EN / 中文

Silicon-Based Industrial Pressure Sensors: Packaging & Testing, the Hidden Bottleneck for China’s Substitution

by zhongguodianzibao·November 18, 2026

Sensors are regarded as the "eyes" and "tactile sense" of industry, serving as irreplaceable basic components in smart manufacturing and process industries. Among them, silicon-based pressure sensors have become the mainstream sensing technology for industrial measurement and control due to the mature IC fabrication processes of silicon materials, excellent mechanical properties, and on-chip integration capabilities.

In recent years, policy windows have opened, ranging from the national "Industrial Strong Foundation" project to the independent establishment of key special projects for intelligent sensors in the "14th Five-Year Plan," and further to the collaborative promotion of intelligent sensors and industrial measurement in the "15th Five-Year Plan." However, after the window opens, the real test lies in the capacity to absorb and utilize these opportunities.

The root cause of China's industrial pressure sensor sector being "complete but not large, large but not strong" does not lie in individual technical indicators, but rather in the packaging and testing stages that have been long underestimated. These stages simultaneously cap performance ceilings and floor cost limits, directly causing the situation where "indicators can be achieved in samples, but mass production fails." Understanding this is key to grasping the breakthrough logic of the entire industry.

Redefinition: The Unit of Competition Shifts from "Pieces" to "Sets"

To understand the silicon-based industrial pressure sensor industry, one must first redefine the term "sensor." GB/T 7665—2005 "General Terms for Sensors" defines it as "a device or apparatus that senses a measured quantity and converts it into a usable output signal according to certain rules, typically composed of a sensitive element and a conversion element." This definition is accurate and prudent, positioning sensors within the "component" category. Currently, with the development of industrial sensor technology, terms like "transmitters" and "actuators" have transcended the traditional concept of simple signal-converting elements, injecting new meanings such as "signal transmission/conversion" and "execution devices" into the definition of sensors. Facing the rapid development of AI, sensors will evolve into "AI + Sensors," becoming interconnected, autonomously controlled intelligent microsystems, where each microsystem functions as an independent network node. This represents not just three leaps in definition, but a re-understanding and empowerment of sensors. In the past, they were simple components valued by piece count, with value attributed to materials and processes. Now, as transmitters and actuators—calibrated measurement systems—their value belongs to design, packaging/testing, and technological empowerment. In the future, as intelligent nodes capable of independent decision-making, their value will reside in algorithms, multi-modal perception data, and the deep integration of edge intelligence ecosystems.

It is worth noting that when "transmitters" and "actuators" are included in the sensor scope, the unit of industrial competition shifts from "pieces" to "sets." Consequently, value increases, and the focus of competition moves from sensitive chips to packaging, compensation, calibration, and system integration. Simultaneously, the boundary between advanced sensors and instrumentation becomes blurred. Current and future sensors definitely fall under the category of advanced instrumentation, but a typical characteristic of sensors must be the ability to convert sensed and processed information into output signals. This conversion and processing function determines that sensors are product "nodes" rather than "terminals," and also dictates that future sensors will earn not just component margins, but primarily through empowering data services and decision systems.

Returning to strategic positioning, the weight of industrial basic sensors far exceeds their output value. Situated within industrial processes, they provide raw information for control, processing, and decision-making, directly influencing production outcomes. They are core basic electronic components highly valued and actively developed by developed countries, serving as the cornerstone for China's strategy to become a manufacturing powerhouse. Their application scale, variety richness, and technical level characterize and support a nation's industrial manufacturing capability. Given such heavy industrial positioning, where exactly have we been stuck in recent years?

After years of development, China has formed six major industrial clusters centered around key cities: The Yangtze River Delta (Shanghai, Zhejiang, Jiangsu, Anhui) has a relatively complete system for force, thermal, magnetic, inertial, and image sensing; The Pearl River Delta (Guangzhou, Fujian, Shenzhen) focuses on force, thermal, magnetic, ultrasonic, and weighing sensors; Northeast China (Shenyang, Dalian, Changchun, Harbin) serves as an R&D base for MEMS force, gas, and humidity sensing; The Bohai Rim (Beijing, Tianjin, Shijiazhuang) specializes in force, thermal, gas, and novel sensors; Central China (Zhengzhou, Wuhan, Taiyuan) integrates industry-academia-research, showing good momentum in thermistors and gas sensors; Western China (Xi'an, Baoji, Tianshui, Chengdu, Chongqing, Mianyang) has established an industrial system for force, thermal, and optical sensors. The 2020 "China Sensor Industry Blue Book" indicates that there are approximately 2,000 domestic industrial sensor manufacturers of significant scale, covering about 16,000 varieties and specifications, essentially covering all fields of industrial manufacturing. China's industrial sensor layout is basically formed, categories are mostly complete, and low-to-mid-range products meet the needs of the entire industry.

However, two easily overlooked issues need clarification. First, while the six clusters have their own focuses, cross-cluster coordination and division of labor are unclear, supporting relationships are not yet established, resulting in more homogeneous regional duplication. Second, the overlap of the three "basics" outlines the typical form of our industrial sensors being "complete but not large, large but not strong."

From an international perspective, industry statistics show that China accounts for about 10% of the global sensor market, compared to 29% for the US, 21% for Japan, and 19% for Germany. While market share differences are not vast, there is a gap of approximately 10–15 years in comprehensive product capabilities. Specifically, design technology lags by 5–10 years, while manufacturing, industrialization, and application technologies lag by 10–15 years. Iconic industrialization enterprises in the US, Europe, and Japan include GE, Honeywell, Siemens, Bosch, STMicroelectronics, Endress+Hauser (E+H), Yokogawa, Omron, and many others, whereas domestic counterparts have long been absent.

Behind the gaps in software, processes, talent, and industrial standardization lies a lack of relevant content, meaning process parameters cannot be stably reproduced, products cannot be fixed, customers cannot accept them based on standards, and the industry cannot accumulate knowledge. This is a soft constraint harder than equipment limitations and is the fundamental shackles preventing any industrial development and application strategies from taking effect.

Core Issue: Packaging and Testing Determine Performance Ceilings and Cost Floors

Packaging and testing of industrial silicon sensors are not equivalent to IC packaging and testing. ICs pursue two-dimensional planar structures maximizing on-chip device counts; sensors require three-dimensional structures to achieve sensing and transmission. Sensor packaging must reflect not only on-chip devices (resistors, capacitors, etc.) but also movable sensitive structures, typically manifested as beam-film-island bulk silicon structures. Based on industrial environment requirements, structures must also achieve isolation of the measured medium and protection of the sensor structure. This leads to a series of special requirements different from IC packaging. For instance, in stress management, packaging stress directly translates to zero-point drift and temperature drift, necessitating stress isolation design; regarding thermal budget, high-strength bonding processes below 500°C are usually required to meet wafer-level bonding needs; in testing strategy, it involves wafer-level screening, compensation calibration, and product aging and batch testing throughout the manufacturing process; in industrial form, IC packaging has a mature OSAT (Outsourced Semiconductor Assembly and Test) foundry system, while sensors lack professional foundries, requiring manufacturers to solve all key technologies and processes for sensor packaging and testing internally, with know-how fully embedded in the sensor production line itself.

Differences in cost structure are even more thought-provoking: IC packaging accounts for about 40–50% of costs, mainly material and capacity, which can be diluted through expansion; MEMS packaging accounts for about 30–40% of costs, mainly non-standard process development and yield loss. Although the percentages are close, their natures are completely different. The cost curve for industrial silicon-based sensor MEMS packaging and testing follows a learning curve rather than economies of scale, making it the hardest part for latecomers to overcome. It is precisely for this reason that during the "14th Five-Year Plan," the state independently listed silicon-based sensors, separating them from the IC field, highlighting their independent functional characteristics.

Reflected in products, silicon-based industrial pressure sensors follow three main technical routes. Diffused silicon piezoresistive types use the piezoresistive effect to prepare Wheatstone bridges on silicon, offering the most mature process, lowest cost, and most complete supply chain, making them mainstream for industrial OEM sensors, but suffering from significant temperature drift requiring complex temperature compensation; silicon capacitive types achieve differential capacitor output via a central movable plate, offering low temperature drift, high sensitivity, and low power consumption, but being sensitive to parasitic capacitance and non-linearity with limited range coverage; single-crystal silicon resonant types utilize the change in natural frequency of resonators with stress to output quasi-digital frequency signals, offering optimal precision and long-term stability, representing the technical high ground for high-end industrial transmitters.

Analyzing current international first-tier products in the process industry pressure sensor field reveals: Yokogawa EJX-A series has a typical accuracy of ±0.025% FS and long-term stability of ±0.1% URL/15 years; Emerson Rosemount 3051S series has a typical accuracy of ±0.025% FS and long-term stability of ±0.15% URL/15 years; FUJI FCX-AIII series has a typical accuracy of ±0.01% FS and long-term stability of ±0.1% URL/10 years. Currently, several domestic companies can produce products with0.075% FS accuracy, but the gap lies in whether they can maintain 0.1% URL or 0.15% URL stability after continuous operation for 10 or 15 years. The former threshold involves design and materials, while the latter requires systematic guarantees in packaging, processes, stress control, and accelerated life verification. This leads to two meaningful conclusions: First, do not expect to "overtake via advanced process nodes"; second, for silicon-based industrial pressure sensors, an 8-inch line is not necessarily better than a 6-inch line; the key lies in process stability and matching specialized equipment. Time is an incompressible investment. Silicon-based MEMS sensors, especially their sensitive core chips, have their own characteristics. As functional devices, the development of MEMS pressure chips does not follow Moore's Law, but requires devices to be not only stable and reliable but also to evolve over time. Reliability data must be accumulated in units of years.

Moving from "can make" to "can sell," industrial-grade sensors must pass the hurdles of compensation, calibration, and accelerated life testing. Industrial-grade products undergo a much longer chain than consumer-grade ones upon shipment: signals go through instrument amplification, 24-bit Σ-Δ conversion, MCU lookup table compensation output, and 2D compensation stored in EEPROM for interpolation calculations; followed by reliability verification including temperature cycling, pressure cycling, long-term drift, media compatibility, and accelerated life tests. These are precisely where "indicator-driven leadership" is most prone to distortion. Achieving 0.05% FS to 0.01% FS products in the lab can be a successful process realization, but ensuring 95% of100,000 products on the production line meet standard requirements requires full engineering of calibration takt time, fixture consistency, temperature field uniformity, and data traceability. The former is a scientific problem; the latter is a manufacturing problem. Currently, China's sensor industry excels at the former but struggles with the latter.

The Real Bottleneck: Lack of an "Application Iteration Closed Loop"

Domestic high-end industrial sensors cannot enter key fields, key industries, or major projects (the "Triple Key Areas"), thus failing to obtain field failure data and build reliability accumulation. This creates a self-reinforcing negative loop: unable to enter → no data → unable to improve → even harder to enter. Domestic substitution ultimately aims to solve industrial demands in the national "Triple Key Areas." Neither relying solely on state-funded technical breakthroughs nor idealistic industrial ambitions alone can break this cycle. Compared to international first-tier products, the product structure is identical, process steps are unchanged, and indicators match, yet yields remain low, batch consistency is poor, and long-term drift is significant. Products are "similar in form but different in spirit." The root cause lies in process parameter details not written on drawings, such as minor distributions of doping concentration, slopes of bonding temperature curves, and coupling of polishing removal rates with assembly residual stresses. These cannot be obtained through reverse engineering and can only be accumulated through long-term systematic process research.

More subtly, distorted statistical metrics are misleading decisions. CCID Consulting provides two conflicting figures for the 2025 Chinese sensor market size: CNY 460.86 billion and CNY 470 billion, while segmenting MEMS pressure sensors at approximately CNY 23.94 billion. The difference between CNY 460 billion and CNY 23.9 billion is nearly 20 times. Within this 20-fold difference, the actual data for sensors applied in the true industrial sector is significantly smaller. The glamorous data of "generalized" sensors masks the reality of "chip and device shortages" in industrial sensors. If decision-makers see the former, resource allocation logic will inevitably be wrong, assuming the industry is already large and only needs "structural optimization" to "add flowers to brocade." In reality, the true industrial sensor sector is still in the startup and ramp-up phase, needing to be "lifted up" and given a "push." Correcting this statistical metric to provide true data for policy formulation has a policy value no less than adding another national special project for sensors.

The fragility of the current industrial chain is also reflected in single-point supply. The supply of sensitive chips for silicon-based industrial sensors has a high threshold, requiring suppliers to provide full-range, full-temperature, and full-category (range) supply. No silicon-based industrial sensor manufacturer, especially leading enterprises, will easily change its entire production process due to a change in a single chip. It is crucial to note that the current domestic industrial silicon-based chip industry concentration relies heavily on single-source supply. The risk lies not in exclusive supply per se, but in the fact that this situation determines the physical upper limit of domestic industrial silicon-based pressure sensor production capacity. Without backup capacity, price competition, or technical route redundancy, once operational or supply fluctuations occur, dozens of downstream sensor manufacturers face the risk of supply cutoff. This is not an isolated case. During the industrialization transition of research institutes in 2012–2013, a veteran national-level sensor institute exited the niche domestic silicon-based industrial pressure-sensitive chip field. Within the same window, foreign manufacturers quickly filled the domestic gap after one year of effort, occupying nearly 90% of the domestic silicon-based pressure sensor chip market in less than five years. A national institute with first-mover advantage faded out of a segmented market during institutional adjustments, instantly losing 90% market share to foreign entities. Rather than a natural result of market competition, this reflects the cost paid due to the lack of connection mechanisms in the domestic silicon-based industrial pressure-sensitive chip industry. To prevent the predicament of "chipless" production, domestic silicon-based industrial pressure sensor enterprises prioritize securing chip inventory (full range, full series, full category) as the top priority for survival. Chip inventory pressure has become the Achilles' heel of enterprise development.

Underestimated Risk: Geopolitical Control Through Technical Source Penetration

The lethality of technology blockades lies not in explicit performance thresholds, but in Entity List + Foreign Direct Product Rules (FDPR) technical penetration. Relevant foreign control actions have been frequent in recent years. Since 2018, the US Bureau of Industry and Security (BIS) has continuously added Chinese enterprises to the Entity List. In 2022, the US signed the CHIPS and Science Act, and the EU launched the European Chips Act. In 2024, BIS issued the third revision of Advanced Computing and Semiconductor Manufacturing Equipment (AC/S & SME) rules. A textbook case occurred in June 2026, when US BIS announced penalties against two German subsidiaries of Bosch for exporting MEMS sensors to entities on the Entity List between 2020 and 2024, involving transactions of approximately USD 72.36 million and fines of USD 36.18 million. The key issue was that the涉案 products themselves were classified as EAR99, the lowest control category, but because their tape-out, calibration, and testing stages used controlled US technology, they were still deemed subject to EAR jurisdiction and severely penalized.

This offers us reversed insights. First, if a domestic pressure sensor uses controlled US equipment or software in its manufacturing process (even just a test bench for calibration), it may fall under EAR jurisdiction when shipping to certain customers. "No contact with controlled technology sources throughout the process" is a ruler far stricter than "import rate of sensitive chips" and a reality that high-end device autonomy must face. Additionally, ECCN 2B230 indeed contains explicit threshold entries for pressure sensors, serving non-proliferation scenarios like nuclear and strong corrosion, but this is not the threshold for general high-end silicon-based MEMS sensors. We must now be vigilant about the FDPR technical source penetration mechanism. This is not a hypothetical performance red line, but a sword hanging over the head that our silicon-based pressure sensor industry development must guard against. Path Selection: Leverage Packaging and Testing as a Fulcrum for Breakthroughs.

Currently, the direction for developing the industrial sensor industry is clear. We need to move from "countermeasures" to "solutions," forming a systematic path.

First, continue cultivating hidden champions, exerting sustained efforts in four actions: strengthening foundations, preliminary research, tackling bottlenecks, and decision-making. One, strengthen foundations by following a path with Chinese characteristics based on the "Industrial Strong Foundation" project; Two, conduct preliminary research by proactively deploying frontier sensors such as graphene, quantum, and flexible sensors; Three, tackle bottlenecks by conducting original R&D in new materials, principles, processes, and structures, implementing "zero-out clearing" attacks on long-standing "chokepoint" products—turning lists into item-by-item checkable ledgers with clear acceptance criteria (e.g., must supply batches for three consecutive years without major quality issues); otherwise, "clearing attacks" degrade to mere "acceptance passes"; Four, provide top-level design and strategic planning consulting for decision-making.

Second, achieve industrial chain advancement through "solidifying, supplementing, and strengthening" chains, in that strict order. The correct sequence is first to "solidify" the chain by fixing processes and inspection norms and establishing corresponding standards; then to "supplement" the chain by filling auxiliary supports like design software; finally to "strengthen" the chain by enhancing localized equipment facilities. A common mistake in reality is buying equipment first (strengthening), then supplementing software, and finally solidifying processes/norms/standards. If standards are not solidified, processes cannot be reproduced, and even advanced equipment will only produce products that are "similar in form but different in spirit." Standardization is a prerequisite for industrial chain advancement, not a result.

Third, establish a new tripartite coordination mechanism involving government, enterprises, and markets to address the localization development of process industries like coal, electricity, gas, chemical, and oil, as well as national "Triple Key Areas," breaking the "application iteration" dead loop.

Fourth, unify consensus across the sensor industry, breaking boundaries between sectors and specialties; elevate the development of the industrial sensor industry to the national level, reflecting national will. Focus on post-production process research and equipment manufacturing, implement joint attacks according to industrial chain requirements, break the barriers of bidding systems, clarify outcome attributes, strengthen performance assessments, and ensure lead roles are taken by unrelated third parties to strictly prevent the "attack" process from becoming a redistribution of interests.

Fifth, view "AI + Sensors" rationally. The contribution of rapidly developing AI to current sensors lies more in calibration and diagnosis, embrace and fusion, rather than edge-side inference. Machine learning fitting temperature drift and non-linear surfaces, participating in mechanical property analysis, can significantly compress calibration points and work hours, predict sensor failure precursors early, transform periodic calibration into on-demand calibration, use production line data to infer process parameter windows, and shorten the cycle for new process finalization. AI is an accelerator, not a substitute; it cannot replace process accumulation, only accelerate its convergence.

Conclusion

China's industrial sensor industry has weathered over sixty years of storms, evolving from nothing to something, from something to completeness, yet remaining "complete but not large, large but not strong." Here are three judgments:

First, the root cause of being "complete but not large, large but not strong" is not insufficient capability, but necessary "capability mismatch" in the early development process. China possesses the most complete product lineage, the most production lines, and the largest application market, yet has shortcomings in high-end industrial basic sensors, particularly in the coal, electricity, gas, chemical, and oil process industries. The main reasons are process accumulation and application iteration, both requiring long-term patience to cultivate.

Second, packaging and testing is an underestimated strategic fulcrum. It simultaneously determines performance ceilings and cost floors, spanning the manufacturing and industrialization segments, acting as the watershed between "can make" and "can sell." Under constraints of limited resources, focusing on packaging and testing yields a higher return on investment than any single-point technical breakthrough. If the industrial chain is likened to a river, design is upstream, manufacturing is midstream, and application is downstream; packaging and testing is not a specific section but the water quality of the entire river.

Third, the policy window has opened, but it will not stay open forever. The "15th Five-Year Plan" proposes process foundation building, quality assurance, lineage supply, and scenario implementation. The next five years will determine whether China's industrial sensor industry can leap from catching up to running alongside, and eventually leading in some areas. The success of this leap depends not on how many samples are made, but on how many products truly enter the installations of the five major fields (coal, electricity, gas, chemical, oil) and operate continuously for five or ten years. The value of a sensor is ultimately determined by how long it lasts in the field.


Author | Liu Qin, Former Deputy Chief Engineer of Shenyang Institute of Instrumentation Technology, Executive Director of National Engineering Research Center for Sensors, Chief Expert of Decision-Making Consultation Group of CAST, Chief Expert of Intelligent Sensor Decision-Making Consultation Team of Chinese Society of Instrumentation and Automation