Machine Learning for Discovery of Structure-Process-Property Relations in Electronic Materials: Machine Learning in Materials Engineering II
Sponsored by: ACerS Electronics Division
Program Organizers: B. Reeja Jayan, Carnegie Mellon University; Aarti Singh, Carnegie Mellon University
Monday 2:00 PM
November 2, 2020
Room: Virtual Meeting Room 43
Location: MS&T Virtual
Session Chair: B. Reeja Jayan, Carnegie Mellon University
2:00 PM Invited
3D Printing and Machine Learning: Anthony Rollett1; Srujana Yarasi1; Christopher Kantzos1; Elizabeth Holm1; 1Carnegie Mellon University
3D printing, aka additive manufacturing, has grown strongly in recent years. For metals, fusion-based powder bed technologies dominate at present because of their ability to make (near) net shape, complex parts. Machine learning (ML) is having a pervasive impact on the field because of the complexity of the processes that are used and the largely empirical development that has occurred. Applications will be given of using ML to classify powders, to identify spreading defects, to relate powder morphologies to their flow characteristics, to classify pore types, to recognize off-normal microstructures, to identify defect formation from acoustic signals, to predict stress hot spots on rough surfaces and to find relationships between surface roughness and fatigue life. Although modern ML methods are making vital contributions particularly to feature extraction from images, traditional data analytics are also useful and it is essential for the materials scientist to apply the full spectrum of methods.
2:30 PM
Fast and Generalizable Detailed Router Using Attention-based Reinforcement Learning: Haiguang Liao1; Levent Kara1; Qingyi Dong1; Xuliang Dong1; Wentai Zhang1; Wangyang Zhang2; Weiyi Qi3; Elias Fallon3; 1Carnegie Mellon University; 2Uber ATG; 3Cadence Design Systems
In the physical design of integrated circuits, global and detailed routing are critical stages involving the determination of the interconnected paths of each net on a circuit while satisfying the design constraints. Existing actual routers as well as routability predictors either have to resort to expensive approaches that lead to high computational times, or use heuristics that do not generalize well. In this work, we propose a new router — attention router, which is the first attempt to solve the detailed routing problem using reinforcement learning. Complex design rule constraints are encoded into the routing algorithm and an attention-model-based REINFORCE algorithm is applied to solve the most critical step in routing. The attention router is applied to solve different commercial advanced technologies analog circuits problem sets. It demonstrates generalization ability to unseen problems and achieves more than 100× acceleration over the genetic router without significantly compromising the routing solution quality.
2:50 PM
Cycle Life Prediction of Lithium Ion Batteries Based on Data Driven Methods: Laisuo Su1; Mengchen Wu1; B. Jayan1; 1Carnegie Mellon University
Machine learning algorithms are much better to learn hidden features for complex, nonlinear systems than human expects. Those hidden features are crucial for many applications, like mode identification and performance prediction. In this study, we compare the ability between human experts and machine learning algorithms for capturing features to predict lifetime of lithium ion batteries (LIBs). We generate a comprehensive dataset with 104 commercial LiNi0.8Co0.15Al0.05O2/graphite 18650-series batteries under wide range of test conditions. Based on charge and discharge curves, we capture 20 different features that relate to the lifetime of LIBs. The best prediction error is around 50% based on those human captured features using linear regression method and neural network model. In comparison, a convolution neural network (CNN) that captures hidden features can predict cycle life with less than 10% error. This study demonstrates the advantages of applying machine learning algorithm for capturing hidden features for complex, nonlinear systems.