A Low Energy Oxide‐Based Electronic Synaptic Device for Neuromorphic Visual Systems with Tolerance to Device Variation

Abstract

Neuromorphic computing is an emerging computing paradigm beyond the conventional digital von Neumann computation. An oxide-based resistive switching memory is engineered to emulate synaptic devices. At the device level, the gradual resistance modulation is characterized by hundreds of identical pulses, achieving a low energy consumption of less than 1 pJ per spike. Furthermore, a stochastic compact model is developed to quantify the device switching dynamics and variation. At system level, the performance of an artificial visual system on the image orientation or edge detection with 16 348 oxide-based synaptic devices is simulated, successfully demonstrating a key feature of neuromorphic computing: tolerance to device variation.

Artificial synapse network on inorganic proton conductor for neuromorphic systems

“…Such EPSC behaviour is quite similar to an EPSC process in a biological excitatory synapse 34 . The energy dissipation of single spike event is estimated to be B45 pJ, which is lower than that of the artificial synapse based on conventional CMOS circuit 35 and is comparable to the energy dissipation of the reported hardware-based artificial synapse 8,9,15 . At present, the channel width of our synaptic transistors is large (B1 mm), and it can be scaled down to submicrometer scale by a photolithography method.…”

“…In the beginning, synaptic functions were emulated by complementary metal oxide semiconductor (CMOS) neuromorphic circuits, but such CMOS circuits consumed substantially more energy than a biological synapse, and it is hard to scale up the circuits to a size comparable with the brain 5 . Recently, resistive switching memory, memristors or atomic switch has been investigated in biologically inspired neuromorphic circuits [6][7][8][9][10][11] . Important synaptic learning rules such as spike-timing-dependent plasticity (STDP) and short-term memory to long-term memory transition have been demonstrated.…”

Memristor with Ag‐Cluster‐Doped TiO2 Films as Artificial Synapse for Neuroinspired Computing

“…However, most memristors reported related references can only continuously modulate conductance in single voltage polarity. [9,21] In our device, bidirectional positive and negative pulses with nanosecond timescale were employed to tune the condution in detail. First, we study the influence of the amplitude of pulse train on variation of conduction when sustained spikes of pulse train are applied to the memristor cell.…”

An energy efficient and high speed architecture for convolution computing based on binary resistive random access memory

“…14,15) Furthermore, various oxide RRAM devices exhibit the multilevel or analog storage, which has been implemented in the bio-inspired computing. 7,[16][17][18][19] Based on RRAM, several non-von Neumann computing systems have been proposed and demonstrated, such as the neuromorphic computing in which the RRAM acts as the synaptic device; [20][21][22][23][24] and the RRAM based in-memory logic computing, which is considered to be one of the solutions of the von-Neumann bottleneck. 25,26) Also the RRAM array has been utilized to accelerate some specific computing operations, for example the vector-matrix multiplication.…”