Subthreshold neuromorphic devices for Spiking Neural Networks applied to embedded A.I

Loyez, Christophe; Carpentier, Kévin; Sourikopoulos, Ilias; Danneville, F.

doi:10.1109/newcas50681.2021.9462779

Cited by 12 publications

(10 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The layout of a neuron circuit designed in [3] and [4] confirms this mathematical behavior as illustrated in Fig. 2.…”

Section: A Nonlinear Behavior Of Spiking Neuronssupporting

confidence: 67%

“…In aspect of hardware, circuit simulation results show gains in energy consumption of at least two orders of magnitude compared with current solutions [3]. Studies on neuromorphic systems for spike processing have considered complementary metal-oxide semiconductor (CMOS) technology operation in the subthreshold region (100 mV and 1 nA) [4]. According to [5], while implemented on a field-programmable gate array (FPGA), the energy efficiency of an SNN reaches over 100 times higher than an ANN.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Dual-band Transmitter Linearization using Spiking Neural Networks

Wang

Ferreira

Benlarbi-Delaï

2022

2022 Asia-Pacific Microwave Conference (APMC)

View full text Add to dashboard Cite

In this paper, a novel way for dual-band power amplifiers (PA) linearization is proposed based on spiking neuron networks (SNN). A dual-input dual-output SNN is interacted with classical memory polynomial model, which can largely reduce the computational complexity. The experimental results on a real PA show that the proposed method can reach similar linearization performance compared with traditional methods but with low energy consumption. This is the first time that the SNN is deployed for multi-band PA linearization. Future work is to develop a trainable SNN model for real-time PA linearization.

show abstract

“…The layout of a neuron circuit designed in [3] and [4] confirms this mathematical behavior as illustrated in Fig. 2.…”

Section: A Nonlinear Behavior Of Spiking Neuronssupporting

confidence: 67%

Section: Introductionmentioning

confidence: 99%

Dual-band Transmitter Linearization using Spiking Neural Networks

Wang

Ferreira

Benlarbi-Delaï

2022

2022 Asia-Pacific Microwave Conference (APMC)

View full text Add to dashboard Cite

show abstract

“…This charge-sharing brings a drop in the voltage at the gate of the first transistor, which results in a small output current. 2) Neural Network: We construct neural networks with neurons designed in [9] and the complementary synapses designed in this paper as illustrated in Fig. 1.…”

Section: A Circuit-level Designmentioning

confidence: 99%

“…The spiking neural network (SNN) becomes a very promising solution since analog systems can be power efficient [6], [7]. In the literature [8], [9], the analog implementation of neuromorphic systems exhibit ultra-low power (ULP) consumption [10], [11], as well as excellent miniaturization perspectives. In this paper, we firstly design, both in schematics and layout, lowlevel circuit blocks (neurons and synapses) for SNN with ultralow power.…”

Section: Introductionmentioning

confidence: 99%

Smart IoT - Implementation of Neuromorphic Circuits

Artur¹,

Wang²,

Sulzbach³

et al. 2022

2022 IEEE 7th Forum on Research and Technologies for Society and Industry Innovation (RTSI)

View full text Add to dashboard Cite

This paper presents a simple analog neuromorphic system in microelectronics suitable for an audio source localization problem. The receptor can find the relative position of an audio source by the information of angles and distances according to acquire acoustic signals. In this paper, we focus on the angle detection, but some information of the distance is also presented. This paper also presents the development of some alternatives to the most important circuit blocks in neuromorphic systems: the neurons and the synapses. The results are validated in two different levels: the system level and the transistor level.

show abstract

“…While the production of ML code for embedded devices has been significantly facilitated over the years, it has retained a few challenges. Since edge devices remain constrained in terms of power, memory and computational resources (Loyez et al, 2021 ), the candidate models must be carefully chosen not only based on the type of input data, but also on the hardware's requirements. After acquiring a large amount of sample data through the same embedded sensors that will be used in the final application, the models undergo a training phase: this phase is usually executed on a server machine due to its high computational demand.…”

Section: Introductionmentioning

confidence: 99%

Spike encoding techniques for IoT time-varying signals benchmarked on a neuromorphic classification task

et al. 2022

View full text Add to dashboard Cite

Spiking Neural Networks (SNNs), known for their potential to enable low energy consumption and computational cost, can bring significant advantages to the realm of embedded machine learning for edge applications. However, input coming from standard digital sensors must be encoded into spike trains before it can be elaborated with neuromorphic computing technologies. We present here a detailed comparison of available spike encoding techniques for the translation of time-varying signals into the event-based signal domain, tested on two different datasets both acquired through commercially available digital devices: the Free Spoken Digit dataset (FSD), consisting of 8-kHz audio files, and the WISDM dataset, composed of 20-Hz recordings of human activity through mobile and wearable inertial sensors. We propose a complete pipeline to benchmark these encoding techniques by performing time-dependent signal classification through a Spiking Convolutional Neural Network (sCNN), including a signal preprocessing step consisting of a bank of filters inspired by the human cochlea, feature extraction by production of a sonogram, transfer learning via an equivalent ANN, and model compression schemes aimed at resource optimization. The resulting performance comparison and analysis provides a powerful practical tool, empowering developers to select the most suitable coding method based on the type of data and the desired processing algorithms, and further expands the applicability of neuromorphic computational paradigms to embedded sensor systems widely employed in the IoT and industrial domains.

show abstract

Subthreshold neuromorphic devices for Spiking Neural Networks applied to embedded A.I

Cited by 12 publications

References 7 publications

Dual-band Transmitter Linearization using Spiking Neural Networks

Dual-band Transmitter Linearization using Spiking Neural Networks

Smart IoT - Implementation of Neuromorphic Circuits

Spike encoding techniques for IoT time-varying signals benchmarked on a neuromorphic classification task

Contact Info

Product

Resources

About