“Data is the foundation of AI at the edge, and perception is the source of that data. Just as a person needs not only eyes to sense the world, but also the ears, which are important organs for perceiving the natural world, machines also need hearing and vision, and various sensors have emerged with the development of technology. The single-chip millimeter-wave radar launched by TI can avoid the drawbacks of traditional cameras in many applications, and at the same time supports multiple data fusion of the system, so that the machine can better obtain data and achieve accurate perception of the target.
In 1956, when McCarthy of Stanford University proposed “artificial intelligence”, he must not have imagined that this concept would be in full swing in China decades later. Artificial intelligence is not only expected to trigger a new industrial revolution, but it is integrated into everyone’s daily life and is triggering social change. In fact, in April 2020, when the National Development and Reform Commission identified three aspects of the new infrastructure, it mentioned artificial intelligence in both information infrastructure and integrated infrastructure, but it only clearly informed the public of the ongoing changes.
With the development of computing power, data, and the Internet, artificial intelligence is at the point of changing from quantitative to qualitative, especially at the edge showing explosive development. Gartner predicts that by 2025, at least 75% of data processing will take place outside the cloud or data center. The tide of artificial intelligence is both an opportunity and a challenge for semiconductor companies. Unlike the cloud, the most important requirements for chips on the edge side still return to the eternal topics of performance, cost and power consumption, and products must have both of them in order to win. In addition, since the development cycle of edge AI products is short and the iteration window is fast, a friendly development environment is also critical.
Howard Jiang, Director of Embedded Product Systems and Applications, Texas Instruments (TI) China
What kind of edge AI system is successful? “Precise perception, fast decision-making, human-machine collaboration, high efficiency and energy saving, safety and reliability”, this is the answer given by Howard Jiang, director of embedded product systems and applications at Texas Instruments (TI) China. As we all know, perception, decision-making and execution are the three links of edge artificial intelligence, and with the development of edge artificial intelligence, the requirements for embedded perception and decision-making technology are stricter and more differentiated than those of non-artificial intelligence.
Perception – The Data Source for Edge AI
Data is the foundation of AI at the edge, and perception is the source of that data. Just as a person needs not only eyes to sense the world, but also the ears, which are important organs for perceiving the natural world, machines also need hearing and vision, and various sensors have emerged with the development of technology. The single-chip millimeter-wave radar launched by TI can avoid the drawbacks of traditional cameras in many applications, and at the same time supports multiple data fusion of the system, so that the machine can better obtain data and achieve accurate perception of the target.
Using the transmission and reception of millimeter-wave radar, the distance and relative speed of objects and obstacles within its field of view can be measured with extremely high accuracy. An important advantage of mmWave sensors compared to vision- and lidar-based sensors is that they are less susceptible to environmental conditions such as rain, dust, smoke, fog, or frost. Additionally, mmWave sensors work in complete darkness or in direct sunlight. Mounted directly behind a plastic housing with no outer lenses, vents or sensor surfaces, these sensors are rugged enough to meet the Ingress Protection (IP) 69K standard.
TI’s single-chip millimeter-wave radar is made of CMOS process technology, which realizes the cost-effective advantage that traditional radar does not have. At the same time, combined with ASIC back-end processing, it can directly reduce BOM cost, reduce product size, and reduce the need for processors. dependency. The product based on TI’s millimeter-wave radar design is one-third the volume and half the weight of the miniature lidar rangefinder.
More importantly, in addition to the field of autonomous driving, millimeter-wave radar can also be applied to a wider range of industries such as smart homes, smart buildings, and medical care. For example, through the combination of millimeter wave radar and air conditioner, many intelligent functions such as wind movement, attitude perception of the target human body, and automatic switch can be realized. In other applications, such as safety monitoring for robotic arm operators, obstacle avoidance detection for logistics robots/UAVs, and monitoring of falls of the elderly, millimeter-wave radar has the advantages of accurate and fast perception that previous image sensors do not have. At the same time, it meets the data desensitization requirements of many applications (can be installed in bedrooms, bathrooms, etc.).
In addition to millimeter-wave radar, TI also offers a wide range of products such as temperature sensors, DLP® technology, ToF, etc., further enriching the way machines and humans interact.
Decision Making – The Brain of Edge AI
Edge AI devices need a smart “brain” for data processing and decision-making. An integrated SoC is often a good choice in edge AI because, in addition to housing the various processing elements capable of performing deep learning inference, the SoC integrates many of the necessary components for the entire embedded application. Some integrated SoCs include Display, graphics, video acceleration, and industrial networking capabilities, enabling single-chip solutions to do more than just run ML/AI.
Jacinto of TI The 7 series processor is such a highly integrated SoC that includes high-performance computing, deep learning engines, dedicated accelerators for signal and image processing, and complies with functional safety ASIL-D/SIL-3 standards. In addition to advanced driver assistance systems (ADAS), processors can also be used in robotics, machine vision, radar, and more.
Integrated dedicated accelerators include “C7x” next-generation DSP cores that combine TI’s industry-leading DSP and EVE cores, add vector floating-point computing capabilities, and support backward compatible code. With the rise of edge artificial intelligence, DSP can significantly improve the efficiency of matrix operations due to its Harvard architecture, which is very suitable for neural network computing acceleration. At the same time, the newly added “MMA” deep learning accelerator can achieve 8 TOPS of computing performance at low power under typical operating conditions.
Common cores include multi-core Arm Cortex-A72, Cortex-R5F and 8XE GE8430 GPUs.
The multi-core heterogeneous processor architecture design of Jacinto 7 series can maximize the selection and optimization of tasks, so as to achieve better performance improvement and cost control. In addition, TI will also hardware mature algorithms, coupled with the evolution of the semiconductor process, so as to achieve the best cost performance and power consumption ratio. For example, TI’s ISP can automatically realize wide dynamic adjustment, image pyramid scaling, stereo depth vision and dense optical flow algorithm acceleration based on the hardware acceleration unit embedded in the chip.
The Jacinto 7 series processors provide a comprehensive security solution involving both hardware and software, an important focus for the automotive and industrial markets. Jacinto 7 series processors are system designed for ASIL-D functionality using a hardware development process certified by an independent functional safety assessment body such as TÜV SÜD. In response to the new challenges of high bandwidth and multi-port brought by ADAS data fusion, the Jacinto 7 series also integrates multi-ports such as CSI-2, which can ensure interconnection with multiple sensors and support high-bandwidth data requirements. The Jacinto 7 series integrates both a PCIe hub and a Gigabit Ethernet switch and can be used as a domain controller for a higher level of integration.
In order to facilitate user development, TI has launched TI-Edge-AI-Cloud, a cloud tool for AI inference on Jacinto processors to evaluate and support many industry-wide and popular deep learning frameworks (including TensorFlow Lite, ONNX Runtime, OpenGL ES, etc. ) to help easily compile and deploy models and accelerate inference.
In addition to the CNN commonly used in visual recognition, the Jacinto 7 processor also provides corresponding support for RNNs required for edge artificial intelligence scenarios such as predictive maintenance. In addition, TI’s industrial applications processor SitaraTMThe series, which integrates Arm Cortex-A series cores, can also implement edge artificial intelligence applications with relatively low computing power requirements through Arm NN, such as predictive maintenance in industrial applications.
From imaginative to fast landing
In addition to perception and decision-making, TI also has many processors, motor drives, and various analog devices in the execution link, which shows that TI can support the key aspects required to realize edge AI through a broad product portfolio. Technology serves people. As a semiconductor company with a history of 90 years, TI’s secret is to continuously study the changes in social life, gain insight into and meet people’s needs, and thus constantly change itself.
In the 35 years that TI entered China, it has helped Chinese customers to achieve innovation time and time again. In the field of artificial intelligence, Howard sees China’s development already ahead in some places. For example, in the field of 4D imaging radar, foreign countries are still in the laboratory proof-of-concept stage, while domestic car manufacturers have proposed a specific time point for mass production. At the same time, in the smart industry, robotics, smart home, medical image analysis and other markets, Chinese customers are also actively importing. Sometimes Howard also inspires his American colleagues with innovations adopted by Chinese clients.
When TI defines products, it will cooperate with some customers in depth, so that the demand for innovation is reflected in the product architecture, which is not only to meet the existing use, but more importantly, to jointly foresee future needs. The rich imagination of Chinese customers also enables the demand from China to be quickly reflected in the development of TI’s next-generation products and for innovative industries around the world.
Thirty-five years ago, TI entered China and started from scratch with local Chinese partners to innovate together in many markets such as industry, consumption, communications, and automobiles. Now, with the expansion of new infrastructure projects represented by artificial intelligence and edge artificial intelligence, TI, together with partners, has once again stepped into a new field, through a complete range of analog and embedded processing products, strong local manufacturing and research and development Capability, product distribution and sales network all over the country to solve the new challenges brought by edge artificial intelligence and meet the new requirements of Chinese customers’ product design. At TI, there are a group of outstanding engineers like Howard, with the vision of “core to China, science and technology to create the world”, and strive to make the world a better place through semiconductor technology.
Not long ago, Midea Group’s Kitchen and Water Heater Division and TI jointly established the “Perception and Interaction Joint Laboratory”, which aims to help Midea use TI’s millimeter-wave radar technology and a wide range of analog and embedded processing products to accelerate Midea’s kitchen. Development of thermoelectric appliance applications. Howard hopes TI can help support more companies like Midea to turn ideas into reality.