SACHIN JAGANADE

Project Title: Chat Bot for Semiconductor Fab (Current Project)

• Exploration and development of Chat Bot for Semiconductor Fab as assistant.
  Role
• Exploration of SoAT techniques like Retrieval-Augmented Generation (RAG), Supervised Finetuning and Reinforcement Learning with Human Feedback (RLHF)
• Detailed mind-map for RAG, SFT and RLHF
• Proof of concept for document chunking, vectorization techniques and RAG. using FAB human logs

Tool

• PyCharm, Jupyter Notebook

Project Title: Long-term Time Series Prediction and Part Failure Analysis

• Developing Deep Learning based solutions for long-term time series prediction.

Role

• Problem formalization, literature review, and feasibility analysis - explored different SoAT techniques and finalized simple and explainable techniques for long term time series prediction techniques.
• Making detailed design document and code documentation.
• Developing pipeline for time series prediction and part failure analysis
• Testing and fine-tuning the pipeline
• Logger support for debugging
• Code review and guiding team members

Tool

• PyCharm, Jupyter Notebook

Project Title: Domain Specific Explainable Feature Extraction

• Extracting features which is explainable and also useful for failure detection during productivity loss.

Role

• Problem formalization, literature review, and feasibility analysis
• Developed a method for checking whether extracted features are sufficient or not and iteratively identifying the features and adding it.
• Making detailed design document and code documentation.
• Testing and fine-tuning the pipeline.
• Code review and guiding team members

Tool

• PyCharm, Jupyter Notebook

Project Title: Commonality Analysis and Tool-to-Tool Matching (TTTM)

• Developing Machine Learning based solutions for Commonality analysis and TTTM for time-series data

Role

• Problem formalization, literature review, and feasibility analysis
• Developing pipeline for failure impact node and commonality analysis
• Testing and fine-tuning the pipeline
• Logger support for debugging
• Code review and guiding team members
• Problem formalization, literature review, and feasibility of Graph Neural Networks (GNN) for multivariate time-series based TTTM. Proof of concept in progress.

Tool

• PyCharm, Tensor-flow

Project Title: Anomaly Detection in Multi-variate Time-series data

• For predictive maintenance, and anomaly detection, measure quality and productivity loss.

Role

• Problem formalization, literature review, and feasibility analysis
• Developing pipeline for anomaly detection for multi-variate time-series data
• Testing and fine-tuning the pipeline. Prepared complex technical reports.
• Logger support for debugging
• Code review and guiding team members

Tool

• PyCharm

Project Title: Machine Learning Studio

• Developing machine learning studio for semiconductor FAB

Role

• Developing different algorithms, categorized into pre-processing, feature selection, feature extraction, machine learning modules, deep learning modules for time-series, and images
• Literature review, feasibility, development, and testing for time-series and image data
• Trained team members for development, testing, and fine-tuning of the pipelines, and improving performance.
• Fine-tuning various pipelines to achieve performance metrics for time-series data
• Solved FAB problems using deep learning techniques
• Coordinated with teams to solve complex and deadline-dependent issues.
• Prepared complex technical reports for engineering teams.

Application of Deep Learning Pipelines for FAB Time-series Data

• Developed Remaining Useful Life (RUL) for semiconductor FAB equipment part.
• Developed an unsupervised technique from scratch using advanced deep learning for Root Cause Analysis (RCA) for multi-variate time-series data
• Developed an unsupervised anomaly detection technique for predictive maintenance and failure analysis

Tool

• PyTorch, Tensor-flow, Tensor-board, PyCharm

Achievement

• Developed multiple pipelines for Semiconductor FAB-related applications
• Proposed solutions helped FAB engineers to reduce root cause analysis time by more than a day, and helped in reducing maintenance costs.

Project Title: Generative Adversarial Network (GAN)

• Generating Time-series data for Semiconductor FAB
• Application of GAN for Semiconductor FAB

Role

• Prepared design document, and requirement document.
• Interacting with counterparts and weekly reporting
• Literature review and preparing technical reports.
• Feasibility study and POC
• Developing and testing different GAN architectures for time-series data
• Quality analysis of generated data
• Problem formalization and application of GAN for anomaly detection
• Guiding a team member on problem understanding and development task
• Developing, testing, and fine-tuning pipelines

Tool

• PyTorch, Tensor-flow, Tensor-board

Achievement

• Application of GANs for time series data in Semiconductor FAB.

Project Title: Wafer Quality Prediction

• Deep learning-based module for wafer quality prediction for time-series data.

Role

• Literature review and feasibility of time-series analysis for wafer quality prediction
• Identifying architectures and developing them for time-series data
• The proposed, imaging time-series technique performed best. It was the first time introduced to the team
• Developing, testing, and fine-tuning the entire pipeline. Prepared complex technical reports.

Tool

• PyTorch, Tensor-flow, and Tensor-board

Achievement

• The first project in deep learning for Semiconductor FAB-related applications

Project Title: Human Log Analytics

• Was working on Human Log Analytics for the semiconductor FAB domain

Role

• Developing algorithms for unstructured human logs in FAB, the system that extracts data with tagging tool and storing in structured DB
• Developing image processing based algorithm for image search based on a search query
• Involved in leading the team of four people and project management (release) activities, communicating with the stakeholders every week
• Proposed word2vec method, first time introduced in the team

Tool

• Java, C#, and Python

Achievement

• The proposed image search solution being used in FAB

Project Title: Touch Technology

• Was working on capacitance-based touch technology

Role

• Literature review and feasibility study
• Developing signal processing algorithms for co-ordinate extraction, noise supersession due to panel and unwanted charger
• Classifying various touch types like glove touch, touch on flip cover, hovering, and touch on the panel
• Developing algorithms from scratch, then fixed point conversion and optimization, and then porting, profiling, and testing performance on board
• Preparing technical documents for developed algorithms

Tool

• MATLAB, and C.

Achievement

• The solution was part of the S7 touch panel.

Project Title:- Vision-based Surveillance

Role

• Literature survey and feasibility study
• Developing image processing, and computer vision-based algorithms from scratch for surveillance application
• Developing face detection, people counting, algorithms
• Integrating into the product and testing the real-time performance

Tool

• All modules were implemented in C
• Used OpenCV

Achievement

• The proposed solution being part of the main product