MEMS Sensors at Industrial Scale: Bosch Sensortec, Automotive and Industrial Demand

A MEMS sensor is an odd thing to find on a bill of materials, because it is the one part of an otherwise solid-state system that is built, on purpose, to move. Logic dies, memory arrays, passives and power devices all do their job sitting still. A MEMS accelerometer or gyroscope works because a microscopic silicon structure inside it deflects, vibrates or rotates in response to the physical world, and an adjacent circuit reads that motion as a signal. A moving part made with the tolerances of a logic transistor is what makes MEMS manufacturing, packaging and testing a discipline of its own.

This note follows that thread from the wafer to the finished automotive and industrial product, using Bosch Sensortec as the worked example. The facts come from the sensors and MEMS chapter of my ESDM book. Where its sources conflict, I say so.

Two ways to carve a moving part

MEMS devices are built with two complementary micromachining disciplines, and the choice depends on where in the silicon the moving structure needs to sit.

Bulk micromachining removes material from deep inside the substrate itself. Historically that meant wet anisotropic etching with potassium hydroxide (KOH) or tetramethylammonium hydroxide (TMAH), and increasingly it means dry plasma etching. It carves cavities, membranes and cantilevers that can span a large fraction of the wafer's thickness. It is the natural route for a structure that flexes over a large area, such as the thin silicon diaphragm in a piezoresistive pressure sensor, whose resistance changes as pressure bends it.

Surface micromachining goes the other way. It builds structural layers, typically polysilicon, up and out from the wafer surface, separated from the substrate by a sacrificial oxide. A release etch then dissolves the oxide and frees a suspended structure. This is the technique behind the interlocking comb fingers of a capacitive accelerometer or gyroscope: acceleration displaces the released structure, the capacitance between fixed and moving fingers changes, and the readout circuit measures that change.

The Bosch process

A variant of bulk micromachining, deep reactive-ion etch (DRIE), was patented by Robert Bosch GmbH and is now used across the industry, which is why so much MEMS vocabulary traces back to one company. The process alternates two steps. A passivation step deposits a thin fluorocarbon polymer film (from a gas such as C4F8) on every exposed surface. An etch step uses a fluorine-based plasma (typically SF6) that removes material preferentially from the trench floor rather than the sidewalls. Cycling quickly between the two produces deep, near-vertical trenches with high aspect ratio. Per Bosch's own patent documentation the process can reach a silicon etch rate of up to 10 micrometres per minute, fast enough to make deep-trench inertial sensors commercially viable rather than a laboratory curiosity.

Packaging is half the design

Fabrication is only half the manufacturing problem. A MEMS die carries a moving element, so it cannot be encapsulated in ordinary mould compound the way a logic die can. Compound flowing over a released comb structure would simply lock it in place. MEMS instead needs wafer-level packaging that seals the moving structure before the wafer is ever diced. The options are a discrete cavity package or wafer-level vacuum packaging (WLVP), where a cap wafer is bonded to the device wafer under vacuum so the cavity holds a controlled low-pressure atmosphere.

The vacuum matters mechanically as much as it does for hermeticity. A gyroscope's suspended proof mass needs low gas damping so it can vibrate at its intended resonant frequency without air resistance swamping the signal. Packaging is therefore a second design parameter for performance, not a downstream afterthought.

Testing carries its own burden. The device under test has to be physically stimulated, shaken, pressurised or exposed to sound, rather than just electrically probed. Outsourced assembly and test houses that handle MEMS need acoustic and motion stimulus equipment that a purely digital line does not.

The economics run opposite to a shrinking logic node. On a logic wafer, a point defect usually kills one die and yield absorbs it statistically. On a MEMS wafer, a defect in the release etch or the cap bond can compromise the mechanical function of every die sharing that step, because the structure either releases cleanly and seals or it does not. There is little of the graceful, partial degradation a digital circuit sometimes tolerates. That is the reason the supplier base stayed narrow even as unit volumes reached the billions. A supplier has to hold bulk micromachining, surface micromachining, DRIE trench control and hermetic cap bonding at the same time, instead of specialising in one narrow step the way a passive-component maker can specialise in a ceramic dielectric.

The sensor taxonomy as a market map

The four sensor categories are worth restating as a market-structure question, because commercial weight, not physics, decides where fabrication and packaging investment goes.

Motion sensing is the most deeply integrated. Three-axis accelerometers and three-axis gyroscopes are routinely combined in one six-axis IMU in a single land-grid-array package, serving image stabilisation, drone control, vehicle stability and wearables from one part number. Pressure and acoustic sensing hides two manufacturing lineages under one label: MEMS microphones, which by Yole Group's acoustic data now ship at more than seven billion units a year, and piezoresistive pressure sensors, which sit in the bulk-micromachining line and range from manifold pressure sensing to thermal-runaway monitoring in EV battery packs and medical ventilators. Optical time-of-flight is grouped with MEMS commercially but is not MEMS in the strict sense. A SPAD array is a silicon photonics structure, not a moving mechanical one. It shares customers, packaging partners and design-in cycles with the rest of the portfolio, which is why it sits in the same bucket.

Supply is concentrated among a small set of integrated device makers: Bosch Sensortec (Germany), STMicroelectronics (Italy and France), TDK InvenSense (Japan and the United States), Analog Devices (United States), Infineon (Germany) and Teledyne DALSA (Canada). Yole Group puts the global MEMS industry at $15 billion to $18 billion. That is modest next to logic and memory, and it understates the strategic weight, since MEMS is the layer through which a vehicle, a factory or a phone perceives the physical world.

The same short list is also a map of where the real barrier sits. A merchant foundry can in principle license or replicate a micromachining flow. Hermetic wafer-level capping, wafer-level vacuum packaging and motion or acoustic stimulus test are harder to buy off the shelf. They need purpose-built equipment and a qualification history that a new entrant must build one product generation at a time. The fabrication step is necessary but not sufficient. Packaging and test is where the entry barrier lives. India's domestic MEMS fabrication capacity is an open data point in the book, flagged as evidence required pending a MeitY component survey, so I do not assert a figure.

Bosch Sensortec at industrial scale

Bosch Sensortec is the clearest available example of running MEMS globally, partly for its volumes and partly because its reporting shows a problem common to MEMS market data: share figures that disagree even within one source. The book's reference document has an executive summary claiming the company controls over 25% to 30% of the global consumer and automotive MEMS market. The same document's data-gaps table, sourced independently to Yole Group, says approximately 10% to 15%. Both describe the same company and they cannot be reconciled from the material provided.

What is not disputed is the manufacturing scale. Bosch's corporate data puts cumulative MEMS shipments above 18 billion units, built on the DRIE process above and deployed across smartphones, automotive stability and airbag systems, smartwatches and drones. The process flow runs from a silicon substrate wafer through DRIE, hermetic cap-wafer bonding and stacking of a CMOS signal-conditioning ASIC to a finished six-axis IMU package.

Read together, the portfolio shows how one maker can lead across such different uses. The BMI323 and BMA400 differ mainly in the power versus performance trade-off built into a similar capacitive structure. The SMA7xx pushes the same comb-accelerometer physics to a measurement range and a functional-safety level that a consumer part has no reason to carry. Same fabrication process, very different qualification bars.

Automotive demand: safety sets the pace

Per SAE International and Yole Group data cited in the book's reference, automotive electronics are the fastest-growing demand vector for semiconductors, and MEMS rides that growth directly. Semiconductor content per vehicle rises from roughly $500 in an internal-combustion vehicle to a range the sources do not agree on. The executive summary cites $1,500 to $2,500-plus for Level 3 and 4 electric vehicles, while a data-gaps table sourced to Yole Group and Strategy Analytics cites $1,200 to $2,000. The book reports the wider $1,200 to $2,500-plus span, and I follow it. Either way that is a three to five times increase over the ICE baseline.

MEMS sits inside that budget in safety-critical roles that predate any electrification trend. Capacitive accelerometers and gyroscopes sit inside electronic stability control and anti-lock braking. Piezoresistive pressure sensors sit inside tyre-pressure monitoring. High-g crash accelerometers, like the SMA7xx, trigger airbags. Electrification adds a layer rather than replacing one: battery management systems increasingly use pressure monitoring inside the pack to catch the early pressure rise that comes with thermal runaway.

These uses are qualified under the same framework as the rest of vehicle electronics: ISO 26262 functional safety up to ASIL-D for airbag and stability sensors, IATF 16949 quality management, and AEC-Q100 component qualification across -40 C to +150 C. That bar is what the SMA7xx is built to clear.

Where MEMS ends and the perception stack begins

The book's automotive reference groups several sensing types under one "sensor fusion" heading, and it is worth being exact. ADAS fusion combines 77 GHz millimetre-wave radar, LiDAR, high-resolution cameras and ultrasonic sensors, processed by accelerators such as NVIDIA DRIVE Thor, Mobileye EyeQ6 and Qualcomm Snapdragon Ride. Radar and cameras are semiconductor devices but not MEMS as defined here, because they contain no released or deeply etched moving silicon. LiDAR spans both, since some designs use MEMS scanning mirrors and others do not, and the reference does not resolve which. So inertial and pressure MEMS remain the vehicle's core motion and pressure layer, and the perception stack is an adjacent, largely non-MEMS layer on the same architecture and the same content growth curve.

That architecture is itself changing. Legacy vehicles spread more than 100 electronic control units around the car. Modern EV designs consolidate into four to six zonal controllers under a central computer, linked by automotive Ethernet (100BASE-T1, 1000BASE-T1 and 10 Gbps multi-gig) instead of CAN and LIN. MEMS sensors then report into a shared zonal controller rather than a single-function unit, which raises the premium on sensor-level diagnostic integrity. One compromised input now has to be trusted by software steering a wider slice of the vehicle.

India's automotive electronics base, per the reference, includes Sona Comstar, Tata AutoComp, Continental India, Bosch India, Kaynes Technology and Spark Minda, supplying ECUs and harnesses to Tata Motors, Mahindra, TVS and Ather Energy. India's share of global automotive ECU exports is flagged as evidence required pending an ACMA audit, and I do not assert it. My inference, which the book supports but does not state as data, is that a Tier 1 sourcing MEMS inertial and pressure sensors for an Indian platform is very likely buying from the same global product families, an SMA7xx-class crash accelerometer or an automotive-adapted BMI-series IMU, given that India's MEMS wafer capacity is an open question. India has depth in ECU integration, harnessing and assembly, while the highest-barrier layer, the MEMS die, arrives from foreign device makers.

Industrial demand: longevity sets the pace

Industrial automation, smart-factory IoT and robotics are a different kind of MEMS demand. It is resilient and high-margin instead of fast-growing, and it is governed by lifetime and environmental robustness rather than a certification regime as strict as ISO 26262. The book's industrial reference puts the global industrial automation market at roughly $200 billion to $230 billion, spanning PLCs, variable-frequency drives, edge gateways, collaborative robots and condition-monitoring nodes. It specifies a 10 to 15 year product lifecycle, operating temperatures of -40 C to +85 C, and 2.5 kV to 5 kV of galvanic isolation on I/O to survive electrical noise and vibration on a factory floor.

The clearest MEMS use is condition monitoring. A vibration accelerometer bolted to a motor, pump or bearing feeds a predictive-maintenance pipeline that flags a developing fault before it becomes unplanned downtime. It is the same capacitive accelerometer technology used for automotive stability and consumer motion sensing, repackaged for a longer service life and wider tolerance. The data travels over an industrial IoT stack of private 5G NR, sub-1 GHz LoRaWAN, NB-IoT and Wirepas mesh, with IEEE 802.1 Time-Sensitive Networking Ethernet where deterministic low latency is needed. The book notes that robotic joint controllers add encoder feedback but does not say whether those encoders are MEMS or optical or magnetic, so I make no claim there.

The global automation leaders are Siemens, Rockwell Automation, Schneider Electric, ABB, Mitsubishi Electric and Omron. India's ecosystem includes Siemens India, Schneider Electric India, Centum Electronics, Kaynes Technology and L&T Semiconductor Technologies, producing controllers, smart meters and solar inverters. India's installed base of domestic smart meters is flagged evidence required pending an EESL audit.

The isolation and temperature figures are the concrete reason a consumer or automotive part cannot simply be relabelled for industrial duty. A sensor on a motor housing sits much closer to switching noise and shock than the same sensor class in a phone or a cabin. The 2.5 kV to 5 kV isolation is there to keep that noise out of the signal path before it reaches a PLC or gateway. Combine that with a 10 to 15 year field life and an industrial-grade sensor is qualified against fewer extreme-event tests than a crash accelerometer but a far longer continuous-duty specification. It is a different point on the reliability curve, not a lesser one.

The two demand vectors pull one fabrication and packaging capability in two qualification directions. Automotive is safety-driven and rewards a supplier who can prove diagnostic integrity inside a product cycle of a few vehicle generations. Industrial is longevity-driven and rewards a supplier who can guarantee availability and stable performance over a service life several times longer. A maker that clears both, as Bosch Sensortec's span from BMA400 to SMA7xx suggests, is running two qualification businesses on one shared process.

What to take away

• MEMS manufacturing is defined by a moving part. Bulk and surface micromachining, DRIE trench control and hermetic cap bonding all have to work together, and a defect at release or bonding can take out every die on the wafer that shares the step.
• The entry barrier is packaging and test as much as fabrication: wafer-level vacuum capping and motion or acoustic stimulus testing are hard to acquire and slow to qualify.
• Bosch Sensortec's market share is somewhere between 10% and 30% depending on the definition, and the source itself contradicts itself. The undisputed number is more than 18 billion cumulative units.
• Automotive demand is qualification-driven (ASIL-D, AEC-Q100, IATF 16949) and content per vehicle of $1,200 to $2,500-plus against about $500 for ICE. Industrial demand is lifetime-driven, at 10 to 15 years and -40 C to +85 C.
• For India, the gap is the die. Integration and assembly are strong, and the MEMS wafer capacity figure is still an open data point.

The physics inside the package does not change between an airbag controller and a factory motor. Everything wrapped around it does, and that wrapping is where a supplier earns its place.