PB173 Domain specific
development: side-channel
analysis
Seminar 10: Main goals grading
Łukasz Chmielewski
chmiel@fi.muni.cz, Consultation: A406 Friday 9:00-11:00
EM SIDE-CHANNEL ANALYSIS:
SEMINAR 1 REMINDER
Example
2
DEMO
Active Side-Channel
3
ORGANIZATIONAL
4
Organization
• Group 1: Alignment
– https://github.com/2lol555/pb173-side-channel/tree/main
– Progress: ?
• Group 2: Parallel computations with acquisition
– https://github.com/makuga01/pb173-sidechannels
– Progress: ?
5
Group 1: Alignment
• Goals:
– Peak-Based Alignment
– Correlation-based Alignment
– Optional: elastic versions
• Look at:
– AES_fixed_rand_input_CAFEBABEDEADBEEF0001020304050607+SAVEEVEN(0,1000).trs
– AES_fixed_rand_input_CAFEBABEDEADBEEF0001020304050607+SAVEEVEN(0,1000)+MIS(100).trs
• First tasks:
– Try to align the traces mentioned above using peak-based
alignment. Note that it might not work for …MIS… traces.
– See all the uploaded scripts till now
• Later task - Correlation-based Alignment
• How is it going?
6
Group 2: Parallel computations with
acquisition
• Implement multithreaded Acquisition + Processing
• Measure Efficiency
• First Task: measure the efficiency of the
acquisition (done?) Do you have some graphs?
• Later tasks: observe the impact of processing and
try to add frequency processing in parallel to the
acquisition
• How is it going? Have you used?
– https://github.com/ikizhvatov/efficient-columnwise-correlation and
– cpa_aes_evol.py (the corr. traces are also uploaded for Seminar08)
7
Remaining Seminars Plan
• 7: evaluation of progress on first steps: 1 point per person
per work done till today also based on the commits in GIT
• 8: evaluation of finished first steps : 3 points per group
(personalized per person based on the Github) + giving the
next tasks
9: work in progress (I will join online for some time)
• 10: 4 points per group (personalized per person based on
the GitHub) + what would say about showing a more official
progress presentations? Decide today.
This seminar: real SCA setup
• 11/12: national holiday / online consultation
• 13: final 2 points for work + 2 points for presentations + 2
points for activity, grading.
8
WHAT WAS DONE + GIVING NEW TASKS?
9
Group 1: Alignment
• Nice!
10
Group 1: Alignment
• Main:
• Commits:
• Explain work division.
11
Group 1 Tasks:
1. Try to misaligned_1000 traces
2. Try alignment on lower peaks (local maximum peaks)
3. Try the Absolute Window Resample + Alignment
approach
4. Try pattern matching as explained during the seminar
5. Longer term: Correlation Alignment
• From my side, computing correlation between the traces:
from scipy.stats import pearsonr
12
Group 2: Parallel computations with
acquisition
• Nice!
13
Group 2: Parallel computations with
acquisition
14
It looks evenly distributed, but please describe the
division.
Group 2 Tasks:
15
1. Perform analysis with jitter enabled.
2. Try Spectrogram + CPA together
3. Perform evaluation when turning on and off various
parallelizations
4. Generate graphs for comparison
• From my side, I will add more ideas for extension for the
next seminar. I am considering asking to add an
alignment code from Group 1.
WALK-AROUND + WORKING IN GROUPS
+ AGREEING ON FINAL GOALS
16
Group 1 Final Tasks:
1. Finalize Correlation Alignment on the provided traces.
– Potentially: investigate optimizations of calculating Normalized
Cross Correlation (NCC) between the static reference and target
traces. Lukasz’s idea: find out how it is efficiently done at
https://github.com/Riscure/Jlsca. There is an efficient
implementation there!
2. Make Peak Correlation + Window Resampling work
also for other trace sets:
– Before 02/05/2024 I will upload two new tracesets to IS.
3. Help Group 2 to incorporate peak alignment into their
acquisition pipeline.
17
Group 2 Final Tasks:
1. Finish comparison of various settings with respect to the
number of threads and the amount of traces acquired.
– Clarify which approach is the best on your system.
– Possibly use a profiler (e.g., cProfile) to identify the most
important bottlenecks of your solution.
– Experiment with various numbers of samples used (or
acquired). Does it matter?
2. Add a peak alignment code from Group 1 to your
pipeline and perform experiments.
3. Optional: add bandpass filtering to your pipeline:
https://stackoverflow.com/questions/12093594/how-to-implement-
band-pass-butterworth-filter-with-scipy-signal-butter
18
Agree on next two weeks
1. I will prepare a presentation/video about DFA in 2
weeks and publish it online.
2. Online consultation.
3. Discuss ?
19
Reading
• For interested people
• Side-Channel Analysis – blue book:
– http://dpabook.iaik.tugraz.at/
– The books is available at the uni.
– Look online
• The Hardware Hacking Handbook:
– https://nostarch.com/hardwarehacking
– I have an epub version.
20
21
| PA193 - Programming in the presence of side-channels /
faults
Questions