AN4230 · Application noteIntroduction to random number generation validation using the NIST statistical test suite for STM32 MCUs and MPUs
Download .pdfOpen in the full document

NIST SP800-90B test suite steps

Step 1: random number generator

Connect the STM32 board to the workstation, via USB Type-A to Micro-B cable.

The STM32 RNG is run via the UART firmware in order to generate a random number as described in STM32 MCU side. Data are stored on the workstation using a terminal emulation application such as PuTTY (free and open-source terminal emulator, serial console, and network file transfer application).

Step 2: NIST statistical tests

The Makefile is used to compile the program as described in the readme file mentioned in NIST SP800-90b test suite side.

For non-IID tests, the user must follow the steps detailed below:
  1. Use the Makefile to compile the program: make non_iid
  2. Run the program with
    ./ea_non_iid [-i|-c] [-a|-t] [-v] [-l <index>,<samples> ] <file_name> [bits_per_symbol]

where

-i indicates that data is unconditioned and returns an initial entropy estimate (default).
-c indicates that data is conditioned.
-a estimates the entropy for all data in the binary file (default).
-t truncates the created bit-string representation of data to the first 1 Mbit.
-l reads (at most) data samples after indexing into the file by * bytes.
-v verbosity flag for more output (optional, can be used multiple times)bits_per_symbol: number of bits per symbol. Each symbol is expected to fit within a single byte.
Example:
/ea_non_iid ../bin/l5.bin 1 -i -t -v
For restart tests, the user must follow the steps detailed below:
  1. Use the Makefile to compile the program: make restart.
  2. Run the program with.
./ ea_restart [-i|-n] [-v] <file_name> [bits_per_symbol] <H_I>
Where
[bits_per_symbol]: Must be between 1-8, inclusive.
<H_I>: Initial entropy estimate.
[-i|-n]: '-i' for IID data, '-n' for non-IID data. Non-IID is the default.
-v Optional verbosity flag for more output.
<file_name>: Must be relative path to a binary file contains the data that consists of 1000 restarts, each with 1000 samples. The data is converted to rows and columns. 
Example:
./ea_restart It_res_1000x1000.bin 1 0.7695 -n -v

Step 3: test report

The NIST statistical tests provide an analytical routine to facilitate the interpretation of the results:
  • For the non IID tests, the result for each IID test is provided and, at the end, the final min entropy.
  • For the restart tests, we determine the Hr and Hc , the entropy estimates of the row and the column datasets, respectively. the final result of the restart tests is min (H_r, H_c, H_I).

Conclusion

This application note describes the main guidelines and steps to verify the randomness of numbers generated by the STM32 microcontrollers RNG peripheral, using either NIST statistical test suite SP800-22rev1a, April 2010 or SP800-90B, January 2018.

NIST SP800-22rev1a statistical test suite

The results are represented as a table with p rows and q columns, where:
  • p, the number of rows, corresponds to the number of statistical tests applied
  • q, the number of columns (q = 13) is distributed as follows:
    • columns 1-10 correspond to the frequency of 10 Pvalues
    • column 11 is the Pvalue that arises via the application of a chi-square test11
    • column 12 is the proportion of binary sequences that passed
    • column 13 is the corresponding statistical test

The example below shows the first and last part of the test results. For more details, refer to the finalAnalysisReport file under sts-2.1.1\experiments\AlgorithmTesting.

Part 1

Part 2

The minimum pass rate for each statistical test, except the random excursion (variant) test, is approximately 8 for a sample size of 10 binary sequences.

The minimum pass rate for the random excursion (variant) test is approximately 4 for a sample size of 5 binary sequences.

For further guidelines, construct a probability table using the MAPLE program provided in the addendum section of the documentation.

NIST SP800-90b statistical test suite

This is an example of the execution of the non-IID tests.

$ ./ea_non_iid ../bin/l5.bin 1 -i -t -v Opening file: '../bin/l5.bin'
Number of Binary Symbols: 1024000
Symbol alphabet consists of 2 unique symbols 
Running non-IID tests...
Running Most Common Value Estimate... 
MCV Estimate: mode = 530185, p-hat = 0.51775878906249995, p_u = 0.51903071907139675 
Most Common Value Estimate (bit string) = 0.946108 / 1 bit(s)
Running Entropic Statistic Estimates (bit strings only)... 
Collision Estimate: X-bar = 2.4954732991667945, sigma-hat = 0.49998011778222412, p = 0.55717151750062044
Collision Test Estimate (bit string) = 0.843807 / 1 bit(s)
Markov Estimate: P_0 = 0.4822412109375, P_1 = 0.51775878906249995, P_0,0 = 0.48618305677846313, P_0,1 = 0.51381694322153693, P_1,0 = 0.47857068759018079, P_1,1 = 0.52142931240981927, p_max = 6.2798397734367098e-37
Markov Test Estimate (bit string) = 0.939536 / 1 bit(s) 
Compression Estimate: X-bar = 5.2113069751685934, sigma-hat = 1.0187463366852749, p = 0.035431813237513987
Compression Test Estimate (bit string) = 0.803135 / 1 bit(s)
Running Tuple Estimates... t-Tuple Estimate: t = 16, p-hat_max = 0.53187518804173173, p_u = 0.53314533218239224 
LRS Estimate: u = 17, v = 38, P_{max,W} = 0.50423097749594759, p_u = 0.50550366496585808 
T-Tuple Test Estimate (bit string) = 0.907399 / 1 bit(s) 
LRS Test Estimate (bit string) = 0.984207 / 1 bit(s)
Running Predictor Estimates... 
MultiMCW Prediction Estimate: N = 1023937, Pglobal' = 0.51634782805743162 (C = 527405) Plocal = 0.42674646958653317 (r = 21) 
Multi Most Common in Window (MultiMCW) Prediction Test Estimate (bit string) = 0.953585 / 1 bit(s)
Lag Prediction Estimate: N = 1023999, Pglobal' = 0.50526439621655594 (C = 516087) Plocal = 0.40830971576974662 (r = 20)
Lag Prediction Test Estimate (bit string) = 0.984890 / 1 bit(s) 
MultiMMC Prediction Estimate: N = 1023998, Pglobal' = 0.51899462537798435 (C = 530147) Plocal = 0.42674521449187308 (r = 21) 
Multi Markov Model with Counting (MultiMMC) Prediction Test Estimate (bit string) = 0.946208 / 1 bit(s) 
LZ78Y Prediction Estimate: N = 1023983, Pglobal' = 0.51899440603936897 (C = 530139) Plocal = 0.42674552311446512 (r = 21) 
LZ78Y Prediction Test Estimate (bit string) = 0.946209 / 1 bit(s)
H_original: 1.000000 
H_bitstring: 0.803135
min(H_original, 1 X H_bitstring): 0.803135

This is an example of the execution of the restart tests.

$ ./ea_restart It_res_1000x1000.bin 1 0.7695 -n -v
Opening file: '/home/Desktop/It_restart_1000x1000_bins/It_res_1000x1000.bin'
Loaded 1000000 samples made up of 2 distinct 1-bit-wide symbols.
H_I: 0.769500
ALPHA: 5.0251553006530614e-06, X_cutoff: 654
X_max: 592
Restart Sanity Check Passed...
Running non-IID tests...
Running Most Common Value Estimate...
Literal MCV Estimate: mode = 524815, p-hat = 0.52481500000000003, p_u = 0.52610132816195332
Most Common Value Estimate (Rows) = 0.926587 / 1 bit(s)
Literal MCV Estimate: mode = 524815, p-hat = 0.52481500000000003, p_u = 0.52610132816195332
Most Common Value Estimate (Cols) = 0.926587 / 1 bit(s)
Running Entropic Statistic Estimates (bit strings only)...
Literal Collision Estimate: X-bar = 2.4939272870560187, sigma-hat = 0.49996374423442524, p = 0.56366499259647185
Collision Test Estimate (Rows) = 0.827090 / 1 bit(s)
Literal Collision Estimate: X-bar = 2.4988330426351748, sigma-hat = 0.49999926291763369, p = 0.54001782688956723
Collision Test Estimate (Cols) = 0.888921 / 1 bit(s)
Literal Markov Estimate: P_0 = 0.52481500000000003, P_1 = 0.47518499999999997, P_0,0 = 0.5286091289311472, P_0,1 = 0.4713908710688528, P_1,0 = 0.52062569446782725, P_1,1 = 0.47937430553217275, p_max = 3.6149969304059282e-36
Markov Test Estimate (Rows) = 0.919808 / 1 bit(s)
Literal Markov Estimate: P_0 = 0.52481500000000003, P_1 = 0.47518499999999997, P_0,0 = 0.52490687194535213, P_0,1 = 0.47509312805464787, P_1,0 = 0.5247146368564598, P_1,1 = 0.4752853631435402, p_max = 1.4806522327909964e-36
Markov Test Estimate (Cols) = 0.929869 / 1 bit(s)
Literal Compression Estimate: X-bar = 5.2053159685055412, sigma-hat = 1.0214276541597365, p = 0.040397638399976898
Compression Test Estimate (Rows) = 0.771598 / 1 bit(s)
Literal Compression Estimate: X-bar = 5.2093623833151179, sigma-hat = 1.0198576801023285, p = 0.037170466884093645
Compression Test Estimate (Cols) = 0.791617 / 1 bit(s)
Running Tuple Estimates...
Literal t-Tuple Estimate: t = 16, p-hat_max = 0.54171776156680396, p_u = 0.54300118613085924
Literal LRS Estimate: u = 17, v = 38, p-hat = 0.50185202103932736, p_u = 0.50313992749997694
Literal t-Tuple Estimate: t = 16, p-hat_max = 0.53421514909068124, p_u = 0.535500045383837
Literal LRS Estimate: u = 17, v = 36, p-hat = 0.50196715258596614, p_u = 0.50325505791399572
T-Tuple Test Estimate (Rows) = 0.880973 / 1 bit(s)
T-Tuple Test Estimate (Cols) = 0.901041 / 1 bit(s)
LRS Test Estimate (Rows) = 0.990968 / 1 bit(s)
LRS Test Estimate (Cols) = 0.990638 / 1 bit(s)
Running Predictor Estimates...
Literal MultiMCW Prediction Estimate: N = 999937, Pglobal' = 0.52385156519694509 (C = 522532) Plocal can't affect result (r = 21)
Multi Most Common in Window (MultiMCW) Prediction Test Estimate (Rows) = 0.932770 / 1 bit(s)
Literal MultiMCW Prediction Estimate: N = 999937, Pglobal' = 0.52391856160939276 (C = 522599) Plocal can't affect result (r = 22)
Multi Most Common in Window (MultiMCW) Prediction Test Estimate (Cols) = 0.932586 / 1 bit(s)
Literal Lag Prediction Estimate: N = 999999, Pglobal' = 0.50560437226186183 (C = 504316) Plocal can't affect result (r = 20)
Lag Prediction Test Estimate (Rows) = 0.983919 / 1 bit(s)
Literal Lag Prediction Estimate: N = 999999, Pglobal' = 0.50229941431576741 (C = 501011) Plocal can't affect result (r = 19)
Lag Prediction Test Estimate (Cols) = 0.993381 / 1 bit(s)
Literal MultiMMC Prediction Estimate: N = 999998, Pglobal' = 0.52608038171682492 (C = 524793) Plocal can't affect result (r = 21)
Multi Markov Model with Counting (MultiMMC) Prediction Test Estimate (Rows) = 0.926645 / 1 bit(s)

Literal MultiMMC Prediction Estimate: N = 999998, Pglobal' = 0.5260743824720614 (C = 524787) Plocal can't affect result (r = 22)
Multi Markov Model with Counting (MultiMMC) Prediction Test Estimate (Cols) = 0.926661 / 1 bit(s)
Literal LZ78Y Prediction Estimate: N = 999983, Pglobal' = 0.52609326184739169 (C = 524798) Plocal can't affect result (r = 21)
LZ78Y Prediction Test Estimate (Rows) = 0.926610 / 1 bit(s)
Literal LZ78Y Prediction Estimate: N = 999983, Pglobal' = 0.52608626262396685 (C = 524791) Plocal can't affect result (r = 22)
LZ78Y Prediction Test Estimate (Cols) = 0.926629 / 1 bit(s)
H_r: 0.771598
H_c: 0.791617
H_I: 0.769500
Validation Test Passed...
min(H_r, H_c, H_I): 0.769500

Revision history

Table 5. Document revision history
DateVersionChanges
13-May-20131Initial release.

22-Jun-2016

2Updated:
  • Introduction.
  • Section 1: STM32 microcontrollers random number generator.
  • Figure 1: Block diagram.
  • Section 3.1: Firmware description.
  • Section 3.2.1: First step: random number generator.
  • Section 3.2.2: Second step: NIST statistical test.
  • Section 4: Conclusion.
  • Added

  • Figure 2: STM32 lines embedding the RNG hardware peripheral.

1-Oct-2019

3Updated:
  • Introduction and Table 1: Applicable products
  • Table 2: STM32 lines embedding the RNG hardware peripheral
  • Figure 1: STM32 true RNG block diagram
  • Section 6: Conclusion
  • Appendix A: NIST SP800-22b statistical test suite

Added:

Section 4: NIST SP800-90b test suite

Section 5: NIST SP800-90b test suite running and analyzing

Appendix B: NIST SP800-90b statistical test suite

10-Oct-2019

4

Updated Table 2: STM32 lines embedding the RNG hardware peripheral.

8-Jan-2020

5

Updated:
  • Table 1: Applicable products
  • Table 2: STM32 lines embedding the RNG hardware peripheral
  • Section 5.1.1: STM32 MCU side

18-Aug-2020

6

Updated with new products STM32H72x/73x:
  • Table 1: Applicable products
  • Table 2: STM32 lines embedding the RNG hardware peripheral
25-Aug-20216Migration to DITA
1-Jul-20227Updated:
  • Table 1: Applicable products
  • NIST SP800-22rev1a test suite
  • NIST SP800-90b test suite
  • Firmware description used for RNG NIST evaluation
09-Mar-20238Updated the following elements to add STM32H5 and STM32WBA series:
  • Table 1. Applicable products.
  • Table 2. STM32 lines embedding the RNG hardware peripheral.
  • Section 5.1.3: NIST compliant RNG configuration.

Updated the document title.

Applied minor changes to the whole document.

10-Jan-20249Updated:
  • Document title.
  • Table 1. Applicable products.
  • Table 2. STM32 lines embedding the RNG hardware peripheral.
  • Figure 2. STM32 true RNG block diagram of the products verified with the SP800-90b.
  • Table 3. RNG register values.
  • Section 5.1.3: NIST compliant RNG configuration..
14-Mar-202410Added STM32U0 series and STM32H7R3/7S3 and STM32H7R7/7S7 lines in Table 1. Applicable products
Updated:
  • Table 3. RNG register values
  • Table 4. NIST SP800-90B certificates
21-Oct-202411Added STM32MP2 series in Table 1. Applicable products
Updated:
  • Table 2. STM32 lines embedding the RNG hardware peripheral
  • Table 3. RNG register values
  • Table 4. NIST SP800-90B certificates
07-Feb-202512Added STM32N6 series and STM32U3 series in Table 1. Applicable products
Updated:
  • Table 2. STM32 lines embedding the RNG hardware peripheral
  • Table 3. RNG register values
  • Table 4. NIST SP800-90B certificates
16-Feb-202613
  • Added STM32C5 series
  • Added Section 5.1.2.1: Interruption extraction mode
  • Section 5.1.2.2: Pooling extraction mode
  • Section 5.1.2.3: How to configure the RNG NIST Compliant configurations

Updated:

  • Section 5.1.2: STM32 raw data extraction firmware
  • Table 2. STM32 lines embedding the RNG hardware peripheral
  • Table 3. RNG register values
  • Table 4. NIST SP800-90B certificates
28-Jul-202614Updated Table 3 RNG register values.
1 A special device must be used to execute the NIST test suite, except for the STM32L4P/Q, STM32H7A3/7B3, and STM32H72x/73x.
2 The bit 6 of RNG_CR enables or disables the bypass of post‑processing/conditioning logic. This bit must be enabled to run the NIST SP800‑90B test suite.
3 "Yes" indicates when the configuration values are part of an official NIST SP800-90B certificate. Links to these certificates can be found in ***.
4 The HTCR value programmed in the product OTP fuse configuration (rng_htcr_value) applies only to the boot ROM. Because the OTP selection supports only a restricted set of HTCR values, set rng_htcr_value to 0x6, which corresponds to HTCR = 0xAAC7. The user application can then use the HTCR value provided in ***.