What does this research mean for the field?

NeuroCipher is a neural network pseudorandom sequence generator that passes all NIST SP 800-22 validation tests and demonstrates superior performance metrics compared to AES-256. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This work aims to introduce and validate NeuroCipher, a pseudorandom sequence generator based on neural networks.

March 13, 2026Open Access

Neurocipher

Key Points

This work aims to introduce and validate NeuroCipher, a pseudorandom sequence generator based on neural networks.
Developed NeuroCipher model with 2048 parameters.
Validated against NIST SP 800-22 using 15 tests.
Compared performance metrics against AES-256 algorithm.
Measured generation speed on multiple threads.
Achieved an entropy of 8.00 bits and avalanche effect of 49.84%.
Exhibited a near-ideal balance of 49.98% with zero collisions.
Displayed competitive speed of 0.64 MB/sec.

Abstract

This paper presents NeuroCipher — a neural network pseudorandom sequence generator that has passed full NIST SP 800-22 validation (15 out of 15 tests). The model occupies 8 KB (2048 parameters) and demonstrates entropy of 8.00 bits, avalanche effect of 49.84% with ±4.42% spread, near-ideal balance of 49.98%, and zero collisions. A comparison with the reference AES-256 algorithm is conducted across 12 parameters, including quantum resistance and side-channel attack resilience. Generation speed reaches 0.64 MB/sec (39,682 blocks/sec) on 4 threads. Results of comparative analysis with AES-256 are presented, and application areas in microcontrollers and hardware security modules are discussed.

Read Full Paperexternally

Ask AI

Helpful

Bookmark

View Full Paper