5 Common Sensors and MEMS

5.1 A Clear First Route

Imagine a smart room needs to know whether it is hot, bright, noisy, occupied, or unsafe. The team must choose a sensor family for each real change. This page starts with one job. Name the thing that must be measured. Then note how heat, light, force, gas, sound, or motion becomes a signal. Look for range, response, drift, power, fit, and a known failure state. Last, choose choose a family, test it in place, or reject it. Keep the limit in view. A family name does not prove that one part will work in the final site.

5.2 Start With the Measurement Story

A smart-room prototype can sense motion, light, temperature, sound, gas, or orientation, but each family tells a different kind of story about the room. The design starts by naming the change that matters and the sensor family that can observe it reliably.

Phoebe’s Field Notes: The Thermocouple Reading That Needs a Second Number

The mathematical gist. A thermocouple measures a difference, $V=S(T_{hot}-T_{cold})$, not an absolute temperature. With the chapter’s catalog-typical Type K sensitivity of 41 µV/°C, 4.10 mV means a 100.0 °C junction difference. If the instrument’s cold junction is 22.0 °C, compensation gives $T_{hot}=122.0$ °C; ignoring that second temperature under-reads by 22 °C.

Math Bridge · guided foundationsWhy does one thermocouple need two temperatures?Let Phoebe turn Seebeck voltage into a compensated hot-junction reading. Learn the maths

5.3 Sensor Families, Not Guarantees

Common IoT sensors include temperature, humidity, pressure, light, motion, distance, position, current, gas, and contact sensors. MEMS devices add tiny mechanical structures for acceleration, rotation, pressure, microphones, and other compact measurements. These families are useful starting points, but a family name does not prove fit.

A sensor becomes acceptable when its measured quantity, operating principle, output form, range, response time, calibration need, interface, power behavior, environmental limit, and failure mode match the application decision. A familiar module can still be wrong if it cannot preserve the evidence the system needs.

The family name is only the first cut, so inspect Figure 5.1 before comparing parts. It makes the physical quantity and the typical sensing method visible together, which is the distinction the selection review needs.

Across the top of Figure 5.1, Temperature pairs thermistors, RTDs, thermocouples, and ICs with a very broad “−40°C to 1000°C+” span; that span belongs to the family, not to every device in it. Compare that with Humidity, whose “0–100% RH” label still says nothing about condensation tolerance or accuracy at the endpoints. On the lower row, Proximity groups ultrasonic, IR, ToF, and capacitive ranging even though they observe different physical effects, while Accelerometer connects MEMS and piezoelectric devices to vibration and tilt. The closing SELECTION GUIDANCE therefore supplies the correct next move: refine the family by accuracy, range, interface, and power budget, then verify operating conditions and calibration in the datasheet.

The same family can hide very different evidence burdens. A temperature reading might come from a thermistor, RTD, thermocouple, or digital IC. A motion claim might come from PIR detection, radar presence, a MEMS accelerometer, or a camera pipeline. Those options can all look familiar on a parts list while carrying different mounting, firmware, privacy, calibration, and maintenance requirements.

For that reason, treat common families as comparison bins rather than answers. First decide what the system must prove, then narrow the family by the physical coupling, signal path, interface, power state, and failure behavior that match the installed device.

5.4 Evidence-Based Comparison

A useful comparison record keeps the design from overfitting to a familiar module. It names candidate sensor families, the evidence each can preserve, what each cannot prove, and which field change would force a retest. The record should be short enough to use during procurement and specific enough to survive maintenance.

Environmental families are a good place to make the comparison concrete. For light, separate the photometric unit from the sensing element: candela describes luminous intensity, lumens describe total visible light emitted by a source, and lux describes illuminance falling on a surface. An LDR or photoresistor changes resistance with light and is useful for simple threshold or daylight-trend decisions. A photodiode produces photocurrent and is better when response speed and linear readout matter. A phototransistor adds gain, so it can be sensitive in simple circuits, but it trades off linearity, saturation behaviour, and low-light performance. A color sensor takes this further by splitting incoming light into filtered red, green, and blue photodiode channels instead of reporting one intensity, so it can support test-strip reading, color sorting, ambient-light calibration, and color-matching decisions that a single-channel LDR or photodiode cannot.

Before acting on evidence-based comparison, inspect Figure because it condenses the evidence path: the TCS3200 board shows what “splitting incoming light” means in hardware: the small central package contains an 8-by-8 filtered photodiode array, and its interface converts the selected color channel’s current into a frequency the controller can count. Photo: SergeMoutou, CC BY-SA 4.0.

Read Figure from the central optical package to the surrounding board connections, then use the adjacent array view to separate the red-, green-, blue-, and clear-filtered photodiodes. The selected channel becomes a frequency output that the controller counts; it is not already a calibrated colour name. That path connects the hardware to the comparison record: illumination geometry, integration time, surface distance, channel calibration, and saturation must be controlled before two candidates can be compared fairly.

For temperature, a thermocouple uses the Seebeck effect and survives very wide temperature spans, an NTC thermistor is cheap and sensitive in a narrower range, and an RTD is more linear and stable when the budget supports the front end. For relative humidity, capacitive, resistive, and thermal devices all report moisture by a different physical effect; their useful record should include response time, hysteresis, long-term stability, condensation exposure, and temperature compensation. These concrete family names help procurement, but the acceptance still depends on the same evidence: range, accuracy, calibration, placement, interface, power state, and retest trigger.

The mechanism matters, and so does the definition it is reporting: relative humidity is the ratio of the moisture the air is actually holding to the maximum it could hold at that temperature, so temperature compensation is part of every humidity mechanism, not an optional extra. A capacitive humidity sensor sandwiches a thin moisture-sensitive dielectric, often a polymer or metal-oxide film, between two electrodes, and capacitance rises as the film absorbs water. A resistive humidity sensor changes resistance through a salt, polymer, or other conductive medium as moisture content changes. A thermal-conductivity humidity sensor pairs two thermal elements — one sealed in dry nitrogen as a reference, one exposed to the ambient air — and compares how much heat each element loses; the exposed element's heat loss tracks the air's actual moisture content, so airflow, temperature compensation, and contamination can dominate its evidence. The record should say which mechanism is installed before assuming that one humidity module can stand in for another.

Two more common families deserve a name of their own. A Hall-effect sensor detects magnetic field strength directly and outputs a voltage that varies with the field, so pairing it with a small magnet supports contactless wheel or shaft rotation-speed measurement, proximity sensing, and positioning, and it can sense current indirectly by reading the magnetic field a current-carrying conductor produces. A microphone converts sound pressure into an electrical signal and supports audio recording, wake-word or speech-recognition front ends, and even echolocation-style distance sensing when paired with an emitted pulse.

A newer set of sensor families push the same evidence discipline into more specialized claims. An electronic nose senses the chemical concentrations behind a scent instead of a single gas type, which fits health-screening and food or safety applications that a plain gas sensor cannot resolve. A tactile sensor reports texture, friction, and thermal conductivity rather than a single force value, which is why robotic grippers and object-recognition systems use it to distinguish materials, not just detect contact. A blood-gas or blood-sugar sensor reports the composition of blood itself — arterial gas levels, pH, or glucose — for wearable and home health-monitoring claims. A brainwave sensor measures the brain's electrical activity and can localize that activity to a region inside the skull, supporting brain-computer-interface and mental-state claims. Each of these adds a calibration, safety, and interpretation burden well beyond a simple threshold, so the same acceptance discipline — range, accuracy, calibration, and retest trigger — applies even though the measured quantity is unfamiliar.

MEMS sensors deserve the same treatment. An accelerometer, gyroscope, pressure sensor, or microphone may be compact and low power, but the evidence still depends on mounting, orientation, noise, temperature behavior, packaging, sampling, and signal processing.

5.5 Common Sensor Boundary Failures

Most sensor-family failures happen at the boundary between the physical world and the digital record. The transducer may be right for the category but wrong for the mounting. The output may be digital but calibrated poorly. The reading may be stable but biased by enclosure heat, airflow, sunlight, vibration, humidity, power state, or target material.

MEMS devices make this boundary easy to miss because the sensing structure, analog front end, ADC, and compensation logic often sit inside one small package. The package may output a clean acceleration, rotation, pressure, or microphone value, but the proof mass, diaphragm, or resonant structure still reacts to stress, temperature, orientation, shock, vibration, and board mounting. Treat the package output as interpreted evidence, not raw truth.

Selection should also protect against false precision. A module may produce a number with many decimal places even when accuracy, placement, or calibration does not support that level of trust. A binary output may look decisive even when the threshold, delay, warm-up, or false-trigger behavior is poorly understood.

The under-the-hood rule is to ask what can make the common sensor lie. If that failure mode would change the application decision, it belongs in the selection record with a diagnostic, fallback, known limit, or retest trigger.

That question should include software boundaries too. Library defaults can hide averaging windows, unit conversions, compensation tables, debounce rules, and invalid-state handling. When those defaults change, the sensor family has not changed, but the evidence path has.

5.6 Match a Greenhouse Need to a Sensor Family

A greenhouse controller needs leaf-zone temperature, damp-air warning, door position, and fan vibration. Those are four different physical effects, so one attractive breakout board cannot stand in for the whole sensor plan. The family view in Figure 5.1 separates temperature, humidity, proximity, motion, light, pressure, and other sensor choices before a part number enters the discussion. In Figure, the coloured channels show that even “light” may mean several wavelength bands rather than one brightness value.

For the hot-water pipe, a Type K thermocouple producing 4.10 mV at 41 µV/°C represents a junction difference of (4100/41=100\ ^\circ\text{C}). If the cold junction is 22 °C, compensation gives (100+22=122\ ^\circ\text{C}). The same calculation would be meaningless for the door reed switch, which reports an open or closed state instead of temperature.

5.7 Summary

Start by common IoT sensors are useful families, not automatic design approvals. Then sensor choice should start with the measurement claim and the decision that will use the reading. Next environmental, motion, position, distance, gas, current, contact, and MEMS sensors differ by operating principle, output, interface, power, calibration, and field limits. After that digital output does not prove physical accuracy, and a familiar module still needs mounting, enclosure, environment, calibration, and lifecycle evidence. Finally a changed sensor, mounting, enclosure, sample rate, threshold, interface, calibration, environment, target, or application decision should reopen the sensor selection review.

5.8 See Also