Encryption of data and communication protocols requires unbiased, true random numbers. However, in most cases, they are actually pseudo-random because they are generated from deterministic and therefore predictable digital hardware.
Active research is underway to explore intrinsically random and non-deterministic processes in physical devices. Among these, thermal fluctuations provide an efficient and virtually cost-free means to generate a random bit stream (0 and 1) from a physical system characterized by two energy minima.
However, these hardware implementations often suffer from an intrinsic bias, leading to unequal energy minima and energy barriers. This leads to significant deviations from a fully balanced distribution of the binary levels 0 and 1. In order to make the two levels equivalent, additional signal processing is needed.

Researchers at CEA-IRIG/SPINTEC have validated a concept for generating true random and intrinsically unbiased bit streams, based on the stochastic phase dynamics of spintronic nano-oscillators. When the oscillator is synchronized to an external signal, whose frequency is twice its own frequency, then its phase stabilizes on one out of two discrete values with equal probability, ​in multiples of Pi​. (cf. figure a). Thus, a Pi​-periodic potential for the phase appears whose minima and maxima are intrinsically identical. In the presence of thermal noise, stochastic transitions of the phase between the minima, such as from a state 0 to ​Pi and from Pi to 2Pi,​ have exactly equal probability, allowing the generation of an unbiased bit stream (cf. figure b). Validation using the National Institute of Standards and Technology ​statistical test suite confirmed the suitability of the generated bit stream for secure encryption applications.

 
Figure: (a) Illustration of phase binarization by synchronization to an external source signal (Drive). The phase of the oscillator locks to that of the external signal.
(b) The two values of the binarized phase are converted into a binary voltage signal, using an electronic circuit specifically developed for this purpose. The generation of an unbiased bit stream is confirmed by the perfectly balanced histogram.
​​

Compared to other concepts, the spintronic nano-oscillator-based implementation does not require additional signal processing to make the bit stream unbiased. The researchers are currently exploring an innovative magnetic tunnel junction configuration to reach a bit-stream-fluctuation-rate in the GHz range. These results are exploited in collaboration with CEA-LETI to implement an Ising machine based on the stochastic phase dynamics of spintronic nano-oscillators.

Collaborations

• Iberian Nanotechnology Laboratory (INL) 
• ​National Institute of Standards and Technology (NIST) 
  
• University of Maryland

• ​ANR-NSF Stochnet 
• Grenoble INP Bourse Présidence 
• ANR MIAI@Grenoble Alpes 
  
• Bourse de these CEA Focus Numérique Frugal