Neuron Circuits for Low-Power Spiking Neural Networks Using Time-To-First-Spike Encoding

Abstract

Hardware-based Spiking Neural Networks (SNNs) are regarded as promising candidates for the cognitive computing system due to its low power consumption and highly parallel operation. In this paper, we train the SNN in which the firing time carries information using temporal backpropagation. The temporally encoded SNN with 512 hidden neurons achieved an accuracy of 96.90% for the MNIST test set. Furthermore, the effect of the device variation on the accuracy in temporally encoded SNN is investigated and compared with that of the rate-encoded network. In a hardware configuration of our SNN, NOR-type analog memory having an asymmetric floating gate is used as a synaptic device. In addition, we propose a neuron circuit including a refractory period generator for temporally encoded SNN. The performance of the 2-layer neural network composed of synapses and proposed neurons is evaluated through circuit simulation using SPICE based on the BSIM3v3 model with 0.35 μm technology. The network with 128 hidden neurons achieved an accuracy of 94.9%, a 0.1% reduction compared to that of the system simulation of the MNIST dataset. Finally, each block's latency and power consumption constituting the temporal network is analyzed and compared with those of the rate-encoded network depending on the total time step. Assuming that the network has 256 total time steps, the temporal network consumes 15.12 times less power than the rateencoded network and makes decisions 5.68 times faster.

A MoS2 Hafnium Oxide Based Ferroelectric Encoder for Temporal‐Efficient Spiking Neural Network

“…Although incorporating an additional comparator could complicate the circuit operation of the TTFS approach compared to the conventional rate encoding method, this technique is highly desirable for temporal-efficient information processing as it only looks for a single spike regardless of the input intensity. [18,19,27] Figure 4e shows the extracted spike time from various input intensities with P th from 0.005 to 0.5. The exponential correlation between the spike time and input signal can be distinctly identified, in which the generated spike times are inversely proportional to the input intensities.…”

“…Our ferroelectric encoder, synergetically exploiting the robust ferroelectric response and TTFS approach, significantly improves the encoding efficiency compared to other hardware encoders using the conventional rate encoding approach. [15][16][17][18][19][20] Although the rate encoding has been widely employed, redundant timesteps (e.g., >200 timesteps) are necessary to accurately observe the input intensity, which reduces the encoding efficiency. [20] Furthermore, the precision of our ferroelectric encoder is quantitatively assessed by the standard MNIST dataset using a simulated SNN to perform the classification task of handwritten digits.…”

“…Although hardware encoders, including complementary metal-oxide-semiconductor (CMOS), memristors, and transistors, have been reported, the challenges remain as the encoding procedure is either sophisticated (e.g., digital circuit implementation) or power inefficient (e.g., low throughput). [15][16][17][18][19][20] Thus, this imposes stringent criteria on the fundamental level where innovative material solutions are practically indispensable. The hafnium oxide-based ferroelectric material system offers exceptional opportunities from the aforementioned aspects.…”

Analyzing time-to-first-spike coding schemes: A theoretical approach

“…Historically, TTFS was first proposed to explain the phenomenal speed of processing in the brain for certain tasks, such as object recognition (Thorpe and Imbert, 1989 ). More recently, TTFS has attracted much attention from the AI community (Mostafa, 2017 ; Rueckauer and Liu, 2018 ; Zhou et al, 2019 ; Kheradpisheh and Masquelier, 2020 ; Park et al, 2020 ; Sakemi et al, 2020 ; Zhang et al, 2020 ; Comsa et al, 2021 ; Mirsadeghi et al, 2021 ), because it can be efficiently implemented on low power event-driven neuromorphic chips (Abderrahmane et al, 2020 ; Nair et al, 2020 ; Srivatsa et al, 2020 ; Göltz et al, 2021 ; Liang et al, 2021 ; Oh et al, 2022 ), leveraging two key features. The first one is sparsity (Frenkel, 2021 ).…”

Spiking-GAN: A Spiking Generative Adversarial Network Using Time-To-First-Spike Coding

“…It significantly increases the sparsity of the output spike train, thereby giving large energy savings. It has recently been used in SNNs for classification problems [12][13][14] and hardware implementation of such schemes is also being explored [15]. There is also building evidence of the biological plausibility of such temporal coding schemes [16,17].…”