MTJ beyond memory element – from logic to neuromorphic computing
Prof. W. J. Park
Date & Time: 15:00 - 16:00, February 19th (Mon), 2018.
Place: Conference room #513-514, Advanced Structural Materials Building, Sengen
Abstract:
With compatibility to CMOS integration developed for magnetic random access memory (MRAM), magnetic tunnel junction (MTJ) has also been expected to be a prominent candidate for use as a basic element of computing architecture with functional and operating advantages that can be reconfigured in real time with non-volatility, high speed, and low power consumption. In this presentation, we will discuss our recent researches that have been newly observed for computational functions in MTJ. Discussions are included as following: 1) Boolean functions implemented by two-input extension of MTJ as a common configuration of computing element using independent physical variables of the Ampere field and the spin transfer torque, 2) construction of the universal logic gates using the MTJs, 3) formation of bit stream emulating the neuron signals representing the rate coded pulses, 4) circuital demonstration with MTJs representing the spike rate-dependent plasticity for emulating neurotransmission system, 5) MTJ neural network performing learning algorism using the SRDP learning rule.