Physical Reservoir Computing Using Spintronic Nano-devices

The rapid development of integrated circuit technology is the foundation of the modern information industry.However,with the continuous improvement of the semiconductor process technology,the size of the basic block of devices is approaching its physical limit,resulting in Moore's law of complementary metal-oxide-semiconductor(CMOS)chip eventually fails.As we all know,traditional CMOS chip technology only uses the charge property of electrons.As the size of devices decreases,another property,the spin property of electrons,is emerging and dominating the physical properties of materials and the functions of devices.The emerging spin-based devices,also named as spintronics,combines both fields of semiconductor and magnetism,uses both the charge and spin properties of electrons in materials to develop a new generation of chip technology,such as magnetic random access memory(MRAM).In recent years,with artificial intelligence boom,some spintronic devices also exhibit the great promising application in the field of artificial intelligence because they possess several advantages,e.g.,fast-speed,low power consumption,nonlinear,and short-term memory properties,which can be used to build Von Neumann architecture brainlike chips.In this master'thesis,we numerically modeled two physical reservoir computing(RC)systems based on one single magnetic skyrmion memristor(MSM)and 24 spin-torque nano-oscillators(STNOs),and adopt the standard image classification and nonlinear dynamic system predictiontasks to verify their functions and performances,respectively.In the third chapter,we first analyzed in detail a theoretical work about how to deal with the complex nonlinear chaotic phenomena via RC,and preliminarily reproduced some of their results.Then we built the two physical RC systems by using two physical spintronic devices based on the nonlinear responses of the MSM and STNO with current pulse stimulation.We found that the MSM-based RC system exhibits excellent performance on image classification,while the STNO-based RC system does well in solving the complex unknown nonlinear dynamic problems,e.g.,a secondorder nonlinear dynamic system and NARMA10(a ten-order nonlinear dynamic system).Our micromagnetic simulation results and analysis of the current-dependent nonlinear dynamic properties of the MSM and STNO,proposed in this thesis,provide the strategy to optimize the experimental parameters in building the better spintronic-based brainlike devices for machine learning-based computing.