Concept and modelling of memsensors as two terminal devices with enhanced capabilities in neuromorphic engineering

Abstract

We report on memsensors, a class of two terminal devices that combines features of memristive and sensor devices. Apart from a pinched hysteresis (memristive property) and stimulus dependent electrical resistance (sensing property) further properties like dynamic adaptation to an external stimulus emerge. We propose a three component equivalent circuit to model the memsensor electrical behaviour. In this model we find stimulus dependent hysteresis, a delayed response to the sensory signal and adaptation. Stimulus dependent IV hysteresis as a fingerprint of a memsensor device is experimentally shown for memristive ZnO microrods. Adaptation in memsensor devices as found in our simulations resembles striking similarities to the biology. Especially the stimulus dependency of the IV hysteresis and the adaptation to external stimuli are superior features for application of memsensors in neuromorphic engineering. Based on the simulations and experimental findings we propose design rules for memsensors that will facilitate further research on memsensitive systems.

Figure 1

Possible device schematics (horizontal and vertical design) for memsensors as well as typically applied stimulus signals (e.g. light, etc.). For the horizontal design, the active layer (e.g. nanowire) is open to any stimulus. In the vertical design, the top electrode must not prevent the stimulus from reaching the active layer. This can be achieved by structuring (e.g. for gas molecules) or the choice of a transparent electrode material (e.g. ITO for light).

Figure 2

(a) Circuit symbol for a UV memsensor as proposed by Chiolerio et al.; (b) generalized memsensor circuit symbol in agreement with the proposed (c) equivalent circuit of a memsensor device featuring memristive elements parallel (with R_m,par) and in series (with R_m,ser) to the sensing element (with R_s). The resistance R_s is sensitive to an external stimulus α_stim, e.g. gas concentration or UV-irradiation.

Figure 3

Adaptation to an external stimulus: (a) Amplitude adaptation: Amplitude of the response is proportional to the stimulus. (b) Spike frequency adaptation as proposed by E.D. Adrian in 1926: The frequency of spikes is proportional to the stimulus and fades over time. Spike frequency adaptation has been experimentally realized in a memristive spiking neuron circuit^–,. The system does not respond to subthreshold stimulus.

Figure 4

Schematic overview over the switching regimes realized by the mathematical representation of the memristive device in Equation (6) (a) and Equation (5) (b). The SET regime (green) corresponds to a switching of the memristive device to LRS at voltages above the threshold voltage. The slow RESET (yellow) describes the finite retention of the LRS. The fast RESET (red) below the negative threshold voltage corresponds to the switching to HRS.

Figure 5

Simulation of adaptation to an external stimulus: (a) The response of the memsensor (black line) decreases with each subsequent stimulus pulse (blue background). After a sufficient time at low stimulus, the memsensor recovers. In depth look at the impact of serial (green line) and parallel (red line) memristive element on the memsensor adaptation to external stimulus pulses (blue background). The adaptation is shown at the example of a short pulse (b) and a single long pulse (c).

Figure 6

Evaluation of simulated IV hysteresis based on the equivalent circuit. (a) Dependency of hysteresis on cycling time, the hysteresis loop narrows for faster cycling times; (b) dependency of hysteresis on external stimulus, the hysteresis loop widens for lower external stimulus.

Figure 7

(a) Schematic drawing (top) and a photographic image (right) of a ZnO microrod device for comparison. The SEM micrograph is showing the diameter of the ZnO microrod is 13 µm (left); The IV hysteresis curves show strong dependence on the illumination by UV light (b) as shown in linear IV plot and logarithmic RV plot. Without UV illumination the ZnO device approached the limit of reliable detection (10 pA). The time constants for the sensing of UV light are determined (c) to be in the range of seconds.