A tunable and versatile 28nm FD-SOI crossbar output circuit for low power analog SNN inference with eNVM synapses

Joao Henrique Quintino Palhares,Yann Beilliard,Jury Sandrini,Franck Arnaud,Kevin Garello,Guillaume Prenat,Lorena Anghel,Fabien Alibart,Dominique Drouin,Philippe Galy

2023-05-25

Abstract:In this work we report a study and a co-design methodology of an analog SNN crossbar output circuit designed in a 28nm FD-SOI technology node that comprises a tunable current attenuator and a leak-integrate and fire neurons that would enable the integration of emerging non-volatile memories (eNVMs) for synaptic arrays based on various technologies including phase change (PCRAM), oxide-based (OxRAM), spin transfer and spin orbit torque magnetic memories (STT, SOT-MRAM). Circuit SPICE simulation results and eNVM experimental data are used to showcase and estimate the neurons fan-in for each type of eNVM considering the technology constraints and design trade-offs that set its limits such as membrane capacitance and supply voltage, etc.

Systems and Control,Emerging Technologies

What problem does this paper attempt to address?

This paper aims to address the low - power consumption and large - scale integration challenges encountered when integrating emerging non - volatile memories (eNVMs) for analog spiking neural network (SNN) inference at the 28 - nanometer fully - depleted silicon - on - insulator (28nm FD - SOI) technology node. Specifically, the paper explores how to improve the cross - bar fan - in of SNN by designing a tunable current attenuator and Leaky Integrate - and - Fire (LIF) neuron circuit, thereby supporting the effective integration of different types of eNVMs such as phase - change random - access memory (PCRAM), oxide - based random - access memory (OxRAM), spin - transfer - torque magnetoresistive random - access memory (STT - MRAM) and spin - orbit - torque magnetoresistive random - access memory (SOT - MRAM). The key contribution of the paper lies in proposing a co - design method. This method not only takes into account the limiting factors in circuit design, such as membrane capacitance and power supply voltage, but also combines SPICE simulation results and eNVM experimental data to evaluate the neuron fan - in for each eNVM technology. In this way, the research team can set an appropriate current attenuation ratio (SDF) for different eNVM technologies, thereby optimizing the neuron performance in SNN hardware, especially improving its ability to handle a large number of synaptic connections. This research is of great significance for promoting the application of eNVM - based SNN in edge computing.

A tunable and versatile 28nm FD-SOI crossbar output circuit for low power analog SNN inference with eNVM synapses

1S1R Optimization for High‐Frequency Inference on Binarized Spiking Neural Networks

Comparative Evaluation of Memory Technologies for Synaptic Crossbar Arrays -- Part I: Robustness-driven Device-Circuit Co-Design and System Implications

Analog circuits for mixed-signal neuromorphic computing architectures in 28 nm FD-SOI technology

A CMOS Spiking Neuron for Brain-Inspired Neural Networks with Resistive Synapses and In-Situ Learning

Analog Spiking Neuron in CMOS 28 nm Towards Large-Scale Neuromorphic Processors

A 45nm CMOS Neuromorphic Chip with a Scalable Architecture for Learning in Networks of Spiking Neurons

A Novel CNFET SRAM-Based Compute-In-Memory for BNN Considering Chirality and Nanotubes

A Neuromorphic Spiking Neural Network Using Time-to-First-Spike Coding Scheme and Analog Computing in Low-Leakage 8T SRAM

A memristive nanoparticle/organic hybrid synapstor for neuro-inspired computing

RRAM-based Spiking Nonvolatile Computing-In-Memory Processing Engine with Precision-Configurable in Situ Nonlinear Activation

Analog ferroelectric domain-wall memories and synaptic devices integrated with Si substrates

HXNOR-PBNN: A Scalable and Parallel Spintronics Synaptic Architecture for Probabilistic Binary Neural Networks

Analog Spiking Neuron in 28 nm CMOS

Hybrid synaptic structure for spiking neural network realization

On-chip learning in a conventional silicon MOSFET based Analog Hardware Neural Network

Monolithically Integrated Complementary Ferroelectric FET XNOR Synapse for the Binary Neural Network

Ultra-Low Power Silicon Neuron Circuit for Extreme-Edge Neuromorphic Intelligence

NEXUS: A 28nm 3.3pJ/SOP 16-Core Spiking Neural Network with a Diamond Topology for Real-Time Data Processing

Design of Synaptic Driving Circuit for TFT eFlash-Based Processing-In-Memory Hardware Using Hybrid Bonding