A few days ago, at the 2022 World Artificial Intelligence Conference (WAIC), Lars Reger, Global Executive Vice President and CTO of NXP, delivered a speech at the theme forum of “AI Future, Sensing the Core World” in the form of video, and shared his views on the intelligent edge device market. and how to responsibly empower edge intelligence.

Lars said that we have realized a world of “on-demand customization”, and you can complete shopping, ordering cars, ordering meals and other convenient services with just one tap on your smart device. In the not-too-distant future, with the development of Internet of Things technology, the new applications we create can not only provide services “on demand”, but also predict what users want to do and execute them automatically. These tasks will be completed by smart devices around us .

According to the forecast of the analysis agency, the number of smart connected devices will reach 75 billion in the next ten years. They are like small robots all over us, helping us manage all aspects of our lives. This also means that semiconductor companies will usher in strong market growth opportunities.

A world where everything is connected and intelligent is exciting and longing. However, as the technology enabler of this world, we also shoulder the responsibility to ensure that people can easily and worry-free enjoy the various conveniences of the intelligent world, and at the same time ensure that the world sustainable development. According to Lars, this responsibility is mainly reflected in three areas: computing energy efficiency, intelligent productivity, data security and privacy protection.

We are excerpting the exciting content of Lars’s speech as follows, hoping to help everyone understand the development trend of edge AI in the future and discover the huge market potential.

1

intelligent future

What core applications will be worthy of our attention in the future? Areas with great potential include industrial automation, smart home, smart city, building control, etc. Examples include smart locks on doors, traffic control systems, manufacturing robots, Industry 4.0 fully connected factories, wheel-legged robots, and more. In addition, there are use cases in areas such as energy processing, smoke detection, and smart building controls such as infrastructure and environmental management.

If we take a large and complex smart device such as a car as an example, its intelligent applications span from car access control to ADAS, from automatic driving to efficient OTA upgrades, and so on. A decade ago, automakers did the major work after a car left the assembly line. Today, after a car leaves the assembly line, the product life cycle has just begun. It needs constant maintenance, software upgrades, and will generate a lot of data. sensor platform.

2

AI at the edge

We need to give more intelligence to this edge device. If we encapsulate all architectural systems in a device, and data does not need to go in and out of the device, it will be much easier to ensure information security and functional safety in this device, and its response time will be faster.

Conversely, if all data needs to be transmitted to the cloud, the device always has to wait for the cloud to make a decision, and then transmit the data back. There is no doubt that this is not smart and not timely enough. In terms of energy, the entire data transmission process consumes a lot of energy and network capacity, and generates a lot of useless data flow. Of course, training is also faster in the local device, which inherently has learning capabilities that can be augmented at runtime when running relatively simple models.

In other words, we will see a huge trend, a shift in architecture, we no longer put intelligence in the cloud, but individual terminal devices, they will do all the work independently, even without a network connection, it only works for us Only connect to the Internet to play audio files from the Internet or to fetch additional data from the Internet. This is a completely different approach from the past, with huge changes in terms of architecture and technical capabilities.

This means that we need to introduce artificial intelligence and machine learning into these decentralized small devices, small systems, and these systems must have certain performance and computing power. Although compared with cloud computing, many terminal devices have limited computing conditions, but as long as they are properly trained, they can also have very high performance, run very robustly and very fast.

We have seen that a large number of artificial intelligence and machine learning applications have been implemented in terms of vision and image recognition, speech recognition and anomaly detection, and even a large-scale market has been formed. For the detection use case, only a chip the size of a fingernail is needed, and the entire job can be done with relatively small accelerators in the system. Of course, if it’s a fully autonomous use case in a car, you have to use large systems that consume thousands of watts, which are much more difficult to build and also to build in a reliable way.

3

Connect securely to the cloud

Smart, connected edge devices are not enough. Take a car as an example. Your car was produced in the south, and now it is sold in the north. In winter, the sun is weaker, there is more snow, and the appearance of traffic signs is also different. You need to get data feedback and upgrade your model. This requires aggregating information from millions of sensors in the cloud, reprocessing, updating the model, and deploying it back to the field devices.

These monitoring, detection, updating and continuous learning, is where we need the cloud, so we need to establish a secure connection between the end device and the cloud, otherwise if someone attacks the system, or the connection is not reliable, it will jeopardize the entire fleet of devices all behaviors. At NXP, we address this with our eIQ software tool, which contains predefined code and models, which we call the eIQ Reserve for Vision, Audio and Anomaly Detection.

In addition to this, we also strive to apply NXP’s experience in security. From bank cards, mobile payments to e-government, we have a lot of security-related expertise and bring it into microcontroller design. You can choose to use an edge processor with embedded security, or a processor with a separate secure element that does the heavy lifting for high-security applications. We build and deliver reference designs to enable high levels of encryption, high levels of performance, and secure connections to the cloud, making sure everything runs seamlessly and easily in a trusted manner.

4

key to success

Simply put, we have been talking about big data for the past decade, and big data is the new oil. What will be discussed in the next decade will no longer be big data, but “relevant data”. Only relevant data will be sent to the cloud for group learning.

We endow edge devices with the inference capabilities of artificial intelligence and machine learning models, but the key factors for the success of these devices in the market are efficient, scalable processing power and high security, coupled with efficient and easy-to-use software capable of OTA device management ecosystem. With these elements met, I believe the future of 75 billion smart connected devices will soon become a reality and we can all play a part in this booming market.

Reviewing editor: Peng Jing

Leave a Reply

Your email address will not be published.