Spiking Neural Networks With Time-to-First-Spike Coding Using TFT-Type Synaptic Device Model

Abstract

In hardware-based spiking neural networks (SNNs), the conversion of analog input data into the arrival time of an input pulse is regarded as a good candidate for the encoding method due to its bioplausibility and power-efficiency. In this work, we trained an SNN encoded by time to first spike (TTFS) and performed an inference process using the behavior of the fabricated TFT-type flash synaptic device. The exponentially decaying synaptic current model required in the inference process was implemented by reading devices in the subthreshold region using triangle pulses. In a high-level system simulation, the TTFS-SNN (two-layer MLP with 512 hidden neurons) reached a high accuracy of 97.94%. Compared to conventional rate-encoded SNNs, TTFS-SNN made 2.9 times faster judgment and consumed ~10 times less energy in the inference process. Additionally, to use the network in a more stable condition, we propose a method to operate it using a rectangle pulse in the saturation region of the synaptic device. The distortion caused by this approximation was minimized by shortening the pulse width. As a result, the modified inference system showed an accuracy of 97.36%, and the prediction time and energy consumption were reduced 3.97-and 83.04-times when compared to those of the rate-SNN. Finally, we analyzed the sensitivity of the network performance due to unexpected issues that may occur in the hardware system and thus explained the competitiveness of the proposed synaptic behavior in the saturation region.

An artificial visual neuron with multiplexed rate and time-to-first-spike coding

“…In contrast, TTFS coding, which resembles natural vision, can provide important spatiotemporal information for the implementation of SNN to process dynamic visual data with high sparsity and high energy e ciency. Hardware-based SNN with TTFS coding are more e cient than SNN with rate coding, demonstrating faster speed and reduced energy consumption [23][24][25][26] . Although precise temporal encoding has been achieved in arti cial visual neuron systems, fusing rate and TTFS coding in a single spike train has not yet been realized in SNN hardware, compromising the capacity of such networks to rapidly and accurately process visual data in complex visual environments [27][28][29][30][31][32] .…”

On-Chip Adaptive Implementation of Neuromorphic Spiking Sensory Systems with Self-X Capabilities

“…As a result, the time resolution improves with CMOS scaling, leading to a growing interest in the time domain, as reported by [8,9]. This motivated the researchers to create electronic sensor systems that use spike or time-coded signals, which possess a technology-agnostic property that remains robust even as technology scales up, as demonstrated in [10][11][12][13][14][15]. A scalable ADC based on the neural engineering framework was proposed by the authors in [10].…”

“…The process of translating analog input values into input pulse frequencies through a large number of spikes can result in increased power consumption as the spiking neural network (SNN) structure grows deeper and larger. Consequently, these traits may not be well-suited for power-efficient and robust devices in edge computing [13,15].…”

Reliability‐aware design of temporal neuromorphic encoder for image recognition

“…TFS coding involves mapping input data into a single spike where the information is contained in the time taken to first spike. Hardware implementation of temporal encoding has focused only on the TFS approach in the past 12,15 due to its simplicity of circuit implementation. However, it can only represent single‐dimensional information 14 .…”