China
Africa
Explore chapters and articles related to this topic
Securing Future Autonomous Applications Using Cyber-Physical Systems and the Internet of Things
Published in Amit Kumar Tyagi, Niladhuri Sreenath, Handbook of Research of Internet of Things and Cyber-Physical Systems, 2022
S. Sobana, S. Krishna Prabha, T. Seerangurayar, S. Sudha
In autonomous industries computing has radically changed nearly every aspect, from business and agriculture to communication and entertainment. For perfect operation of any system, we relay computing in the design of system for energy, defense, and transportation. Computing innovations includes faster algorithms, statistical models, programming abstractions, high performance networks [85]. Autonomous systems today have to process massive amount of data to construct an AI, ML, and make use of decision-making techniques like deep learning. The computing power thus required to implement these technologies are higher than that needed for previous technologies. Usually, the data science and ML tasks are more resource concentrated and implemented with the help of using large number of CPU and graphic processing unit (GPUs). The solution to this problem is utilizing cloud services such as Google Cloud, Microsoft Azure, and Amazon AWS with an expense of high cost. The computation cost can be reduced by the use of crypto mining farms and exchange of computing power only between participants in the same community. The energy efficiency of dark silicon used for the construction of CPUs and GPUs can be improved by utilizing the accelerators in an efficient way. The accelerators can be connected with the single-instruction-multiple data (SIMD) hardware, can be placed on the conventional GPUs, and attaching them with direct-memory-access (DMA) engines [85].
On-Chip Regulators for Low-Voltage and Portable Systems-on-Chip
Published in Fei Yuan, Krzysztof Iniewski, Low-Power Circuits for Emerging Applications in Communications, Computing, and Sensing, 2018
Near-threshold computing has received significant attention due to enhanced energy efficiency, particularly for mobile systems-on-a-chip (SoCs) [36]. Highly parallelized architectures based on near-threshold operation have been proposed as a possible solution to dark silicon [37]. Developing an integrated voltage regulator module with application to near-threshold operation is challenging due to low output voltages in the range of 0.5 V. The regulator should simultaneously satisfy high power efficiency and power density (to minimize area overhead). Furthermore, the output ripple should be minimized since near-threshold circuits are highly sensitive to power supply variations (due to near-exponential dependence).
Processor Physics and Moore’s Law
Published in Vivek Kale, Parallel Computing Architectures and APIs, 2019
The catalyst for the proliferation of heterogeneous architectures is a possible development in the near future, called dark silicon. Given a fixed power budget, only a small portion of the chip will be powered up at any point in time. This, however, allows unprecedented flexibility for specialization, because a large number of accelerators can be built on the same chip to be woken up only when needed.
Hybrid buffers based coarse-grained power gated network on chip router microarchitecture
Published in International Journal of Electronics, 2020
Yogendra Gupta, Lava Bhargava, Ashish Sharma, M.S. Gaur
For several years, IC manufacturers have focused on Moore’s law, which says doubling the number of transistors (increasing the transistor density within the same die) in every eighteen months with the fixed cost. Dennard’s Scaling law states that as transistors get smaller and consequently channel length of the transistor scale down. Both, voltage and current scale with the transistor channel length, hence their power density remains constant, and by this, the power dissipation stays in proportion with the area. In the dark silicon era where all the NoC components cannot be powered on simultaneously to avoid overheating and burning the chip. Dennard’s Scaling establishes that scaling down the transistor dimensions allows operating them at a lower voltage and lower power, keeping power density constant ‘(Esmaeilzadeh, Blem, St. Amant, Sankaralingam, & Burger, 2011)’. Dennard’s scaling has failed as voltage could not be scaled down in keeping with the transistor length, this leads to increase the power density, by it more heat is generated per unit area and causes hotspot and thermal/reliability issues. To avoid hotspot a certain part of the integrated circuit remains inactive, the inactive part is termed as ‘Dark Silicon’ ‘(Rahmani, Pasi Liljeberg, Hemani, & Tenhunen, 2016)’. Figure 1 depicts 4 × 4 network on chip based multi-core architecture. Each Tile consists of a processing element which is attached to the router through a network interface. Routers interconnected through links to another router to form an NoC.