Accelerated implementation for testing IID assumption of NIST SP 800-90B using GPU.

Area of Science:

• Cryptography
• Computer Science
• Hardware Acceleration

Background:

• Insufficient entropy in random number generator (RNG) noise sources can compromise cryptographic keys.
• Accurate entropy estimation is vital for RNG security.
• NIST SP 800-90B provides methods for entropy estimation, but current implementations are time-consuming.

Purpose of the Study:

• To accelerate the entropy estimation process for RNGs.
• To reduce the significant runtime burden of NIST's entropy estimation tools.
• To enhance the security analysis of cryptographic modules.

Main Methods:

• Developed a GPU-based parallel implementation for the independent and identically distributed (IID) assumption testing, a key part of entropy estimation.
• Proposed a scalable method for optimizing GPU global memory allocation and workload balancing.
• Created a hybrid CPU/GPU approach, incorporating OpenMP for a statistical test unsuitable for GPU.

Main Results:

• Achieved a speedup of 3 to 25 times compared to the NIST package for entropy estimation.
• Demonstrated the effectiveness of GPU acceleration for computationally intensive cryptographic tasks.
• Validated a hybrid approach for comprehensive entropy estimation.

Conclusions:

• The proposed GPU-based method significantly accelerates entropy estimation, addressing a critical bottleneck.
• This acceleration facilitates more rigorous security evaluations of RNGs and cryptographic modules.
• The optimized GPU implementation offers a practical solution for developers and evaluators in the field.