Amitoz Singh Dandiana
San Jose
Summary
Passionate about applying machine learning to complex computer vision challenges, I have over 10 years of experience in designing and implementing innovative AI solutions for semiconductor process control systems. I lead high-impact, cross-functional teams to deploy production grade solutions for Scanning Electron Microscope (SEM) image restoration, defect detection and classification. I am seeking opportunities to drive high-impact computer vision solutions through hands-on contributions or by leading high-performing teams.
Overview
12
12
years of professional experience
Work History
Algorithms Manager, Deep Learning & Computer Vision
KLA
04.2023 - Current
- Established a results-driven team with expertise in AI, computer vision, and compute optimization through structured hiring and career development processes.
- Engineered and directed transformer-based foundation model R&D for zero shot denoising, sharpening and super-resolution on unseen SEM images-reducing imaging setup time by 80% and improving transformer model inference time by 8x to meet compute budget requirements.
- Brainstormed and architected an end-to-end workflow for design-to-SEM image translation using GANs, including sampling diverse design-SEM pairs for model training, introducing metrics to detect model hallucinations, and applying post processing to ensure structural and histogram-level consistency between real and generated SEM images.
- Led research on 2-shot AutoFocus using MobileNet-based regression model for predicting focus offset from two unfocused images with known offsets, thereby reducing the number of required focus captures from nine to two and increasing system throughput by 15%.
- Championed integration of the Segment Anything Model (SAM) to enable auto-annotation for large and complex objects, thus reducing manual labeling time in supervised learning pipelines by 40%.
Algorithms Product Lead, SEM Review Platform
KLA
02.2020 - 12.2023
- Managed project timelines, schedules, risks, and resources for the SEM Review platform, with a focus on deep learning based defect detection and classification solutions for buried and subtle defects in advanced 2nm process nodes.
- Explored supervised, unsupervised and traditional approaches for multi-modal defect detection in SEM images using design as a reference while evaluating tradeoffs across performance, compute cost, and setup time, resulting in 2x throughput.
- Guided research on the application of zero shot diffusion method DiffPIR using fine-tuned Stable Diffusion and DDPM models for denoising extremely low SNR SEM images with Poisson noise characteristics.
- Architected a GAN-based solution for SEM image artifact removal caused by electron beam distortion and stage vibration, enabling customer acceptance of SEM review tools for in-fab deployment.
- Collaborated with cross-divisional algorithm teams to link Optical and SEM tools in the fab, reducing false positives by 90% through improved alignment using design as a reference.
Lead Analyst, Deep Learning Research and Development
KLA
09.2016 - 02.2020
- Developed deep learning solutions for defect detection for SEM inspection platforms, enabling reliable capture of defects as small as 2 pixels with a signal-to-noise ratio of 3, and ensuring repeatable results with as few as tens to hundreds of labelled examples.
- Spearheaded R&D for novelty defect classification, combining traditional and deep learning features with a Random Forest classifier for outlier detection.
- Conducted extensive R&D to study the impact of the loss functions, class weighting, ground truth labelling errors and training data volume on model performance, establishing key heuristics to guide model development iterations.
- Engaged field teams to ensure model robustness to process drifts and excursions, increasing confidence in post-deployment AI performance.
- Deployed explainability tools such as t-SNE, Layer-Wise Relevance Propagation (LRP) and k-Nearest Neighbors for model interpretability and diagnosis.
Algorithms Engineer
KLA
07.2013 - 09.2016
- Architected a scalable Random Forest-based Automatic Defect Classification system with active learning across customer-deployed classifiers.
- Led feature engineering, minimized training data storage, defined data sufficiency metrics, and designed the user experience for classifier setup and monitoring.
- Published patent on automatic defect classification, novelty detection and sequential classifiers for improved accuracy.
Education
Bachelors - Electrical Engineering
Indian Institute of Technology Delhi
01-2013
Masters - Information and Communication Technology
Indian Institute of Technology Delhi
01-2013
Skills
- Python
- C
- Matlab
- PyTorch
- TensorFlow
- Computer Vision
- Image Processing
- Probability
- Statistics
- Machine Learning
- Feature Engineering
- Random Forests
- CNN
- GANs
- Diffusion Models
- Transformers
- Image Restoration
- Object Detection
- Classification
- Segmentation
- Electron Microscopy Imaging
Timeline
Algorithms Manager, Deep Learning & Computer Vision
KLA
04.2023 - Current
Algorithms Product Lead, SEM Review Platform
KLA
02.2020 - 12.2023
Lead Analyst, Deep Learning Research and Development
KLA
09.2016 - 02.2020
Algorithms Engineer
KLA
07.2013 - 09.2016
Bachelors - Electrical Engineering
Indian Institute of Technology Delhi
Masters - Information and Communication Technology
Indian Institute of Technology Delhi
Similar Profiles
Jacob BowenJacob Bowen
Customer Support Engineer at KLACustomer Support Engineer at KLA
Yuval TsidkiayhuYuval Tsidkiayhu
Senior Software Engineer at KLASenior Software Engineer at KLA
Joanne TsaoJoanne Tsao
Sr. Account Manager at KLASr. Account Manager at KLA