scholarly journals Reconfigurable Architecture and Dataflow for Memory Traffic Minimization of CNNs Computation

Micromachines ◽  
2021 ◽  
Vol 12 (11) ◽  
pp. 1365
Author(s):  
Wei-Kai Cheng ◽  
Xiang-Yi Liu ◽  
Hsin-Tzu Wu ◽  
Hsin-Yi Pai ◽  
Po-Yao Chung
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
Data Reuse ◽  
Memory Access ◽  
Data Migration ◽  
Layer By Layer ◽  
Exploration Process ◽  
Array Architecture ◽  
Reuse Methodology ◽  
On Chip

Computation of convolutional neural network (CNN) requires a significant amount of memory access, which leads to lots of energy consumption. As the increase of neural network scale, this phenomenon is further obvious, the energy consumption of memory access and data migration between on-chip buffer and off-chip DRAM is even much more than the computation energy on processing element array (PE array). In order to reduce the energy consumption of memory access, a better dataflow to maximize data reuse and minimize data migration between on-chip buffer and external DRAM is important. Especially, the dimension of input feature map (ifmap) and filter weight are much different for each layer of the neural network. Hardware resources may not be effectively utilized if the array architecture and dataflow cannot be reconfigured layer by layer according to their ifmap dimension and filter dimension, and result in a large quantity of data migration on certain layers. However, a thorough exploration of all possible configurations is time consuming and meaningless. In this paper, we propose a quick and efficient methodology to adapt the configuration of PE array architecture, buffer assignment, dataflow and reuse methodology layer by layer with the given CNN architecture and hardware resource. In addition, we make an exploration on the different combinations of configuration issues to investigate their effectiveness and can be used as a guide to speed up the thorough exploration process.

МЕТОДЫ ДОСТИЖЕНИЯ МАКСИМАЛЬНОЙ ЭФФЕКТИВНОСТИ ПЛАТФОРМЫ ПРОТОТИПИРОВАНИЯ ВЫСОКОПРОИЗВОДИТЕЛЬНЫХ СИСТЕМ НА КРИСТАЛЛЕ НА ЗАДАЧАХ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА

2020 ◽  
Vol 96 (3s) ◽  
pp. 585-588
Author(s):  
С.Е. Фролова ◽  
Е.С. Янакова
Keyword(s):  
Neural Network ◽  
Artificial Intelligence ◽  
Computer Vision ◽  
High Performance ◽  
Systems On Chip ◽  
High Performance Systems ◽  
On Chip ◽  
Network Technologies ◽  
Neural Network Technologies

Предлагаются методы построения платформ прототипирования высокопроизводительных систем на кристалле для задач искусственного интеллекта. Изложены требования к платформам подобного класса и принципы изменения проекта СнК для имплементации в прототип. Рассматриваются методы отладки проектов на платформе прототипирования. Приведены результаты работ алгоритмов компьютерного зрения с использованием нейросетевых технологий на FPGA-прототипе семантических ядер ELcore. Methods have been proposed for building prototyping platforms for high-performance systems-on-chip for artificial intelligence tasks. The requirements for platforms of this class and the principles for changing the design of the SoC for implementation in the prototype have been described as well as methods of debugging projects on the prototyping platform. The results of the work of computer vision algorithms using neural network technologies on the FPGA prototype of the ELcore semantic cores have been presented.

scholarly journals Graphene-based 3D XNOR-VRRAM with ternary precision for neuromorphic computing

2021 ◽  
Vol 5 (1) ◽  
Author(s):  
Batyrbek Alimkhanuly ◽  
Joon Sohn ◽  
Ik-Joon Chang ◽  
Seunghyun Lee
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
Recognition Accuracy ◽  
Material Selection ◽  
Weighted Sum ◽  
Device Design ◽  
Key Factors ◽  
Neuromorphic Computing ◽  
Device Scaling ◽  
The Impact

AbstractRecent studies on neural network quantization have demonstrated a beneficial compromise between accuracy, computation rate, and architecture size. Implementing a 3D Vertical RRAM (VRRAM) array accompanied by device scaling may further improve such networks’ density and energy consumption. Individual device design, optimized interconnects, and careful material selection are key factors determining the overall computation performance. In this work, the impact of replacing conventional devices with microfabricated, graphene-based VRRAM is investigated for circuit and algorithmic levels. By exploiting a sub-nm thin 2D material, the VRRAM array demonstrates an improved read/write margins and read inaccuracy level for the weighted-sum procedure. Moreover, energy consumption is significantly reduced in array programming operations. Finally, an XNOR logic-inspired architecture designed to integrate 1-bit ternary precision synaptic weights into graphene-based VRRAM is introduced. Simulations on VRRAM with metal and graphene word-planes demonstrate 83.5 and 94.1% recognition accuracy, respectively, denoting the importance of material innovation in neuromorphic computing.

Implementation of an On-Chip Learning Neural Network IC Using Highly Linear Charge Trap Device

2021 ◽  
pp. 1-13
Author(s):  
Jong-Moon Choi ◽  
Do-Wan Kwon ◽  
Je-Joong Woo ◽  
Eun-Je Park ◽  
Kee-Won Kwon
Keyword(s):  
Neural Network ◽  
Linear Charge ◽  
Charge Trap ◽  
On Chip

scholarly journals Freely scalable and reconfigurable optical hardware for deep learning

2021 ◽  
Vol 11 (1) ◽  
Author(s):  
Liane Bernstein ◽  
Alexander Sludds ◽  
Ryan Hamerly ◽  
Vivienne Sze ◽  
Joel Emer ◽  
...  
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
Data Transfer ◽  
Power Delivery ◽  
Optical Data ◽  
Optical Multicast ◽  
Optical Neural Network ◽  
Fully Connected ◽  
Management Power

AbstractAs deep neural network (DNN) models grow ever-larger, they can achieve higher accuracy and solve more complex problems. This trend has been enabled by an increase in available compute power; however, efforts to continue to scale electronic processors are impeded by the costs of communication, thermal management, power delivery and clocking. To improve scalability, we propose a digital optical neural network (DONN) with intralayer optical interconnects and reconfigurable input values. The path-length-independence of optical energy consumption enables information locality between a transmitter and a large number of arbitrarily arranged receivers, which allows greater flexibility in architecture design to circumvent scaling limitations. In a proof-of-concept experiment, we demonstrate optical multicast in the classification of 500 MNIST images with a 3-layer, fully-connected network. We also analyze the energy consumption of the DONN and find that digital optical data transfer is beneficial over electronics when the spacing of computational units is on the order of $$>10\,\upmu $$ > 10 μ m.

scholarly journals Machine Learning-Based Energy System Model for Tissue Paper Machines

Processes ◽  
2021 ◽  
Vol 9 (4) ◽  
pp. 655
Author(s):  
Huanhuan Zhang ◽  
Jigeng Li ◽  
Mengna Hong
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
Gradient Boosting ◽  
Paper Machine ◽  
Steam Consumption ◽  
Tissue Paper ◽  
Paper Machines ◽  
Energy Consumption Model ◽  
Extreme Gradient Boosting ◽  
Consumption Model

With the global energy crisis and environmental pollution intensifying, tissue papermaking enterprises urgently need to save energy. The energy consumption model is essential for the energy saving of tissue paper machines. The energy consumption of tissue paper machine is very complicated, and the workload and difficulty of using the mechanism model to establish the energy consumption model of tissue paper machine are very large. Therefore, this article aims to build an empirical energy consumption model for tissue paper machines. The energy consumption of this model includes electricity consumption and steam consumption. Since the process parameters have a great influence on the energy consumption of the tissue paper machines, this study uses three methods: linear regression, artificial neural network and extreme gradient boosting tree to establish the relationship between process parameters and power consumption, and process parameters and steam consumption. Then, the best power consumption model and the best steam consumption model are selected from the models established by linear regression, artificial neural network and the extreme gradient boosting tree. Further, they are combined into the energy consumption model of the tissue paper machine. Finally, the models established by the three methods are evaluated. The experimental results show that using the empirical model for tissue paper machine energy consumption modeling is feasible. The result also indicates that the power consumption model and steam consumption model established by the extreme gradient boosting tree are better than the models established by linear regression and artificial neural network. The experimental results show that the power consumption model and steam consumption model established by the extreme gradient boosting tree are better than the models established by linear regression and artificial neural network. The mean absolute percentage error of the electricity consumption model and the steam consumption model built by the extreme gradient boosting tree is approximately 2.72 and 1.87, respectively. The root mean square errors of these two models are about 4.74 and 0.03, respectively. The result also indicates that using the empirical model for tissue paper machine energy consumption modeling is feasible, and the extreme gradient boosting tree is an efficient method for modeling energy consumption of tissue paper machines.

scholarly journals Predicting the Energy Consumption of a Robot in an Exploration Task Using Optimized Neural Networks

Electronics ◽  
2021 ◽  
Vol 10 (8) ◽  
pp. 920
Author(s):  
Liesle Caballero ◽  
Álvaro Perafan ◽  
Martha Rinaldy ◽  
Winston Percybrooks
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
Mobile Robot ◽  
Energy Budget ◽  
Dynamic Models ◽  
Pearson Correlation ◽  
Experimental Conditions ◽  
Grid Map ◽  
Proposed Model ◽  
Exploration Task

This paper deals with the problem of determining a useful energy budget for a mobile robot in a given environment without having to carry out experimental measures for every possible exploration task. The proposed solution uses machine learning models trained on a subset of possible exploration tasks but able to make predictions on untested scenarios. Additionally, the proposed model does not use any kinematic or dynamic models of the robot, which are not always available. The method is based on a neural network with hyperparameter optimization to improve performance. Tabu List optimization strategy is used to determine the hyperparameter values (number of layers and number of neurons per layer) that minimize the percentage relative absolute error (%RAE) while maximize the Pearson correlation coefficient (R) between predicted data and actual data measured under a number of experimental conditions. Once the optimized artificial neural network is trained, it can be used to predict the performance of an exploration algorithm on arbitrary variations of a grid map scenario. Based on such prediction, it is possible to know the energy needed for the robot to complete the exploration task. A total of 128 tests were carried out using a robot executing two exploration algorithms in a grid map with the objective of locating a target whose location is not known a priori by the robot. The experimental energy consumption was measured and compared with the prediction of our model. A success rate of 96.093% was obtained, measured as the percentage of tests where the energy budget suggested by the model was enough to actually carry out the task when compared to the actual energy consumed in the test, suggesting that the proposed model could be useful for energy budgeting in actual mobile robot applications.

Memory Access Optimization for On-Chip Transfer Learning

2021 ◽  
Vol 68 (4) ◽  
pp. 1507-1519
Author(s):  
Muhammad Awais Hussain ◽  
Tsung-Han Tsai
Keyword(s):  
Transfer Learning ◽  
Memory Access ◽  
On Chip
Sign in / Sign up

Export Citation Format

Share Document