Hardware Implementation of Differential Oscillatory Neural Networks Using VO 2-Based Oscillators and Memristor-Bridge Circuits

Shamsi, Jafar; Avedillo, M.J.; Linares-Barranco, B.; Serrano-Gotarredona, Teresa

doi:10.3389/fnins.2021.674567

Cited by 25 publications

(17 citation statements)

References 44 publications

(57 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Coupled oscillator networks have been successfully exploited in the field of pattern recognition (Kantner et al, 2015;Hölzel and Krischer, 2011). However, larger systems have not yet been successfully developed due to noise-stability problems and device constraints in the new class of compact oscillators based on VO 2 or NbO x , for example (Shamsi et al, 2021;D. Lee et al, 2018).…”

Section: Advanced Computing Architecturesmentioning

confidence: 99%

Matter & Mind Matter

Birkoben¹,

Kohlstedt²

2022

Preprint

View full text Add to dashboard Cite

As a result of a hundred million years of evolution, living animals have adapted extremely well to their ecological niche. Such adaptation implies species-specific interactions with their immediate environment by processing sensory cues and responding with appropriate behavior. Understanding how living creatures perform pattern recognition and cognitive tasks is of particular importance for computing architectures: by studying these information pathways refined over eons of evolution, researchers may be able to streamline the process of developing more highly advanced, energy efficient autonomous systems. With the advent of novel electronic and ionic components along with a deeper understanding of information pathways in living species, a plethora of opportunities to develop completely novel information processing avenues are within reach. Here, we describe the basal information pathways in nervous systems, from the local neuron level to the entire nervous system network. The dual importance of local learning rules is addressed, from spike timing dependent plasticity at the neuron level to the interwoven morphological and dynamical mechanisms of the global network. Basal biological principles are highlighted, including phylogenies, ontogenesis, and homeostasis, with particular emphasis on network topology and dynamics. While in machine learning system training is performed on virgin networks without any a priori knowledge, the approach proposed here distinguishes itself unambiguously by employing growth mechanisms as a guideline to design novel computing architectures. Including fundamental biological information pathways that explore the spatiotemporal fundamentals of nervous systems has untapped potential for the development of entirely novel information processing systems. Finally, a benchmark for neuromorphic systems is suggested.

show abstract

Section: Advanced Computing Architecturesmentioning

confidence: 99%

Matter & Mind Matter

Birkoben¹,

Kohlstedt²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…An example of solving the vertex coloring optimization problem, using a computational engine based on memristor ONN, is presented in [ 25 ]. ONN can be successfully applied to solving emerging problems in the field of pattern recognition, image and audio signal processing, and robot control, as well as for the development of neuromorphic systems [ 26 , 27 , 28 , 29 ].…”

Section: Introductionmentioning

confidence: 99%

The Coupled Reactance-Less Memristor Based Relaxation Oscillators for Binary Oscillator Networks

2023

View full text Add to dashboard Cite

This paper discusses the application of coupled reactance-less memristor-based oscillators (MBO) with binary output signals in oscillatory networks. A class of binary-coupled memristor oscillators provides simple integration with standard CMOS logic elements. Сombining MBOs with binary logic elements ensures the operation of complex information processing algorithms. The analysis of the simplest networks based on MBOs is performed. The typical reactance-less MBO with current and potential inputs is considered. The output responses for input control signals are analyzed. It is shown that the current input signal impacts primarily the rate of memristor resistance variation, while the potential input signal changes the thresholds. The exploit of the potential input for the synchronization of coupled MBOs and current control input in order to provide the necessary encoding of information is suggested. The example of the application of coupled MBOs in oscillatory networks is given, and results of simulation are presented.

show abstract

“…As a coupling component, a memristor can be used to interconnect the oscillators. Differential ONNs (DONNs) have recently been introduced that use differential oscillators, each providing one in-phase and one anti-phase signal [5]. Differential outputs allow a memristor-bridge synaptic circuit to be used to implement positive, negative, or zero weights [6].…”

Section: Introductionmentioning

confidence: 99%

Effect of Device Mismatches in Differential Oscillatory Neural Networks

Shamsi

Avedillo

Linares-Barranco

et al. 2023

IEEE Trans. Circuits Syst. I

Self Cite

View full text Add to dashboard Cite

Analog implementation of Oscillatory Neural Networks (ONNs) has the potential to implement fast and ultralow-power computing capabilities. One of the drawbacks of analog implementation is component mismatches which cause desynchronization and instability in ONNs. Emerging devices like memristors and VO 2 are particularly prone to variations.In this paper, we study the effect of component mismatches on the performance of differential ONNs (DONNs). Mismatches were considered in two main blocks: differential oscillatory neurons and synaptic circuits. To measure DONN tolerance to mismatches in each block, performance was evaluated with mismatches being present separately in each block. Memristor-bridge circuits with four memristors were used as the synaptic circuits. The differential oscillatory neurons were based on VO 2 -devices. The simulation results showed that DONN performance was more vulnerable to mismatches in the components of the differential oscillatory neurons than to mismatches in the synaptic circuits. DONNs were found to tolerate up to 20% mismatches in the memristance of the synaptic circuits. However, mismatches in the differential oscillatory neurons resulted in non-uniformity of the natural frequencies, causing desynchronization and instability. Simulations showed that 0.5% relative standard deviation (RSD) in natural frequencies can reduce DONN performance dramatically. In addition, sensitivity analyses showed that the high threshold voltage of VO2-devices is the most sensitive parameter for frequency non-uniformity and desynchronization.

show abstract

Hardware Implementation of Differential Oscillatory Neural Networks Using VO 2-Based Oscillators and Memristor-Bridge Circuits

Cited by 25 publications

References 44 publications

Matter & Mind Matter

Matter & Mind Matter

The Coupled Reactance-Less Memristor Based Relaxation Oscillators for Binary Oscillator Networks

Effect of Device Mismatches in Differential Oscillatory Neural Networks

Contact Info

Product

Resources

About