Slide 1: Machine learning for design optimizations and prediction of optical chip performance
Machine learning for design optimizations and prediction of optical chip performance K. Yadav, S. Bidnyk, A. Balakrishnan Enablence Technologies Inc. Photonics West February 1, 2023
Slide 2: Introduction: Optical communication
Introduction Optical communication • The deployment of wide-scale optical communication systems led to a phenomenal growth in information exchange. • High-capacity optical fibers, combined with the use of integrated optical devices to control light, allow for optical networks with advanced routing a
Slide 3: Introduction: Artificial intelligence and machine learning
Introduction Artificial intelligence and machine learning • Artificial intelligence (AI) and machine learning (ML) emerged as a powerful new approach for solving previously intractable problems. 3
Slide 4: Introduction: Machine learning
Introduction Machine learning • Artificial intelligence (AI) and machine learning (ML) emerged as a powerful new approach for solving previously intractable problems. 4
Slide 5: Introduction: Machine learning
Introduction Machine learning • Artificial intelligence (AI) and machine learning (ML) emerged as a powerful new approach for solving previously intractable problems. 5
Slide 6: Introduction: Machine learning
Introduction Machine learning • Artificial intelligence (AI) and machine learning (ML) emerged as a powerful new approach for solving previously intractable problems. ML is able to capture essential features from vast amounts of high-dimensional data 6
Slide 7: AI/ML in photonics
Introduction AI/ML in photonics • The photonics industry has began adopting AI and ML techniques to further both research and deployment of optical technologies. 7
Slide 8: AI/ML in photonics
Introduction AI/ML in photonics • The photonics industry has began adopting AI and ML techniques to further both research and deployment of optical technologies. • Advances have been made in: • Deep learning for inverse design 8
Slide 9: AI/ML in photonics
Introduction AI/ML in photonics • The photonics industry has began adopting AI and ML techniques to further both research and deployment of optical technologies. • Advances have been made in: • Deep learning for inverse design • Deep learning microscopy 9
Slide 10: AI/ML in photonics
Introduction AI/ML in photonics • The photonics industry has began adopting AI and ML techniques to further both research and deployment of optical technologies. • Advances have been made in: • Deep learning for inverse design • Deep learning microscopy • Machine learning in optical communication an
Slide 11: AI/ML in photonics
Introduction AI/ML in photonics • The photonics industry has began adopting AI and ML techniques to further both research and deployment of optical technologies. • Advances have been made in: • Deep learning for inverse design • Deep learning microscopy • Machine learning in optical communication an
Slide 12: Photonic integrated circuits
Introduction Photonic integrated circuits • Photonic integrated circuits have grown into a powerful and versatile platform that is able to meet the challenging demands of today’s high-speed communication and advanced vision systems. • Photonic circuits possess high optical performance and are well s
Slide 13: Volume production of photonic circuits
Introduction Volume production of photonic circuits • The key to accelerated adoption of photonics is achieving reproducible performance – a significant challenge in the photonics industry. • Traditionally, the problem of performance inhomogeneity stemming from process variations has been handled by
Slide 14: Design optimizations: Volume production of photonic circuits
Design optimizations Volume production of photonic circuits • Photonic integrated circuits have been widely used to realize high-performance wavelength division multiplexing (WDM) devices for datacom and telecom applications. • A typical production wafer contains hundreds of optical devices: 14
Slide 15: Design optimizations: The challenge of process uniformity
Design optimizations The challenge of process uniformity • Process uniformity and consistency is critical in the manufacturing of photonic chips. • Traditionally, standard statistical methods are used to compensate for systematic process non-uniformities: systematic index variation 15
Slide 16: Adjustments of design parameters through ML
Design Optimizations Adjustments of design parameters through ML TRAINING SET INPUT x(i) LABEL d(i) ◼ ✿ ⬤ ⏣ ✦ ✚ ⌘ ★ ⌽ ► ▲ ♠ ❖ ⌾ ◀ ⚑ ♣ ♦ MODEL TRAINING simulated spectrum design parameters VALIDATION SET MODEL VALIDATION 16
Slide 17: Adjustments of design parameters through ML
Design Optimizations Adjustments of design parameters through ML INPUT x(i) d#(i) ◼ ✿ ⬤ ⏣ TRAINED ✦ ✚ ⌘ ★ MODEL ⌽ ► ▲ ♠ ❖ ⌾ ◀ ⚑ ♣ ♦ measured inferred design spectrum parameters 17
Slide 18: Adjustments of design parameters through ML
Design Optimizations Adjustments of design parameters through ML Ini6al ML-enhanced mask mask INPUT x(i) d#(i) ◼ ✿ ⬤ ⏣ TRAINED ✦ ✚ ⌘ ★ MODEL ⌽ ► ▲ ♠ ❖ ⌾ ◀ ⚑ ♣ ♦ measured inferred design spectrum parameters 18
Slide 19: Adjustments of design parameters through ML
Design Optimizations Adjustments of design parameters through ML • To validate the approach, we applied it to a production mask with 600 devices: • Devices on the mask were designed to be identical, except for a refractive index distribution correction computed by traditional statistical means. • De
Slide 20: Adjustments of design parameters through ML
Design Optimizations Adjustments of design parameters through ML • To validate the approach, we applied it to a producUon mask with 600 devices: • We used the model predicLons to insert correcLons into each of the chips on the mask, thereby producing a ML-enhanced version of the producLon mask. Init
Slide 21: Performance Predictions
Performance Predic4ons Classifica9on based on a wafer probe measurement 21
Slide 22: Performance Predictions
Performance Predictions Classification based on a wafer probe measurement Probed locations on the wafer 22
Slide 23: Performance Predictions
Performance Predic4ons Classifica9on based on a wafer probe measurement Probed locations Typical spectroscopic on the wafer signature 23
Slide 24: Performance Predictions
Performance Predictions Classification based on a wafer probe measurement Probed loca7ons Typical spectroscopic Predicted performance of on the wafer signature hundreds of chips on a wafer 24
Slide 25: Performance Predictions
Performance Predictions Classification based on a wafer probe measurement Specification parameters: 1. IL 2. IL uniformity 3. Grid detuning 4. Channel spacing uniformity 5. 0.5 dB passband 6. 1 dB passband 7. 3 dB passband 8. PDL 9. Ripple 10. Adjacent crosstalk 11. Non-adjacent crosstalk 12. Total
Slide 26: Performance Predictions: Traditional vs wafer probe testing
Performance Predictions Classification based on a wafer probe measurement Tradi&onal chip tes&ng Wafer probe tes&ng Time to measure a 4-channel chip 2 minutes Time to perform a wafer probe measurement 12 minutes Cost of measuring a 4-channel chip (North America) $4.25 Time to infer the PASS/FAIL of
Slide 27: Performance Predictions
Performance Predictions Classification based on a wafer probe measurement Probed locations Typical spectroscopic on the wafer signature 27
Slide 28: Performance Predictions
Performance Predictions Classification based on a wafer probe measurement Predicted PASS/FAIL over 26 specifica7on parameters 28
Slide 29: Performance Predictions
Performance Predictions Classification based on a wafer probe measurement Predicted PASS/FAIL Actual PASS/FAIL over 26 specification over 26 specifica7on parameters parameters 29
Slide 30: Optimizations Through ML
Optimizations Through ML Robust designs for volume manufacturing • AI/ML has become instrumental in our work in extending the reach of the photonics technology. • We continue using “human-machine collaboration” to achieve outstanding uniformity of performance of mass-produced photonics chips despite
Slide 31: Conclusions
Conclusions • Machine learning has the capacity to capture features from vast amounts of high- dimensional data. • We described how AI/ML was used in the field of photonic integrated circuit design and manufacturing: • We used deep neural network multivariate regression model to optimize the individ
Slide 32: Custom Optical Design
Custom Optical Design We have buit systems-on-a-chip for avionics, medical robotics, ‘automotive LIDAR, 3D mapping, and optical sensing. We can do commercial-grade prototyping or high-volume production of chips. Our ‘mechanical design engineers can also assist with fiber pigtalling and packaging. Th