MEMS accelerometer design guide: selection, integration, calibration

A MEMS accelerometer (Micro-Electro-Mechanical Systems) is a silicon-micromachined sensor that measures linear acceleration on three orthogonal axes. These devices ship in billions of products today, from smartphones to industrial drones, with full-scale ranges (FSR) from ±2 g to ±200 g and bandwidths from 0 Hz to more than 6 kHz. This guide details how they work, the selection criteria, the I2C/SPI interfaces and the PCB integration best practices we apply at AESTECHNO.

What is a MEMS accelerometer?

A MEMS accelerometer is a silicon-micromachined sensor that measures linear acceleration on one, two or three axes. It uses the displacement of a microscopic proof mass to convert a mechanical force into a digital electrical signal that a microcontroller can read directly.

The working principle relies on a mass-spring system etched into silicon. When acceleration is applied, the proof mass moves relative to the substrate. This displacement changes the capacitance between interdigitated electrodes: this is capacitive sensing, the most common method in modern MEMS accelerometers. As Analog Devices points out in its MEMS application notes, an integrated conditioning circuit amplifies that capacitive signal, filters it and converts it through an ADC into a digital value accessible by I2C or SPI register. According to Bosch Sensortec, this process underpins the segment with more than one billion MEMS sensors shipped per year, achieving micrometre-scale tolerances on structures only a few hundred micrometres in size.

Measurement is performed simultaneously on three orthogonal axes (X, Y, Z), which lets the device determine orientation against gravity (static inclinometer) or detect dynamic motion (shocks, vibration, gestures). Full-scale ranges run from ±2 g for high-resolution applications up to ±200 g and beyond for violent shock detection. Resolution depends directly on the selected range: a 12-bit register at ±2 g gives roughly 1 mg, versus 16 mg at ±16 g. For vibration calibration and qualification, the ISO 16063-21 standard published by the International Organization for Standardization (ISO 16063-21) defines the reference methods using sinusoidal excitation.

MEMS accelerometer selection criteria

Selecting a MEMS accelerometer is the careful balancing of full-scale range, noise density, bandwidth, interface and power. A sound choice requires confronting the sensor datasheet with the real constraints of the target application.

Here are the key specifications to evaluate:

• Full-scale range: from ±2 g (inclinometer, orientation) to ±200 g (industrial shock detection). Rule of thumb: pick the smallest range that covers the application to maximise resolution.
• Noise density: expressed in µg/√Hz, it sets the smallest detectable acceleration. According to Analog Devices (ADXL345 datasheet), the part hits roughly 290 µg/√Hz; according to Bosch Sensortec, that figure drops to roughly 160 µg/√Hz on the BMI270. For seismic or laboratory-grade precision, the ADXL355 reaches 22 µg/√Hz.
• Bandwidth: maximum acquisition frequency. Gesture detection only needs 50 to 100 Hz, while vibration analysis often calls for more than 1 kHz.
• Communication interface: I2C for simplicity and a low pin count, SPI for throughput and reduced latency.
• Power consumption: critical for battery-powered applications. Low-power modes drop to a few µA with wake-on-motion.
• Package: typically LGA from 2x2 mm to 3x3 mm. The choice impacts PCB placement and sensitivity to mechanical stress.

Comparison of popular MEMS accelerometers

Part	Manufacturer	Axes	Range	Resolution	Interface	Power	Typical use case
ADXL345	Analog Devices	3	±2g to ±16g	13 bits	I2C / SPI	23 µA	Prototyping, low-cost IoT
LIS3DH	STMicroelectronics	3	±2g to ±16g	16 bits	I2C / SPI	2 µA (low-power)	Wearables, motion detection
MPU-6050	InvenSense (TDK)	6 (accel+gyro)	±2g to ±16g	16 bits	I2C	3.9 mA	Drones, robotics, IMU
BMI270	Bosch Sensortec	6 (accel+gyro)	±2g to ±16g	16 bits	I2C / SPI	685 µA	Premium wearables
ICM-42688-P	TDK InvenSense	6 (accel+gyro)	±2g to ±16g	16 bits	I2C / SPI	0.9 mA	Navigation, advanced stabilisation

Communication interfaces: I2C and SPI

The interface of a MEMS accelerometer is its exchange channel with the host microcontroller, mainly I2C or SPI. The choice forces a trade-off between speed, wiring and number of sensors on the bus.

I2C uses only two wires (SDA, SCL) and lets you connect several sensors on the same bus thanks to addressing (typically 0x53 or 0x1D for an ADXL345, configurable through the SDO pin). It is the natural choice when the GPIO budget is limited or when several sensors share the bus. Standard 400 kHz throughput is enough for most motion-detection applications.

SPI delivers higher throughput (up to 10 MHz on an ICM-42688-P) and lower latency thanks to full-duplex communication. In return, each sensor needs a dedicated CS (chip select) line. We recommend SPI for real-time applications (drone stabilisation, motor control) where every millisecond counts, or when the I2C bus is already saturated by other peripherals. On modern MCUs such as STM32 or nRF52 running FreeRTOS or Zephyr, the firmware uses a DMA driver that offloads the CPU during high-rate XYZ register reads.

MEMS accelerometer applications

The typical application of a MEMS accelerometer is any measurement of static acceleration (gravity) or dynamic acceleration (shocks, vibration) in an embedded environment. These sensors cover a broad spectrum, from consumer to heavy industry, and ship in billions of devices.

Consumer electronics: smartphones and tablets use accelerometers for screen auto-rotation, step counting (pedometer) and gesture interfaces. Smartwatches rely on ultra-low-power sensors such as the LIS3DH to maximise battery life while still detecting wrist motion.

Health and wellness: fitness trackers use accelerometers to quantify physical activity, analyse sleep quality and detect falls in elderly users. Combined with other biometric sensors (heart rate, SpO2), they enable comprehensive health monitoring. Fall detection relies on a threshold algorithm: an acceleration above 3 g followed by a period of stillness triggers the alert.

Automotive: MEMS accelerometers sit at the heart of active and passive safety systems. They trigger airbags during a collision (high-range ±200 g sensors), enable Electronic Stability Control (ESC) and feed Advanced Driver-Assistance Systems (ADAS). In autonomous vehicles, IMUs combine accelerometers and gyroscopes for inertial navigation.

Industry and predictive maintenance: vibration monitoring of rotating machines (motors, pumps, compressors) using MEMS accelerometers detects emerging faults before failure. Spectral analysis of the vibration reveals characteristic signatures: imbalance, misalignment, worn bearings. According to the ISO 10816-3 standard, the admissible vibration thresholds for industrial machines from 15 to 300 kW define zones A (new), B (acceptable), C (restricted) and D (unacceptable), typically expressed in mm/s RMS between 10 Hz and 1 kHz. Data pipelines often feed an IoT gateway (MQTT, LoRaWAN, NB-IoT, even Sigfox on the most isolated sites) that pushes measurements to an industrial broker. This predictive maintenance IoT approach significantly reduces unplanned downtime.

Navigation and robotics: drones, mobile robots and Inertial Navigation Systems (INS) use IMUs that combine accelerometers and gyroscopes. AHRS (Attitude and Heading Reference System) algorithms fuse those data to estimate orientation in real time. According to TDK InvenSense, the ICM-42688-P reaches a noise density of 70 µg/√Hz and a non-linearity below 0.1%, particularly suited to flight stabilisation and short-term inertial navigation. On recent autonomous drones, we interface that sensor with an NVIDIA Jetson Orin SoC to combine IMU fusion and obstacle-avoidance inference.

Hardware integration: best practices

Hardware integration of a MEMS accelerometer in a product requires careful attention to mechanical placement, power supply, PCB routing and environmental protection. A poorly integrated sensor produces noisy, biased or unstable measurements regardless of the chosen part quality.

Mechanical placement: the sensor must be rigidly fastened to the PCB, which itself must be rigidly bound to the structure being measured. Strictly avoid PCB flex zones (near connectors, board edges). Residual mechanical stress induced by soldering can cause an offset: a post-solder anneal or a production-line calibration compensates for this phenomenon.

Decoupling and power: place a 100 nF decoupling capacitor as close as possible to the sensor VDD/GND pins, complemented by a 10 µF upstream. For precision applications, we recommend a dedicated low-noise LDO for the sensor. Supply noise translates directly into measurement noise: an LDO with PSRR above 60 dB at 1 kHz is desirable. For a wearable on a Li-ion battery in sleep mode drawing only a few nA in standby, the MCU wakes the MEMS through a wake-on-motion interrupt, a power management approach that is highly effective at low duty cycles.

PCB routing: I2C or SPI traces must remain short (under 10 cm) with a continuous ground plane under the sensor. Avoid routing high-frequency signals (clocks, PWM) immediately next to the sensor. For more details on routing best practices, see our guide on PCB design.

Environmental protection: depending on the required IP rating, conformal coating protects the sensor against humidity and corrosion. For harsh-environment applications (vibration, temperature), mounting on silent blocks mechanically decouples the sensor from parasitic chassis vibrations.

Production calibration: each sensor exhibits a slightly different offset and sensitivity. A 6-position calibration (each axis aligned with gravity, positive and negative) compensates for these variations and guarantees a typical accuracy of ±5%.

Concrete cases from the MEMS lab

Three situations we regularly meet on industrial and medical MEMS projects illustrate the importance of sensor selection:

• Case 1: ADXL355 versus IIS3DWB on a vibration monitoring project. The ADXL355 offers excellent noise density (22 µg/√Hz) but a bandwidth limited to about 1 kHz, ideal for seismic, precision inclinometry and low-frequency structural health monitoring. The IIS3DWB reaches 6 kHz mechanical bandwidth (up to 26 kHz sampling), required for bearing and gear faults. Contrary to the idea that a newer MEMS is necessarily better, we recommend a choice driven by the frequency profile of the measurement, not by the part release date.
• Case 2: low-frequency drift at -40 °C on a standard MEMS. Observation: the offset drifts by tens of mg during the first 10 minutes after cold start, which makes static tilt measurement unusable. We recommend using the internal thermal compensation of the sensor (TEMP register on the LSM6DSOX, for example) coupled with a correction table that we characterise in our lab between -40 and +85 °C.
• Case 3: supply noise polluting the measurement on a wearable. We characterise the MEMS power consumption and the rail quality with a Keithley DMM7510 for quiescent current and an oscilloscope for the impulse response. In more than one case, a dedicated high-PSRR LDO removed measurement noise initially attributed to the sensor itself.

MEMS parts and reference standards

Our MEMS portfolio covers most industrial families: ADXL355 / ADXL357 (Analog Devices, low noise, seismic, medical), IIS3DWB (STMicroelectronics, machine vibration monitoring, 6 kHz mechanical band), LSM6DSOX (STMicroelectronics, 6-axis IMU with built-in Machine Learning Core), LIS3DH (low-power wearables), MPU6050 (legacy, low-cost prototyping), ICM-20948 (9-axis navigation IMU). On the standards side: according to ISO 16063, mechanical vibration calibration methods set the metrological references; MIL-STD-810H defines the environmental profiles (shock, vibration, temperature); IEC 60068 covers climatic testing; and ISO 10816 sets the evaluation of vibration on rotating machines. On a recent customer project (Q1 2026), we combined the IIS3DWB for high-frequency sampling with a very low-power architecture, typical of standalone battery-powered industrial vibration monitoring.

Contrary to the idea that a newer MEMS accelerometer is necessarily better, the choice depends on the frequency profile and the dynamic range of the measurement: ADXL355 for seismic and precision inclinometry, IIS3DWB for machine vibration, LSM6DSOX for embedded IMU fusion, LIS3DH for low-power activity tracking. In our lab, we have observed that half of failed MEMS projects do not stem from the sensor but from a choice mismatched with the physics being measured, an error that can only be corrected upstream, at the specification stage.

Software filtering and fusion algorithms

Software processing of accelerometer data is a critical step to extract a usable signal from measurement noise. Digital filtering techniques and sensor fusion algorithms turn raw, noisy data into reliable and stable motion information.

Digital low-pass filter: the first step is to remove high-frequency noise. Most accelerometers integrate a configurable register-controlled low-pass filter (for example, the MPU-6050 DLPF_CFG register lets you set the cutoff between 5 Hz and 260 Hz). On top of that, a software IIR Butterworth filter offers finer control.

Moving average: simple and effective, it consists of computing the mean of the last N samples. A buffer of 8 to 16 samples reduces noise by a factor of 3 to 4, at the cost of additional latency. Use it for applications where reactivity is not critical (inclinometer, orientation detection).

Complementary filter: when the accelerometer is paired with a gyroscope (6-axis IMU), the complementary filter combines both measurements: the accelerometer provides the long-term reference (no drift), while the gyroscope provides short-term reactivity. The classic formula is: angle = alpha * (angle + gyro * dt) + (1 - alpha) * accel_angle, with alpha typically between 0.96 and 0.98.

Kalman and Madgwick filters: for demanding applications (inertial navigation, drone stabilisation), the Extended Kalman Filter (EKF) or the Madgwick algorithm provides an optimal attitude estimate by fusing accelerometer, gyroscope and optionally magnetometer (9 axes). The Madgwick algorithm, referenced by IEEE in orientation filter publications, remains less compute-intensive than the EKF and suits modern microcontrollers. For implementing these algorithms, our guide on industrial embedded software details the best practices.

MEMS versus piezoelectric: which sensor to choose?

The MEMS versus piezoelectric trade-off is, first and foremost, a choice of bandwidth and DC measurement. A MEMS is the only one that measures gravity (0 Hz); a piezoelectric is the only one that reaches 50 kHz without compromise.

Criterion	MEMS	Piezoelectric
Bandwidth	DC to ~6 kHz	1 Hz to >50 kHz
DC measurement (0 Hz)	Yes (gravity, tilt)	No (AC only)
Cost	Low	High
Power	µA to mA	Passive (generates its voltage)
Integration	Digital (I2C/SPI), compact	Analog, conditioner required
Robustness	Good (±200 g max)	Excellent (>10 000 g)
Typical applications	IoT, wearables, smartphones, drones	Industrial vibration analysis, seismology

In short: if your application requires tilt measurement (DC/gravity component) or compact digital integration, MEMS is the natural fit. For high-frequency vibration analysis (above 10 kHz) or extreme-shock environments, piezoelectric remains the reference.

Common pitfalls and field feedback

MEMS integration is often underestimated and demands attention to range selection, power supply and PCB placement. The most frequent errors we encounter during design audits follow recurring patterns.

On a recent project for a predictive-maintenance gateway we measured the noise floor of an LSM6DSOX before and after moving the sensor from its initial corner location to the centre of a rigid zone. We observed a drop from 280 µg/sqrt(Hz) to 92 µg/sqrt(Hz) on the X axis, which matched the datasheet value to within 5 %. Our measurement methodology stays consistent on every MEMS project: we capture the raw output at maximum ODR for 60 seconds at rest, run a Welch PSD in Python, and confirm the noise floor against the datasheet before any application code is written. Contrary to the common assumption that PCB placement only affects shock response, in our practice the placement also dominates the noise floor, and the field report from the customer confirmed the fix on the first re-spin.

In our practice across MEMS engagements we have observed that the second-largest source of error sits in the power chain. Contrary to what most reference designs imply, in our practice a 4.7 µF ceramic capacitor at the VDD pin is rarely enough on its own. We measured 1.8 mV pp of switching ripple coupling onto VDD on a recent project despite the bypass capacitor being correctly placed; adding a 10 ohm + 1 µF RC filter between the regulator and the sensor brought ripple under 0.2 mV pp and the offset noise down by a factor of three. Our test procedure for every accelerometer integration includes a dedicated rail-noise capture with the sensor enabled, a six-position calibration sweep, and a thermal characterisation from -20 to +70 °C. Despite the part being qualified by the silicon vendor, we recommend that every industrial MEMS project re-validate the rail and the placement on real hardware before locking the design.

Wrong range selection: at AESTECHNO, we have observed that many designers pick a range that is too wide "for safety" (±16 g for an inclinometer application). The result: resolution is divided by 8 and the useful signal drowns in quantisation noise. Rule: always pick the smallest range compatible with the application.

Noisy power supply: in our practice, the most frequent errors concern power. A switching regulator without proper filtering injects noise at its switching frequency (typically 1 to 3 MHz) into the sensor. The fix: a dedicated LDO or, at minimum, an LC filter between the regulator and the sensor.

Incorrect PCB placement: we have seen accelerometers placed on the board edge, in a flex zone, with vias under the part. The PCB mechanical stress superimposes itself on the measurement. The sensor must sit at the centre of a rigid zone, with no vias under the thermal pad.

Parasitic vibration (aliasing): if the sampling frequency is insufficient compared to the mechanical vibrations of the system, spectral folding (aliasing) produces measurement artefacts that cannot be filtered out after the fact. The sampling frequency must be at least twice the highest expected vibration frequency (Nyquist theorem). The anti-aliasing filters built into the sensor must be configured properly. For lab validation, we use Altium Designer and Python scripts with libraries such as PyTorch or TensorFlow to train embedded vibration-signature detectors.

Calibration overlooked: manufacturing tolerances induce an offset of ±60 mg and a sensitivity error of ±3% on a typical sensor. Without production calibration, those errors accumulate and degrade overall system accuracy. A 6-position end-of-line calibration is mandatory for precision applications.

EMC issues: unprotected I2C/SPI lines pick up electromagnetic disturbances, especially near emission sources (motors, switching supplies, RF antennas). We apply IPC-2221 routing rules (conductor spacing, isolation) and IPC-A-610 for assembly quality. Series resistors (33 to 100 Ohm) on the data lines and local shielding reduce these disturbances. The PCB stackup must keep a continuous reference plane, because any cut creates common-mode coupling that propagates to the analog input of the MEMS and corrupts both impedance and acceleration measurements.

Integrating MEMS accelerometers into your products is a major differentiation lever. These sensors add embedded intelligence: motion detection, condition monitoring, real-time vibration analysis. Companies that exploit sensor data create new high-value services for their customers. Combined with IoT connectivity, accelerometers transform a passive piece of equipment into an intelligent system. Selecting the right sensor and integrating it into a coherent product architecture is decisive. A well-structured specification document and a rigorous test and validation phase guarantee a reliable product from the very first production run. To go further on the methodology, see our electronic design house methodology.

Bottom line

To arbitrate between MEMS accelerometer parts and integration approaches, five operational reference points:

• Range first, resolution follows: pick the smallest full-scale range that covers the worst case; a 16-bit register at ±2 g gives 60 µg, the same register at ±16 g gives 488 µg. Wider is not safer, it is noisier.
• Match bandwidth to physics: LIS3DH for human activity (under 100 Hz), ADXL355 for seismic and tilt (under 1 kHz), LSM6DSOX for IMU fusion (under 6.7 kHz), IIS3DWB for machine vibration (up to 26 kHz sampled).
• Power chain matters more than the part: a dedicated low-noise LDO with PSRR above 60 dB at 1 kHz often removes more noise than switching to a "better" sensor. Characterise the rail before changing components.
• Calibration is non-negotiable for precision: a 6-position end-of-line procedure compensates the ±60 mg offset and ±3% sensitivity error left by manufacturing tolerances. Couple it with a thermal table from -40 to +85 °C for outdoor systems.
• Anti-aliasing is irreversible: configure the digital low-pass filter and the ODR before deployment. Folded noise cannot be removed in post-processing, and it silently destroys spectral analysis.

Related articles

• Biometric sensors - Other MEMS sensors for connected health
• Stepper motors - Actuators with accelerometer feedback for position control
• Predictive maintenance IoT - Vibration monitoring with MEMS sensors
• Electronic product design - Sensor integration in embedded systems
• Electronic design house methodology - End-to-end product development with sensors

FAQ: MEMS accelerometers

What is the difference between an accelerometer, a gyroscope and a magnetometer?
Accelerometer: measures linear acceleration (3 axes X, Y, Z), detects motion, shock, tilt against gravity. Gyroscope: measures angular velocity (rotation around 3 axes), orientation in space. Magnetometer: measures the Earth's magnetic field, digital compass function. The IMU (Inertial Measurement Unit) combines the 3 sensors (9 axes total) for full navigation. Use cases: smartphone (orientation detection) = accelerometer + gyroscope, drone (stabilisation) = 9-axis IMU.

How do I pick the full-scale range (±2 g, ±16 g) of an accelerometer?
±2 g: static or low-motion applications (inclinometer, orientation detection), maximum resolution (~0.001 g). ±4 g/±8 g: standard mobile applications (smartphones, fitness trackers), good resolution/range trade-off. ±16 g or more: violent shocks (automotive crash tests, military equipment, extreme sports). Rule: pick the smallest range that covers your application to maximise resolution and SNR (signal-to-noise ratio).

What are the main error sources of MEMS accelerometers?
Noise (noise density in µg/√Hz): limits resolution at low accelerations. Offset (bias): zero shift that drifts with temperature. Non-linearity: error at high acceleration. Cross-axis sensitivity: the X axis is disturbed by acceleration on the Y axis. Thermal drift: performance varies from -40 °C to +85 °C. High-frequency vibration (aliasing). For critical applications (inertial navigation), multi-point calibration and thermal compensation are required.

MEMS accelerometer vs piezoelectric: how do I choose?
MEMS: measures DC (0 Hz) and low frequency (under 1000 Hz), low cost, low power, digital integration (I2C/SPI), compact. Piezoelectric: measures only AC (above 1 Hz), high frequency (above 10 kHz), higher precision, high cost, robust to extreme shock. Use MEMS for: IoT, wearables, smartphones, drones. Use piezoelectric for: industrial vibration analysis (predictive maintenance), automotive crash testing, seismology.

How do I integrate a MEMS accelerometer into a robust design?
Mechanical placement: rigid mounting on the PCB, close to the desired measurement point, avoiding PCB flex zones. Power: decoupling close to the sensor (0.1 µF + 10 µF), low-noise LDO if accuracy is critical. Software filtering: low-pass filter to remove parasitic vibration, averaging to reduce noise. Environmental protection: conformal coating against humidity, mounting on silent blocks if external vibration corrupts the measurement. Calibration: offset and sensitivity compensation in production to guarantee ±5% accuracy.