Selected Research Results

Causal feature selection allows for HDD failure prediction with a reduced set of sensor data.
Reference: A. Yang, A. Ghassami, M. Raginsky, N. Kiyavash and E. Rosenbaum, “Model-Augmented Conditional Mutual Information Estimation for Feature Selection,” 2020 UAI, http://proceedings.mlr.press/v124/yang20b.html
Circuit aging simulation may be performed with RNN models of library cells or IP blocks rather than transistor level models (labeled RelXpert in figure). The continuous time RNN is implemented in Verilog-A.
Reference: E. Rosenbaum, J. Xiong, A. Yang, Z. Chen and M. Raginsky, “Machine learning for circuit aging simulation,” 2020 IEDM. https://doi.org/10.1109/IEDM13553.2020.9371931
Comparison of cross-device attacks on Frodo and NewHope post-quantum cryptography algorithms. 1D-CNN+Reg follows the approach of Kim et al. DS indicates that a downsampled power trace was used to train/evaluate the model.
Reference: Kashyap, Priyank, Furkan Aydin, Seetal Potluri, Paul Franzon, and Aydin Aysu. “2Deep: Enhancing Side-Channel Attacks on Lattice-Based Key-Exchange via 2D Deep Learning.” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (2020),  doi: 10.1109/TCAD.2020.3038701.
Sufficient summary plot showing relationship between eye width and a 1-dimensional active variable reduced from a 16-dimensional problem.
Reference: H. Ma, E. -P. Li, A. C. Cangellaris and X. Chen, “High-Speed Link Design Optimization Using Machine Learning SVR-AS Method,” 2020 IEEE 29th Conference on Electrical Performance of Electronic Packaging and Systems (EPEPS), 2020, pp. 1-3, doi: 10.1109/EPEPS48591.2020.9231368.
The Candidate algorithm for analog circuit optimization outperforms human designers and Bayesian optimization.
Reference: Y. Wang and P. D. Franzon, “RFIC IP Redesign and Reuse Through Surrogate Based Machine Learning Method,” 2018 IEEE MTT-S International Conference on Numerical Electromagnetic and Multiphysics Modeling and Optimization (NEMO), Reykjavik, Iceland, 2018, pp. 1-4. doi: 10.1109/NEMO.2018.8503446
Clock tree synthesis (CTS) using a generative adversarial network (GAN) outperforms commercial tools.
Reference: Y.-C. Lu, J. Lee, A. Agnesina, K. Samadi and S. K. Lim, “GAN-CTS: A Generative Adversarial Framework for Clock Tree Prediction and Optimization,” 2019 ICCAD, doi: 10.1109/ICCAD45719.2019.8942063

Replace a receiver model with an easy to build NNARMAX model.
Reference: B. Li, B. Jiao, M. Huang, R. Mayder and P. Franzon, “Improved System Identification Modeling for High-speed Receiver,” 2019 IEEE 28th Conference on Electrical Performance of Electronic Packaging and Systems (EPEPS), Montreal, QC, Canada, 2019, pp. 1-3.
doi: 10.1109/EPEPS47316.2019.193212

Predicted eye height and confidence intervals obtained from Bayesian active learning with drop out (BALDO) are compared to actual values. Reference: H. M. Torun, J. A. Hejase, J. Tang, W. D. Beckert and M. Swaminathan, “Bayesian Active Learning for Uncertainty Quantification of High Speed Channel Signaling,” 2018 EPEPS, doi: 10.1109/EPEPS.2018.8534251