liboqs/docs/algorithms/kem/ml_kem.md

ML-KEM

• Algorithm type: Key encapsulation mechanism.
• Main cryptographic assumption: Module LWE+R with base ring Z[x]/(3329, x^256+1).
• Principal submitters: Peter Schwabe.
• Auxiliary submitters: Roberto Avanzi, Joppe Bos, Léo Ducas, Eike Kiltz, Tancrède Lepoint, Vadim Lyubashevsky, John M. Schanck, Gregor Seiler, Damien Stehlé.
• Authors' website: https://pq-crystals.org/kyber/ and https://csrc.nist.gov/pubs/fips/203
• Specification version: ML-KEM.
• Primary Source:
  • Source: 048fc2a7a7
  • Implementation license (SPDX-Identifier): CC0-1.0 or Apache-2.0
• Optimized Implementation sources: 048fc2a7a7
  • cupqc-cuda:
    • Source: b026f4e547
    • Implementation license (SPDX-Identifier): Apache-2.0

Parameter set summary

Parameter set	Parameter set alias	Security model	Claimed NIST Level	Public key size (bytes)	Secret key size (bytes)	Ciphertext size (bytes)	Shared secret size (bytes)	Keypair seed size (bytes)
ML-KEM-512	NA	IND-CCA2	1	800	1632	768	32	64
ML-KEM-768	NA	IND-CCA2	3	1184	2400	1088	32	64
ML-KEM-1024	NA	IND-CCA2	5	1568	3168	1568	32	64

ML-KEM-512 implementation characteristics

Implementation source	Identifier in upstream	Supported architecture(s)	Supported operating system(s)	CPU extension(s) used	No branching-on-secrets claimed?	No branching-on-secrets checked by valgrind?	Large stack usage?‡
Primary Source	ref	All	All	None	True	True	False
Primary Source	x86_64	x86_64	Linux,Darwin	AVX2,BMI2,POPCNT	True	True	False
Primary Source	aarch64	ARM64_V8	Linux,Darwin	None	True	False	False
cupqc-cuda	cuda	CUDA	Linux,Darwin	None	False	False	False

Are implementations chosen based on runtime CPU feature detection? Yes.

‡For an explanation of what this denotes, consult the Explanation of Terms section at the end of this file.

ML-KEM-768 implementation characteristics

Implementation source	Identifier in upstream	Supported architecture(s)	Supported operating system(s)	CPU extension(s) used	No branching-on-secrets claimed?	No branching-on-secrets checked by valgrind?	Large stack usage?
Primary Source	ref	All	All	None	True	True	False
Primary Source	x86_64	x86_64	Linux,Darwin	AVX2,BMI2,POPCNT	True	True	False
Primary Source	aarch64	ARM64_V8	Linux,Darwin	None	True	False	False
cupqc-cuda	cuda	CUDA	Linux,Darwin	None	False	False	False

Are implementations chosen based on runtime CPU feature detection? Yes.

ML-KEM-1024 implementation characteristics

Implementation source	Identifier in upstream	Supported architecture(s)	Supported operating system(s)	CPU extension(s) used	No branching-on-secrets claimed?	No branching-on-secrets checked by valgrind?	Large stack usage?
Primary Source	ref	All	All	None	True	True	False
Primary Source	x86_64	x86_64	Linux,Darwin	AVX2,BMI2,POPCNT	True	True	False
Primary Source	aarch64	ARM64_V8	Linux,Darwin	None	True	False	False
cupqc-cuda	cuda	CUDA	Linux,Darwin	None	False	False	False

Are implementations chosen based on runtime CPU feature detection? Yes.

Explanation of Terms

• Large Stack Usage: Implementations identified as having such may cause failures when running in threads or in constrained environments.