Micro-LED backlight module by deep reinforcement learning and micro-macro-hybrid environment control agent

Che-Hsuan Huang; Yu-Tang Cheng; Yung-Chi Tsao; Xinke Liu; Hao-Chung Kuo

doi:10.1364/PRJ.441188

Photonics Research, Volume. 10, Issue 2, 269(2022)

Micro-LED backlight module by deep reinforcement learning and micro-macro-hybrid environment control agent

Che-Hsuan Huang¹, Yu-Tang Cheng², Yung-Chi Tsao³, Xinke Liu⁴, and Hao-Chung Kuo⁵

¹College of Materials Science and Engineering, Guangdong Research Center for Interfacial Engineering of Functional Materials, Institute of Microelectronics (IME), Shenzhen University, Shenzhen 518060, China

²Department of Photonics & Institute of Electro-Optical Engineering, College of Electrical and Computer Engineering, Taiwan Yang Ming Chiao Tung University & Taiwan Chiao Tung University, Hsinchu 30010, China

³Department of Computer Science, University of Liverpool, Liverpool, UK

⁴College of Materials Science and Engineering, Guangdong Research Center for Interfacial Engineering of Functional Materials, Institute of Microelectronics (IME), Shenzhen University, Shenzhen 518060, China

⁵Department of Photonics & Institute of Electro-Optical Engineering, College of Electrical and Computer Engineering, Taiwan Yang Ming Chiao Tung University & Taiwan Chiao Tung University, Hsinchu 30010, China

show less

Figures & Tables(15)

Fig. 1. (a) Schematic diagram of micro-LED backlight module; (b) schematic diagram of LED with DBR structure; (c) highly-reflective surface substrate; (d) etching structure of the receiver.

Download full size

View in Article

Fig. 2. Workflow for optimizing micro-LED backlight module: (a) the process of deep reinforcement learning and (b) the process of the virtual-realistic experiment.

Download full size

View in Article

Fig. 3. (a) Workflow of environment control agent and schematic diagram of the virtual-realistic experiment; (b) principle of kernel1: Gaussian and Lambertian reflection; (c) principle of kernel2: BSDF properties.

Download full size

View in Article

Fig. 4. Workflow of DDQN network.

Download full size

View in Article

Fig. 5. Virtual-realistic experiment: (a) single light pattern analysis; (b) module pattern analysis.

Download full size

View in Article

Fig. 6. Result of reinforcement learning: (a) uniformity for every iteration with reward function1; (b) uniformity for every iteration with reward function2; (c) uniformity for every iteration with reward function3; (d) the best result by reward function1; (e) the best result by reward function2; (f) the best result by reward function3.

Download full size

View in Article

Fig. 7. Demonstration of the micro-LED backlight module.

Download full size

View in Article

Fig. 8. Influence of DBR structure: (a) light pattern with state Sb3′, (b) light pattern with state Sb3, and (c) uniformity of Sb3 and Sb3′.

Download full size

View in Article

Fig. 9. Schematic of resonant loss for the micro-LED backlight module.

Download full size

View in Article

Table 1. Definition List of Action

View table

View in Article

Table 1. Definition List of Action

Action No.	Action Definition	Variation Symbol
$a_{1}$	Add distance	$D$
$a_{2}$	Reduce distance	$D$
$a_{3}$	Add width	$W$
$a_{4}$	Reduce width	$W$
$a_{5}$	Add spacing	$S$
$a_{6}$	Reduce spacing	$S$
$a_{7}$	Add thickness	$T$
$a_{8}$	Reduce thickness	$T$
$a_{9}$	Add DBR pairs	DBR
$a_{10}$	Reduce DBR pairs	DBR

Table 2. Definition List of State
View table
View in Article
Table 2. Definition List of State
State No. State Definition Variation symbol
$S_{1}$ Value of distance $D$
$S_{2}$ Value of spacing $W$
$S_{3}$ Value of DBR pairs $S$
$S_{4}$ Value of thickness $T$
$S_{5}$ Value of width DBR

Table 3. Definition List for Range of Parameters
View table
View in Article
Table 3. Definition List for Range of Parameters
No. Parameters Variation Symbol Range
1 Distance from receiver to LED $D$ 100–180 μm
2 LED width $W$ 10–100 μm
3 LED spacing $S$ 300–700 μm
4 Thickness of LED $T$ 5–55 μm
5 DBR-BSDF pairs DBR 4.5–9.5 pairs

Table 4. Results of Virtual-Realistic Experiment

View table

View in Article

Table 4. Results of Virtual-Realistic Experiment

No.	Parameters	Value
1	Bulk of GaN-based light-emitting diodes	Refraction $index 2.4869$
2	High-reflectivity white bottom surface of module	Gaussian type 5° with 97%
3	Receiver reflectivity	Lambertian with 35%
4	Index of DBR material (AIN/GaN)	$n_{AIN} = 2.1793 {/ n}_{GaN} = 2.4869$
5	Thickness of DBR material (AIN/GaN)	$t_{AIN} = 51.6 nm {/ t}_{GaN} = 45.2 nm$

Table 5. Best Uniformity for Different Reward Functions

View table

View in Article

Table 5. Best Uniformity for Different Reward Functions

Best State	Reward Formula	Parameters	Number of Iterations	Best Uniformity
$S_{b 1}$	$(Uniformity new - Uniformity old) / 100$	$D = 0.18 mm$ , $W = 0.04 mm$ , $S = 0.4 mm$ , $T = 0.02 mm$ , $DBR = 5.5 pairs$	289	83.97%
$S_{b 2}$	${(Uniformity - 75)}^{3} / 1000$	$D = 0.16 mm$ , $W = 0.02 mm$ , $S = 0.46 mm$ , $T = 0.015 mm$ , $DBR = 5.5 pairs$	183	86.51%
$S_{b 3}$	${(Uniformity - 79)}^{3} / 1000$	$D = 0.18 mm$ , $W = 0.028 mm$ , $S = 0.5 mm$ , $T = 0.035 mm$ , $DBR = 6.5 pairs$	249	90.32%

Table 6. Work Efficiency of Designing Agent
View table
View in Article
Table 6. Work Efficiency of Designing Agent
Number of Iterations Time (h) Optimal Result
Entire loop 106,920 297 Uniformity: 89.61%
Designing agent 16,000 53.3 Uniformity: 90.32%

Tools

Get Citation

Copy Citation Text

Che-Hsuan Huang, Yu-Tang Cheng, Yung-Chi Tsao, Xinke Liu, Hao-Chung Kuo, "Micro-LED backlight module by deep reinforcement learning and micro-macro-hybrid environment control agent," Photonics Res. 10, 269 (2022)

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites