MEMS — Micromachining, Sensors & Actuators
1. At a glance
MEMS (Micro-Electro-Mechanical Systems) are devices that integrate mechanical structures — beams, diaphragms, proof masses, channels, mirrors — with electronic circuitry on a silicon substrate at the micron scale (characteristic dimensions 1–1000 µm; minimum features typically 0.5–5 µm; typical die areas 0.5–25 mm²). They are fabricated by adapting the planar lithographic process developed for integrated circuits: oxidation, deposition, photolithography, etching, and — uniquely for MEMS — release steps that remove sacrificial layers to free the mechanical element.
The 2026 MEMS market is roughly USD 20–25 billion, dominated by four high-volume product families:
- Inertial sensors — accelerometers, gyroscopes, magnetometers, IMUs. ~25 % of revenue. Every smartphone (ICM-42688, BMI270, LSM6DSO), every car airbag (ADXL series), every drone, every VR headset.
- Pressure sensors — manifold-absolute-pressure (MAP), tire-pressure (TPMS), barometric altimeters, blood-pressure cuffs, industrial. ~15 %. Bosch BMP series, ST LPS, Sensata, Honeywell.
- MEMS microphones — capacitive backplate-diaphragm transducers. ~15 %. Knowles, Infineon, Goertek, AAC, TDK InvenSense. ~4–6 per smartphone; ~10–20 per smart-speaker beamforming array.
- RF MEMS — SAW, BAW, FBAR filters and switches. ~25 %. Qorvo, Broadcom, Skyworks, Murata. Every 4G/5G radio has ~30–80 acoustic filter resonators between antenna and transceiver.
Emerging high-growth segments: micromirrors for LiDAR and AR/VR display (TI DLP, Mirrorcle, STMicro projection-MEMS, Microvision), ultrasonic transducers (pMUT, CMUT — replacing PZT in medical ultrasound and fingerprint sensors), environmental gas sensors (Bosch BME688, Sensirion SGP), micropumps for drug delivery, BioMEMS (Utah neural electrode arrays, Neuralink threads, microfluidic lab-on-chip), and THz/mmWave passive components for 6G research.
The defining characteristic of MEMS is mechanical motion at micron scale — and the corollary that classical macroscopic intuitions about gravity, inertia, friction, and damping mostly fail at that scale. Surface forces dominate volume forces; air becomes molasses; thermal Brownian motion is a noise source; pull-in instability lurks in every electrostatic actuator. MEMS design is therefore as much applied surface-physics as it is mechanical engineering.
2. Why it matters
MEMS is the bridge that made consumer electronics physical. Before the late-1990s, an “electronic device” was a black box that ran a program; sensing the outside world meant external transducers, analog signal conditioning, and discrete components. MEMS collapsed accelerometer + ASIC + package into 2 mm × 2 mm × 1 mm at $0.30 in volume. The downstream consequences:
- Smartphones (2007–): 8–15 MEMS sensors per device — accel, gyro, mag, baro, multiple microphones, ambient-light, proximity, sometimes humidity/temperature/gas. Without MEMS, neither screen-rotation nor turn-by-turn navigation works.
- Automotive (1980s–): 50–100 MEMS sensors per vehicle by 2026 — MAP for engine management, accels for airbag/ESC, gyros for rollover detection and lane-keep, TPMS pressure, oxygen sensors, microphones for ANC and voice. The airbag accelerometer (ADXL50, 1991) was the first MEMS device sold in mass volume.
- Wearables and IoT: accel + PPG + temp + skin-impedance in every watch; environmental sensors in every smart-home device.
- Industrial (CbM, condition monitoring): vibration accels for predictive maintenance; pressure for hydraulics; flow for HVAC.
- Medical: pacemaker accels (rate-responsive pacing), insulin micropumps, MEMS-based DNA sequencing (Oxford Nanopore), neural probes.
- 5G / Wi-Fi 6E / Wi-Fi 7: BAW/FBAR acoustic filters at every band of every radio. Without MEMS RF filtering, the spectral congestion of modern wireless would be unmanageable.
The flip side: MEMS combines the worst constraints of semiconductor manufacturing (yield, thermal sensitivity, batch processing) with the worst of mechanical engineering (fatigue, fracture, stiction, packaging-induced stress). A MEMS development program typically takes 5–8 years and USD 50–200M to bring a new device to qualified mass production. The barrier to entry is what keeps the field consolidated to ~15 major foundries worldwide.
3. First principles
Scaling laws — why MEMS physics is non-intuitive
Scale all linear dimensions by a factor s (s ≪ 1 for miniaturisation). Forces scale by the geometric power of their underlying mechanism:
| Force | Scaling | At s = 10⁻³ (mm → µm) |
|---|---|---|
| Gravity / inertia | s³ (mass ∝ volume) | 10⁻⁹ |
| Friction (Coulomb, ∝ normal force) | s² | 10⁻⁶ |
| Electrostatic (∝ area × E²) | s² | 10⁻⁶ |
| Magnetic (∝ volume) | s³ | 10⁻⁹ |
| Capillary / surface tension (∝ length) | s¹ | 10⁻³ |
| Van der Waals (∝ area) | s² | 10⁻⁶ |
Surface and electrostatic forces fall three orders of magnitude slower than gravity/inertia under miniaturisation. At µm scale, surface tension is enormous (a water droplet can lift a 100 µm cantilever), electrostatic actuation becomes practical at 10–50 V (impossible at macro scale because of dielectric breakdown of the air gap), and magnetic actuation becomes impractical (no room for windings or permanent magnets of useful volume).
Reynolds number for fluid flow in a 10 µm channel at 1 mm/s is Re ≈ 0.01 — pure laminar Stokes flow with no turbulence. Knudsen number Kn = λ/L (λ = mean free path, ~70 nm in air at STP) approaches unity in narrow MEMS gaps; air no longer behaves as a continuum and slip-flow corrections to Reynolds equation apply.
Spring-mass-damper at the micron scale
The basic MEMS mechanical element — proof mass m supported by a flexure of stiffness k, immersed in a fluid (usually air) providing damping b:
m·ẍ + b·ẋ + k·x = F(t)
Resonant frequency ω_n = √(k/m). Quality factor Q = √(k·m)/b. For a typical capacitive accelerometer:
- m ≈ 10⁻¹⁰ to 10⁻⁶ kg (100 pg to 1 µg)
- k ≈ 0.1 to 100 N/m (set by flexure geometry; silicon E = 170 GPa)
- ω_n ≈ 10⁴ to 10⁶ rad/s (1.6 to 160 kHz natural frequency)
For a gyro tuning-fork mode at 20 kHz, m ≈ 1 ng implies k = m·ω_n² = 10⁻¹² · (1.26 × 10⁵)² = 0.016 N/m. The flexure is a hair-thin silicon beam ~2 µm wide × 50 µm long.
Squeeze-film damping
A proof mass moving normal to a fixed substrate at small gap h pushes air out through the narrow gap; this dominates b in most parallel-plate MEMS. The Reynolds-equation result for a rectangular plate of length L, width W, gap h, viscosity µ:
b_squeeze ≈ 0.42 · µ · L · W³ / h³
At h = 2 µm, W = 1 mm, L = 1 mm, µ_air = 1.85 × 10⁻⁵ Pa·s:
b ≈ 0.42 · 1.85e-5 · 1e-3 · (1e-3)³ / (2e-6)³ = 0.97 N·s/m
That damping is enormous relative to typical k ≈ 1 N/m — Q is small (~1–10) in air, which is desirable for accelerometers (flat low-frequency response) and fatal for gyros (Coriolis signal is proportional to Q). Solution: gyros are vacuum-encapsulated at ~10 Pa or below, where damping is set by gas kinetics rather than viscous flow and Q rises to 1000–50000.
Thermomechanical (Brownian) noise
Equipartition theorem assigns ½ k_B T of energy to every degree of freedom. The fluctuation-dissipation theorem gives the spectral density of force noise on a damped mass:
S_F = 4·k_B·T·b [N²/Hz]
Equivalent acceleration noise density:
a_n = √(S_F)/m = √(4·k_B·T·b)/m [(m/s²)/√Hz]
For m = 100 µg, b giving Q = 10 at ω_n = 16 kHz (b = √(k·m)/Q = √(1·1e-7)/10 = 31.6 µN·s/m) at T = 300 K:
a_n = √(4·1.38e-23·300·3.16e-5) / 1e-7 = √(5.23e-25) / 1e-7 = 7.23e-13 / 1e-7 = 7.23 µg/√Hz
Below this floor no readout can improve the SNR — it is set by thermal physics. Commercial consumer accels run 50–200 µg/√Hz (ASIC-noise-limited); navigation-grade MEMS push to 5–20 µg/√Hz (Brownian-floor-limited).
Capacitive sensing
Two parallel plates of area A separated by gap g have capacitance C = ε₀ε_r·A/g. Small displacement Δx of one plate changes C by:
ΔC/C ≈ Δx/g
Differential sensing (comb fingers — one moving electrode between two fixed) doubles the sensitivity and rejects common-mode. Typical comb accel sensitivity: 0.1–1 fF per g of acceleration. ASICs read this with a switched-capacitor charge amplifier or a continuous-time chopper-stabilised front-end, both achieving ~aF-level noise floors.
Piezoresistive sensing
Strain on a doped silicon resistor changes its resistivity (and slightly its geometry). The fractional resistance change is ΔR/R = G·ε where G is the gauge factor. For lightly-doped p-type silicon along ⟨110⟩: G ≈ 50–100 (vs G ≈ 2 for a metal foil strain gauge). The gauge factor decreases with doping and with temperature (TCR adds an unwanted signal). Piezoresistive readout needs a Wheatstone bridge (4 elements, 2 in tension and 2 in compression) to cancel temperature and excitation drift.
Piezoelectric sensing and actuation
A piezoelectric thin film (AlN, ScAlN, ZnO, PZT) generates a voltage proportional to applied strain, or a strain proportional to applied voltage. Coupling coefficients:
| Material | d₃₃ (pC/N) | k_t² (%) | TCF (ppm/°C) | MEMS use |
|---|---|---|---|---|
| AlN | 3.4 | 6.5 | −25 | FBAR, RF resonators, pMUT, microspeaker |
| ScAlN (Sc 30 %) | 14 | 26 | −18 | High-coupling FBAR (5G n77/n79) |
| ZnO | 5–10 | 8 | −60 | Earlier RF MEMS; less common today |
| PZT (sputtered) | 60–250 | 35 | varies | High-strain actuators; not CMOS-compatible |
| LiNbO₃ | (k² up to 30 %) | 30 | −94 | SAW, single-crystal piezo |
Piezoelectric devices are AC-only (no DC response — charge leaks off) and have a small but nonzero pyroelectric effect (signal from temperature change). AlN-on-CMOS is the dominant integrated piezo MEMS technology in 2026 because aluminium nitride is CMOS-compatible (no lead, no Pb-process contamination), thermally stable, and patterns cleanly with Cl-based plasma.
Electrostatic actuation and pull-in
Two parallel-plate electrodes of area A separated by gap g with voltage V across them generate attractive force:
F_e = ε₀ · A · V² / (2 · g²)
The mass-spring system reaches equilibrium where F_e = k·Δx; for one third of the original gap, the system becomes unstable — F_e grows faster than the restoring spring force can compensate — and the moving plate snaps to the fixed one. The pull-in voltage:
V_pi = √(8·k·g₀³ / (27·ε₀·A))
This is the fundamental limitation of every gap-closing electrostatic actuator and the destruction mode of any device whose drive amplifier doesn’t current-limit. Comb-drive actuators avoid pull-in by moving parallel to the fixed electrodes (constant gap, force proportional to V² but independent of x within the working range) — invented by Tang and Howe in 1989 and still the default for in-plane MEMS actuation.
Other actuation principles
- Electrothermal (Joule heating of differential beams): large stroke (5–50 µm), large force (mN), slow (ms), high power (mW–W). Used in MEMS optical switches, microvalves.
- Magnetic / Lorentz force: F = B·I·L. Requires either external permanent magnet (DLP micromirrors used to combine Hall + magnet) or on-chip coil (low B, low force). Generally unattractive at MEMS scale.
- Shape-memory alloy (SMA): TiNi thin film; phase transition at ~70 °C produces 5 % strain. Slow, high-stroke; used in microgrippers and emerging surgical tools.
- Pneumatic / hydraulic: external pressure source; large force, slow; used in microfluidic valves and pumps.
- Piezoelectric: sub-nm to µm displacement, kHz–MHz bandwidth, high force, low stroke. Dominant in scanning-probe microscopes and high-precision MEMS positioners.
4. Micromachining processes
MEMS fabrication overlaps heavily with CMOS but adds release steps (sacrificial-layer removal) and high-aspect-ratio etch steps (DRIE) that ordinary IC fabs do not run. Three process families dominate.
Bulk micromachining
The bulk of the silicon substrate itself becomes the mechanical structure. Etching is done through-wafer or to controlled depth.
| Etch | Selectivity | Aspect ratio | Sidewall | Use |
|---|---|---|---|---|
| KOH (potassium hydroxide), 30 %, 80 °C | {100}/{111} ≈ 400:1 | ~7:1 limited by {111} angle | Pyramidal 54.7° | Diaphragms, V-grooves, nozzles, fluidic chambers |
| TMAH (tetramethylammonium hydroxide), 25 %, 90 °C | {100}/{111} ≈ 30:1 | ~7:1 | Pyramidal 54.7° | CMOS-compatible alternative to KOH (no K contamination) |
| HF / HNO₃ / acetic (isotropic Si etch) | Non-selective | n/a | Rounded | Spherical cavities, polishing |
| XeF₂ (gas-phase isotropic Si etch) | Si/SiO₂ ≈ 1000:1 | n/a | Rounded | Sacrificial Si removal under structures; vapour-phase, no stiction |
| DRIE Bosch process (SF₆ etch + C₄F₈ passivation, alternating) | Si/oxide ≥ 100:1 | 30:1 typical, 100:1 demoed | Vertical (scalloped ~100 nm) | Through-wafer trenches, high-AR features, deep cavities |
| DRIE cryogenic (SF₆ + O₂ at −110 °C) | Si/resist ≥ 100:1 | 40:1 | Vertical, smooth | Lower scalloping than Bosch; harder to control |
| RIE (CHF₃ / SF₆ at room T, single chemistry) | moderate | < 10:1 | ~80° tapered | Patterning thin films |
The Bosch process (Lärmer/Schilp, Robert Bosch GmbH, US patent 5501893, 1996) alternates a few seconds of SF₆ plasma (isotropic Si etch) with a few seconds of C₄F₈ plasma (deposits Teflon-like passivation). The passivation on the trench sidewalls protects them while ion bombardment clears it from the trench floor, where SF₆ then etches. The result is a vertical sidewall with ~100 nm “scalloping” from each cycle. Bosch DRIE is the enabling technology for nearly all modern MEMS — without it, deep narrow trenches in silicon are unobtainable.
SOI (Silicon-On-Insulator) wafers — a thin “device” silicon layer (2–100 µm) on a buried oxide layer (1–4 µm) on a thick handle wafer — let DRIE stop precisely on the oxide, giving thickness control to ±0.5 µm across the wafer. This is the dominant substrate for high-precision MEMS (gyros, micromirrors, RF resonators).
Surface micromachining
A sacrificial layer (typically PSG/oxide or polysilicon) is deposited and patterned on the substrate, followed by a structural layer (typically polysilicon or metal) that anchors to the substrate through holes in the sacrificial. A wet or vapour etch then removes the sacrificial layer, releasing the structural layer to move freely.
Sandia’s SUMMiT V process is the classic reference — five layers of polysilicon (each ~2 µm thick) with four sacrificial oxide layers between them, allowing complex 3D mechanical structures (gear trains, rack-and-pinion, mirrors with hinges). Released by HF vapour or supercritical CO₂ to avoid stiction.
The stiction problem is the central pathology of surface micromachining: during a wet release, capillary forces from the receding liquid pull freed structures against the substrate, where van der Waals + hydrogen bonding hold them down permanently. Mitigations:
- Supercritical CO₂ drying — replaces the rinse with liquid CO₂, then raises it past the critical point (31 °C, 7.4 MPa) where surface tension vanishes; release achieved without capillary collapse.
- Vapour HF release — never wets the structure; standard for industrial-volume surface MEMS.
- Anti-stiction coatings — self-assembled monolayers (SAM, e.g. octadecyltrichlorosilane OTS, or perfluorinated FOTS) that lower surface energy and prevent re-adhesion if shock causes momentary contact.
- Geometric stiction prevention — dimples on the underside of moving structures reduce contact area below the van der Waals threshold.
LIGA and X-ray micromachining
LIGA (Lithographie, Galvanoformung, Abformung — German for lithography, electroplating, moulding; developed at KfK Karlsruhe, 1980s) uses synchrotron-source X-ray lithography of thick (100 µm – 1 mm) PMMA resist, then electroplates metal (Ni, Cu, Au) into the developed pattern. Produces all-metal high-aspect-ratio (≥ 100:1) structures with smooth vertical sidewalls — used for gears, microneedles, RF transmission lines, and (originally) reactor-monitoring devices. The synchrotron requirement makes it niche; UV-LIGA with SU-8 photoresist replaces the synchrotron at the cost of slightly lower aspect ratio and lower sidewall verticality.
Wafer bonding
Permanently joining two (or more) wafers — to encapsulate MEMS cavities, to laminate dissimilar materials, to build 3D-stacked devices.
| Method | Temperature | Materials | Use |
|---|---|---|---|
| Anodic (Si–glass with high V) | 300–450 °C, 500–1000 V | Si to borosilicate (Pyrex 7740) | Pressure sensor caps; sealed cavities |
| Si fusion (direct) | 1000–1200 °C anneal | Si–Si or Si–oxide | SOI wafer manufacture; gyro encapsulation |
| Au–Au thermocompression | 300–400 °C + force | Au-on-Au | Vacuum-sealed cavities for resonators |
| Cu–Cu hybrid bonding | 200–300 °C | Cu pads | 3D IC stacking; backside-illuminated CMOS image sensors |
| Eutectic (Au–Si, Al–Ge, In–Au) | 380, 425, 156 °C | Au+Si liquidus | Hermetic seals at modest temperature |
| Glass-frit | 400–450 °C | Lead-glass paste | Low-cost MEMS encapsulation; not vacuum-tight |
| Adhesive (BCB, polyimide) | 200–300 °C | Polymer | Low-T, non-hermetic; flex MEMS |
| Plasma-activated room-T fusion | 200–400 °C anneal | Si, oxide | Modern SOI; reduces thermal budget |
Through-silicon vias (TSV)
Vertical electrical interconnect through the wafer, enabling chip stacking. TSV diameter typically 5–50 µm, depth 50–300 µm (often the full wafer thickness after backside thinning). Fabrication: DRIE the via, line with isolation oxide, deposit barrier (Ta/TaN) + seed (Cu), electroplate Cu fill, CMP planarise. Critical for 3D MEMS-ASIC integration (e.g. CMUT arrays for ultrasound, where the ASIC sits beneath the transducer array and connects through TSVs).
Wafer-level packaging (WLP)
A capping wafer with etched cavities is bonded to the device wafer at the wafer level — every MEMS die is packaged simultaneously, before dicing. The cap-wafer bond can be done in vacuum (sub-Pa for resonator-grade gyros) or in controlled atmosphere (5 kPa for damping-controlled accels). WLP cuts package volume by 5–10×, enables hermeticity without ceramic packages, and drove the cost-down curve that put MEMS in smartphones.
Foundries and shared multi-project wafer (MPW) services
| Foundry | Process specialities | Notes |
|---|---|---|
| Bosch Reutlingen | Auto MEMS (accel, gyro, MAP, TPMS) | Vertically integrated with Bosch sensor brand |
| STMicroelectronics (Catania, Crolles, Agrate) | Smartphone IMU, mic, MAP | Captive + foundry |
| Murata | Pressure, mic, RF MEMS | Acquired VTI Technologies |
| TDK / InvenSense | IMU, mic | Acquisition trail (TDK→InvenSense→Chirp) |
| Robert Bosch (Charleston, USA) | Auto MEMS Tier-1 | 8” + 12” lines |
| X-FAB (Erfurt, Itzehoe, Sarawak) | Open MEMS foundry (analog/mixed-signal + MEMS PDK) | Multi-customer |
| Silex Microsystems (Sweden) | Open MEMS foundry; piezo, RF | TSV specialist |
| TSMC | CMOS-MEMS integration | 0.18 µm MEMS PDK |
| GlobalFoundries (Singapore) | 350 nm MEMS-CMOS | BAW filter co-fab |
| Sony (Kumamoto, Yokohama) | Image sensors; some MEMS | BSI CMOS image-sensor leader |
| MEMSCAP (Bernin, France) | MPW shared runs (PolyMUMPs, MetalMUMPs, PiezoMUMPs, SOIMUMPs) | Academic + startup entry point |
| CMC Microsystems / CMP / IMT | MPW broker for university designs | Multi-foundry aggregator |
A typical MEMSCAP PolyMUMPs shared-MPW run costs USD 5–15 K for ~1 cm² of die area on a shared wafer — the standard entry point for university research and small-volume MEMS startups.
5. MEMS device families
Inertial sensors
Accelerometers measure linear acceleration along one or more axes. The proof mass deflects under inertial force F = m·a; deflection is sensed capacitively (differential comb fingers) or piezoresistively (doped silicon strain gauges on the flexure).
| Class | Bias instability | Noise density | g range | Vendors |
|---|---|---|---|---|
| Consumer (smartphone, wearable) | 20–100 µg | 100–250 µg/√Hz | ±2/4/8/16 g | Bosch BMA, ST LIS2DH, TDK ICM, Kionix |
| Automotive airbag / ESC | 5–20 mg | 100–500 µg/√Hz | ±10/100/250 g | ADI ADXL, Bosch SMI |
| Industrial / CbM | 5–25 µg | 25–60 µg/√Hz | ±5/30 g | ADI ADIS, Murata, Silicon Designs |
| Tactical / navigation | 0.5–5 µg | 5–25 µg/√Hz | ±5–15 g | Honeywell QA, Innalabs, Northrop Grumman |
| Seismic / gravity | 0.01–0.1 µg | < 5 µg/√Hz | ±2 g | Sercel, Sercel ION, custom |
Gyroscopes measure angular rate using the Coriolis effect: a vibrating proof mass under rotation experiences a force F = −2m(Ω × v) perpendicular to both rotation axis and drive velocity. Tuning-fork resonators are the dominant topology; ring or disk resonators emerging for high-performance applications.
| Class | Bias instability | ARW | Vendors |
|---|---|---|---|
| Consumer (phone, gaming) | 5–25 °/hr | 0.05–0.3 °/√hr | Bosch BMG, ST L3GD, TDK ICM |
| Auto ESC / rollover | 0.5–10 °/hr | 0.02–0.1 °/√hr | Bosch SMG, ADI ADXRS |
| Industrial / unmanned | 0.1–1 °/hr | 0.005–0.05 °/√hr | ADI ADIS, Honeywell HG, Silicon Sensing |
| Tactical | 0.01–0.1 °/hr | 0.001–0.01 °/√hr | Honeywell HG4930, Northrop LN-200 |
| Navigation (HRG, FOG, RLG) | 0.0001–0.01 °/hr | < 0.001 °/√hr | Northrop SIRU, KVH, Honeywell GG1320 |
Magnetometers — Hall (AKM, ams), AMR (Honeywell), GMR/TMR (NVE, MultiDimension). Integrated in most consumer IMUs to provide heading reference.
Pressure sensors
A thin silicon diaphragm bows under differential pressure. Sensed by piezoresistors at the flexure regions (max stress) or capacitively (deflection-vs-fixed electrode). Standard configurations: absolute (vacuum reference under diaphragm), gauge (atmospheric reference), differential (two ports). Typical ranges 0–100 kPa (barometric) to 0–250 MPa (hydraulic). Bosch BMP series, ST LPS, Honeywell HSC, Sensata for industrial.
MEMS microphones
A capacitive transducer: thin silicon-nitride or polysilicon diaphragm above a perforated backplate, separated by 1–4 µm air gap. Acoustic pressure deflects the diaphragm; ASIC reads ΔC and outputs analog (audio amplifier) or digital (PDM/I²S) signal. Sensitivity −38 to −42 dB re 1 V/Pa, SNR 65–74 dBA, AOP (acoustic overload point) 130–140 dB SPL. Knowles, Infineon, Goertek, Cirrus Logic, TDK. Dominated phone market since iPhone 4 (2010).
RF MEMS
- SAW (Surface Acoustic Wave) filters — interdigital metal transducers on piezoelectric substrate (LiNbO₃, LiTaO₃) launch and receive Rayleigh waves. Frequency limit ~3 GHz; low cost; used at sub-3 GHz cellular bands. Murata, TDK, Skyworks.
- BAW (Bulk Acoustic Wave) / FBAR (Film Bulk Acoustic Resonator) — AlN or ScAlN thin film between two electrodes; resonates at thickness mode (1.5–8 GHz). Higher power handling, better stopband rejection, smaller die than SAW. Qorvo (Avago heritage), Broadcom, Akoustis, Resonant. Every 5G FR1 phone contains 30–60 FBAR filters and duplexers.
- MEMS switches — cantilever or shunt-bridge that mechanically closes/opens a transmission line. Sub-µs switching, near-zero insertion loss, isolation > 40 dB to mmWave. Analog Devices/Hittite ADGM, Menlo Micro. Long-promised, low-volume; finally entering 5G base-station beamforming.
- MEMS varactors — variable-gap capacitors for antenna matching. Cavendish Kinetics (Qorvo), WiSpry.
Optical MEMS
- DMD (Digital Micromirror Device) — Texas Instruments DLP. Each pixel is a 7.6 µm aluminium mirror on a torsional hinge that tilts ±12° between two stable positions. Up to 8 million mirrors per chip at 60 Hz. Used in cinema projectors, business projectors, AR HUDs, 3D printers (DLP-stereolithography), and direct-write lithography. Hornbeck (TI, 1987–1989) original development.
- Scanning micromirrors — 1- or 2-axis MEMS mirrors for LiDAR, laser projection, retinal scanning, optical coherence tomography. Mirrorcle Technologies, Microvision (now closed/acquired), Hesai (MEMS-flash LiDAR), STMicro projection. Quasi-static (low-frequency, large angle) or resonant (kHz, larger fields-of-view).
- Optical switches and VOAs — fibre-network components by Calient, Polatis, Lumentum.
Microfluidic MEMS
Channels, valves, pumps, mixers, droplet generators — covered in detail in [[Engineering/microfluidics]] (companion note). Key MEMS technologies: glass-PDMS bonding, deep-RIE silicon channels, piezo / electromagnetic / pneumatic micropumps, PCR-on-chip, droplet microfluidics for single-cell sequencing.
Energy harvesters
- Piezoelectric vibration — AlN cantilever harvesting 10 µW – 1 mW from machine vibration at 50–200 Hz. MIDÉ Volture, perpetual-mobile prototypes.
- Electromagnetic — moving-magnet coil; better for low-frequency motion (human gait, swing).
- Electrostatic — variable-capacitance harvesters; need pre-charge (electret). Niche.
- Thermoelectric MEMS — Bi₂Te₃ thin-film couples; on-chip thermal-to-electric, body-heat harvesting for wearables.
BioMEMS
- Lab-on-chip — PCR, capillary electrophoresis, droplet sequencing. Agilent Bioanalyzer, Fluidigm.
- Neural electrodes — Utah array (Blackrock Microsystems, 96-shank silicon probes), Michigan probes, Neuralink threads. For BCI research and clinical use.
- Microneedle arrays — silicon or polymer; transdermal drug delivery, vaccination.
- DNA sequencing MEMS — Oxford Nanopore solid-state nanopores; Illumina flow cells.
Environmental sensors
- Temperature / humidity — Sensirion SHT, Bosch BME, TI HDC.
- MOX gas sensors (Bosch BME688, Sensirion SGP, ams CCS) — heated SnO₂/WO₃ films change resistance with VOC/CO/NOx exposure.
- Photoacoustic CO₂ — Sensirion SCD4x; MEMS resonator detects pressure modulation from optically-pumped CO₂.
6. Worked examples
Example A — Capacitive accelerometer sensitivity (consumer-grade design)
Design a ±2 g capacitive accelerometer with DRIE-fabricated silicon proof mass:
- Proof mass m = 100 µg = 1 × 10⁻⁷ kg (rectangular slab 1 mm × 1 mm × 50 µm of Si, ρ = 2330 kg/m³ gives m = 1.165 × 10⁻⁷ kg — close).
- Comb fingers: N = 20 pairs of differential capacitors, each finger 200 µm long × 4 µm thick × 50 µm high (out-of-plane), gap g₀ = 2 µm.
- Single-finger nominal C₀ = ε₀·A/g₀ = 8.85e-12 · (200e-6 · 50e-6) / 2e-6 = 8.85e-12 · 1e-8 / 2e-6 = 44.3 fF. Total static C = 20 × 44.3 fF = 886 fF ≈ 0.89 pF.
- Spring stiffness from four parallel folded-beam flexures, each beam 200 µm long × 2 µm wide × 50 µm thick (out-of-plane), Si E = 170 GPa: k_beam = E·b·h³/L³ = 170e9 · 50e-6 · (2e-6)³ / (200e-6)³ = 170e9 · 50e-6 · 8e-18 / 8e-12 = 8.5 N/m per beam. Four in parallel: k = 34 N/m.
- Natural frequency ω_n = √(k/m) = √(34/1e-7) = √(3.4e8) = 18.4 krad/s; f_n = 2.93 kHz.
Under 1 g = 9.81 m/s² acceleration, deflection x = m·a/k = 1e-7 · 9.81 / 34 = 28.9 nm.
Differential ΔC per finger pair: ΔC = ε₀·A·(1/(g₀−Δx) − 1/(g₀+Δx)) ≈ 2·ε₀·A·Δx/g₀² = 2 · 8.85e-12 · 1e-8 · 28.9e-9 / (2e-6)² = 1.28 fF for all 20 pairs combined ≈ 25.6 fF per g. ΔC/C ≈ 25.6/886 ≈ 2.9 %/g — easily readable.
Bandwidth (in air, Q ≈ 1 at squeeze-film damping): −3 dB at f_n ≈ 3 kHz, flat response from DC to ~1 kHz. Brownian-noise floor with this geometry: ~20 µg/√Hz; ASIC-limited noise typically ~80 µg/√Hz. Total RMS noise in 200 Hz BW: 80 µg · √200 ≈ 1.1 mg — consistent with consumer-grade spec.
Example B — Pressure-sensor diaphragm deflection
Square silicon diaphragm: side 2a = 1 mm (a = 0.5 mm), thickness h = 20 µm, fabricated by KOH backside etch of a {100} Si wafer. Material: single-crystal Si, E = 170 GPa, ν = 0.064 (anisotropic; using isotropic approximation E* = E/(1−ν²) = 170.7 GPa).
Centre deflection under uniform pressure P (clamped square diaphragm, small-deflection theory):
w_max = 0.0138 · P · (2a)⁴ / (E·h³)
At P = 100 kPa (1 atm differential, typical absolute-pressure sensor full-scale):
w_max = 0.0138 · 1e5 · (1e-3)⁴ / (170e9 · (2e-5)³) = 0.0138 · 1e5 · 1e-12 / (170e9 · 8e-15) = 1.38e-9 / 1.36e-3 = 1.02 × 10⁻⁶ m = 1.02 µm
Maximum bending stress (at clamped edge mid-side):
σ_max ≈ 0.308 · P · (2a)² / h² = 0.308 · 1e5 · 1e-6 / 4e-10 = 76.9 MPa
Well below silicon fracture strength (~1 GPa for high-quality DRIE Si; the diaphragm has a fatigue safety factor of ~10).
Piezoresistive readout: a p-type silicon resistor along ⟨110⟩ at the high-stress edge sees ΔR/R = π_l·σ_long + π_t·σ_trans, with π_l ≈ 71.8 × 10⁻¹¹ Pa⁻¹ and π_t ≈ −66.3 × 10⁻¹¹ Pa⁻¹ at moderate doping. With σ_long ≈ 77 MPa, σ_trans ≈ 0:
ΔR/R ≈ 71.8e-11 · 77e6 ≈ 5.5 %/100 kPa = 0.055 %/kPa
A Wheatstone bridge with 5 V excitation gives V_out = 0.055 % · 5 V = 2.75 mV per kPa — needs ASIC amplification (gain ~500–1000) and 12–16 bit ADC for useful resolution.
Example C — MEMS gyro Coriolis signal
Tuning-fork resonator at drive frequency f_d = 20 kHz; two proof masses each m = 1 ng = 1 × 10⁻¹² kg; drive amplitude x₀ = 5 µm; sense-mode resonance at f_s = 20.1 kHz (slight detuning) with Q_s = 1000 in vacuum.
Drive velocity amplitude: v_d = 2π·f_d·x₀ = 2π · 20e3 · 5e-6 = 0.628 m/s.
For an input rotation rate Ω = 1 °/s = 1 · π/180 = 0.01745 rad/s (a typical handheld-shake amplitude):
Coriolis force per mass: F_c = 2·m·Ω·v_d = 2 · 1e-12 · 0.01745 · 0.628 = 2.19 × 10⁻¹⁴ N.
Sense-mode deflection: x_s = F_c · Q_s / k_s, where k_s = m·ω_s² = 1e-12 · (2π · 20.1e3)² = 0.0159 N/m.
x_s = 2.19e-14 · 1000 / 0.0159 = 1.38 × 10⁻⁹ m = 1.38 nm.
A 1 nm displacement on a 1 µm capacitive gap produces ΔC/C ≈ 10⁻³ — translating into a sub-fA signal current at the ASIC charge-amp input. This is why MEMS gyros need vacuum-encapsulated high-Q resonators and chopper-stabilised lock-in detection at the carrier frequency. Below ~0.1 °/hr the readout is limited by the Brownian rotation noise ω_n = √(4·k_B·T·b)/(m·v_d) of the resonator itself.
7. Edge cases and gotchas
- Stiction during release — capillary forces collapse cantilevers permanently if released wet. Use supercritical CO₂ drying or vapour HF. Once stiction occurs, the device is dead — no field rework possible.
- Pull-in instability — every gap-closing electrostatic actuator self-destructs at V > V_pi = √(8·k·g₀³/(27·ε₀·A)). Drive amplifiers must clamp; closed-loop charge control (not voltage control) extends the range.
- Brownian noise floor — set by the fluctuation-dissipation theorem; cannot be improved by ASIC redesign. Lowering damping (vacuum encapsulation) is the only way to push it below ~10 µg/√Hz on accels or below ~0.05 °/hr bias on gyros.
- Etch non-uniformity — DRIE etch rate varies 5–15 % across a 200 mm wafer (centre-to-edge); proof-mass thickness uniformity directly affects sensitivity. SOI wafers fix this by stopping on buried oxide; tight-margin designs need wafer-level trim or per-die calibration.
- Wafer warpage and packaging-induced stress — encapsulation moulds, solder reflow, and CTE mismatch with PCBs stress the MEMS die, shifting bias by mg or %FS. Stress-isolated packaging (gel-fill, internal flexures, low-modulus die-attach) is mandatory for precision parts.
- Bias temperature coefficient — typically 1–10 mg/°C for accels, 1–50 °/hr/°C for consumer gyros. Polynomial calibration to 4th order is standard in modern IMU ASICs; residual TC ~10× lower than uncalibrated.
- Cross-axis sensitivity — orthogonality between sense axes is set by lithography alignment, typically 0.5–2 % per axis. Higher-grade IMUs (Honeywell HG4930, ADIS16505) factory-calibrate the full 3×3 cross-axis matrix.
- Vibration rectification (VRE) — accels exhibit a small DC offset proportional to RMS vibration² because the squared-V electrostatic feedback nonlinearity rectifies AC inputs. Datasheet spec µg/g²; matters in vibration-rich environments (drones, vehicles).
- Lock-in detection / chopper stabilisation — capacitive readout signal is at sense-mode frequency (10–30 kHz); ASIC mixes with the drive carrier, lowpass-filters, and recovers the (slow) rate signal. 1/f noise at baseband becomes irrelevant; only thermal noise around the carrier matters.
- ASIC integration vs SiP — die-stacked CMOS+MEMS (BMI270, ICM-42688) requires TSVs and bonded cap; lower cost per part, hardest qualification. Side-by-side multi-chip module (older Bosch BMA, ADXL) is cheaper to develop but larger and slower.
- Yield, test, and trim — every MEMS device gets individually tested in package — Wafer-Level Probe is insufficient because packaging stress shifts performance. Test cost is often 30–50 % of total manufacturing cost. Per-part trim of offset, gain, and TC coefficients is burned into OTP at end of line.
- Reliability and qualification — drop test 5000–10000 g half-sine 0.5 ms (consumer); 50000 g full-pulse (auto airbag); temperature cycling −40 to +125 °C (auto, AEC-Q103); HALT/HASS; humidity 85 °C/85 % RH 1000 h; ESD HBM ±2 kV / CDM ±500 V. Hermetic seal integrity by helium fine-leak (MIL-STD-883 method 1014).
- Outgassing in resonator cavities — over 5–10 years, residual gases adsorbed on cavity walls slowly release, raising pressure inside the encapsulated MEMS from ~1 Pa to ~10–100 Pa; this drops Q of the gyro resonator and shifts both scale-factor and bias. Getter materials (SAES Group ST171, ST172 thin-film getters) on the inside of the cap absorb H₂, O₂, CO, CO₂ over the device lifetime.
- Stiction after shock — even devices designed for stiction-free release can stick after a high-shock event slams structures together. Some IMUs include a built-in “shake test” — short electrostatic pulse intended to free any stuck masses; failure flags the part dead.
- Foundry IP transfer and process portability — MEMS processes are not portable. A design developed in Bosch’s auto-process cannot be ported to ST’s process; lithography, etch chemistries, and thin-film stresses are foundry-specific. A 2-year process-transfer engagement is typical when moving products between fabs.
Reference geometry data for the common building blocks
Selected closed-form expressions used in MEMS hand-calculation, with the regime where each is accurate enough for design (FEM verification still required before silicon).
Cantilever beam (length L, width b, thickness h, modulus E):
- Spring constant at tip, point load: k = E·b·h³ / (4·L³)
- First flexural resonance: f_1 = 0.162 · (h/L²) · √(E/ρ)
- Max tip stress at clamped end: σ = 6·F·L / (b·h²)
- Useful range: L/h > 10 (Euler–Bernoulli valid)
Folded beam (serpentine flexure) — favoured for low-stiffness x-translation with relief from lithographic stress:
- k_x ≈ E·b·h³·(N/(2L³)) where N is the number of folds; significantly softer than cantilever for the same footprint.
Clamped-clamped beam under axial buckling (problematic in MEMS resonators):
- Critical buckling load P_cr = π²·E·b·h³ / (3·L²)
- Residual film stress σ_res shifts f_1 by Δf/f ≈ 0.293·σ_res·L²/(E·h²) — a few-MPa stress in deposited polysilicon shifts a 20 kHz resonator by hundreds of Hz.
Square diaphragm (side 2a, thickness h, simply-supported edges, uniform pressure P): w_max = 0.0444·P·(2a)⁴/(E·h³); approximately 3× the clamped-edge deflection above. Real diaphragms anchored to bulk silicon are between simply-supported and fully-clamped depending on the boundary geometry.
Comb-drive lateral force (N fingers, finger thickness h, gap g, voltage V): F = N·ε₀·h·V²/g. Independent of x within the engagement range — the property that distinguishes comb drives from parallel-plate actuators and the reason they are the workhorse of in-plane MEMS actuation.
Engineering judgement and design margin
- Design for the slow corner of the wafer-process spread. Lithographic CD variation of ±0.2 µm on a nominal 2 µm beam changes k by ±30 %. Either trim per-die, or design k margin into the system (typically 2× margin on f_n at design centre).
- Never share an interconnect bus between fragile MEMS and high-voltage drive. Pull-in and ESD destruction are 90 % of MEMS field failures; protect inputs with TVS arrays sized for the actuator clamp voltage, not for the digital supply.
- Vacuum-package gyros are time-bombs. Even with a getter, cavity pressure drifts upward with age; reference-grade designs include an internal Q-monitor pin that an ASIC reads at boot to compensate scale-factor drift over the device’s 10-year life.
- Trim every part. Wafer-spread is too large for batch-calibration; each die’s offset, gain, and TC coefficients are measured at end-of-line test and burned into on-chip OTP. Without trim, every MEMS sensor is consumer-grade at best.
- Drop-test before you tape out. Mechanical-shock survival is set by stop-bumps, dimple geometry, and stress-concentration at flexure clamps — features that FEM struggles to model. A 50000 g drop test on a prototype reveals which corners actually fail in 200 ms; design iteration is unavoidable.
8. Packaging
| Package | Hermeticity | Volume | Typical use | Notes |
|---|---|---|---|---|
| Wafer-Level Package (WLP) cap | Hermetic vacuum or controlled atm | < 1 mm³ | High-volume consumer (phone IMU, MEMS mic) | Lowest cost in volume; ASIC stacked or side-by-side |
| LGA (Land Grid Array) | Plastic mould, semi-hermetic | 2–10 mm² | Consumer MEMS standard package | Plastic encap over WLP-capped die |
| QFN (Quad Flat No-lead) | Plastic | 3 × 3 to 6 × 6 mm | Industrial MEMS (ADIS, BMI) | Visible leads; better solderability than LGA |
| CSP (Chip Scale Package) | Plastic | ~ die size | Minimum footprint | RDL bumps directly on WLP cap |
| TO can (TO-46, TO-39) | Hermetic metal-can | ~ 6 mm dia | Legacy mics, special-purpose | Ka-band-compatible; expensive |
| Ceramic LCC / DIP | Hermetic ceramic with seam-welded lid | 10–25 mm² | Aerospace, space, military (Sigma Hi-Rel) | MIL-STD-883 qualified |
| Glass interposer | Hermetic | various | High-frequency MEMS, mmWave passives | Lower loss than Si interposer |
| 3D-stacked TSV (MEMS over ASIC) | WLP-hermetic | < die area | Modern phone IMU | TDK ICM-42688, Bosch BMI270 |
| Pre-mold cavity (gel-fill or open) | Non-hermetic | 6–10 mm² | Pressure sensors (gel-coated diaphragm) | Port allows pressure access |
9. Specialised and emerging directions
- Piezoelectric MEMS (pMUT, ScAlN, ScAlN with high Sc%) — AlN doped with up to 40 % scandium quadruples the piezo coefficient while remaining CMOS-compatible; key enabler for 5G n77/n79 BAW filters and emerging acoustic-MEMS speakers (Pyle, xMEMS Cowell). pMUT arrays (Chirp/TDK Microsystems) replace PZT in fingerprint sensors and short-range ultrasonic ranging.
- CMUT (Capacitive Micromachined Ultrasonic Transducer) — sealed-cavity capacitor used as ultrasound transducer. Lower drive voltage than PZT, integrable with CMOS imaging electronics, conformable arrays. Butterfly Network’s handheld whole-body ultrasound is CMUT-based; Hitachi, Philips also developing.
- THz / mmWave MEMS — micromirrors for THz beam-steering (300 GHz–3 THz), MEMS filters for mmWave radar (77 GHz, 140 GHz), reconfigurable intelligent surfaces (RIS) for 6G.
- Quantum MEMS — chip-scale atomic clocks (CSAC, Microsemi/Microchip SA.45s), trapped-ion qubit transports, MEMS-actuated optical-cavity tuning for cold-atom systems.
- Bio-integrated MEMS — Utah array (96-shank silicon penetrating arrays, Blackrock), Neuralink threads (thousands of 5 µm polyimide electrodes, robotically inserted), retinal prosthetics (Argus II — discontinued, Second Sight Medical).
- Stretchable / flex MEMS — polyimide-substrate sensors for wearables, skin patches, implants. Mc10 Biostamp, Epicore Biosystems sweat patch.
- Optical MEMS scanners for AR/VR — laser-beam-scanning displays (Microvision, ST B-Pro family) projecting full-colour images onto retinal or waveguide combiner.
- MEMS atomic vapour cells — chip-scale atomic magnetometers (QuSpin OPM), nuclear-magnetic-resonance gyros (Northrop). Sub-pT sensitivity in shoebox-sized packages.
10. Tools and software
Simulation
- COMSOL Multiphysics with MEMS Module — the dominant academic and industrial MEMS simulator; FEM solver supporting coupled electrostatics, elastodynamics, fluid–structure interaction, piezoelectricity, joule heating, contact mechanics.
- Coventor MEMS+ (Lam Research / Coventor) — system-level modelling with parametric library of MEMS building blocks (comb drives, flexures, plates); integrates with Cadence Virtuoso for MEMS+ASIC co-simulation.
- ANSYS Mechanical / Maxwell / HFSS — strong for mechanical FEA, electromagnetics, and RF respectively; less integrated for MEMS-specific multiphysics than COMSOL but widely used.
- Sentaurus / Silvaco TCAD — semiconductor process simulation; for MEMS used to model doping profiles, junction temperatures, and piezoresistor response.
- MEMSPro (Mentor / SoftMEMS) — discontinued but still used in legacy MEMS design flows.
Design and layout
- L-Edit (Mentor / Siemens) — layout editor with MEMS-specific shapes (curved geometries, fillets).
- Cadence Virtuoso MEMS Design Platform — CMOS-MEMS co-design with parametric cells.
- KLayout (open-source) — Python-scriptable layout editor; widely used for MEMSCAP MPW submissions.
- MEMSCAP design-rule decks — published process design kits for PolyMUMPs, MetalMUMPs, PiezoMUMPs, SOIMUMPs.
- Foundry PDKs — Bosch, Murata, X-FAB, Silex publish their own PDKs to qualified customers under NDA.
Test instrumentation
- Laser Doppler Vibrometer (Polytec MSA-100, MSA-600, OFV-505) — non-contact measurement of MEMS-structure velocity and displacement; pm-resolution at MHz bandwidth.
- White-light interferometer (Zygo, Bruker NPFLEX, Filmetrics) — sub-nm-vertical 3D surface mapping; the standard for MEMS profile characterisation.
- SEM (Zeiss, FEI/Thermo, JEOL) — scanning electron microscopy for sidewall, scallop, and feature-size inspection.
- FIB (FEI Helios, Zeiss Crossbeam) — focused-ion-beam cross-sectioning to expose buried MEMS features.
- Helium leak test (Pfeiffer, Inficon) — hermeticity per MIL-STD-883 method 1014.
- Drop tower / shock test fixture — calibrated mechanical-shock generation for qualification.
- Centrifuge rate-table — gyro calibration up to ~1000 °/s with arc-second pointing accuracy (Acutronic, Ideal Aerosmith).
Process equipment vendors
ASML (DUV/EUV lithography), Lam Research (deep-RIE, ALE), Applied Materials (CVD/PVD/CMP), Tokyo Electron (coater/developer, ALD), EVG and SUSS MicroTec (wafer bonding, nanoimprint), DISCO (dicing), Kulicke & Soffa (wire bonding).
11. Cross-references
[[Engineering/semiconductor-devices]]— MOSFET/BJT/diode physics underlies every readout ASIC and every CMOS-MEMS integration.[[Engineering/materials-ceramics]]— silicon, silicon nitride, alumina, AlN, ScAlN — all the structural materials of MEMS are ceramic-class.[[Engineering/materials-aluminum]]— Al metallisation, AlCu and AlSiCu interconnect; AlN piezoelectric.[[Engineering/pcb-design]]— MEMS-IC integration via SiP, COB, and substrate packaging; PCB layout for low-noise capacitive readout.[[Engineering/electromagnetics-engineering]]— RF MEMS S-parameters, EM modelling of resonators and filters.[[Engineering/op-amps]]— charge amplifier and transimpedance topologies for fF-level capacitive readout.[[Engineering/rf-design]]— SAW/BAW/FBAR filters and MEMS switches as integral RF building blocks.[[Engineering/microfluidics]]— companion note (same batch) covering BioMEMS and lab-on-chip systems.[[Engineering/biomechanics]]— planned; mechanical loads on bio-integrated MEMS.[[Engineering/bioinstrumentation]]— planned; instrumentation for BioMEMS and clinical devices.[[Robotics/sensors-pose-motion]]— practical IMU/gyro/accel selection and integration, building on the MEMS device physics in this note.
12. Citations
- Senturia, S. D. (2001). Microsystem Design. Kluwer. The canonical MEMS textbook — coupled-domain modelling from first principles.
- Madou, M. (2011). Fundamentals of Microfabrication and Nanotechnology (3rd ed., 3 vols). CRC Press. The encyclopedic process reference; volumes cover solid-state physics, fabrication, and BioMEMS.
- Beeby, S., Ensell, G., Kraft, M. & White, N. (2004). MEMS Mechanical Sensors. Artech House. Application-focused on accel/gyro/pressure design.
- Yazdi, N., Ayazi, F. & Najafi, K. (1998). Micromachined Inertial Sensors. Proceedings of the IEEE, 86(8), 1640–1659. Foundational taxonomy of silicon inertial sensors.
- Lyshevski, S. E. (2002). MEMS and NEMS: Systems, Devices, and Structures. CRC Press. Systems-level treatment.
- Tang, W. C., Nguyen, T.-C. H. & Howe, R. T. (1989). Laterally driven polysilicon resonant microstructures. Sensors and Actuators, 20, 25–32. Origin of comb-drive electrostatic actuation.
- Lärmer, F. & Schilp, A. (1996). Method of anisotropically etching silicon. US Patent 5501893 (Robert Bosch GmbH). Bosch DRIE process patent.
- Saif, M. T. A. & MacDonald, N. C. (1996). A milli-scale fully differential lateral resonator. Proc. IEEE MEMS, 109–114. Early DRIE-based lateral resonator.
- Hornbeck, L. J. (1989). Deformable-mirror spatial light modulators. Proc. SPIE, 1150, 86–102. Origin of the TI DLP / DMD device.
- Greenwood, J. C. (1984). Etched silicon vibrating sensor. Journal of Physics E: Scientific Instruments, 17, 650. Early micromachined Si resonator.
- Tilmans, H. A. C. (1996). Equivalent circuit representation of electromechanical transducers: I. Lumped-parameter systems. Journal of Micromechanics and Microengineering, 6, 157–176.
- Bao, M. (2005). Analysis and Design Principles of MEMS Devices. Elsevier. Strong on squeeze-film damping, electrostatic actuation, pull-in.
- Lobontiu, N. & Garcia, E. (2004). Mechanics of Microelectromechanical Systems. Springer. Compliance-matrix and flexure-design reference.
- JEDEC JESD22-A104. Temperature Cycling. Standard reliability test for packaged semiconductors and MEMS.
- JEDEC JESD22-B104. Mechanical Shock. Drop and shock-test methodology.
- MIL-STD-883. Test Method Standard, Microcircuits. Method 1014 (seal — hermeticity), 2002 (mechanical shock), 1010 (temperature cycle), 2001 (constant acceleration).
- AEC-Q103. Failure Mechanism Based Stress Test Qualification for MEMS Pressure Sensors. Automotive MEMS qualification standard.
- SEMI standards M-series (silicon wafer specifications), MS-series (MEMS-specific standards).
- ISO 19261. Microelectromechanical Systems (MEMS) — Vocabulary and Definitions.
- IEC 60749 / IEC 60068-2 series. Environmental testing for electronic components.
- Sandia National Laboratories SUMMiT V design manual (sandia.gov). Public-domain reference for surface micromachining design rules.
- MEMSCAP MPW user manuals: PolyMUMPs, MetalMUMPs, PiezoMUMPs, SOIMUMPs design handbooks (memscap.com).
- Manufacturer datasheets and application notes: Bosch Sensortec (BMI, BMP, BME), STMicroelectronics (LSM, LIS, LPS), TDK InvenSense (ICM, MPU), Analog Devices (ADXL, ADXRS, ADIS), Knowles Acoustics (SPH-series microphones), Texas Instruments (DLP optical-MEMS), Qorvo and Broadcom (BAW filters), Wolfspeed/Coherent (SiC substrates for high-T MEMS), Sensirion (SHT, SCD, SGP).