Information theoretic models for signatures in VLSI power delivery systems

We propose several abstract models for power distribution systems (PDSs) and show how the physics of PDNs pose limits on our ability to protect against power attacks. Integrated circuits increasingly use integrated voltage regulators (IVRs) to condition the on-chip power signal. IVRs present new opportunities for both decreasing observability of logic activity (power attacks) and monitoring the details of logic activity (Trojan horse detection). We propose three different models for the information in regulated power signals. Longer regulation periods improve the regulator's ability to hide information about the behavior of the logic; unfortunately, longer regulation periods may result in poorer power quality or larger regulator area.