SPINART: Topological SPINn textures for true random number generation with applications to ARTificial intelligence

Spintronic devices are promising building blocks for the development of low consumption hardware dedicated to artificial intelligence. Indeed, Magnetic Tunnel Junctions (MTJ) technologies used in Magnetic Random-Access Memory technologies have been recently successfully applied to build True Random Number Generators (TRNGs) to implement artificial intelligence algorithms. In the SPINART, project we propose to go toward a further improvement of the energy consumption involved in random bit generation. For this we propose to replace the homogeneously magnetized layer (macrospin) in the MTJ by a Topological Spin Texture (TST). TST are seen as promising systems to decrease the energy consumption in the writing process for TRNG due to the ultrafast processes which are involved in their nucleation and the possibility to use optics to write and read them. In the SPINART project we will fabricate a TRNG based on Topological Change in TST triggered by thermal noise. In order to unveil the potentiality of this system for TRNG, we will carry out National Institute of Standards and Technology (NIST) statistical test on a TC-TRNG device with a switching probability tuned close to 0.5.

In the SPINART project, we will carry out a detailed experimental study, using Magneto Optical Kerr Effect (MOKE) microscopy, on the statistics of Topological Changes in a Pt/Co/Al2O3 system whose properties have been tuned to present a large number of thermally activated topological changes. Our system presents two phases, a skyrmion phase and a stripe phase.Figure 2 : Preliminary results. Filtered MOKE images (out of plane magnetization) in a Pt/Co/Al2O3 system. (a)(b) Skyrmion images under an applied magnetic field. (d)(e) Stripe domains recorded at zero applied magnetic field. Images (a)/(d) and (b)/(e) are recorded a few seconds apart. (c)/(f) Image difference between (a)/(d) and (b)/(e). The topological changes are highlighted by a green circle while a domain motion is emphasized by a red circle.In Figure 2 we present preliminary results to illustrate the thermally activated properties of the sample and demonstrate the presence of topological changes in the skyrmion and stripe phases. In Figure 2(a)-(b), we show two MOKE microscopy images obtained in a polar geometry where white skyrmion appears in a black ferromagnetic background. The images in (a) and (b) are recorded at the same position and separated by a few seconds. In Figure 2(c), we present the image difference (a)-(b). We observe three types of events: skyrmion nucleation, skyrmion annihilation and skyrmion displacement. The events that are associated with a topological change are highlighted with a green circle, while the domain displacement is shown in red. In Figure 2 (d)- (e), we present images of the same sample at zero magnetic field in the stripe phase state. Topological changes are also observed in this phase as highlighted by green circles. In order to go beyond these preliminary results and study the potential of the observed topological changes to be used as TRNG, we propose the following program:

1. Obtain the statistics of topological change on a large number of topological change eventsusing image analysis (classical algorithms, machine learning algorithms) on large images(500 µm x 500 µm) for various applied magnetic fields and temperatures (equipmentrequired).

2. Design, fabricate and tune individual devices with a topological changes probability [15]𝑃𝑇𝐶 = 1 – exp⁡(- 𝑡𝜏𝑇𝐶 ) close to 0.5, where t is the observation time and 𝜏𝑇𝐶 is the meantopological change time.

3. Run National Institute of Standards and Technology (NIST) statistical test on the TC-TRNGdevice.In order to carry out these objectives, we will benefit from an existing MOKE setup developed by the PI, as well as from the PI experimental and theoretical expertise on TST. In addition, the SPINTEC laboratory in Grenoble (Hélène Béa), already provided us with samples particularly adapted to this study. This system consist of a Pt/Co/Al2O3 trilayer which has been tuned by depositing a double wedge of Co and Al (prior to oxidation) on a 4 in. wafer. This fine-tuning enables us to obtain potentially much higher topological changes rates compared to previous works [16], [17]. Within the framework of SPINART, we plan to increase the detection rate of our MOKE microscope, which is currently limited to the millisecond range. For this we will use a system of photodiodes to locally detect the topological changes (equipment required). For this experimental development of our setup we will benefit from the expertise at CEMES in the development of optical setups with Sebastien Weber. This will allow us to detect TC optically at the microscale and frequency range up to the MHz. Concerning the energy consumption, as our TC-TRNG is harvesting thermal energy to reach a switching probability of 0.5, there is no energy associated with the writing process at the level of our study. The drawback will be a high sensitivity to temperature variations, and in more integrated versions of our device a writing process will include the use of current/voltage to temporally increase the topological change probability (beyond the scope of this project).