In an increasingly data-rich world the need for developing computing systems that cannot only process, but ideally also interpret big data is becoming continuously more pressing. Brain-inspired concepts have shown great promise towards addressing this need. Here we demonstrate unsupervised learning in a probabilistic neural network that utilizes metal-oxide memristive devices as multi-state synapses. Our approach can be exploited for processing unlabelled data and can adapt to time-varying clusters that underlie incoming data by supporting the capability of reversible unsupervised learning. The potential of this work is showcased through the demonstration of successful learning in the presence of corrupted input data and probabilistic neurons, thus paving the way towards robust big-data processors.

No abstract

Neuromorphic architectures offer great promise for achieving computation capacities beyond conventional Von Neumann machines. The essential elements for achieving this vision are highly scalable synaptic mimics that do not undermine biological fidelity. Here we demonstrate that single solid-state TiO2 memristors can exhibit non-associative plasticity phenomena observed in biological synapses, supported by their metastable memory state transition properties. We show that, contrary to conventional uses of solid-state memory, the existence of rate-limiting volatility is a key feature for capturing short-term synaptic dynamics. We also show how the temporal dynamics of our prototypes can be exploited to implement spatio-temporal computation, demonstrating the memristors full potential for building biophysically realistic neural processing systems.

No abstract

A report is presented on the operation of an analogue programming circuit for accurately setting the state of a memristor. The circuit exploits the dynamic modulation of resistance under a constant DC bias while real-time measurements of the memristance are performed using an AC signal. The circuit employs feedback for converging the state of a device at any required level within a decade. This allows the memristor to act as an analogue potentiometer, with its resistance corresponding to an input analogue voltage. This implementation was tested with the HP memristor model revealing an accuracy of less than 0.4% (8 bit precision) in relation to the full dynamic range.Introduction: The memristor is a description of a basic phenomenon of nature [1] that was postulated as the fourth missing element by Chua [2] and has attracted significant research interest after HP identified this phenomenon in nanoscale resistive-RAM [3]. Its unconventional properties, non-volatile memory that depends upon the biasing history and nonlinearity, enable the development of new circuit architectures that can facilitate low precision analogue computation [4], neuromorphic circuits [5] and resistive memory [6].In all cases, programming a memristor with increased accuracy is therefore essential. Several circuits have been proposed for this reason that can be divided into two main categories: (i) utilising distinct width and amplitude pulses to read, write and reset a device to an intermediate state [7,8], and (ii) using one or several linear resistors as references in order for the memristance to 'latch' onto the reference resistor [9,10]. The drawback of the first method is associated with external processing needed to calculate the pulse timings, for compensating the highly nonlinear dynamics of the device, while the latter scenario is typically avoided as it is component-hungry, due to the reference resistors needed for setting the programming precision. In this Letter, we propose an analogue approach to memristor programming that does not suffer from the limitations described above, and most importantly, can provide a linear relationship between an analogue input voltage and the final programmed value of memristance. Our approach was evaluated with SPICE simulations and we demonstrate that programming the device within 8 bit precision is possible.