Author:

Abstract:

In the context of neuromorphic computation, spintronic memristors are investigated for their use as synaptic weights [1]. Here, we propose and demonstrate a scalable synaptic device based on 10 magnetic tunnel junctions (MTJs) connected in series (Fig. 1(a)). Our device allows for multi-level switching, showing a large number of resistance states. This coincides well with the desired analog-like behavior in ideal memristors. When driving the magnetization reversal with magnetic field, this device can reach 10 resistance levels, switching the MTJs consecutively as shown in Fig. 1(b). For spin-transfer torque (STT)-driven reversal with positive voltage pulses, half of MTJs switch from parallel (P) state to anti-parallel (AP) while the other half switches from AP to P. However, by applying bipolar voltage pulses and by selecting the appropriate external magnetic field, we can intentionally decrease (Fig. 1(c)) or increase (Fig. 1(d)) the resistance of the chain. In the ideal case, a total of 2^N resistance levels could be accessible when connecting N MTJs in series [2]. Here, we show experimentally that we can distinguish at least 14 levels in our memristor (Fig. 2). These levels are obtained by sweeping the external magnetic field and adding a bipolar voltage pulse at every field step. We performed this experiment a dozen times and could access at least 14 discrete synaptic weights. This device shows promise as a scalable synapse in neuromorphic hardware. References: [ 1] S. Lequeux et al., "A magnetic synapse: multilevel spin-torque memristor with perpendicular anisotropy." Scientific Reports 6 (2016). [2] Zhang, Deming,et al. "All spin artificial neural networks based on compound spintronic synapse and neuron." IEEE transactions on biomedical circuits and systems 10.4 (2016): 828-836