Number Generation from Different Types of Magnetic Domain Images

Abstract

1. IntroductionRandom numbers are an important resource in a science and engineering, such as numerical simulations, cryptography and machine learning. A Faster and simpler true random number generators (TRNG), which employs physical phenomena as a noise source, is needed. However, the TRNGs that is compact and faster than GHz is not realized yet.We focused on the magnetic-domain patterns in magnetic garnets as a noise source for TRNG. The magnetic domain can be generated repeatedly and chaotically, and converted to digital value at few Gbit/s by image sensors. The random numbers that generated from the labyrinth magnetic-domain patterns were passed the statistic randomness test (NIST SP800-22) [1]. In this study, other types of magnetic domains were investigated as the noise source.2. ExperimentalSputtered-polycrystalline bismuth substituted yttrium iron garnet (Bi:YIG) films on glass and gadolinium gallium garnet (GGG) substrates, and liquid phase epitaxy (LPE) holmium iron garnet (HoIG) film were investigated. About 3,000 magnetic domain images of each garnet were captured using a polarized light microscope, as shown in Fig. 1.The obtained images were processed to random sequence. The data process was composed of binarization, stable-pattern removal, bit-shift, and XOR. A hundred of the 10 Mbit sequences were generated and evaluated by the randomness test (NIST SP800-22) [2].3. Result and discussionA hundred of the 10 Mbit sequences were generated and tested. The results of the randomness test is shown in table 1. The NIST SP800-22 includes 15 periodicity tests. The p-value more than 0.01 indicates that the sequence is non-periodic (random). The pass rates on the table 1 is a ratio of the statistically random sequence. The pass rate more than 0.97 is passed the randomness test. The all samples passed almost of all tests. The low-score with the FFT test could be improved by optimization of the data processing. These results suggest that the random sequences could be generated from many types of magnetic domains.  Fig. 1 Polarized microscope image of the garnet films  Table. 1 Result of SP800-22 for 100×10Mbit sequences