Reproduce “RTNinja: A generalized machine learning framework for analyzing random telegraph noise signals in nanoelectronic devices” on OpenResearch

About this paper

Anirudh Varanasi, Robin Degraeve, Philippe Roussel, Clement Merckling

A generalized machine learning framework, RTNinja, was developed by researchers at imec and KU Leuven for automated and unsupervised analysis of complex Random Telegraph Noise (RTN) signals in nanoelectronic devices. This framework accurately deconvolutes multi-source signals to identify individual defect amplitudes and activity, demonstrating robust performance even as detectability and duration accuracy are influenced by increasing noise and complexity.

Key insights

• Framing RTN signal analysis as an unsupervised learning problem enables data-driven extraction of hidden defect characteristics without prior assumptions about source numbers or distributions.
• A modular approach, integrating Bayesian level identification, Markov modeling of transitions, and Affinity Propagation clustering, effectively deconvolutes complex RTN signals into their constituent independent sources.
• The use of a Monte Carlo simulator to generate labeled synthetic RTN datasets is crucial for objectively validating the performance of an unsupervised learning framework.

Summary by alphaXiv.