Electrode Material Dependence of Resistance Change Behavior in Ta2O5 Resistive Analog Neuromorphic Device

In a human brain, our closest and very low-power information processor, one neuron transmits electrical signals depending on the electrical stimulation from other neurons. Therefore, it can be regarded as the multi-input one-output system that is the same as an elementary perceptron in the pattern recognition system. In analogy with the actual neurons whose coupling strength varies on a moment-to-moment basis, a flexible control of the weight for each individual input signal is required in order to realize the correct recognition. The analog change of the resistance values observed in the resistance change device is quite suitable for such application. In this contribution, we successfully demonstrate the high-speed analog resistance change in the TiN/TaOx/Ta2O5/TiN resistive analog neuromorphic device (RAND). An introduction of the TiN electrode smoothed the discontinuity in both the resistance switching processes by dc and pulse voltages. On the other hands, digital resistance switching was dominant in TiN/TaOx/Ta2O5/Pt device. We deduce that the electrodes reactivity with oxygen and the interface resistance play a key role for the analog resistance switching. The analog resistance change speed of 200 ns is much faster than the signal transmission speed between neurons and is thought to increase the number of operations per unit energy consumption. By introducing present RAND, the human-brain inspired information processor is expected to become energetically more efficient.