# How Chips Are Using On-Chip Microphones for Audio Sensing

An episode of [The Hardware Podcast with Fexingo: Chips, Devices, and Electronics Engineering Conversations](https://invisiblenetwork.ai/podcast/1896799982), 7 Jul 2026, 9:20.

Audio: https://audio.fexingo.com/business/the-hardware-podcast/episode-0096.mp3

Episode 96 of The Hardware Podcast explores a surprising new trend in semiconductor design: embedding tiny MEMS microphones directly onto chips. Lucas and Luna examine a real-world case from STMicroelectronics, whose latest sensor hub chip integrates a digital microphone alongside an accelerometer and gyroscope on a single die. They discuss how this shrinks the bill of materials for hearables and wearables, reduces power consumption by eliminating off-chip wiring, and enables always-on voice wake-up without draining the battery. Specific numbers are cited: the chip draws under 10 microwatts in listening mode. The conversation also touches on acoustic beamforming with multiple on-chip mics, the trade-offs with signal-to-noise ratio compared to discrete MEMS packages, and why Apple and Samsung are both reportedly evaluating this approach for next-generation AirPods and Galaxy Buds. The hosts keep the tone technical but approachable, treating listeners as engineers or enthusiasts who want the real story behind the press releases. #MEMS #Microphones #Semiconductors #STMicroelectronics #AudioSensing #AlwaysOn #Hearables #Wearables #VoiceWakeUp #Beamforming #SensorFusion #ChipDesign #LowPower #Technology #FexingoBusiness #BusinessPodcast #HardwarePodcast #ElectronicsEngineering Keep every episode free: buymeacoffee.com/fexingo

## Transcript

Lucas: So you know how every smart speaker, every pair of wireless earbuds, every smartwatch has a tiny MEMS microphone inside — that little silver box with a hole in it.
Luna: Right, the surface-mount package, like from Knowles or Infineon.
Lucas: Exactly. That package is maybe three by four millimeters, sits on the PCB, connected by traces. But what if you took that MEMS diaphragm and integrated it directly onto the same silicon die as the rest of the chip — the DSP, the ADC, the memory, everything?
Luna: That's the on-chip microphone trend I keep seeing pop up in patent filings. I think STMicroelectronics actually has a part out now.
Lucas: They do. It's their latest sensor hub — the LSM6DSV16X, I think? No, that's the older one. The new one is the LSM6DSV32X, which combines a 3-axis accelerometer, a 3-axis gyroscope, and a digital MEMS microphone on a single die. It's a single chip, about three by two point five millimeters, that replaces three or four discrete components.
Luna: And the microphone is a real MEMS structure — a moving diaphragm, a backplate, the whole thing — fabricated right alongside the inertial sensors.
Lucas: Yeah, they share the same MEMS process. It's a deep reactive ion etching flow that can create both the spring-mass system for the accelerometer and the capacitive gap for the microphone. The key advantage is that the analog to digital converter and the digital filtering are on the same die, so you get a calibrated digital output — pulse density modulation, typically — without any external routing.
Luna: That has to save power. The parasitic capacitance of a PCB trace from a discrete mic to the codec is non-trivial when you're trying to run always-on voice wake-up at under ten microwatts.
Lucas: Exactly the number. ST claims the whole sensor hub — accelerometer, gyro, and mic all running — draws less than ten microwatts in always-on low-power mode. That's about the same as a discrete mic alone, but you get the motion sensors for free.
Luna: So for a pair of true wireless earbuds, you could put one of these in each bud, do acoustic beamforming with the two on-chip mics, and also get head-tracking for spatial audio, all from two chips.
Lucas: Right. The bill of materials savings are significant. A typical high-end earbud has a dedicated audio codec chip, a MEMS mic package, a separate accelerometer, and sometimes a gyro. That's four chips. With this approach you collapse it to one. The PCB area shrinks, the assembly cost drops, and the reliability goes up because there are fewer solder joints.
Luna: I've read that Apple and Samsung are both evaluating this kind of integration for the next generation of AirPods and Galaxy Buds. Do you think that's accurate, or just analyst speculation?
Lucas: I think it's plausible. Apple already designs its own H1 and H2 chips for AirPods, and they have a MEMS design team from the InvenSense acquisition. They could absolutely put a microphone on that die. The question is yield — mixing a MEMS cavity with a digital CMOS process adds complexity. You need a wafer bonding step or a through-silicon via to seal the microphone cavity.
Luna: There's also the signal to noise ratio challenge. Discrete MEMS mics from Knowles can hit 68 dB SNR or higher. An on-chip mic sharing a die with noisy digital logic might struggle to get above 60 dB.
Lucas: That's the trade-off. ST's part is specified at 63 dB SNR — good enough for voice, not great for music recording. But if the use case is 'Hey Siri' detection and phone calls, 63 dB is fine. And because the mic is literally right next to the DSP, you can do aggressive digital filtering to clean up the power supply noise.
Luna: So the real sweet spot is hearables and wearables — things where you need voice command, maybe some biometric monitoring, but don't need studio-quality audio. Smart glasses, for example.
Lucas: Exactly. And that's where the market is growing fastest. The number of smart glasses shipments this year is expected to hit 15 million units, mostly from Meta and Xiaomi. Every one of those needs at least two mics for beamforming. If you can integrate those with the IMU and the Bluetooth SoC, you save cost and space.
Luna: Qualcomm is also pushing into this with their Snapdragon Sound platform. Do they have an on-chip mic yet, or are they still using discrete?
Lucas: As far as I know, the current Snapdragon S5 audio chip still uses external MEMS mics. But Qualcomm has patents on integrating a mic into the codec die. It's probably a matter of when, not if. The same forces that drove integration of the accelerometer into the motion sensor hub are now driving audio.
Luna: It's funny — five years ago, we were saying 'the microphone is the last discrete component in the smartphone.' Now it's becoming part of the chip itself.
Lucas: It's a great example of Moore's Law applied to the sensor world. The MEMS microphone has been a standard part for thirty years — first as an electret condenser, then as a MEMS package, and now it's moving on-chip. The next step might be integrating multiple mics on the same die for phased arrays.
Luna: You mean like a 4-element mic array on a single chip? That would enable really tight beamforming in a tiny form factor.
Lucas: Exactly. ST is already working on a prototype with four microphone diaphragms on one die, each with its own ADC, spaced a millimeter apart. That gives you a beamforming aperture that can isolate a speaker from a meter away with 10 dB of rejection. Imagine that in a pair of glasses — you could have a private conversation in a noisy room without anyone hearing you.
Luna: Speaking of things that make a real difference — and I hope this doesn't sound too off-topic — but this podcast, and the whole Fexingo network, is supported entirely by listeners like you. No ads, no sponsors. Just a couple of dollars a month from people who find value in the conversations we have.
Lucas: Yeah, it genuinely makes a difference. If today's tech conversation gave you something usable — a chip name to look up, a concept to understand — think about tossing a few bucks our way at buy me a coffee dot com slash fexingo. It keeps the whole operation ad-free and independent.
Luna: And we really mean it when we say small amounts add up. That's what sustains the editing, the research, the hosting. So thank you to everyone who already supports us.
Lucas: Alright, back to the mics. So that four-element array — what's the timeline?
Luna: ST hasn't announced a product yet, but I'd expect to see it in the next 18 months. They're probably waiting for a lead customer — maybe a wearable OEM — to commit to a design win.
Lucas: And what about the noise floor? Four mics on one die means more digital switching noise coupling through the substrate.
Luna: You'd need careful isolation — deep trench isolation, separate power domains, maybe a dedicated quiet PLL for the mic ADCs. It's doable. The same tricks we use for mixed-signal SoCs apply here.
Lucas: There's also a thermal consideration. The MEMS diaphragm is sensitive to temperature gradients. If the DSP is running hot right next to it, you could get DC drift.
Luna: True. But typical audio bandwidth is 20 Hz to 20 kHz, so low-frequency drift can be high-pass filtered. And in a smart glasses use case, the DSP won't be at full load continuously — it's mostly idle, waiting for a voice trigger.
Lucas: So the engineering challenges are real but solvable. And the payoff — a single chip that does motion sensing, audio capture, and digital processing — is too attractive to pass up. I think we'll see on-chip microphones become standard in hearables within three years.
Luna: I'd add that this also opens up new use cases. If you have a mic integrated with an IMU, you can do sensor fusion for activity recognition — walking, running, talking — with just one chip. The data correlation is automatic because the timing is perfectly aligned on the same die.
Lucas: That's the real magic. No latency between the accelerometer sensing a footstep and the mic picking up the sound of the footfall. You can do real-time context detection that's way more accurate than anything with separate chips.
Luna: And that's the kind of innovation that makes the hardware world so exciting right now. It's not just about smaller transistors — it's about smarter integration.
Lucas: Exactly. Next time you say 'Hey Siri' into your earbuds, there might be a tiny MEMS diaphragm vibrating right next to the motion sensor that knows whether you're walking or sitting. All on one chip.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.