CPA on 32-bit AES (Lab 2_1) fails to recover full key on CW308T-STM32F4HWC

Embedded Security Side Channel Power Analysis
1

I’m running Lab 2_1 – CPA on 32-bit AES on the CW308T-STM32F4HWC target, but even with 10,000 traces the full key is not recovered. 14 out of 16 bytes are recovered correctly, but 2 bytes are wrong (byte 2 ranks 2nd / PGE=2, byte 6 ranks 3rd / PGE=3). The correlation peaks come out quite low overall, around 0.1–0.2. Could anyone explain why this happens and how to improve it?
Finished traces 9975 to 10000

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
PGE= 0 0 2 0 0 0 3 0 0 0 0 0 0 0 0 0
0 2B
0.229		 7E
0.060		 26
0.105		 16
0.207		 28
0.168		 AE
0.139		 E1
0.110		 A6
0.137		 AB
0.194		 F7
0.216		 15
0.083		 88
0.139		 09
0.131		 CF
0.199		 4F
0.180		 3C
0.182
1 50
0.085		 4D
0.052		 C2
0.094		 E0
0.121		 DF
0.105		 88
0.094		 24
0.106		 14
0.104		 70
0.122		 88
0.110		 2E
0.073		 3E
0.090		 FE
0.120		 C5
0.099		 41
0.100		 8B
0.114
2 F0
0.083		 C5
0.052		 15
0.093		 E1
0.119		 27
0.103		 95
0.092		 BD
0.104		 70
0.102		 68
0.116		 D4
0.102		 F7
0.071		 37
0.082		 63
0.108		 79
0.098		 BC
0.099		 EA
0.110
3 F8
0.079		 89
0.050		 7A
0.093		 38
0.117		 53
0.103		 A8
0.091		 D2
0.099		 51
0.093		 54
0.114		 10
0.102		 EB
0.069		 D7
0.082		 DF
0.107		 69
0.094		 71
0.097		 EF
0.108
4 E0
0.078		 A8
0.048		 8F
0.091		 F8
0.114		 0E
0.100		 D8
0.090		 E4
0.097		 E1
0.088		 30
0.106		 70
0.102		 2A
0.064		 A2
0.080		 27
0.102		 30
0.093		 AC
0.097		 6A
0.104
2

Do you have the 1.2V source enabled on the F4? There’s an internal regulator that needs to be bypassed, so it’s possible to have the device function but not give good power measurements.