As Richard Kingston, vice president of market information at CEVA, states, human society is entering the fourth industrial revolution, just as the previous three industrial revolutions (mechanization, mass production and electrification, computers and automation), the fourth industrial revolution Change people's lives again.
Humankind has landed on the earth. The science fiction that once used to be an ancients has now come true. Today we are faced with the scenes of automatic driving and human-computer conversation. For the future, we may as well conceive of it.
The transformation of forms of knowledge and information is the most important aspect of how the fourth industrial revolution changed the future, in the past, knowledge was written on paper, stored in universities and libraries, and processed and utilized by people. Today, knowledge is stored in In modern data centers, a variety of machinery and equipment analyze and apply them, Kingston said: 'This is a tremendous shift in the digitalization process and the world is changing.'
From financial services, healthcare to retail, home, and many of these industries and applications, cloud computing-based information and data are everywhere, and questions are coming in and who should handle that information - in the cloud or with artificial intelligence In fact, these two approaches will coexist. "Kingston added: 'The terminals in the future will autonomously process data and make decisions, and more importantly, they will continue to learn and feed back the results to the cloud's knowledge base, The Next Generation Network is being launched for specific end markets, and some of the data I've seen lately indicate that about 50% of AI devices will have built-in machine learning capabilities by 2022 - which can be a significant amount if the data is correct.
Deploying artificial intelligence on the far side can help to solve the following problems:
Latency: This is especially important in safety-oriented scenarios
Privacy: Devices should not send private data to the cloud
Security: Cloud data is more vulnerable to hacking
Network coverage: There is no network connection to the cloud
cost problem
Kingston goes on to say: 'If you rely on connecting to the cloud to do a lot of AI processing, there are a number of issues that are going to happen, and in the automotive industry, dealing with latency is a big deal and you do not want to put private data Stored in the cloud. '
Front-end AI has a large number of application scenarios, such as:
Intelligent monitoring
face detection
Voice biometrics
Sound detection
Motion sensing
Connectivity (Connectivity)
vehicle electronics
visual sensor
Communication
Radar and laser radar
GPS
Connectivity features - Bluetooth, WiFi, cellular network
Data fusion and data analysis
Since front-end AI was added to smartphones, neural network processors and hardware accelerators have become mainstream technologies and AI far-end processing has taken a major step forward in the industry. Both Apple and Huawei have introduced in their end products Special neural network processor for face recognition, this all completed by the terminal program, the advantages of security, privacy and direct.
Qualcomm and NVIDIA have also released neural network processors for smartphones and other mobile devices CEVA predicts that in a few years each of the camera-equipped devices will have built-in artificial intelligence for vision and neural network processors. One-third of smartphones will support artificial intelligence.
In addition, data processing performance and power efficiency are expected to increase tenfold over the same period, which is crucial because smartphones have a much faster need for data processing performance than battery capacity. "Kingston continues: 'Battery capacity There's a big gap between the development and the (data) processing power upgrade, and it will only get worse.When you need to add a neural network processor to these devices and run AI functions in real time, you'll have to If your battery technology continues to grow at a constant rate, your device will not be able to stand still for a long time, so you'll have to deal with both the processing power and the battery technology.
CEVA program
The front-end AI covers four main sections - communications, sensor processing, data fusion and data analysis, as well as applications / implementations. "CEVA offers solutions for the first three parts.
"At a glance, we focus on three challenges facing the front-end AI market-power consumption, price, and soaring performance requirements." CEVA achieves this by technology empowering devices from end-to-end base stations, Including sound equipment, connected equipment and wireless network equipment.
In any case, the current focus is on the vision system, including computer vision DSPs, neural networks, accelerators for different applications ranging from cell phones to cars, and neural network frameworks.
The number of global cameras is expected to grow by 216% from 2016 to 2022. There will be about 44 billion camera-enabled devices this new vertical trend is due to a growing number of different devices The integrated camera has started (see below).
CEVA also recently announced an alliance with LG to develop smart 3D cameras that will use the multi-core CEVA-XM4 vision DSP while LG effectively reduces costs by deploying self-researching algorithms on existing commercial chips. CEVA also concluded with startup Brodmann 17 Partnership, the company develops deep learning software on embedded devices that use the CEVA-XM processor to meet processing accuracy requirements at up to 100 frames per second, which is faster than NVIDIA's Jetson X2 platform Almost 170% faster
Kingston concludes: 'We will see more forms of AI devices coming to market, including smartphones, drones, ADAS and surveillance equipment. CEVA is working on the next generation of computer vision and neural network product development, specifically The potential of neural network technology, and how to establish the most effective network training system in the data center.
'We hope to continue investing in the artificial intelligence ecosystem based on current products and hope to purchase some additional technologies to enrich our products so that our customers can get to market sooner than ever before.CEVA is no longer a DSP- Nuclear communications company, and now we provide a wider range of different areas of technology, let us wait and see together the development of the next few years!