A MEMS microphone is a 3.5 mm by 2.65 mm part that costs between $0.40 and $2.50, and it is the component most likely to make a connected product feel broken. Voice control that misses every third command, a recorder that clips when someone speaks close, a doorbell that whistles in the wind — those failures are almost never the microphone die. They are the hole in front of it, the gasket around it, and the output format chosen months earlier without much thought.

The die is a silicon diaphragm a few microns thick over a perforated backplate, with an ASIC in the same package doing bias and buffering. Sound reaches it through a port hole on the package's bottom or top face. Everything you build in front of that port is part of the acoustic system whether you designed it or not.

Analog, PDM, or I2S

The output format is the first decision because it drives the microcontroller you need, the routing rules, and how many microphones you can support.

Analog microphones output a single-ended signal, typically around -38 dBV sensitivity, into an external codec or ADC. They draw the least current, roughly 100 to 250 microamps, and cost the least. The cost lands on the board: an analog trace of any length picks up switching noise from regulators, backlights, and radios, so the part has to sit near its ADC on a short, guarded route.

PDM microphones digitize on-package and stream a one-bit signal at 1 to 3.2 MHz. Board-noise immunity is dramatically better, and two microphones share one data line by clocking on opposite edges. The cost is a decimation filter on the receiving end — a hardware PDM interface, or CPU cycles you may not have. Current runs 600 microamps to 1.5 mA, with a reduced-clock sub-mode for always-listening wake-word detection.

I2S microphones output finished PCM samples: easiest to integrate, most power hungry at 1 to 2 mA, largest package, and the natural choice when an I2S bus already exists for a codec. The wider bus tradeoffs sit alongside those in choosing an embedded communication bus.

SNR and AOP: The Two Numbers That Bound Your Product

Signal-to-noise ratio is measured against a 94 dB SPL, 1 kHz reference and expressed in dBA. A 62 to 64 dBA part is fine for a phone held near the mouth. Far-field voice needs 65 to 68 dBA, because at three meters the talker's contribution at the diaphragm has dropped 20 dB and the microphone's own noise floor sets your recognition rate. Each 3 dB of SNR roughly doubles usable distance in a quiet room.

Acoustic overload point is the level at which distortion reaches 10 percent. Standard parts sit at 120 to 124 dB SPL — enough for conversation, not for anything loud. A power tool, a concert, or an infant crying into a monitor exceeds 120 dB routinely. High-AOP parts reach 130 to 135 dB SPL for about a dollar more, far cheaper than a product that distorts on the exact events it exists to capture.

Sensitivity matching matters the moment you use more than one microphone. Beamforming depends on the array elements agreeing; specify parts binned to ±1 dB rather than the standard ±3 dB, or the array's null steers to the wrong place.

The Acoustic Port Is a Resonator

The channel from the outside world to the microphone package behaves as a Helmholtz resonator. Get it wrong and you build a 6 to 12 dB peak somewhere in the voice band, plus a rolloff above it.

  • Keep the port channel short. Under 1.5 mm is comfortable; past 3 mm the resonance drops into speech frequencies.
  • Hole diameter of 0.8 to 1.2 mm is the usual compromise between acoustic transparency and debris ingress.
  • Keep channel volume small and cross-section constant. A funnel, a countersink, or a cavity behind the wall all add volume and move the resonance down.
  • Provide a sealed path. Any leak between the housing hole and the package port lets sound take an unintended route and destroys the response and any array phase relationship.

The Gasket

The seal between housing boss and package is a die-cut silicone or foam ring, 0.3 to 0.8 mm thick, compressed 25 to 40 percent. Under-compression leaks; over-compression stresses the solder joints on a 1 mm-tall part and cracks them during drops. Design boss height and board standoff so compression lands in range across the full tolerance stack-up, not just at nominal.

Bottom-port parts need a PCB hole aligned to the package port with no via inside the gasket footprint. Top-port parts avoid the board hole but move the sealing plane to the package top, meaning a taller boss and a different stack in the enclosure design.

Wind, Water, and Dust

Wind noise is not sound. It is turbulent pressure fluctuation at the port, and it saturates the front end at wind speeds a pedestrian encounters. The fixes are geometric: recess the port out of the free stream, avoid sharp upstream edges that shed vortices, add a mesh or open-cell foam layer, and use two microphones so correlated speech and uncorrelated wind separate in firmware.

For ingress, the standard solution is an acoustic vent membrane — a thin expanded PTFE disc adhered over the port that passes sound while blocking liquid water, rated by water entry pressure and by insertion loss of 1 to 3 dB. It protects the microphone; it does not by itself give the housing a rating. A product claiming IPX7 needs the whole enclosure sealed, as laid out in what IP ratings actually mean and in waterproof electronics design. MEMS microphones are also damaged by ultrasonic cleaning and high-pressure air blowoff — tell your assembler the part is sensitive and keep the protective film on until after wash.

Test It in the Product, Not on the Bench

A microphone measured on an evaluation board proves the part meets its datasheet and nothing about your acoustic port. Build the real housing, sweep frequency response in a quiet space, and look for the resonant peak. Then run intelligibility tests at the distances and noise levels users will be in, with the enclosure closed. At the prototype stage this costs a week; after tooling it costs a mold change.

Designing the Audio Path With Us

Projects House designs acoustic paths as part of the mechanical and electronic package together: port geometry, gasket stack, part selection against your real distance and level requirements, and the firmware side of the array. Send your product concept through the contact form.