scholarly journals Hardware-Efficient Stochastic Binary CNN Architectures for Near-Sensor Computing

2022 ◽  
Vol 15 ◽  
Author(s):  
Vivek Parmar ◽  
Bogdan Penkovsky ◽  
Damien Querlioz ◽  
Manan Suri
Keyword(s):  
Artificial Intelligence ◽  
High Precision ◽  
Energy Efficient ◽  
Pathogen Detection ◽  
Proof Of Concept ◽  
Stochastic Sampling ◽  
Stochastic Computing ◽  
Microscopy Image ◽  
With Memory

With recent advances in the field of artificial intelligence (AI) such as binarized neural networks (BNNs), a wide variety of vision applications with energy-optimized implementations have become possible at the edge. Such networks have the first layer implemented with high precision, which poses a challenge in deploying a uniform hardware mapping for the network implementation. Stochastic computing can allow conversion of such high-precision computations to a sequence of binarized operations while maintaining equivalent accuracy. In this work, we propose a fully binarized hardware-friendly computation engine based on stochastic computing as a proof of concept for vision applications involving multi-channel inputs. Stochastic sampling is performed by sampling from a non-uniform (normal) distribution based on analog hardware sources. We first validate the benefits of the proposed pipeline on the CIFAR-10 dataset. To further demonstrate its application for real-world scenarios, we present a case-study of microscopy image diagnostics for pathogen detection. We then evaluate benefits of implementing such a pipeline using OxRAM-based circuits for stochastic sampling as well as in-memory computing-based binarized multiplication. The proposed implementation is about 1,000 times more energy efficient compared to conventional floating-precision-based digital implementations, with memory savings of a factor of 45.

scholarly journals Energy-efficient memcapacitor devices for neuromorphic computing

2021 ◽  
Author(s):  
Kai-Uwe Demasius ◽  
Aron Kirschen ◽  
Stuart Parkin
Keyword(s):  
Neural Network ◽  
Artificial Intelligence ◽  
Neural Networks ◽  
Energy Efficiency ◽  
High Precision ◽  
Energy Efficient ◽  
Lateral Size ◽  
Data Intensive Computing ◽  
Data Intensive ◽  
Specialized Hardware

AbstractData-intensive computing operations, such as training neural networks, are essential for applications in artificial intelligence but are energy intensive. One solution is to develop specialized hardware onto which neural networks can be directly mapped, and arrays of memristive devices can, for example, be trained to enable parallel multiply–accumulate operations. Here we show that memcapacitive devices that exploit the principle of charge shielding can offer a highly energy-efficient approach for implementing parallel multiply–accumulate operations. We fabricate a crossbar array of 156 microscale memcapacitor devices and use it to train a neural network that could distinguish the letters ‘M’, ‘P’ and ‘I’. Modelling these arrays suggests that this approach could offer an energy efficiency of 29,600 tera-operations per second per watt, while ensuring high precision (6–8 bits). Simulations also show that the devices could potentially be scaled down to a lateral size of around 45 nm.

An Energy Efficient Multi-Rail Architecture for Stochastic Computing: A Bayesian Sensor Fusion Case Study

2021 ◽  
Author(s):  
Jeremy Belot ◽  
Abdelkarim Cherkaoui ◽  
Raphael Laurent ◽  
Laurent Fesquet
Keyword(s):  
Sensor Fusion ◽  
Energy Efficient ◽  
Stochastic Computing

Stochastic Computing via In Operando Modulation of Rectification in Molecular Tunneling Junctions

2020 ◽  
Author(s):  
Xinkai Qiu ◽  
Sylvia Rousseva ◽  
Gang Ye ◽  
Jan C. Hummelen ◽  
Ryan Chiechi
Keyword(s):  
Self Assembly ◽  
Variable Temperature ◽  
Self Assembled Monolayers ◽  
Microfluidic Channels ◽  
Proof Of Concept ◽  
Glycol Ethers ◽  
Stochastic Computing ◽  
Assembled Monolayers ◽  
In Operando ◽  
Tunneling Junctions

This paper describes the reconfiguration of molecular tunneling junctions during operation via the self-assembly of bilayers of glycol ethers. We use well-established functional groups to modulate the magnitude and direction of rectification in assembled tunneling junctions by exposing them to solutions containing different glycol ethers. Variable-temperature measurements establish that rectification occurs by a bias-dependent tunneling-hopping mechanism and that glycol ethers, beside being an unusually efficient tunneling medium, behave identically to alkanes. We fabricated memory bits from crossbar junctions prepared by injecting eutectic Ga-In into microfluidic channels. Two 8-bit registers were able to perform logical AND operations on bit strings encoded into chemical packets as microfluidic droplets that alter the composition of the crossbar junctions through self-assembly to effect memristor-like properties. This proof of concept work demonstrates the potential for fieldable molecular-electronic devices based on tunneling junctions of self-assembled monolayers and bilayers.

scholarly journals On the Investigation of Energy Efficient Torque Distribution Strategies through a Comprehensive Powertrain Model

2021 ◽  
Vol 13 (8) ◽  
pp. 4549
Author(s):  
Sara Salamone ◽  
Basilio Lenzo ◽  
Giovanni Lutzemberger ◽  
Francesco Bucchi ◽  
Luca Sani
Keyword(s):  
Energy Storage ◽  
Electric Vehicles ◽  
Energy Efficient ◽  
Storage System ◽  
Energy Storage System ◽  
Torque Distribution ◽  
Driving Cycles ◽  
Energy Models ◽  
Battery State Of Charge

In electric vehicles with multiple motors, the torque at each wheel can be controlled independently, offering significant opportunities for enhancing vehicle dynamics behaviour and system efficiency. This paper investigates energy efficient torque distribution strategies for improving the operational efficiency of electric vehicles with multiple motors. The proposed strategies are based on the minimisation of power losses, considering the powertrain efficiency characteristics, and are easily implementable in real-time. A longitudinal dynamics vehicle model is developed in Simulink/Simscape environment, including energy models for the electrical machines, the converter, and the energy storage system. The energy efficient torque distribution strategies are compared with simple distribution schemes under different standardised driving cycles. The effect of the different strategies on the powertrain elements, such as the electric machine and the energy storage system, are analysed. Simulation results show that the optimal torque distribution strategies provide a reduction in energy consumption of up to 5.5% for the case-study vehicle compared to simple distribution strategies, also benefiting the battery state of charge.

Sign in / Sign up

Export Citation Format

Share Document