I’ve been having an issue with the correlation values found during a CPA attack. Recently, I’ve been experimenting with the attack size and I’ve found that when I use a large number of traces, there is always an abrupt drop in the correlation vs trace plot at roughly 11500 traces. Do you have any ideas why this is happening?
I’m using the Chipwhisperer-Pro to attack the CW305 Artix FPGA target. The CW305 has been using the default bitfile and some custom bitfiles. The attack is based on the CW305 AES demo and I have been using the last_round_state_diff leakage model. The plotting is based on Lab 4_3 of sca101.
I included an example plot below.