MEMSIC Attends International MEMS Symposium, Explaining the Industrialization Path of MEMS in the AI Era
At the 18th International Symposium on Sensors and MEMS Industrialization Technology, MEMSIC delivered a presentation titled "MEMS Industrialization in the AI Era: From Physical Measurement to Trusted Perception," proposing that while AI enhances reasoning capabilities, the quality of physical measurement and systems engineering are what determine whether it can safely enter the real world. Its accelerometers have cumulative shipments exceeding 2 billion units, with a global Android market share of over 30%, while its AMR geomagnetic sensors have cumulative shipments exceeding 2.5 billion units, with a market share of over 40%.
Body · AI translation
From November 12 to 13, the 18th (2026) International Symposium on Sensor and MEMS Industrialization Technology (and Achievement Exhibition) (referred to as the "MEMS International Symposium") was held in Shanghai. On November 12, MEMSIC attended the conference and delivered a keynote speech titled "MEMS Industrialization in the AI Era: From Physical Measurement to Trusted Perception," which, drawing on MEMSIC's 27 years of industrialization practice, systematically elaborated the engineering path for MEMS in the AI era to move from device performance toward trusted perception.
This year's symposium was hosted by the Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, the National Key Laboratory of Sensor Technology, and others. It brought together more than 40 renowned experts from home and abroad and representatives of leading MEMS industry enterprises, focusing on the integration of AI and MEMS sensing. Discussions were held around the complete industry chain of sensing chips, wafer foundry, micro-nano processes, packaging and testing, and industrialization implementation, promoting the deep integration of perception technology with frontier directions such as physical AI, embodied intelligence, and robotics, helping to connect scientific research, industry, and capital resources, and jointly exploring a new landscape for the development of the MEMS industry in the intelligent era.
From "able to compute" to "able to perceive":
The speech began with the question "If the robot sees it correctly, why does it still grab off-target?": the camera identifies the position of a transparent water cup on the table, yet the robotic arm grabs off-target—the reason may not necessarily lie in the vision algorithm, but may instead be that the camera and robotic arm coordinates are not aligned, that there is a deviation in timestamps, or that after the cup moves, the system is still using slightly earlier position information. Correct recognition ≠ trusted measurement ≠ reliable action. What AI improves is reasoning capability; the quality of physical measurement and systems engineering determine whether it can safely and stably enter reality. The system needs to simultaneously ensure three progressive layers: "measurable, estimable, executable." The raw quality of sensors, fusion and state estimation, and the closed loop of system action are all indispensable—"being able to measure" does not equal "the system being usable."
Behind this proposition is definite market momentum: according to the IFR's World Robotics 2025, in 2024 global new installations of industrial robots reached 542,000 units, with Asia accounting for 74%, including about 295,000 units in China; SEMI expects that in 2026 spending on MEMS and sensor manufacturing equipment will reach 13 billion US dollars, an increase of 26% over the previous year, and a further increase of 44% in 2027.
The speech specifically noted that equipment spending reflects overall changes in market demand, not sensor sales themselves. Automation demand continues to rise, and perception is becoming the entry point for physical AI—but growth only brings opportunities; verifiable and reproducible system capabilities are what determine whether opportunities can be transformed into long-term value.
Industrialization runs through from chip to application:
It is not about making the chip, but about repeatedly making it right
Industrialization is not about making the chip, but about repeatedly making it right under defined conditions and enabling the system to verify it.
From requirement definition, device design, process and packaging, to testing and calibration, system verification, and mass production delivery, the six stages are interlinked: packaging stress can change device behavior, installation method affects coordinate interpretation, calibration conditions determine how data are corrected, and system interfaces affect time synchronization. To discuss "whether accuracy is good," one must first ask what quantity is being measured, under what conditions of use, and how the system will accept it—discussing accuracy apart from conditions of use is like evaluating vehicle speed without looking at road conditions.
This approach also runs through MEMSIC's 27 years of product practice.
As the designer and manufacturer of the world's first two-axis gas thermal accelerometer, MEMSIC has focused for more than 20 years on two platforms: inertial and magnetic sensing. The thermal accelerometer uses a convective structure without a solid proof mass, achieving impact resistance exceeding 10,000 g, extremely low noise, and per-unit temperature compensation; capacitive accelerometers cover a range of 2 G to 150 G; the fully self-developed six-axis IMU integrates a 3-axis gyroscope and a 3-axis accelerometer; and AMR magnetic sensors and magnetic encoders rely on its own wafer fab to ensure process autonomy. Inertial and magnetic sensing are naturally complementary—inertial zero bias drifts over time, while magnetic signals are susceptible to interference from nearby metals, currents, and magnetic materials. Only after calibration of coordinates, time, and error models and fusion can a more trusted state estimate be formed. Taking AI glasses as an example, motion and orientation information under lightweight and low-power constraints is provided jointly by the six-axis IMU and AMR magnetic sensing, and application value is formed jointly by the device, system integration, and actual experience.
From single-point parameters to system trustworthiness:
Turn verification into evidence, and performance into trust
Industrial value is moving from single-point parameters to system trustworthiness: device specifications answer "what can be measured," application boundaries answer "when it can be used," and system evidence answers "whether it can be sustained." From product datasheets, laboratory testing, system verification, customer verification, to mass production delivery, this is a complete chain of evidence—completing laboratory testing does not mean the system has passed verification, and passing sample verification does not mean mass production conditions are met. Every performance value should carry test conditions, sample scope, and version status. Rigorously stating "which step verification has reached" builds trust better than numbers without boundaries.
MEMSIC's products are precisely evidence of this logic: the fully self-developed six-axis IMU underwent five years of research and development and successively passed stringent verifications such as 60,000 drop tests by a leading customer and impacts on the order of 20 kg while maintaining zero failures; currently, MEMSIC's accelerometers have cumulative shipments exceeding 2 billion units and a global Android market share exceeding 30%, while its AMR geomagnetic sensors have cumulative shipments exceeding 2.5 billion units and a global Android market share exceeding 40%—large-scale, cross-batch stable delivery is the most direct footnote to "a batch of stable performance."
This shift also requires cross-stage collaboration: MEMS sensors are integrating analog and digital circuits and even AI cores, and from design transfer to production to process matching across different fabs, multiple risks exist; many collaborative projects only discover late in the process that the transfer-to-production ramp-up is more time-consuming than front-end development. Application parties, sensor companies, system teams, and verification partners need to jointly maintain the requirements matrix, joint debugging plans, and version management from an early stage, starting with defining scenarios, co-building verification, and sharing feedback, and closing the loop on deviations and issues into design, calibration, and application solutions—good collaboration is not frequent meetings, but rather giving different teams a clear, shared definition of the same result.
"A successful chip tape-out is only one milestone; only repeated verification at the system level and stable operation can truly support industrialization." MEMSIC insists on letting real application metrics drive R&D and uses collaborative verification to accelerate the transformation of technological achievements; relying on its Fab-Lite self-developed manufacturing system, it promotes the reliable embedding of sensing technology into systems—starting from precise and reliable measurement, it empowers intelligent systems to perceive and understand the physical world and achieve stable and controllable physical interaction.
MEMSIC has been deeply engaged in the MEMS industry for 27 years and has built self-developed capabilities spanning process design, core wafer manufacturing, and calibration and verification. With the Fab-Lite model, it efficiently collaborates with industry chain partners and provides one-stop services from sensing chips, signal conditioning, algorithms, software, and system solutions to modules. Its products cover thermal and capacitive accelerometers, AMR magnetic sensors, Hall devices, six-axis IMUs, and more, and are widely used in automotive, industrial, intelligent terminal, and robotics fields.
From the world's first two-axis gas thermal accelerometer to today's full-category sensing portfolio covering automotive, industrial, intelligent terminals, and robotics, MEMSIC's 27 years are precisely a microcosm of the MEMS industry's shift from "making the chip" to "making the system trustworthy." As AI accelerates from the digital world into the physical world, from robotic arms and smart glasses to humanoid robots, behind every stable action are countless measurements worthy of reliance. Distilling device capabilities into verifiable, deliverable system trustworthiness—this ongoing long march of industrialization is one MEMSIC will continue.