Stochastic Learning in Neuromorphic Hardware via Spike Timing Dependent Plasticity With RRAM Synapses

Abstract

Hardware processors for neuromorphic computing are gaining significant interest as they offer the possibility of real in-memory computing, thus bypassing the limitations of speed and energy consumption of the von Neumann architecture. One of the major limitations of current neuromorphic technology is the lack of bio-realistic and scalable devices to improve the current design of artificial synapses and neurons. To overcome these limitations, the emerging technology of resistive switching memory has attracted wide interest as a nanoscaled synaptic element. This paper describes the implementation of a perceptron-like neuromorphic hardware capable of spiketiming dependent plasticity (STDP), and its operation under stochastic learning conditions. The learning algorithm of a single or multiple patterns, consisting of either static or dynamic visual input data, is described. The impact of noise is studied with respect to learning efficiency (false fire, true fire) and learning time. Finally, the impact of stochastic learning rule, such as the inversion of the time dependence of potentiation and depression in STDP, is considered. Overall, the work provides a proof of concept for unsupervised learning by STDP in memristive networks, providing insight into the dynamics of stochastic learning and supporting the understanding and design of neuromorphic networks with emerging memory devices.

Memristive Synapses for Brain‐Inspired Computing

“…Although hardware artificial neural networks based on various memristive synapses have been successfully constructed, the neuronal functions are implemented either by CMOS circuits or in software running on the processors . It severely limits further improvements on scalability, stackability, and energy efficiency of the networks .…”

“…Artificial neural networks based on 1T1M arrays have been constructed using memristors with various memristive materials such as Al 2 O 3 /HfO 2 , TaO x /HfO 2 , HfO x , TaO x /HfAl y O x , and Pr 0.7 Ca 0.3 MnO 3 . A bilayer network has also been developed by employing HfO 2 as the memristor material in 1T1M arrays .…”

Bio-mimetic synaptic plasticity and learning in a sub-500 mV Cu/SiO2/W memristor

“…The time constants are functions of the device conductance and converges to the same value at the boundary points such that the conductance change gradually becomes zero. Such state dependent behavior is akin to the saturating conductance responses observed in many of the gradual conductance and STDP demonstrations in the memristive devices 26,35,36,[40][41][42][43] . The STDP response from the phenomenological model is shown in Fig.3c, where the model conductance is initialized to 1 µS and is programmed with a sequence of random ∆ts.…”

Stochastic Memory Devices for Security and Computing

“…In all these applications, guaranteeing secure data-transmission is of paramount importance. Several emerging computing paradigms, such as stochastic [17,18,26] and brain-inspired computing, [19,20] also rely on large bitstreams of random analog/digital signals for their operation, thus requiring on-chip entropy sources. For implementation in resource constrained IoT systems, RNGs should be compact and reliable, while featuring high-quality entropy, high throughput, and low power consumption.…”