PaperlayerView exact revision on arXiv

Source-linked AI summary

A Sub-mm$^3$ Ultrasonic Free-floating Implant for Multi-mote Neural Recording

Mohammad Meraj Ghanbari, David K. Piech, Konlin Shen, Sina Faraji Alamouti, Cem Yalcin, Benjamin C. Johnson, Jose M. Carmena, Michel M. Maharbiz, Rikky Muller

arXiv:1905.09386v2eess.SPcs.ET

TL;DR

Sub-mm wireless implants face challenges in reaching deep neural targets while delivering power and data and supporting multi-site recording. This paper presents an ultrasonically powered recording mote with linearized echo modulation and orthogonal subcarriers. Two implants were synchronously powered at 50 mm, and the reconstructed signal showed 1.2% maximum static nonlinearity.

  • Problem

    Sub-mm implants face challenges in achieving centimeters-deep operation while delivering power and data, and deep-tissue multi-site recording with free-floating implants had not been demonstrated in the cited prior art.

    Source: §I · p. 2

  • Method

    The paper presents an ultrasonically powered recording mote with linearized analog echo modulation and orthogonal subcarriers for simultaneous multi-implant interrogation.

    Source: §I · p. 3, C. Linear echo modulator · p. 8

  • Results

    At 50 mm, two implants were synchronously powered for simultaneous uplink transmission, and reconstructed input showed 1.2% maximum static nonlinearity.

    Source: §V · pp. 9, 11

  • Takeaways & Limitations

    The measured depth supports the stated scope of recording most peripheral nerve targets and deep brain targets through thinned skull with a single-element unfocused transducer.

    Source: §I · p. 3

    Carrier noise dominated total input-referred noise; the estimated chip-only noise was 5.37 µV_RMS in a 5 kHz bandwidth.

    Source: §V · p. 9

Abstract

from arXiv · show

A 0.8 mm$^3$ wireless, ultrasonically powered, free-floating neural recording implant is presented. The device is comprised only of a 0.25 mm$^2$ recording IC and a single piezoceramic resonator that is used for both power harvesting and data transmission. Uplink data transmission is performed by analog amplitude modulation of the ultrasound echo. Using a 1.78 MHz main carrier, >35 kbps/mote equivalent uplink data rate is achieved. A technique to linearize the echo amplitude modulation is introduced, resulting in <1.2\% static nonlinearity of the received signal over a $\pm$10 mV input range. The IC dissipates 37.7 $\mu$W, while the neural recording front-end consumes 4 $\mu$W and achieves a noise floor of 5.3 $\mu$V$_{rms}$ in a 5 kHz bandwidth. This work improves sub-mm recording mote depth by >2.5x, resulting in the highest measured depth/volume ratio by $\sim$3x. Orthogonal subcarrier modulation enables simultaneous operation of multiple implants, using a single-element ultrasound external transducer. Dual-mote simultaneous power up and data transmission is demonstrated at a rate of 7 kS/s at the depth of 50 mm.

I. INTRODUCTION

Sub-mm implants promise less invasive access to deep neural sites, but depth, power and data delivery, and multi-site communication remain difficult together. This work presents a 0.8 mm^3 ultrasonically powered mote verified at 50 mm and designed for simultaneous multi-mote operation through one unfocused transducer.

I. INTRODUCTION

  • Sub-mm implants targeting deep structures must limit size to reduce implantation trauma and long-term scarring, while centimeters-deep operation and multi-site communication pose power and data challenges.Smaller devices may enable catheter-based, laparoscopic, or injection-based implantation.Source: §I · p. 2
  • 6 mm was the maximum theoretical depth of the smallest prior free-floating implant, while another approach reached 20 mm using a tertiary coil for transcranial power transmission.The 20 mm approach was described for epicortical implants; its coil form factor and coplanar implant requirement limit extension to deep-tissue recording.Source: §I · p. 2
  • 8.8 mm was the operating depth of a prior ultrasonically powered implant, restricted by its lack of a low-noise gain stage and need for a focused transducer.The introduction also notes that deep-tissue, multi-site recording with free-floating implants had not been demonstrated in the cited prior art.Source: §I · p. 2
  • 0.8 mm^3 implants were verified at 50 mm depth in a tissue phantom and designed for simultaneous power-up and parallel telemetry using a single-element unfocused transducer.The reported scope includes most peripheral nerve targets and deep brain targets through thinned skull.Source: §I · p. 3

II. SYSTEM OVERVIEW

The system uses an interleaved pulse-echo link for both power and data, with analog modulation of the returning echo. It adds echo linearization and orthogonal subcarriers to support cleaner signals and simultaneous multi-implant interrogation.

II. SYSTEM OVERVIEW

  • A single ultrasound link provides power and uplink data in an interleaved pulse-echo scheme, avoiding a secondary resonator and off-chip storage or matching capacitors.The echo is received through the same external transducer after the implant harvests energy and wakes to record neural signals.Source: §II · p. 3
  • A low-noise analog front end reduces input-referred noise by 34x, while a reflection-coefficient linearization technique enables linear analog amplitude modulation of the echo.The design also enables simultaneous multi-implant interrogation without sacrificing interrogation frequency, using one external transducer.Source: §II · pp. 3–4
  • Orthogonal coded subcarriers let multiple implants transmit amplitude-modulated echoes simultaneously, with synchronized code multiplication and averaging separating a selected channel from others.Each implant can use a unique code, and decoding is possible when the finite echo contains an instance of a CDM frame.Source: §II · p. 4
  • Analog echo modulation carries more information per cycle than digital modulation and allows shorter minimum echo periods, enabling higher interrogation frequencies and uplink data rates.The comparison assumes the same number of ultrasound cycles for amplitude settling or digital state switching.Source: §II · p. 4, Figure 3 · p. 4
  • The IC rectifies and regulates piezo voltage to 1 V, extracts a clock, and uses a chopper-stabilized amplifier followed by a linear gm-cell to modulate echo amplitude.The external interrogator conditions and reconstructs the received echo digitally.Source: §II · p. 4, Figure 4 · p. 4
  • Because the pulse-powered implant is memoryless and cannot cancel electrode DC offset, its input linear range extends beyond ±10 mV to minimize distortion.The output echo modulation linearity is achieved by linearly modulating the piezo voltage.Source: §II · p. 5

III. LINEAR ECHO MODULATION

The section explains how nonlinear piezoelectric reflection distorts analog echo modulation and develops a simpler resonance-based model to guide linearization. The measured and modeled reflection curves agree, and at series resonance the reflection coefficient is approximately proportional to piezo voltage.

III. LINEAR ECHO MODULATION

  • Nonlinear Γ versus termination resistance RE distorts analog echo modulation, while concurrent energy harvesting imposes a minimum RE that constrains the modulation range.The section motivates co-design of the piezo and modulating IC to reduce this distortion.Source: §III · p. 5
  • Fig. 7 compares the analytical expression, its single-parameter approximation, and measurements; the expressions match the measured values well.The approximation uses RS, which can be obtained empirically or by FEM simulation.Source: Figure 7 · p. 5, §III · p. 5
  • The piezo model treats ports 1 and 2 as acoustic ports and port 3 as the electrical port, with the port-1 acoustic impedance depending on the terminations at ports 2 and 3.The section also provides a list of typical piezoelectric parameters in Table I.Source: §III · p. 5, Table I · p. 6
  • At series resonance, Γ is approximately proportional to the voltage across the piezo, so linear voltage modulation can linearize Γ.The simplified model uses the piezo’s internal series resistance, RS; measurements on a 0.75x0.75x0.75 mm3 piezo gave RS = 1.5 kΩ.Source: §III · p. 5

IV. INTEGRATED CIRCUIT IMPLEMENTATION

The implementation must power up, acquire neural signals, and transmit within the brief interval of an ultrasound pulse. Its power-management circuitry charges the storage capacitor and establishes references before amplifier initialization and signal acquisition.

IV. INTEGRATED CIRCUIT IMPLEMENTATION

  • At 50 mm depth, the IC has less than 66 µs (<2ToF) to power up, record neural signals, and transmit data.The timing diagram in Fig. 8 presents the power-management block interconnection and chip timing.Source: §IV · p. 6, Figure 9 · p. 6
  • About 5 µs are needed to charge the 130 pF storage capacitor and generate the on-chip 1 V and 0.5 V references.The charging interval reflects the storage-capacitor time constant, piezo series resistance, and rectifier conduction angle.Source: §IV · p. 6
  • After reference generation, the amplifier initializes in 3 µs; without ultrasound, the storage capacitor discharges in about 10 µs, setting the minimum inter-pulse interval for reinitialization.This corresponds to a minimum operating depth of 14 mm without an acoustic spacer.Source: §IV · p. 6

A. Low-noise Amplifier

The low-noise amplifier uses a capacitive-feedback, fully differential architecture with offset cancellation and chopping to manage noise. Echo averaging at the interrogator limits the effective signal and noise bandwidth to 5 kHz at a 10 kHz interrogation rate.

A. Low-noise Amplifier

  • The capacitive-feedback LNA uses 0.44 pF feedback and 4.7 pF load capacitors, with feedback factor β ∼1/16, to set its noise.The amplifier is initialized by switches, and its common-mode voltage is set to the 0.5 V mid-rail.Source: A. Low-noise Amplifier · p. 6
  • Auto-zeroing samples amplifier offset on the feedback capacitors, while chopping shifts the signal and moves kT/C and low-frequency amplifier noise out of band after demodulation.The measured electrode impedance in saline is below 100 kΩ above 100 Hz, versus an AFE input impedance of about 1.3 MΩ at the highest chopping frequency.Source: A. Low-noise Amplifier · p. 7
  • The LNA has a ±160 mV output range and 180 kHz bandwidth, sufficient to pass the third harmonic of the highest 55 kHz subcarrier.Its effective noise bandwidth is reduced by post-processing to 5 kHz at a 10 kHz interrogation frequency.Source: A. Low-noise Amplifier · p. 7
  • The fully complementary differential amplifier is selected for an excess noise factor α close to 1 and high Gm/ID; the amplifier noise depends on α, γ, β, and CT.The figure captions identify the LNA topology and its fully complementary OTA topology.Source: A. Low-noise Amplifier · p. 7, Figure 10 · p. 7
  • The architecture includes a linear gm-cell following the LNA, and Fig. 11 identifies its architecture.The supplied figure caption does not specify additional operating characteristics.Source: Figure 11 · p. 7
  • At a 10 kHz interrogation frequency, echo averaging reduces the signal and noise bandwidth to 5 kHz.Averaging over each pulse applies a sinc low-pass filter with 3 dB bandwidth 1/(2Tintegration).Source: A. Low-noise Amplifier · p. 7

B. Linear Gm-cell design

The linear gm-cell converts the amplified voltage into current while maintaining a wide input range and low nonlinearity. Its PFET-based design saves area, and the final current mirror connects to the rectifier output for echo modulation.

B. Linear Gm-cell design

  • The gm-cell accepts ±160 mV with better than 0.5% nonlinearity by placing a PFET in triode using a differential super source follower.The PFET replaces a resistor to save area without sacrificing linearity.Source: B. Linear Gm-cell design · p. 7
  • Fig. 3 identifies the packaged implant and IC micrographs, implant dimensions, and power-consumption breakdown.The figure caption supplies no dimensions or power values.Source: Figure 13 · p. 8
  • Fig. 14 presents measured ENIG electrode impedance and an inset single-electrode model.The caption does not state the measured impedance values.Source: Figure 14 · p. 8
  • The gm-stage transconductance is 120 µS, and its final current mirror connects directly to the rectifier output to provide the uplink modulation signal.The final mirror uses 2.5 V devices because the rectifier voltage is higher.Source: B. Linear Gm-cell design · p. 8

C. Linear echo modulator

The modulator linearizes echo amplitude modulation by linearly modulating the piezo voltage. It uses the active rectifier as a synchronized up-conversion current mixer, with rectifier nonlinearity below 0.5% at 10% modulation depth.

C. Linear echo modulator

  • The synchronized current mixer modulates piezo voltage by up-converting the modulation current in phase with the piezo, producing voltage peaks and valleys that drop by ImRs.The relationship between reflection coefficient Γ and piezo voltage V3 makes linear echo amplitude modulation possible.Source: C. Linear echo modulator · p. 8
  • The active rectifier is reused to implement the synchronized up-conversion mixer with minimal hardware overhead.High-frequency common-gate RF amplifiers in the rectifier serve as comparators.Source: C. Linear echo modulator · p. 8
  • At 10% modulation depth, rectifier nonlinearity is below 0.5%, keeping the piezo voltage—and ultimately the reflection coefficient—nearly linear with modulation current.The single-ended gm-cell is connected to the rectifier output because rectifier nonlinearity is minimal.Source: C. Linear echo modulator · p. 8

V. MEASUREMENT RESULTS

Measurements characterize the assembled mote’s power, linearity, noise, single-mote transmission, depth and alignment tolerance, and simultaneous operation of two implants. In vitro tests demonstrate signal transmission at 45 mm and dual-mote operation at 50 mm.

V. MEASUREMENT RESULTS

  • The implant consumed 37.7 µW during normal operation with a 50% duty cycle; its chip occupied 0.25 mm2.The IC was fabricated in a TSMC 65nm LP CMOS process and assembled with a bulk piezo on a flex PCB interposer.Source: §V · p. 8
  • The chip measured 1.2% maximum static nonlinearity, 50 dB SFDR, and −44 dB THD for a reconstructed 313 Hz, 20 mVPP sine wave.The measurement also showed no visible harmonic tones for a 10 mVPP input signal.Source: §V · p. 9, Figure 17 · p. 9
  • The total input-referred noise density was 328 nV/√Hz, while the estimated chip-only noise was 76 nV/√Hz, or 5.37 µVRMS in 5 kHz bandwidth.Carrier noise measured without the chip was 319 nV/√Hz; chopping partially removes its 1/f noise.Source: §V · p. 9
  • For a 20 mVPP input range, the single-mote link achieved 47.96 dB SNR; without carrier noise, SNR would improve by approximately 10–12 dB.The carrier noise dominated overall link noise; the figure summarizes SNR and equivalent uplink data rate across configurations.Source: §V · pp. 9–10, Figure 19 · p. 10
  • At 45 mm depth, the mote wirelessly transmitted an 800 ms prerecorded rat motor-cortex signal while interrogated at 10 kS/s.The signal was acquired from an awake-behaving rat and compared with the reference recording.Source: §V · p. 10, Figure 20 · p. 10
  • At approximately 50 mm depth, harvested voltage and echo modulation depth were concurrently large; acceptable harvested voltage (>4.5 Vpp) and modulation depth (>20%) were measured at 70 mm.Beyond the transducer’s Fresnel distance, modulation depth decayed by 6% per cm.Source: §V · p. 10, Figure 21 · p. 11
  • Horizontal misalignment up to ±1.75 mm was acceptable with negligible modulation-depth loss, and the margin can double with a 1-inch external transducer.The chip remains fully operational above 4.2 Vpp harvested voltage.Source: §V · p. 11
  • At 50 mm depth and 2 mm implant separation, two implants transmitted simultaneously, with transmission quality maintained over ±2 mm vertical misalignment.The implants used orthogonal 55 kHz and 27.5 kHz subcarriers; the paper estimates four implants can be interrogated at 35–70 mm and up to eight at depths ≥70 mm.Source: §V · pp. 11–12, Figure 22 · p. 11

VI. SUMMARY

The work presents a 0.8 mm3 free-floating implant using one ultrasound link for power harvesting and analog data back telemetry. It reports improved depth and noise performance over prior sub-mm3 implants and simultaneous operation of two motes with a single-element transducer.

VI. SUMMARY

  • The 0.8 mm3 implant uses a single ultrasound link for wireless power harvesting and analog data back telemetry.The work reports a 20 mVPP linear input range for its echo-modulation technique.Source: §VI · p. 12
  • Compared with prior art –, the work reports >2.5x greater operating depth and an approximately 3x higher measured depth/volume ratio.It also reports the lowest nonlinearity at the highest input range and a comparable NEF to the state of the art.Source: §VI · p. 12
  • Two free-floating motes were simultaneously powered and communicated with without a specialized external transducer.Source: §VI · p. 12

APPENDIX PIEZO Γ VS. VOLTAGE LINEARITY

The appendix derives the piezo reflection model using a thickness-mode resonating three-port network. It relates the electrical-port impedance to acoustic terminations and simplifies the model at series resonance.

APPENDIX PIEZO Γ VS. VOLTAGE LINEARITY

  • The piezo is modeled as a thickness-mode resonating three-port network to derive the earlier result.Its input-output port relationships are described by the cited network model.Source: APPENDIX PIEZO Γ VS. VOLTAGE LINEARITY · p. 12
  • The acoustic impedance seen into port 1 is specified with ports 2 and 3 terminated by ZB and ZE, respectively.The appendix also defines Z3,NL as the impedance into port 3 when ports 1 and 2 are acoustically unloaded (ZB = 0).Source: APPENDIX PIEZO Γ VS. VOLTAGE LINEARITY · p. 12
  • At the series resonant frequency, the acoustically unloaded piezo’s electrical-port impedance is real, and the expression is further simplified.The appendix states that the piezo is then modeled by a voltage source and series resistance.Source: APPENDIX PIEZO Γ VS. VOLTAGE LINEARITY · p. 12
Loading 1905.09386v2…