NEUROKDF
Maria Surkova · Zenodo (CERN European Organization for Nuclear Research) · 2026
This paper presents NeuroKDF, a key derivation function based on the NeuroHash neural network architecture with a total code size of 8.93 KB. The algorithm has successfully completed full NIST SP 800-22 certification: 15/15 tests on 800 sequences of 1 million bits each (total volume 100 MB). Key cryptographic characteristics: avalanche effect 50.03% (IQR 4.30%), bit balance 49.99% (bias 0.0037%), Shannon entropy 7.9999 bits/byte, no collisions (0/100,000), resistance to second preimage attacks. Protection against timing attacks is ensured by constant execution time (variation 1.78% IQR/median). The algorithm is deterministic on any CPU (unified scalar implementation). The 8.93 KB code size allows the use of NeuroKDF in embedded systems and IoT devices.