Slide 1: Enabling high-volume production of photonics chips with machine learning
Enabling high-volume production of photonics chips with machine learning Ksenia Yadav Enablence Technologies Photonics West 2024 January 30, 2024
Slide 2: Introduction: Advances in machine learning
Introduction Advances in machine learning 2
Slide 3: Advances in machine learning
Introduction Advances in machine learning ML is able to capture essential features from vast amounts of high-dimensional data 3
Slide 4: AI and ML in photonics
Introduction AI and 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: - Nanophotonics inverse design - Deep learning microscopy - Machine learning in optical communication an
Slide 5: Integrated Photonics
Integrated Photonics • The widespread adoption of optical communication systems is a key driving force of recent surge in information exchange. • Optical solutions dominate long-distance communication, and have the potential to transform short-reach links through advanced optical interconnect soluti
Slide 6: Integrated Photonics
Integrated Photonics • Today, we present our progress in leveraging the power of machine learning to overcome the biggest hurdles for the widespread adoption of photonic integrated circuits: - use of multi-path neural networks as a key tool for transitioning from low-volume prototype designs to high
Slide 7: Prototype design vs. large-scale adoption
Integrated Photonics Prototype design vs. large-scale adoption • The transition from prototype design to achieving uniform manufacturing in high volumes is a pivotal phase in any product development process. Systems-on-chip in monolithically integrated silica-on-silicon platform Integrated Optics: D
Slide 8: Challenges for large-scale adoption
Integrated Photonics Challenges for large-scale adoption Good performance chip Poor performance chip Variation in performance of identically designed chips 8
Slide 9: Multi-Path Deep Learning parameters
Multi-Path Deep Learning Photonics chip design optimizations design A 4-channel multiplexer: parameters spectrum design • Two filter stages • 7 asymmetric MZIs 9
Slide 10: Photonics chip design optimizations
Multi-Path Deep Learning Photonics chip design optimizations TRAINING SET INPUT x(i) LABEL d (i) ◼ ✿ ⬤ ⏣ ✦ ✚ ⌘ ★ ⌽ ► ▲ ♠ ❖ ⌾ ◀ ⚑ ♣ ♦ MODEL TRAINING simulated design spectrum parameters Delays and coupling coefficients of the 7 VALIDATION SET MZIs in the lattice filter MODEL VALIDATION 10
Slide 11: Photonics chip design optimizations
Multi-Path Deep Learning Photonics chip design optimizations Initial ML-enhanced mask mask INPUT x(i) d0(i) ◼ ✿ ⬤ ⏣ TRAINED ✦ ✚ ⌘ ★ MODEL ⌽ ► ▲ ♠ ❖ ⌾ ◀ ⚑ ♣ ♦ measured inferred design spectrum parameters 11
Slide 12: Photonics chip design optimizations
Multi-Path Deep Learning Photonics chip design optimizations Good performance chip Poor performance chip Initial ML-enhanced mask mask Variation in performance of Homogeneity of performance identically designed chips after ML optimization 12
Slide 13: Prediction of Device Performance
Prediction of Device Performance Classification based on a wafer probe measurement Specification parameters: 1. Insertion loss 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 Traditional testing PASS/FAIL map 10. A
Slide 14: Prediction of Device Performance
Prediction of Device Performance Classification based on a wafer probe measurement sensitivity 95% specificity 68% accuracy 90% precision 92% F1 score 94% ROC AUC 85% Traditional testing PASS/FAIL map SVM predicted PASS/FAIL map support vector machine Typical spectroscopic signature Probed locations
Slide 15: Conclusions
Conclusions • AI/ML has become instrumental in our work in extending the reach of the photonics technology. • We described how AI and ML have revolutionized the way photonic integrated circuits are designed and fabricated in a high-volume environment: • Multi-path neural network optimizes the indivi
Slide 16: Custom Optical Design
Custom Optical Design ‘We have bul systems-on-a-cip for avoncs, medical robots, ‘automotive LIDAR, 30 mapping, and optical sensing, We can do ‘commercial-grade prototyping or high-volume production of ches. Our ‘mechanical design engineers can also assist with fiber pigaling and ‘packaging. Through