41377-022-00717-8 [27] Fandiño JS, Muñoz P, Doménech D et al. A monolithic integrated photonic microwave filter. Nat Photonics 11, 124–129 (2017). doi: 10.1038/nphoton.2016.233 [28] Kouloumentas C, Tsokos C, Groumas P et al. Multi-rate and multi-channel optical equalizer based on photonic integration. IEEE Photonics Technol Lett 32, 1465–1468 (2020). doi: 10.1109/LPT.2020.3035506 [29] Zhou HL, Zhao YH, Wang X et al. Self-configuring and reconfigurable silicon photonic signal processor. ACS Photonics 7, 792–799 (2020). doi: 10.1021/acsphotonics.9b01673 [30] Pérez D, Gasulla I, Crudgington L et al. Multipurpose silicon photonics signal processor core. Nat Commun 8, 636 (2017). doi: 10.1038/s41467-017-00714-1 [31] Lu LJ, Zhou LJ, Chen JP. Programmable SCOW mesh silicon photonic processor for linear unitary operator. Micromachines (Basel) 10, 646 (2019). doi: 10.3390/mi10100646 [32] Zhou HL, Zhao YH, Xu GX et al. Chip-scale optical matrix computation for pagerank algorithm. IEEE J Sel Top Quantum Electron 26, 8300910 (2020). [33] Zheng Y, Zhai CH, Liu DJ et al. Multichip multidimensional quantum networks with entanglement retrievability. Science 381, 221–226 (2023). doi: 10.1126/science.adg9210 [34] Tian Y, Zhao Y, Liu SP et al. Scalable and compact photonic neural chip with low learning-capability-loss. Nanophotonics 11, 329–344 (2022). doi: 10.1515/nanoph-2021-0521 [35] Zhang H, Gu M, Jiang XD et al. An optical neural chip for implementing complex-valued neural network. Nat Commun 12, 457 (2021). doi: 10.1038/s41467-020-20719-7 [36] Feng CH, Gu JQ, Zhu HQ et al. A compact butterfly-style silicon photonic–electronic neural chip for hardware-efficient deep learning. ACS Photonics 9, 3906–3916 (2022). doi: 10.1021/acsphotonics.2c01188 [37] Shen YC, Harris NC, Skirlo S et al. Deep learning with coherent nanophotonic circuits. Nat Photonics 11, 441–446 (2017). doi: 10.1038/nphoton.2017.93 [38] Annoni A, Guglielmi E, Carminati M et al. Unscrambling light—automatically undoing strong mixing between modes. Light Sci Appl 6, e17110 (2017). doi: 10.1038/lsa.2017.110 [39] Tanomura R, Tang R, Soma G et al. All-optical dual-polarization MIMO processor based on integrated optical unitary converter. In 2022 European Conference on Optical Communication (ECOC) 1–4 (IEEE, 2022). [40] Tanomura R, Tang R, Ghosh S et al. Robust integrated optical unitary converter using multiport directional couplers. J Lightwave Technol 38, 60–66 (2020). doi: 10.1109/JLT.2019.2943116 [41] Hughes TW, Minkov M, Shi Y et al. Training of photonic neural networks through in situ backpropagation and gradient measurement. Optica 5, 864–871 (2018). doi: 10.1364/OPTICA.5.000864 [42] Zhang H, Thompson J, Gu ML et al. Efficient on-chip training of optical neural networks using genetic algorithm. ACS Photonics 8, 1662–1672 (2021). doi: 10.1021/acsphotonics.1c00035 [43] Shao R, Zhang G, Gong X. Generalized robust training scheme using genetic algorithm for optical neural networks with imprecise components. Photonics Res 10, 1868–1876 (2022). doi: 10.1364/PRJ.449570 [44] Zhang T, Wang J, Dan YH et al. Efficient training and design of photonic neural network through neuroevolution. Opt Express 27, 37150–37163 (2019). doi: 10.1364/OE.27.037150 [45] Cong GW, Yamamoto N, Inoue T et al. On-chip bacterial foraging training in silicon photonic circuits for projection-enabled nonlinear classification. Nat Commun 13, 3261 (2022). doi: 10.1038/s41467-022-30906-3 [46] Geng C, Luo W, Tan Y et al. Experimental demonstration of using divergence cost-function in SPGD algorithm for coherent beam combining with tip/tilt control. Opt Express 21, 25045–25055 (2013). doi: 10.1364/OE.21.025045 [47] Vorontsov MA, Sivokon VP. Stochastic parallel-gradient-descent technique for high-resolution wave-front phase-distortion correction. J Opt Soc Am A 15, 2745–2758 (1998). doi: 10.1364/JOSAA.15.002745 [48] Reck M, Zeilinger A, Bernstein HJ et al. Experimental realization of any discrete unitary operator. Phys Rev Lett 73, 58–61 (1994). doi: 10.1103/PhysRevLett.73.58 [49] Harris NC, Carolan J, Bunandar D et al. Linear programmable nanophotonic processors. Optica 5, 1623–1631 (2018). doi: 10.1364/OPTICA.5.001623 [50] Harris NC, Braid R, Bunandar D et al. Accelerating artificial intelligence with silicon photonics. In Optical Fiber Communication Conference (OFC) 2020 W3A. 3 (Optica Publishing Group, 2020);http://doi.org/10.1364/OFC.2020.W3A.3. -
Access History
Figures(8)
Tables(1)
Article Metrics
Export File
Citation
Wan YJ, Liu XD, Wu GZ et al. Efficient stochastic parallel gradient descent training for on-chip optical processor. Opto-Electron Adv 7, 230182 (2024). doi: 10.29026/oea.2024.230182
Format
Content
DownLoad:
-
Figure 1.
(a) Conceptual diagram of the on-chip optical processor for optical switching and channel descrambling in MDM communication systems. (b) Schematic configuration of the integrated reconfigurable optical processor. θ and ϕ mean the phase shift of the phase shifters. MDM: mode-division multiplexing; MUX: multiplexer; DEMUX: demultiplexer.
-
Figure 2.
Flow chart of Stochastic Parallel Gradient Descent (SPGD) algorithm.
-
Figure 3.
Training results in electronic computer for optical switching, optical channel descrambling, and optical channel descrambling and switching. (a) Emulated light power distributions and (b) normalized light intensity distributions after training when the switching state is I1−O2, I2−O1, I3−O5, I4−O6, I5−O3, I6−O4. (d, e) Normalized light intensity distributions (d) before and (e) after training when randomly generating a set of phases in the part (1) of our chip to emulate crosstalk. (g, h) Normalized light intensity distributions (g) before and (h) after training with crosstalk when the switching state is: I1−O5, I2−O3, I3−O2, I4−O4, I5−O1, I6−O6. (c, f, i) The evaluation function changing with iteration rounds.
-
Figure 4.
(a) Schematic of experimental configuration. (b) Microscopy image of optical processor. VSA: voltage source array; PD: photodetector array.
-
Figure 5.
Online training results for optical switching at a wavelength of 1550 nm. (a) The evaluation function changing with iteration rounds when the switching state is I1−O3, I2−O1, I3−O4, I4−O6, I5−O2, I6−O5. The insets figures show the light power distributions when the round of iteration equals 50, 300, and 600, respectively. (b) The measured light power distributions after training. (c) The normalized light intensity distributions of measured results. (d, e) The measured light power distributions and normalized light intensity distributions when the switching state is I1−O3, I2−O6, I3−O4, I4−O2, I5−O1, I6−O5.
-
Figure 6.
Online training results for optical channel descrambling at a wavelength of 1550 nm. (a) The evaluation function changing with iteration rounds. The insets show the light power distributions when the round of iteration equals 1, 300, and 600, respectively. (b) The light power distributions before training. (c) The light power distributions after training. (d, e) The results of training when generating another matrix
$\widetilde {\boldsymbol{U}}$ -
Figure 7.
Online training results for optical channel descrambling and switching at a wavelength of 1550 nm. (a) The evaluation function changing with iteration rounds when the switching state is I1−O4, I2−O1, I3−O5, I4−O6, I5−O3, I6−O2. The insets show the light power distributions when the round of iteration equals 1, 100, and 400, respectively. (b) The light power distributions before training. (c) The light power distributions after training. (d, e) The results of training when generating another matrix
$\widetilde {\boldsymbol{U}}$ -
Figure 8.
Experimental setup and measured results for optical channel descrambling. (a) Experimental setup for the 6×6 optical descrambling systems. (b) The measured BER performance for back-to-back, optimization without crosstalk, before optimization with crosstalk, and after optimization with crosstalk systems. (c) The measured constellation chart at the back-to-back. (d) The measured constellation chart without crosstalk. (e) The measured constellation chart before optimization with crosstalk. (f) The measured constellation chart after optimization with crosstalk. PC: polarization controller; AWG: arbitrary waveform generator; EDFA: erbium-doped fiber amplifier; VOA: variable optical attenuator; OSC: oscilloscope; DSP: digital signal processing.
