Running SOLN_LAB_3_3 with a Husky doesn't yield correct results

So SchV2 = HWREV 1.0 and SchV3 = HWREV1.1?

What is the difference between A/B/C?

I think B was introduced for the differently behaving MOSFET (same model and everything but different date code IIRC from CrowdSupply update).

Is it similar for C or did I get that wrong?

I have 1.0 and 1.1B so I am trying to test on my end and understand what is which schematic and how they differ :slight_smile:

Thanks @Alex_Dewar for sharing the traces. Indeed the plots look different, but I am not able to distinguish whether this is resulting from jitter, noise, or different gain.
I quickly checked the V2 and V3 schematics, but also here I haven’t been able to spot anything relevant other than the VCCPLL fix in V3, which shouldn’t make things worse for V3.

If it is not the settings, the schematics are the same and assuming that all your devices run on the same firmware, do we have to assume that the lower performance of the V3 units are resulting from differences in component selection or PCB design?

Not sure ATM what the different 1.1 codes are, will have to talk with Colin. My guess is that everything 1.1 is the rev 3 board and everything 1.0 is rev 2. If you try a simple CPA attack against TINYAES on the SAM4S, the rev 2 should work after ~30 traces while the rev 3 should be 60-70.

The current plan is to look more into this next week after CHES is over. Kind of hard to speculate on what the root cause is at this point.

4 Likes

I have tried the CPA attack in Lab 4_3 and Lab 5_1 with my V3 and both run successfully, but at a much higher trace count of about 120 to 130 traces (see below). Also here the adjustment of adc_phase does not change the result significantly.

I really appreciate your support!

EDIT: I have checked the calculate_snr function leading to the following results:
Max. SNR: 40.7715022133584
Min. SNR: -139.35504745677258
Avg. SNR: -39.56641947950609
Not sure how these are calculated and how reliable these are, but the results themselves are not ideal.

Any update regarding Husky?

We’re still trying to narrow down what’s causing this. I was sick last week, so I wasn’t able to go into the office for testing, but JP seems to have a v1.0 board with lowered performance.

3 Likes

Hi @Alex_Dewar, anything we can do on user/customer side to support the analysis?

We should be alright for right now at least. Still looking into this, should hopefully have an update soon.

Alright, so I think I’ve found the culprit here. Some of the rev 2 boards have an 0402 100pF capacitor at C49. I’ve got a v3 board with that installed now and performance is similar to the rev 2 boards that I have.

EDIT: if anyone wants to try soldering this part onto the board, it’s right by the ADC (smallest chip with a heatsink)

2 Likes

Thanks, I will try shortly…

Hello, how easy was it to solder it on the CW ? I have only worked with through-hole boards before.

It works very well now! Thanks a lot for the solution.
The HW AES key can be completely recovered.

1 Like

Amazing! Could you please send a picture of your soldering spot on the pcb?

Sure.

Thanks so much! We will try this

1 Like

FYI, you might see a bit of degradation in performance up around 200MHz (tested with HWAES running on an FPGA at 200MHz) with this setup, so if you need the higher bandwidth, 47pF seems like a good compromise for both lower and higher frequency inputs.

2 Likes

Thank you for your great support, @Alex_Dewar!
I’ve tried the solution for impatient with a 100pF 0603 on the terminals of R29 and R32:

Improvement for CPA in Lab 4_3 is quite impressive (left without 100pF at C49 and right with C49 populated)

Nevertheless Lab 3_3 still doesn’t successfully guess the right key. Any further suggestions?

No, it wouldn’t be a bandwidth issue there. What sort of results do you get? Are you recovering some bytes here, or nothing at all? Also, are you running into any ghost peak issues?

The results seem to be completely random. In none of my recent attempts any of the correct subkeys appeared in the respective Top 5. Very much like @Shanly in the first post here.
The traces look like this:

Not much worse than the other labs I would say, but I haven’t seen traces from other CW devices yet.

@Shanly, I hope your modification was successful. Any improvements of Lab 3_3 on your end?