Connecting spiking neurons to a spiking memristor network changes the memristor dynamics

Abstract

Abstract-Memristors have been suggested as neuromorphic computing elements. Spike-time dependent plasticity and the Hodgkin-Huxley model of the neuron have both been modelled effectively by memristor theory. The d.c. response of the memristor is a current spike. Based on these three facts we suggest that memristors are well-placed to interface directly with neurons. In this paper we show that connecting a spiking memristor network to spiking neuronal cells causes a change in the memristor network dynamics by: removing the memristor spikes, which we show is due to the effects of connection to aqueous medium; causing a change in current decay rate consistent with a change in memristor state; presenting more-linear I − t dynamics; and increasing the memristor spiking rate, as a consequence of interaction with the spiking neurons. This demonstrates that neurons are capable of communicating directly with memristors, without the need for computer translation.

Emergent spiking in non-ideal memristor networks

“…If these changes are not synchronised then these small ∆V 's can move around the network causing the individual memristors to spike and propagate a different ∆V (remember the entire network is subject to a constant voltage so any small change in resistance on one memristor will affect the voltage across the others): this situation is called the 'roving ∆V in [34]. This idea explains the loss of the voltage spikes seen in ideal networks, the dampening of the slow a.c. voltage in figure 5 and the sudden emergence of bursting spikes from an almost flat baseline that has been observed (see the control in [16]). Thus, the oscillations can then be explained as the 'ringing' of the network that results from constructive and destructive interactions of spikes as a ∆V is passed around and the bursting spikes would occur when the spikes constructively interact.…”

“…artificial limbs that interface directly with the nervous system). We have started to investigate this by combining spiking memristor networks with spiking neural cell culture to see if the two spiking networks can influence each other electrically [16]. Another area of interest is biomimetic robotics where spiking memristor networks could be used to process sensory inputs and control a robot.…”

“…It is interesting to compare this work to our ongoing work with neural cell cultures connected to single memristors [16]. In those experiments, the single memristor underwent '1', 'w' and '1s' type behaviour, demonstrating that this behaviour can arise from single memristor circuits.…”

“…Simulations have shown that memristive connections could be used to reproduce spike-time dependent plasticity [10] (the process by which synapses adjust their connection weight to implement Hebbian learning [14]). Experiments have modelled neural action with memristors [15] and investigated the interactions between living neural cells and memristors [16].…”

Competitive Learning in a Spiking Neural Network: Towards an Intelligent Pattern Classifier

“…The neural synchronization in such systems can involve both living and artificial neurons. There are reports showing the first experimental implementations of hybrid networks consisting of living neuronal cells and artificial spiking neurons [22,23]. Thus, the new generation of HMI can be implemented entirely by neural networks where neurons of the brain interact with their artificial counterparts that work as part of prostheses or external devices.…”