Reveal the invisible secret: chosen-ciphertext side-channel attacks on NTRU

NTRU is a well-known lattice-based cryptosystem that has been selected as one of the four key encapsulation mechanism finalists in Round 3 of NIST’s post-quantum cryptography standardization. This paper presents two succinct and efficient chosen-ciphertext side-channel attacks on the latest variants of NTRU, i.e., NTRU-HPS and NTRU-HRSS as in Round 3 submissions. Both methods utilize the leakage from the polynomial modular reduction to recover the long-term secret key. For the first attack, although the side-channel leakage does not directly reveal the secret polynomial f , we recover differences between adjacent coefficients using appropriately chosen ciphertexts, and finally reconstruct f  through linear algebra. The second attack is based on the inherent relation between the secret key and the public key in NTRU-HPS: we first reveal the “invisible” secret polynomial g  with chosen ciphertexts and then use g  and the public polynomial h  to compute f . In theory, these attacks only need 4 and 2 ciphertexts, respectively. We then practically apply those attacks on all reference implementations of four instances in the PQClean library and show that the accuracy of secret-key recovery can reach 100% with only few traces (4 to 24 and 2 to 6, respectively). We also observe similar leakage in optimized implementations in the pqm4 library and propose an according analysis scheme.