Figure 1
At the beginning of the speech, Huangfu Jie first provided a brief introduction. Bosch Group was established in 1886 by Mr. Robert Bosch and has developed over the past 137 years, mainly involving four major business areas: the first and largest is the automotive and intelligent transportation sector; the second is industrial technology; the third is energy and building technology; and the last is consumer goods.
As we all know, Bosch is a giant in automotive semiconductors, with various components in vehicles, including chassis systems, extensively using Bosch modules or sensors. Bosch semiconductors belong to Bosch Automotive’s market segment, which includes four major product modules: first, power devices (mainly automotive-grade power devices), including silicon carbide chips; second, ASICs chips (also automotive-grade products); third, automotive-grade semiconductors, automotive-grade MEMS sensors, involving accelerometers, gyroscopes, and other sensors; fourth, consumer-grade MEMS sensors, mainly focusing on consumer electronic products, including smartphones, wearables, and IoT devices.
Huangfu Jie introduced the key milestones in the development of Bosch semiconductor MEMS sensors on a timeline (see Figure 2). In 1996, Bosch launched its first MEMS sensor based on a 6-inch MEMS wafer fab; starting in 2005, Bosch established Bosch Sensortec GmbH, marking Bosch’s expansion from automotive to consumer sectors; in 2007, Bosch Sensortec established its Asia-Pacific headquarters in Shanghai, and that year, Bosch’s MEMS sensor shipments exceeded 1 billion units; in 2010, Bosch launched an 8-inch wafer fab in Dresden; in 2018, Bosch’s MEMS sensor shipments surpassed 10 billion units; by 2021, a 12-inch wafer fab was launched in Dresden, with shipments exceeding 15 billion units (including automotive and consumer sensors); to date, MEMS sensor product shipments have exceeded 18 billion units.
Figure 2
The development of Bosch’s MEMS sensors began in the automotive sector, then expanded to the consumer sector, and now includes the IoT sector. The Internet of Things (IoT) cannot be separated from sensors’ perception of physical signals, and MEMS sensors make IoT devices smarter.
Today, MEMS sensors have penetrated all aspects of life. In buildings, sensors can detect parking spaces and provide indoor navigation. This is particularly useful in underground parking lots, where GPS signals are often weak. To achieve more precise vehicle navigation that complements indoor maps, sensors can be used for vehicular inertial navigation.
Additionally, sensors are used for air quality detection inside vehicles, security intrusion detection, sleep monitoring, and more. In factories, sensors aid in tracking equipment and the application of AR glasses, providing more convenient and efficient tools for production and research.
Finally, in consumer applications, sensors can calculate calories more accurately, enable image stabilization in smartphone photography, and track steps in smartphones or wearables, all based on accelerometer data. Thus, MEMS sensors have infiltrated every aspect of our lives, even if we do not perceive them directly; they are always present.
Next, Huangfu Jie shared case studies related to Bosch’s consumer sensor products with the seminar attendees.
Huangfu Jie explained that earlier, the role of sensors was merely to detect physical signals, with the primary function being to convert physical signals into electrical signals for higher-level processing, which was relatively single-functional. Over time, sensors have integrated more algorithms and software capabilities, allowing for customized configurations and more functionalities. Today, edge AI algorithms can also be integrated into small sensor products, enabling sensors to handle more tasks and create smarter applications.
Unlike traditional AI algorithms, Bosch Sensortec’s edge AI algorithms are integrated into the sensors themselves. By incorporating an ultra-low power processor into the sensor, this functionality can be achieved.
Huangfu Jie further explained the benefits of integrating AI algorithms into the hardware: first, it allows for device customization, enabling sensors to adapt to each individual user; second, the sensor data resides only on the device itself, or even within the sensor, eliminating the need to upload to the cloud, thus ensuring user data security; third, it allows for real-time responses, as data does not need to be uploaded to the cloud and then fed back to the device, resulting in faster response times; fourth, it further reduces device power consumption, extending battery life or usage time.
For example, the latest BHI380 sensor is compact, making it easy to integrate into various consumer electronic products. Additionally, the BHI380 has very low power consumption, ensuring microamp-level power usage during data fusion, and it comes with some basic functions and AI algorithms pre-integrated, allowing users to easily perform secondary development.
Next, Huangfu Jie listed several typical application scenarios for the BHI380 smart sensor: the first is the common 3D audio effect in TWS earphones, which detects the user’s head movement to adjust the corresponding audio scene based on gyroscope data; the second is gesture or action recognition, such as during walking, cycling, or running, or while using transportation like cars, high-speed trains, or planes—these scenarios can be recognized through data fusion; the third application is common human-computer interaction; the fourth is PDR (Pedestrian Dead Reckoning) navigation, as most people now have smartwatches or smartphones, which can record movement trajectories based on ACC and gyroscope data combined with algorithms, without the need for GPS, even indoors where GPS signals are weak, allowing for path planning based on indoor maps; lastly, it involves AI fitness tracking, such as tracking exercise movements, including swimming posture algorithms. The BHI380 smart sensor integrates AI motion monitoring algorithms, enabling the device to automatically detect the user’s current movements and how many repetitions were performed without any intervention. In addition to the built-in fitness movements, users can define new fitness movements. This is a typical example of AI learning capability. For instance, if a user performs a dumbbell curl, which is not pre-programmed in the device, they can trigger a learning mode, then repeat the action about 3-5 times for the sensor to recognize the movement characteristics. Once the characteristics are captured, the device quickly builds a model based on the user’s features, storing it directly in the sensor. The next time the user performs the same action, it can directly recognize it as a custom new action and record the number of repetitions (see Figure 3). Notably, the BHI380 smart sensor’s power consumption has been reduced by 54% compared to the previous generation, making it highly suitable for wearable devices.
Figure 3
Next, Huangfu Jie introduced the latest environmental sensor—the BME 688 four-in-one sensor, which is currently the smallest four-in-one sensor in the world, integrating temperature, pressure, humidity, and gas detection functions. Based on this sensor, we can provide users with microenvironment monitoring, even enabling microenvironment weather forecasting and smart home interactions.
For example, the BME688 can detect different gases. If a user wants to detect the aroma of a specific coffee, they can use the BME688 sensor to collect the target gas signals. Then, based on the captured target gas sensor signals, a corresponding simplified tool on the PC can be used to capture data features, and once the data features are captured, a data model can be generated and integrated into the user-defined algorithm.
The application fields for VOC gas detection are extensive, including forest fire detection, detection of harmful gases to humans, detection of baby diaper freshness, monitoring food freshness in refrigerators, and outdoor air quality index detection, among others.
The working principle of the BME688 is to heat metal oxides internally to react with the target gas. By collecting the resistance values corresponding to multiple temperature points during the reduction reaction, features can be captured and distinguished. For example, if three types of volatile gases are present, the sensor captures different features and establishes models based on the resistance values at different temperature points.
Here is a successful case of using the BME688 gas sensor for forest fire detection in North America. We powered it with solar energy and equipped it with a wireless communication module, placing these devices throughout the forest to monitor air quality in different locations. Before a forest fire occurs, a large amount of smoke is generated, containing high levels of VOC gases. The cloud can capture sensor signals in real-time, and if the VOC gas levels in a certain area rise sharply, it can quickly locate and address the situation (see Figure 4).
Figure 4
Next, Huangfu Jie introduced the latest pressure sensor—the BMP581.
First, barometers are standard in mobile phones in North America, as local mobile operators require all phones to know their coordinates when making emergency calls, which is achieved through barometers.
Actual data collection from the barometer is shown in Figure 5, with the left side showing different floors where the barometer data differs significantly, and the right side shows the changes in barometer readings while climbing stairs, indicating a one-to-one relationship between data changes and steps, demonstrating the sensor’s low noise and high data accuracy.
Figure 5
The BMP581 barometer has typical scenarios in ear-worn or wearable products. For example, when a user wears smart headphones to do push-ups, the BMP581 can detect the height changes during the push-up and recognize how many repetitions were performed. The same logic applies to pull-ups, where the BMP581 can detect the height difference when the user stretches up and fully extends their arms, quickly recognizing the number of pull-up repetitions. There are many other application cases involving barometers, such as height recognition in conjunction with GPS (GPS typically recognizes only XY directions, while adding a barometer allows for height recognition), as well as calorie calculation, fall detection, and blockage detection in robotic vacuum cleaners.
After the speech, attendees showed great interest and attention to Bosch products, engaging in related questions and discussions with Huangfu Jie. Meanwhile, Bosch Sensortec GmbH’s programmable AI sensor BHI380 received the “2023 Hard Technology Industry Comprehensive Award” and the “E” Ma Dangxian New Product Award, with medals presented at the seminar (see Figure 6).
Figure 6
Looking ahead, Bosch Semiconductor will continue to expand the AI ecosystem. As a technological leader in the industry, Bosch Semiconductor has never stopped innovating.