Nonvolatile Memories in Spiking Neural Network Architectures: Current and Emerging Trends

Abstract

A sustainable computing scenario demands more energy-efficient processors. Neuromorphic systems mimic biological functions by employing spiking neural networks for achieving brain-like efficiency, speed, adaptability, and intelligence. Current trends in neuromorphic technologies address the challenges of investigating novel materials, systems, and architectures for enabling high-integration and extreme low-power brain-inspired computing. This review collects the most recent trends in exploiting the physical properties of nonvolatile memory technologies for implementing efficient in-memory and in-device computing with spike-based neuromorphic architectures.

Exploring the Effects of Caputo Fractional Derivative in Spiking Neural Network Training

“…These results of SNN experiments are similar to ANN training results by [25]. These ANN experiments found that the suitable range for Caputo derivative orders is between 6 9 and 8 9 .…”

“…As described in Section 2, Caputo derivative can be utilized to calculate the partial derivative of the Tempotron cost function with respect to synaptic weights. To acquire the equation by which each weight could be updated using this novel Caputo-based Tempotron (Caputron) optimizer, first the Caputo derivative of the cost function should be considered over a restricted interval of the weight values [c, w ji ], and with a derivative order between 0 and 1, similarly to Equation (8).…”

“…Even modern TPUs cannot achieve as low training and inference energies as specifically designed SNN processors; therefore, SNN technology favors energy-efficient applications [5][6][7]. Recently developed memristive approaches might widen this gap even more [8].…”

Nonvolatile Memristive Materials and Physical Modeling for In‐Memory and In‐Sensor Computing

“…The idea of integrating analog spikes or stimuli is an extension of the summation scheme, which is a vital aspect for integrate‐and‐fire neurons in a spiking neural network (SNN) (Figure 7c). [ 385,386 ] The idea of an artificial synapse that obtains spike signals from various presynaptic inputs has been demonstrated. For example, in PCM layers, input spikes are combined and computed through the Kirchhoff's law of summing currents at a neuron circuitry source (Figure 7d).…”

Computing of neuromorphic materials: an emerging approach for bioengineering solutions

“…The implementation of parallelism has the potential to greatly enhance the efficiency of both neural network training and inference activities. 229…”