I tried an experiment on NANO where I collected power traces while performing multiplication, and attempted to classify traces depending on the Hamming weight of the result. I got an accuracy of approximately 60%. I also tried averaging the traces (for the same operands), but could not get better accuracy. Then I tried the same experiment on HUSKY, and the averaging worked here (keeping the sampling frequency 4x). Then I tried to check whether the issue with NANO is sampling frequency (which is 1x), so I kept the sampling frequency on HUSKY the same as that of NANO (1x), but I can not get results even as good as NANO. I want to know the reason behind this.
Are you using the same target for both? That is a huge variable.
If you are using the same target, the next biggest variable is probably the ADC sampling clock phase. It could be that you are “lucky” with a phase that catches more leakage on Nano vs Husky.
We recently added a notebook which teaches how to adjust the ADC sampling phase on Husky: chipwhisperer-jupyter/demos/husky/07 - Husky Sampling Phase.ipynb at main · newaetech/chipwhisperer-jupyter · GitHub
Thank you for answering. I am using different targets, for NANO, it’s STM32, and for HUSKY, it is SAM4S. I will check the notebook you mentioned also.