Review of On‑Chip Integrated Optical Neural Networks (Invited)

Tingzhao Fu; Run Sun; Yuyao Huang; Jianfa Zhang; Sigang Yang; Zhihong Zhu; Hongwei Chen

doi:10.3788/CJL231227

Journals >Chinese Journal of Lasers >Volume 51 >Issue 1 >Page 0119002 > Article

Chinese Journal of Lasers
Vol. 51, Issue 1, 0119002 (2024)

Review of On‑Chip Integrated Optical Neural Networks (Invited)

Tingzhao Fu^1、4、5, Run Sun^2、3, Yuyao Huang^2、3, Jianfa Zhang^1、4、5, Sigang Yang^2、3, Zhihong Zhu^1、4、5, and Hongwei Chen^2、3、*

Author Affiliations

¹College of Advanced Interdisciplinary Studies, National University of Defense Technology, Changsha 410073, Hunan, China

²Department of Electronic Engineering, Tsinghua University, Beijing 100084, China

³Beijing National Research Center for Information Science and Technology, Beijing 100084, China

⁴Hunan Provincial Key Laboratory of Novel Nano-Optoelectronic Information Materials and Devices, National University of Defense Technology, Changsha 410073, Hunan, China

⁵Nanhu Laser Laboratory, National University of Defense Technology, Changsha 410073, Hunan, China

show less

DOI: 10.3788/CJL231227 Cite this Article Set citation alerts

Tingzhao Fu, Run Sun, Yuyao Huang, Jianfa Zhang, Sigang Yang, Zhihong Zhu, Hongwei Chen. Review of On‑Chip Integrated Optical Neural Networks (Invited)[J]. Chinese Journal of Lasers, 2024, 51(1): 0119002 Copy Citation Text

show less

On-chip optical neural networks based on MZI interference structure. (a) MZI topology cascaded array[25]; (b) ONN supporting in-situ online training and gradient backpropagation[27]

Fig. 1. On-chip optical neural networks based on MZI interference structure. (a) MZI topology cascaded array^[25]; (b) ONN supporting in-situ online training and gradient backpropagation^[27]

Download full size

$Improved on-chip optical neural networks based on MZI interference structure. (a) Complex ONN based on MZI array[41]; (b) ONN based on MZI array and diffractive units[47]$

Fig. 2. Improved on-chip optical neural networks based on MZI interference structure. (a) Complex ONN based on MZI array^[41]; (b) ONN based on MZI array and diffractive units^[47]

Download full size

Fig. 3. On-chip optical neural networks based on MRR wavelength division structure. (a) ONN based on MRR wavelength division system and PCM units^[31]; (b) ONN for fiber nonlinear compensation^[43]

Download full size

Fig. 4. On-chip optical neural networks based on MRRs and other auxiliary optical devices. (a) ONN based on MRR cross array and supporting gradient backpropagation^[57]; (b) ONN system integrating light sources, data loading areas, and data processing units on single chip^[59]

Download full size

$On-chip diffractive optical neural networks verified by simulation. (a) DONN with computational unit characterized by single subwavelength diffractive structure [35]; (b) DONN with computational unit characterized by subwavelength diffractive structure group[44]$

Fig. 5. On-chip diffractive optical neural networks verified by simulation. (a) DONN with computational unit characterized by single subwavelength diffractive structure ^[35]; (b) DONN with computational unit characterized by subwavelength diffractive structure group^[44]

Download full size

$On-chip diffractive optical neural networks verified by experiment. (a) DONN with computational unit composed by two identical subwavelength diffractive structures[50]; (b) DONN with computational unit composed by three identical subwavelength diffractive structures [60]$

Fig. 6. On-chip diffractive optical neural networks verified by experiment. (a) DONN with computational unit composed by two identical subwavelength diffractive structures^[50]; (b) DONN with computational unit composed by three identical subwavelength diffractive structures ^[60]

Download full size

Fig. 7. On-chip optical neural networks based on other structures. (a) ONN based on 3D integrated waveguide array^[85]; (b) designed ONN based on inverse design method^[36]

Download full size

Fig. 8. ONN based on waveguide attenuation modulators^[53]

Download full size

Reference	Basic unit	Theoretical integration /（NBU /mm²）	Operational power consumption /（J/operation）	Throughput /TOPS
Ref.［25］	MZI	<10	7.66 $\times$ 10^-14	6.4
Ref.［41］	MZI	<10	2.14 $\times$ 10^-13	21.6
Ref.［42］	MRR covered with PCM	<5	5.9 $\times$ 10^-15	28.8
Ref.［47］	MZI and Diffractive cell	<20	1.41 $\times$ 10^-15	32
Ref.［50］	Subwavelength unit	~6.7 $\times$ 10³	4.2 $\times$ 10^-19	4.05×10⁴
Ref.［60］	Subwavelength unit	~2 $\times$ 10³	1.1 $\times$ 10^-17	1.38×10⁴

Table 1. Partial performance comparison of on-chip integrated ONNs

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information