Extracting a Bootloader's AES Key via a Power Analysis Side Channel Attack

The talk looks at why third-party accessory makers struggle with GoPro's Hero5 protections, and why encrypted firmware updates made key extraction hard through software alone. The target was a Microchip SAMD21 microcontroller with a small attack surface, so common routes such as debug interfaces, fuzzing and known vulnerabilities did not get far. Timing attacks, voltage glitching and electromagnetic fault injection were set aside in favour of power analysis.

The middle section covers the cryptographic background: the AES decryption rounds, the key schedule, and why the inverse shift step can be ignored for this purpose. Hamming weight models link power draw to the data moving across the chip's buses, and correlation power analysis matches guessed key bytes against measured traces. The setup used a ChipWhisperer, a shunt resistor with a differential probe, and an oscilloscope that oversampled the target's internal clock. Traces were aligned with a sum of absolute differences filter, and the round-ten key was recovered from about 150 traces in the demonstrated run, with any stubborn bytes brute forced.

The closing advice is to build a mockup of the target first, learn the tooling on it, and watch for rabbit holes, since simpler routes or faster methods such as electromagnetic fault injection may exist.