Cryptanalysis of RSA Factorization - Library(ISI Kolkata) - Indian ...
Cryptanalysis of RSA Factorization - Library(ISI Kolkata) - Indian ...
Cryptanalysis of RSA Factorization - Library(ISI Kolkata) - Indian ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Chapter 5: Reconstruction <strong>of</strong> Primes given few <strong>of</strong> its Bits 80<br />
Here, we assume that the tree narrows down to a γ i fraction (0 < γ i ≤ 1) from the<br />
earlier level if both the bits <strong>of</strong> the primes are known. One may note that Heninger<br />
and Shacham [50, Conjecture 4.3] conjectures the average value <strong>of</strong> γ i (call it γ) to<br />
be 1 . We shall discuss this in more details later.<br />
2<br />
Suppose that randomly chosen α fraction <strong>of</strong> bits <strong>of</strong> p and β fraction <strong>of</strong> bits<br />
<strong>of</strong> q are known (by some side channel attack, e.g., cold boot). Then the joint<br />
probability distribution table for the bits <strong>of</strong> the primes will be as follows.<br />
↓ q[i], p[i] → UNKNOWN KNOWN<br />
UNKNOWN (1−α)(1−β) α(1−β)<br />
KNOWN (1−α)β αβ<br />
As shown before, the growth <strong>of</strong> the search tree depends upon the knowledge <strong>of</strong><br />
the bits in the primes. Hence, we can model the growth <strong>of</strong> the tree as a recursion<br />
on the level index i:<br />
W i = (1−α)(1−β)2W i−1 +α(1−β)W i−1 +(1−α)βW i−1 +αβγ i W i−1<br />
= (2−α−β +αβγ i )W i−1 .<br />
IfwewanttorestrictW i (thatisthegrowth<strong>of</strong>thetree)asapolynomial<strong>of</strong>i(thatis<br />
thenumber<strong>of</strong>level),wewouldlike(roughlyspeaking)thevalue<strong>of</strong>(2−α−β+αβγ i )<br />
close to 1 on an average. Considering the average value γ (instead <strong>of</strong> γ i at each<br />
level), we get, 2 − α − β + αβγ ≈ 1 which implies 1 − α − β + αβγ ≈ 0. If<br />
we assume that the same fraction <strong>of</strong> bits are known for p and q, then α = β<br />
and we get 1 − 2α + α 2 γ ≈ 0 ⇒ α ≈ 1−√ 1−γ<br />
. If we assume [50, Conjecture<br />
γ<br />
4.3], then γ ≈ 0.5 and hence α ≈ 2 − √ 2 ≈ 0.5858, as obtained in [50, Section<br />
4.4]. One may note that our idea is simpler compared to the explanation in [50].<br />
This simplification is achieved here by using average value for γ i in the recurrence<br />
relation <strong>of</strong> W i .<br />
The most natural strategy is to first apply Algorithm 7 to retrieve the least<br />
significant half <strong>of</strong> any one <strong>of</strong> the primes and then apply the result <strong>of</strong> Boneh et<br />
al [16, Corollary 2.2] to factorize N. One may note that [50] utilizes their prime<br />
reconstruction algorithm to reconstruct the whole primes p,q whereas our idea is<br />
to use lattice based results after reconstructing just one half <strong>of</strong> any prime. This is<br />
more practical as it requires the knowledge <strong>of</strong> lesser number <strong>of</strong> random bits <strong>of</strong> the<br />
primes, namely, just about 0.5858×0.5 ≈ 0.3 fraction <strong>of</strong> bits (from the LSB half)<br />
instead<strong>of</strong>0.5858fraction<strong>of</strong>theprimesasexplainedin[50]. Moreover,factorization