How Machine Learning Is Reshaping Side-Channel and TEMPEST Threats

Machine learning has turned compromising emanations from a problem for national laboratories into one a smartphone and published models can exploit — and it is simultaneously the strongest defense available against them.

For fifty years the unintended electromagnetic, acoustic, optical and power signals leaking from electronics were a niche concern, held in check by shielding, separation distances and the sheer difficulty of turning noise into intelligence. The survey argues that machine learning has dissolved that difficulty: the core problem in every side-channel attack is pattern recognition under noise, which is precisely what deep learning does best, and commodity software-defined radios and smartphones have collapsed the cost of collection to match.