Algorithmic methods to infer chip behavior from off‑chip observations

Develop complete algorithmic solutions for inferring internal chip behavior and workload execution from off-chip observations (e.g., network taps and analog sensors) with accuracy sufficient for verification, including methods that quantify uncertainty and resist adversarial manipulation.

Background

Off‑chip verification aims to avoid on‑chip security issues by relying on external devices and trusted analysis clusters. This requires inferring internal behavior "from the outside" using sampled communications and analog measurements.

The authors emphasize that several of the needed algorithmic problems do not yet have complete solutions, limiting current precision and robustness.