Executive Summary PRICE USID

 

 

 

Executive Summary

Implantable “USID” wireless audio modules (not traditional RFID) require extremely miniaturized components. Based on current commercial parts, the best solution is a micro-BLE (Bluetooth Low Energy) audio SoC paired with a tiny MEMS microphone and MEMS loudspeaker. For example, a Nordic Semiconductor nRF52840 (WLCSP package ~3.6×3.5 mm) combined with a digital MEMS microphone (e.g. TDK InvenSense ICS-43434, 3.50×2.65×0.98 mm) and a MEMS speaker (e.g. xMEMS Cowell,

3.2×6.0×1.15 mm) fits the 9×6×2.5 mm envelope【24†L43-L50】【40†L49-L57】. This assembly could sample 8 kHz narrowband audio and transmit it wirelessly (e.g. BLE) for G.729 encoding on the host. The total active current can be just a few mA (µW) for audio. However, these COTS parts require a tiny battery (or energy harvesting) and are not medical-grade; biocompatible encapsulation is mandatory.

 

No currently available chip can continuously stream voice audio without a power source. Ultra-low-power BLE SoCs with energy-harvesting (e.g. Atmosic ATM3330E) exist, but even they need an external power reservoir (capacitor/battery) to stream voice for more than brief bursts【58†L72-L81】【58†L82-L90】.

Inductive or ultrasonic powering is possible in principle but would impose severe range and latency limits,

making continuous speech impractical.

 

Below we list candidate components, map compatibility, estimate costs, and outline host-side requirements and risks. Our recommendation is a BLE audio SoC (like Nordic’s nRF52840/WLCSP or similar) + MEMS mic + MEMS speaker + small battery. This can deliver intelligible G.729-coded speech over a BLE link to a PC or satellite gateway. Trade‑offs include the need for a battery (or large capacitor), and the lack of any approved implantable wireless telecom device (all are consumer ICs requiring encapsulation and clearance).

 

1.    Candidate USID (Wireless Audio) SoCs

We define “USID” chips as very-small wireless transceivers or audio SoCs (typically BLE) that could fit inside a 9×6×2.5 mm implant. Key candidates are BLE/2.4 GHz SoCs and ultra-low-power modules. Table below summarizes likely parts (all Bluetooth Low Energy unless otherwise noted):

 

Manufacturer	Model	Package Size (L×W×H mm)	Supply Voltage (V) & Power (active)	Wireless Std.	Audio I/O	G.729 support / note	Notes (batteryless
Nordic Semic.	nRF52805 WLCSP	2.48×2.46×0.75 [1]	1.7–5.5 V; ~4–10 mA RX/TX	BLE 5.2	PDM in, PWM out	Does not natively encode G. 729 (BLE PHY)	No batteryless; requires battery/PMI

 

Manufacturer	Model	Package Size (L×W×H mm)	Supply Voltage (V) & Power (active)	Wireless Std.	Audio I/O	G.729 support / note	Notes (batteryless
Nordic Semic.	nRF52810 WLCSP	2.50×2.50×0.75 mm 【25†L37-L44】	1.7–5.5 V; ~3–5 mA active	BLE 5.0	PDM in, PWM out, 12-bit ADC	BLE link can carry encoded bits	Very low- power; no batteryless option
Nordic Semic.	nRF52820 WLCSP	2.531×2.531×0.85 mm 【27†L44-L51】	1.7–5.5 V; few mA active	BLE 5.2	PDM in, I²S/I²S out, ADC	BLE link; no built-in codec for G.729	(Requires minimal PMIC)
Nordic Semic.	nRF52832 WLCSP	~3.0×3.2×0.8 mm 【30†L129-L133】	1.7–3.6 V; ~5–6 mA TX(0dBm)	BLE 5.0	PDM in, I²S, PWM, ADC	BLE link carrier; no internal G. 729	Popular BLE audio SoC (I²S support ICS-43434)
Nordic Semic.	nRF52840 WLCSP	3.6×3.5×0.7 mm 【24†L43-L50】	1.7–5.5 V; ~5–6 mA TX(0dBm)	BLE 5.2 + legacy	PDM in, I²S, USB, PWM	BLE link; no built-in codec; I²S for digital mic 【24†L27- L30】	Full-feature highest power/ higher cost
NXP Semic.	NXH3675 (LE Audio)	WLCSP ~3.0×3.0×0.6 mm 【22†L224-L231】	1.8–3.3 V; ~8.8 mW (typ gaming/ ble) 【22†L239- L247】	BLE Audio 5.3	PDM in, I²S, analog	Supports LC3 audio (no G.729) 【22†L239- L247】	Very low latency; higher integration (BLE Audio)
NovelBits/ EM	EM9305 WLCSP	1.8×1.8×0.3 mm 【18†L5-L9】	1.1–5.0 V; 3.4 mA TX, 3.1 mA RX 【18†L5- L12】	BLE 5.4	SPI, I²C, GPIO (no I²S/PDM)	BLE link; no native audio codec	Ultra-mini; digital mic via SPI possible
Silicon Labs	EFR32BG21 QFN32	4×4×0.85 mm 【35†L409-L418】	1.7–3.6 V; ~9.9 mA TX(0dBm) 【35†L369- L378】	BLE 5.4	I²S, PDM (2x PDM)	BLE link; no built-in G.729	Larger but low-latency and DSP (M33 core)

 

Manufacturer	Model	Package Size (L×W×H mm)	Supply Voltage (V) & Power (active)	Wireless Std.	Audio I/O	G.729 support / note	Notes (batteryless
Atmosic	ATM3330E*	Module: ~10×9 mm (SoC die unknown)	1.1–4.2 V; supports energy harvest 【58†L74- L83】	BLE 5.3 (EH)	Up to 2× PDM (stereo) 【58†L82- L90】	BLE link; can store & stream (audio via I²S/PDM)	Energy harvesting BLE SoC (photovoltai RF, etc.)

 

上記のBLE SoCはすべてG.729を内蔵していません。代わりに、8 kHzサンプリングのPCM音声を取り込み（PDM or I²S）、無線で転送し、受側PC等でG.729にエンコードする方法が現実的です【24†L27-L30】【22†L239-

L247】。NordicのnRF528xやNXP NXH3675はI²S/PDM入力があり、デジタルMEMSマイクを直結できます。数

mW以下で動作しますが、バッテリなし運用は不可です。一方Atmosic ATM3330eはエネルギーハーベスティング対応（光、熱振動、RF）なので、超低パワー用途で条件次第では単三電池不要の設計も可能ですが、それでもコンデンサ等の電源蓄積なしでは連続音声送信は難しい【58†L72-L81】【58†L82-L90】。RFIDパッシブタグ

（NXP/ATMELなど）や音響バックキャンセラーも検討できますが、これらはID用途のみで音声ストリームには使えません。

2.    Tiny MEMS Microphones & Speakers

 

Type              Model/Part       Dimensions

(L×W×H mm)

Interface           SNR/Other Specs

Supply (V) & Current

Notes

 

Mic (Digital)	TDK InvenSense ICS-43434	3.50×2.65×0.98 【40†L49- L57】	I²S (24-bit)	SNR 64 dBA; Sens. – 26 dBFS 【40†L49- L57】	1.6–3.6 V; 230–490 μA 【40†L49- L57】	Widely used; interfaces directly to I²S/PDM audio SoCs
Mic (Analog)	TDK InvenSense ICS-40720	4.00×3.00×1.20 【43†L349- L357】	Differential Analog	SNR 70 dBA; Sens. – 32 dBV 【43†L361- L370】	1.5–3.63 V; 375 μA 【66†L1152- L1160】	High SNR; larger footprint
Mic (Analog)	TDK InvenSense T4064	2.70×1.60×0.89 【10†L2-L7】	Differential Analog	SNR 61.5 dBA 【10†L2- L7】	1.6–3.6 V; ~180 μA	Ultra-small size (2.7×1.6); lower SNR

 

Type	Model/Part	Dimensions (L×W×H mm)	Interface	SNR/Other Specs	Supply (V) & Current	Notes
Speaker	xMEMS Cowell (full-range)	3.20×6.00×1.15 【70†L245- L250】	Analog (requires amp)	≈20 Hz– 20 kHz; 105 dB SPL@1kHz (at 10 mm)	–	Bottom- ported MEMS speaker 【70†L245- L250】 for in-ear; needs ~0.5–1 V drive at ~1– 2 Ω
Speaker	xMEMS Muir (tweeter)	3.20×5.00×1.15 【70†L219- L224】	Analog (requires amp)	2–20 kHz focus; high-fi tweeter	–	Very high- frequency response; used in 2- way earbuds 【70†L219- L224】
Speaker	Fraunhofer MEMS Speaker*	(Research chip) ~1–2 mm	Analog (piezo MEMS)	–	–	Experimental silicon speakers (not commercial)

 

TDK/InvensenseのICS-43434はI²S出力でデジタル・マイコンと直結可能です【40†L49-L57】。アナログマイクではICS-40720が高SNR(70 dB)で優秀ですが4×3×1.2 mmと最大なので、スペースが許せば検討できます

【43†L349-L357】【66†L1152-L1160】。超小型例としてT4064（2.7×1.6×0.89 mm）もありますがSNRは低め

【10†L2-L7】。

 

スピーカーはxMEMS社製が唯一の市販レベルMEMSで、Cowell(3.2×6.0×1.15)とMuir(3.2×5.0×1.15)がサイズ制限を満たします【70†L219-L224】【70†L245-L250】。どちらも1.15 mm厚で、Cowellは20 Hzまで出せるフルレンジ、Muirは高音専用ツイータです。駆動にはアンプ（または専用IC、xMEMS Aptos2等）が必要です。これらは裸チップ（ボンディングパッド付き）なので、歯槽内実装時は十分な絶縁/防水が必須です。

 

3.    Compatibility & Integration (Chip + Mic + Speaker)

The table below maps each candidate wireless SoC to compatible mic/speaker pairs, required interfaces, and an estimated board outline (≤9×6 mm). In practice, the “implant PCB” would stack/semi-stack these components within 2.5 mm height (see notes):

 

SoC Chip	Mic Model	Speaker Model	Interfaces/Front- End	Notes on Packaging (L×W×H)
nRF52810 (2.5 mm)【25†L37- L44】	ICS-40720 (analog)	xMEMS Cowell (spe.)	Analog ADC input (+mic bias), PWM or DAC+amp out	Chip 2.5×2.5×0.7; Mic 4.0×3.0×1.2; Spk 3.2×6.0×1.15.
(2.5×2.5×0.8)	or T4064 (smaller)			Possible mounting: Chip and mic on PCB, speaker possibly adjacent or under, total stack ≤2.5 mm.
nRF52820 (2.53 mm) 【27†L44-L51】	ICS-43434 (digital)	xMEMS Cowell	PDM/I²S input (direct for ICS-43434), PWM output/amp	Chip 2.53×2.53×0.7; Mic 3.5×2.65×0.98; Spk 3.2×6.0×1.15.
(2.53×2.53×0.7)				Stacking: I²S wiring to mic pads; speaker on opposite face if stacking 2-layer, <2.5 mm total.
nRF52832 (3.0×3.2 mm)	ICS-43434 or ICS-40720	xMEMS Cowell	PDM/I²S or ADC in, PWM out	Chip 3.0×3.2×0.8; Mic 3.5×2.65×0.98; Spk 3.2×6.0×1.15.
nRF52840 (3.6×3.5 mm)	ICS-43434 + XLoud (DAC) or ICS-40720	Cowell	I²S or ADC in, USB/PWM or I²S out	Chip 3.6×3.5×0.7; Mic 3.5×2.65×0.98; Speaker 3.2×6.0×1.15.
NXH3675 (3×3 mm) 【22†L224-L231】	ICS-43434 (I²S)	Cowell (analog)	I²S in (mic), I²S/ PWM out (speaker via amp)	Chip ~3×3×0.5; Mic 3.5×2.65×0.98; Spk 3.2×6.0×1.15.
ATM3330E (module)*	ICS-43434	Cowell	I²S/PDM in, PWM out	Module ~10×9; SoC inside; (Used for energy-harvesting demonstration only.)

 

•   Interfaces: Nordic SoCs like nRF52810/20/32 have PDM input (for I²S mics) and/or ADC inputs (for analog mic). All have PWM timers or I²S output that can feed a small Class-D amplifier (for the MEMS speaker). For instance, ICS-43434 (I²S digital mic) can connect directly to the SoC’s I²S/PDM interface

【24†L27-L30】【70†L245-L250】. Analog mics (ICS-40720/T4064) require a bias resistor and the SoC’s ADC or a small front-end amp. The xMEMS speaker must be driven by an external amplifier (e.g. <10 mW class-D) – this requires additional area for the amp IC.

 

•   Board/Stack dimensions: A plausible 9×6 mm implant could be a 4-layer PCB: one face holds the SoC and mic (chip-scale WLCSP packages are <1 mm tall), the other face holds the speaker (1.15 mm tall). For example, nRF52810 (2.5×2.5×0.8 mm) plus ICS-40720 mic (4.0×3.0×1.2) could lie side by side (total ~6×4 mm area), with Cowell speaker (3.2×6.0×1.15) placed below or alongside. Layers could

share ground plane and traces. The max thickness: 0.8 +1.15 ≈ 1.95 mm for chip+speaker, within

2.5 mm. Gaps and encapsulation add volume, but 2.5 mm is just enough.

 

•   Electrical front-ends: Mic bias resistors and decoupling required for analog mics; anti-alias filters for ADC (if used); speaker driver amplifier with small LC output filter. A tiny flash memory (for firmware) and PMIC might be needed. All this must fit; realistic PCB area might be slightly larger (e.g. 9×6 mm board) and then flexed or folded to 9×6 footprint.

 

() ATM3330E note: This module includes Atmosic’s EH-capable SoC (RistrettoBin GTI-ATM3330E). It supports 2 PDM mics【58†L78-L86】, but packaging is large (10×9 mm) – it’s more a proof of concept for batteryless BLE, not an implant candidate due to size.

 

4.    Batteryless (“No-battery”) USID Chips

Truly battery-free continuous voice is infeasible today. However, some chipsets/tags can operate without a dedicated battery if recharged or constantly driven by external fields:

 

•   Energy-Harvesting BLE SoCs: e.g. Atmosic ATM3330E/ATM2x【58†L72-L81】【58†L82-L90】. These incorporate power management to scavenge light, heat, motion or RF energy. They can support BLE and PDM microphones, but only low-duty or burst operation. In practice, they must charge an internal capacitor or supercap to stream audio. Continuous conversational voice (tens of kbps) would deplete harvest faster than it accrues (especially inside a tooth pocket with limited light and motion).

 

•   RFID/NFC Tags with Energy Harvester: Chips like NXP NTAG + integrated PV cell or EM4325 (NFC tag with solar) can operate briefly on harvested light. But RFID standards are passive and only respond to reader pulses – not a continuous wireless link for audio. Similarly, NFC audio modules exist (RF acoustic tags), but none support real-time G.729-coded streaming.

 

•   Ultrasonic/Inductive Power: Acoustic or inductive power transfer can supply ~1–10 mW to implants (e.g. cardiac or cochlear implants). A hypothetical USID module could harvest ultrasonic energy (MHz) and extract power via a piezo. This could power a SoC briefly, but sustaining even narrowband voice (~10 kbps) would require ~10^(-5) J/s, hard to maintain with safe ultrasound intensities. Practically, these methods only enable ultra-low-duty sensors, not continuous speech.

 

•   Feasibility & Limitations: All batteryless schemes face strict limits: the implant must be extremely energy-frugal, and external powering (light, RF, ultrasonic) must be continuous and strong, which is either unrealistic (in-body) or unsafe (power density limits). The user should assume a small battery or supercap; continuous voice streaming on “no battery” is effectively impossible with current tech.

 

5.    Cost Estimates (per Unit)

Approximate component prices (MSRP or distributor) and implant costs:

 

•   BLE SoC:

•   nRF52810 (CSP): ~$3–4 in quantities (Digi-Key~$3.3 each【25†L37-L44】).

•   nRF52840 (CSP): ~$7–10.

•   NXH3675: likely ~$10–15 (est.).

•   EM9305: small volume, maybe $5–$10.

•   Silabs EFR32BG21: ~$5–10 (QFN 4×4 mm).

•   MEMS Microphone:

•   ICS-43434: ~$2.4 (bulk ~5000 qty)【63†L1-L4】.

•   ICS-40720: ~$5.5 (each)【67†L7-L10】, ~$3.9 in 25pc.

•   T4064: ~$2.0–$3.

•   MEMS Speaker:

•   xMEMS Cowell: ~$20 per unit【68†L0-L2】 (single-unit price via xMEMS store).

•   xMEMS Muir: ~$15–18.

•   Power Mgmt/Amplifiers: (~$1–3 each) e.g. TI/Analog Devices Class-D amp for speaker, buck regulator.

•   PCB + Encapsulation: A few cm² of multi-layer FR4 is low cost (~$1). Biocompatible encapsulation (medical silicone, Parylene coating) might add ~$5–$10 per implant in materials. Surgical fixture (e.g. dental acrylic) ~ negligible per unit.

•   Battery: If used, a tiny LiPo (<5 mm thick) ~ $2–$5 each. Or supercap ~$1.

Total per implant (parts): Roughly $30–$60 for medium production volumes. Cowell speaker and SoC dominate. With encapsulation and assembly, estimate $50–$100 each in quantity. (Custom implant electronics are expensive at low volumes.)

 

6.    Host-Side Hardware & Software Requirements

To receive and process the G.729 audio from the implant:

 

•   Receiver Hardware:

•   A BLE radio (often built into PC/mobile). For example, a USB BLE dongle (~$5–10) or a smartphone/ Tablet with BLE. If uplinking to satellite, a gateway board is needed (see below).

•   For Internet connection: any PC or SBC (Raspberry Pi ~$35) with network (Ethernet/Wi-Fi).

•   For satellite: a satellite modem (see below) connected to PC/SBC.

 

•   Software:

 

•   BLE stack / drivers (e.g. BlueZ on Linux, Bluetooth LE API on Windows). The implant may use a custom GATT or raw radio protocol carrying audio data. The PC runs firmware to collect and forward this data stream.

•   G.729 Codec: A G.729 encoder/decoder library. G.729 is patented/licensed (ITU-T G.729 Annex A/B are license-free variants, others require fees). Open-source implementations exist (e.g. ffmpeg’s

with GPL license, or spandsp library). These run on PC/SBC to encode/compress incoming PCM audio to G.729 frames (8 kHz sampling, 8–12 kbps).

•   Streaming/Networking: If voice is to go over Internet, use VoIP or streaming software. Options include: Asterisk (open VoIP PBX) with G.729 support (~$15 license), OpenCodec libraries, or custom code (e.g. GStreamer, PJSIP). For satellite, the PC can treat the satellite link as a broadband network interface.

•   Satellite Equipment: (if required)

 

•   Inmarsat BGAN/L-band modem: ~$3000+ (Thuraya or Inmarsat) for IP data (~432 kbps) which can carry VoIP (including G.729)【58†L74-L83】.

•   Iridium Short Burst Data / Certus: Iridium 9602/9603 modem ($200–$300) offers very low data

(~2.4 kbps), only feasible for very low-bitrate voice or short bursts (G.729 at ~8 kbps might barely fit with overhead). Certified voice modems (Iridium 9522B) are ~$1000.

•   Globalstar, Starlink, etc.: Alternative satcoms similarly cost hundreds to thousands.

•   Costs: E.g. a Starlink terminal is ~$500+ (for broadband) – ample bandwidth, but not intended for implants.

 

•   Computer: A modern PC or SBC ($35–$100) is sufficient. Real-time latency and processing for 8 kHz audio is very modest; even an 100 MHz MCU could handle G.729.

 

•   Misc.: Possibly a mixing board or audio interface if live playback is needed.

 

For example, one could use a Raspberry Pi 4 ($50) with built-in BLE, running Linux, connecting via Ethernet

or 4G modem to Internet, with                 + G.729 plugin ($0–$15) to send/receive calls. Or a Windows laptop

with a BLE dongle and a softphone (e.g. X-Lite with G.729 codec license).

 

7.    Regulatory & Technical Risks

 

•   Biocompatibility/Medical: None of these chips are medical-grade. They lack hermetic seals and biocompatibility. Full encapsulation in inert material (medical epoxy, Parylene, silicone) is mandatory, and eventual FDA/CE medical device clearance would be needed for humans. Risks include infection, corrosion, immune reaction, and chip failure inside the body.

•   Safety (EMF): Transmitting RF inside the mouth (close to tissues) requires ensuring Specific Absorption Rate (SAR) limits. BLE is low-power (<10 dBm) so likely safe, but this must be evaluated. Inductive/ultrasonic power carries safety risks (tissue heating).

•   Power/Lifetime: A small battery might last <1 day of continuous talk (few mA). Recharge (inductive through cheek?) or ultra-low-power sleep modes are needed. Battery/gas leakage risk if implant fails.

•   Wireless Range & Reliability: BLE’s indoor range is typically <10 m in free space, likely <5 m in- mouth. Host device must be nearby (e.g., wearable phone). Satellite uplink requires secondary gateway box, so range then depends on that link. Latency: BLE adds ~10–20 ms, plus G.729 (~15ms frame) plus network (~50–150 ms) → ~100–200 ms round-trip, acceptable for voice.

•   Legal/Privacy: Implant microphones raise serious privacy concerns. Unauthorized use (secret

listening) is illegal in most jurisdictions. Medical implants have strict labeling, but a self-made one could violate laws (health and wiretapping). Also, transmitting G.729 (patented) commercially may require licensing fees.

•   Technical Risks: Component yields at this scale are low; assembly and interconnections are very fragile. Body fluids could short circuits. Multi-component design <9×6×2.5 mm is extremely challenging to fabricate reliably.

•   Conclusion: The system is only theoretical. Major hurdles are power supply and safety approvals. In practice, battery-powered BLE audio SoCs are the only feasible short-term solution; all batteryless ideas remain research topics【58†L72-L81】【58†L82-L90】.

 

8.    Recommendations

The most feasible combination today is a micro-BLE audio SoC + digital MEMS mic + MEMS speaker, with a tiny battery. For instance:

 

•   SoC: Nordic nRF52840 or nRF52810 (WLCSP) – supports PDM/I²S, PWM, ample flash/RAM, BLE 5.x

【24†L27-L30】【25†L25-L33】.

•   Mic: TDK InvenSense ICS-43434 (3.50×2.65×0.98 mm, 64dB SNR)【40†L49-L57】.

•   Speaker: xMEMS Cowell (3.2×6.0×1.15 mm)【70†L247-L250】 plus a mini amp.

•   Rationale: These are all off-the-shelf, can handle 8 kHz PCM data, and fit together within the implant volume. A small rechargeable Li-ion button cell or supercap would power it. On the PC side, a BLE dongle and a G.729 software codec complete the chain.

We stress that batteryless voice implants are not currently practical for continuous streaming. Any short-range (RFID-like) scheme cannot sustain voice data rates. Therefore, we recommend including at least a minimal battery or capacitive power store.

 

Lastly, note regulatory/ethical constraints: implanting such a device in a person would face major medical device regulations (IEC 60601, ISO 10993 biocompatibility, FCC/CE radio, etc.) and privacy laws. This analysis is purely technical and does not imply any endorsement of such an implant in humans.

 

Sources: Nordic nRF528xx datasheets【24†L43-L50】【25†L37-L44】; NXP NXH3675 spec sheet【22†L224-

L231】; InvenSense (TDK) MEMS mic datasheets【40†L49-L57】【43†L349-L357】; xMEMS speaker data

【70†L219-L224】【70†L247-L250】; Atmosic energy-harvesting BLE press release【58†L72-L81】【58†L82-

L90】; component distributor listings【66†L1152-L1160】【63†L1-L4】. These are manufacturer/distributor datasheets and tech press references.