How is the AI application in MCU and sensor?
With the Cube.AI software feature pack, it is now possible to support the machine's artificial intelligence/deep learning with STM32 microcontroller products. Sensors can also be more intelligent. The use of artificial intelligence on the sensor can reduce the energy consumption of the entire system, and secondly, it can make its own noise, stability and precision better.
SENSOR:SENSORS SAMPLE KI
MCU+ software feature pack for machine learning?
STM32's future plans have six macro trends: more affordable; more secure; more hardware accelerators and more analog peripherals; higher energy efficiency, lower power consumption; more wireless connectivity; and greater computing performance.
ST will continue to increase single-core frequencies and introduce more products in dual-core.
Aside from the performance indicators, today we mainly look at the application of AI in MCU and sensors.
At the beginning of STM32, it was to improve everyone's productivity and make everyone's work easier. ST is very concerned about the development of artificial intelligence. STM32 microcontroller products can now be used to support artificial intelligence/deep learning of machines. “Deep learning is to create a neural network model, and at the same time to simulate the human brain. STM32 Cube.AI can realize the creation of a neural system. This way, STM32 products help to achieve deep learning that supports artificial intelligence. I hope to be as committed as possible to the development of artificial intelligence." Now Cube.AI is available, using floating point calculations to achieve data calculation.
Let's take a look at an example of future artificial intelligence and what this feature package includes. The Cube.AI software package enables audio and motion analysis, character recognition, and image classification. Nowadays, character recognition and motion analysis are still very simple. For example, the recognition of human activities is based on motion perception. In the future, we will see more and more immersive, early warning or preventive new maintenance applications. Machine learning through STM32 can help us realize the induction of the motor. After the motor is sensed, anomaly detection can be achieved, and the relevant voltage can be directly learned from the key data, as well as the control of the motor. In the future, this feature pack can also be improved in terms of vision, such as gender recognition, facial recognition and other technologies. Of course, it is also possible to implement speech recognition, such as voice control, keyword recognition, and context recognition, which can help end products to continuously improve and improve in deep learning.
From a security perspective, ST is constantly improving the security or security architecture and security platform of its features. In the next scalable upgrade process, more security features will be brought, especially from the Cortex-based L5 products, which can achieve enhanced IP isolation and ensure product security.
Sensor + AI, the dual benefits of reducing power consumption and improving performance
As the 5G era is getting closer, the application and experience of AR/VR will gradually be realized. The connection technology in the product is very mature. It is also very strong on the processor side. Sensors have a long history, how will innovation be made in the future?
Sensors can also be more intelligent. Taking AR/VR applications as an example, the high demand for sensors is because the product is directly connected to the head. The human body is a very important sensor unit, and the AV/VR helmet can capture peripheral physical phenomena and information. When you get this information, it will be simpler if it is intelligently processed and then passed to the master. This is the smart sensor.
The use of artificial intelligence on the sensor can reduce the energy consumption of the entire system, and secondly, it can make its own noise, stability and precision better.
He said that ST has a new six-axis accelerometer + gyroscope combo product, which customers have applied to smart wearable products. It combines all the features of the original six-axis product, but adds a machine learning core to it, plus an anti-shake core. The core of machine learning can learn some basic functions and common functions and solidify them in the sensor. The benefits are reduced system power consumption and the use of fewer resources of the master MCU.
For the specific function, the first is to realize the function of the most basic sensor. At the same time, the device can read the signals of other sensors together through the standard interface for data fusion. The second step is to have the function of learning, to do the filtering function, and finally form the concept, forming the function of the judgment tree. The output of the sensor is no longer the original acceleration value or the value of the gyroscope, but some systems can Directly invoked instructions.
In addition, there is a so-called flight mode when flying, but sometimes users may forget to turn off the flight mode. Sensors with artificial intelligence can recognize changes in different states and directly detect the occurrence of similar events.
Next, look at an example. Traditional wearable products often call the APP when they are doing sports. The APP will read various data to do a lot of processing. Sensors with artificial intelligence can be operated directly in the device.
Then there is the fusion of audio and vibration. TWS true wireless headphones have become the market's explosive products this year.
Then why put these two sensors together? What new features can I implement?
The accelerometer realizes speech recognition, and the microphone still performs radio reception. Traditional headphones can't really tell if they are in a noisy environment. However, after the acceleration of the ear bone conduction, when not speaking, it can switch the earphone to a mode, which can make the sound received by the microphone not be processed, and can be shielded. Under the wind noise environment, only when people talk, the other party will hear. If you don't talk, the whole system can shield the microphone from the sound of the digital processor.
The mainstream mid-to-high-end TWS headsets currently on the market already have such features. It is expected that many manufacturers will put such a concept into this year.
There is also a shock sensation, and the battery is the main application. The battery itself is affected by the environment and temperature, and there are certain risks in specific use cases, especially impacts. Acceleration sensors can therefore also be used for such safety applications
In addition to motion sensors, ST also has imaging products and special optics, such as ToF, in addition to traditional mobile applications, can also be used in a variety of smart home and industrial applications. Imaging products can be used in cars to protect the car or to detect the driver's behavior...
Instead of being a sensor alone, ST is integrated with its MCU ecosystem. ST Sensors All development boards, from hardware to software to development tools to partners, are fully integrated into the STM32 ecosystem. In this way, it is very convenient to integrate it into the development.
ST's range of sensors is very rich. As a provider of solutions, the company is able to provide hardware, software and solutions, and continuously innovate sensors. The innovative solution is a low power, high precision and intelligent sensor.
Finally, the traditional, common sensor is to collect the data and report it to the master to calculate and execute. This process takes into account the power consumption and latency of the system. The concept of the edge is that after the sensor inside the product receives the signal, it can handle some specific applications, such as step counting, detecting different motion states, or flight mode/ground mode, etc., which can find certain. law.
As long as we collect the original data, learn it wi