Historical Overview of Physically Observable Side Channels

Question 1

TEMPEST

Answer 1

NSA program that looked at physical side channels => looked at not just crypto, but monitors, cables, etc (1st)

Question 2

SPA

Answer 2

simple power analysis. attack on secret-dependent RSA control flow (square and sometimes multiply)

Question 3

DPA

Answer 3

differential power analysis. MUL produces different signal based on inputs. train data to tell between 0 and 1. works well if can send the same msg for training and attack

Question 4

2 DPA mitigations

Answer 4

• Exponent randomization (use d1 and d2 instead of secret d)
  =>Attack can still work if it can recover d1 and d2 from a single trace
• Message randomization (blinding)

Question 5

Chosen Ciphertext Attacks

Answer 5

Attacker picks ciphertext to help determine the exponent, d

Question 6

Adaptive chosen ciphertext attack

Answer 6

• Multiplying by 0 creates a smaller signal than when multiplying by anything else
• Want to manipulate the temporary values so that there are a lot of zeros if d is 0 and not a lot of zeros if d is 1
• Must send a message for each bit of the exponent

Question 7

Non-adaptive chosen ciphertext attack

Answer 7

• For window RSA implementation
• Send a message for each window:
  - Set message so that when wval is x it looks distinctive from all other wvals
• Need much less attempts than adaptive attacks

Question 8

three mitigations for SC after chosen ciphertext attacks

Answer 8

• exponent randomization
• message randomization
• Secret-independent control flow

Question 9

One&Done

Answer 9

• Can be done in a single trace
• Can be done outside of BN_mod_mul
• Attacks BN_is_bit_set ==> harder because length is function is much shorter than a MUL
• BN_mod_mul is very distinct signal => can isolate BN_is_bit_set

Question 10

Mitigation for One&Done

Answer 10

Use all bits of the window at a time => now the attacker has to distinguish between 2^5 different values in the signal (instead of 0 or 1)