cancel
Showing results for 
Search instead for 
Did you mean: 

TRNG in STM32L4R9I - EVAL pass ALL NIST Test ?

TSang.3
Associate

Hello !

STM32L4R9I support True Random Number ( 32 bits per 1 time generation)

I run an example and then make an program to generate 100.000.000 bits

1 st iteration (100M bit) : HW is 0.5 and Passed NIST Test

2 nd iteration (100M bit) : HW is 0.5001 and Failed NIST Test ( 3 Non-overlapping )

Did TRNG pass All NIST Test ?

If yes, how to set up NIST Test ? ( I attached my setup as below)

If no, why?

P/s: https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-22r1a.pdf

Thanks so much !

1 ACCEPTED SOLUTION

Accepted Solutions
TZARDI
Associate III

Hello TSang.3,

In the "On the Interpretation of Results from the NIST Statistical Test Suite" it shows that data can be considered random even if it does not pass all the tests :

"Computed reference cumulative probabilities indicate that a single sequence can be considered non-random if it fails (p-values are smaller than α = 1%) 7 or more NIST STS tests. According to the uniformity test, 100 sequences can be considered non-random if they fail 7 uniformity tests (α = 1%) or 3 uniformity tests (for α = 0.1% or α = 0.01%). According to the test of proportions, 1000 (100) sequences can be considered non-random if they fail 4 (10) tests of proportion."

You can find attached the file interpreting the results.

Br,

View solution in original post

1 REPLY 1
TZARDI
Associate III

Hello TSang.3,

In the "On the Interpretation of Results from the NIST Statistical Test Suite" it shows that data can be considered random even if it does not pass all the tests :

"Computed reference cumulative probabilities indicate that a single sequence can be considered non-random if it fails (p-values are smaller than α = 1%) 7 or more NIST STS tests. According to the uniformity test, 100 sequences can be considered non-random if they fail 7 uniformity tests (α = 1%) or 3 uniformity tests (for α = 0.1% or α = 0.01%). According to the test of proportions, 1000 (100) sequences can be considered non-random if they fail 4 (10) tests of proportion."

You can find attached the file interpreting the results.

Br,