Implantable medical devices are widely used to provide the monitoring and mapping of biological signals, the support and enhancement of physiological functions, and the mitigation and treatment of diseases (1
). They are having transformative impact on health care and improving the quality of life for millions of people (2
). Particularly, implantable devices for monitoring physiological parameters, such as temperature (3
), blood pressure (4
), glucose (11
), and respiration (12
), to inform an individual’s health state are of great importance and interest for both the diagnostic and therapeutic procedures (9
). These devices perform in vivo sensing and recording of relevant signals directly at the target locations for early diagnosis of health issues (15
), allowing necessary interventions to be deployed at the onset of adverse events (1
Conventional implanted electronics are highly volume inefficient, generally requiring multiple chips, packaging, wires, and external transducers; batteries are often required for energy storage. A constant trend in electronics has been tighter integration of electronic components, often moving more and more functions onto the integrated circuit (IC) itself. This has usually been driven by lower cost and improved electronic functions through the reduction in interconnect parasitics. In the context of implanted electronics, this integration brings additional values through marked increases in this volume efficiency, defined as the amount of functions per unit displaced implant volume.
Here, we seek to push this volume efficiency to the ultimate limit with the monolithic integration of functions, including sensing, energy harvesting and storage, and data telemetry, onto a single complementary metal-oxide-semiconductor (CMOS) IC chip with no external elements. The use of capacitors for energy storage requires a continuous external wireless powering source but eliminates the need for batteries (16
). In our case, the device volume is less than 0.1 mm3
, comparable to a grain of table salt, improving biocompatibility by reducing foreign body rejection and tissue damage (1
), allowing access to limited interstitial spaces (18
), and interfering less with the physiological functions to be monitored (19
). Implantation procedures reduce to injection, which can be made easier and less invasive (9
). At the length scales of these mote devices (linear dimensions less than 600 μm), efficient coupling to radio-frequency or millimeter-wave electromagnetic energy is not possible because wavelengths are substantially larger than achievable antenna sizes. We instead use ultrasound, which attenuates in soft tissues on the scale of only ~0.5 to 1 dB/(cm·MHz) (20
). At ~8.3 MHz, a wavelength of only ~185 μm allows efficient coupling to an integrated piezoelectric transducer on the mote. Use of these motes in the context of ultrasound imaging also allows biogeographic locations for the motes to be determined (22
). While there are several recent examples of ultrasound-powered implants (20
), these devices have much lower levels of integration than that developed here and displaced volumes more than 10 times higher in the best case (26
We demonstrate use of our motes for sensing temperature, both as a vital sign of human health, essential in regulating metabolism and maintaining homeostasis (5
), and as a means to understand the thermal effects arising from medical procedures (27
). Examples of the latter include characterization of heating-based cancer therapies (28
) and therapeutic focused ultrasound (29
) (FUS). An emerging therapeutic use of FUS is in neuromodulation, where high-intensity, short-burst FUS (29
) is used to activate the nerve bundle. How FUS affects neural activity is still a matter of dispute, but thermal effects are certainly present (30
) and must be at least controlled in most cases. Conventional temperature measurement devices, which generally take the form of thermocouple-based temperature probes (28
), fail to accurately relay nerve temperature because of their bulky form factors and intrusive nature that interfere with the nerves of interest. While noninvasive techniques, such as magnetic resonance imaging, have been considered for measuring temperature (32
), the instrumentation requirements lead to limited applicability (34
This work presents a low-power, fully wireless mote with a monolithically integrated PZT ultrasound transducer for temperature monitoring. With a sub–1-nW power consumption and a volume of only 0.065 mm3, sub–50-mK temperature resolution and accuracy are achieved, with ultrasound energy harvesting for powering and backscatter for data telemetry. In vivo use of the device as an implantable temperature sensor was demonstrated in mice at two anatomical locations, the brain and the hindlimb.
Biocompatibility of an implant is improved both by minimizing displaced volume and by encapsulant coatings. Aggressive miniaturization was one of the major goals of this effort. Direct integration of the transducer on a CMOS IC chip achieves the most volume efficient design possible while enabling syringe injection with a needle as small as 18 G (see Materials and Methods and fig. S9) and operation at depths of up to 2 cm in tissue (see Materials and Methods and fig. S10). Functionality at greater tissue depths could be achieved with further reduction in the device power consumption, a reduction in the ultrasound operating frequency (which could lead to an increase in device size), and the application of two-dimensional ultrasound imaging array for more effective focusing of ultrasound energy to the implanted motes and more effective capture of the backscattered signals.
Biocompatible encapsulation with parylene ensures device functionality in the electrolytic environment of interstitial fluid. Parylene, as a common protective coating for biomedical devices, such as stents, pacemakers, and neural interfaces (37
), is chemically and biologically stable, mechanically flexible, and CMOS compatible (38
). It can be conformally deposited as thin films (2
). The thicker the parylene film, the more effective the layer can be expected as an encapsulant. Parylene thickness, however, is limited by ultrasound attenuation through this film (24
). We choose an 8-μm-thick parylene coating, which has minimal impact on ultrasound transmission (thickness is much smaller than the ultrasound wavelength at the operating frequency). Continuous soaking of a device in 1× PBS for >21 days showed no functional or performance impacts with long-term stability for temperature measurement of <138.6 parts per million (36
Angular orientation between the imager and the mote does affect the delivered power to the mote, which harvests maximum pressure when it is well aligned to the ultrasound source with a 0° incidence angle as the preferred configuration. A higher level of ultrasound power is needed to account for lower received pressure at a nonzero incidence angle [an incidence angle of ±20° requires eight times higher ultrasound intensity for activating the mote (36
); see fig. S11]. However, aligning the ultrasound source to the mote is challenging in an implantation scenario as mote orientation is not easily controlled during injection and can vary with time in vivo. This will be improved with the future use of a two-dimensional imaging array, which will allow the mote to work over a wide range of incidence angles at considerably lower power levels.
We have shown the use of these devices for monitoring temperature at the site of stimulation during FUS therapeutics, a rapidly growing area of investigation in medicine. FUS therapeutics depend on both thermal and mechanical effects and include, for example, thermal ablation and drug delivery through the blood-brain barrier (43
). Depending on the duty cycle, pulse width, and intensity, thermal effects can be minimized in favor of mechanical effects, such as cavitation (44
) and acoustic radiation forces (45
), to achieve desired therapeutic outcomes. Local temperature monitoring at or near the site of stimulation could play an important role in these therapeutic applications (43
The motes developed here demonstrate the volume efficiency possible with devices that fully exploit transducer integration, wireless powering, and backscatter-based telemetry with ultrasound. These motes can easily be extended to in vivo wireless sensing of other types of biological parameters including pH and chemical sensing.
MATERIALS AND METHODS
The fabrication of the sensing motes (see fig. S1 for a schematic illustration) uses a sheet of commercially available PZT material (7.24 cm by 7.24 cm by 267 μm; PZT-5H, Piezo Systems Inc.) and CMOS dies (4 mm by 3.2 mm by 300 μm; TSMC), each containing an array of 10 replicas of the same temperature sensor chip for mass-producing the motes. The PZT sheet, covered in 50 nm of nickel on both sides, is first dipped into ferric chloride for 5 s to remove the nickel layer, followed by photolithography with a 1.2-μm-thick AZ-1512 photoresist (MicroChemicals) to pattern arrays of 300 μm by 300 μm openings on the top side of the PZT sheet (fig. S1A). Electron-beam evaporation of a 10-nm chromium (Cr) adhesion layer and a 50-nm Au layer is performed (fig. S1B), followed by a lift-off process in acetone to create Cr/Au contacts matching the ground pads of the temperature sensor chips in the CMOS dies (fig. S1C). Such process is repeated on the bottom side of the PZT sheet to create contacts vertically aligned with those on the top side. A dicing saw (DAD3220; Disco) with a blade suitable for PZT (Z09-SD2000-Y1-90) with a kerf of 50 μm (14000 spindle, 3 mm/s feed speed) is used to dice the sheet into 4 mm by 3.2 mm pieces. Subsequently, a piece of ACF (4 mm by 3.2 mm, TFA220-8, H&S High Tech) is placed on the bottom side of a diced PZT piece with complete coverage and adhered to the PZT using a die bonder (Fineplacer Lambda, Finetech) with 0.1 N of force at 80°C for 5 s (fig. S1D). The PZT piece is then diced ~60 μm into its bottom side (fig. S1E) and precisely aligned and bonded to a CMOS die through the ACF to match the metal contacts on PZT with the corresponding ground pads on the die (fig. S1F). The ACF allows conduction in the vertical direction but not lateral directions to avoid shorting adjacent pads. This bonding process is performed using the same die bonder with a 100-N force at 150°C for 5 s. The parts of PZT not directly on top of the grounding pads are diced away from the top side to create free-standing microscale PZT transducers (fig. S1G). Afterward, sputter deposition of a 9-nm Cr adhesion layer and a 1.2-μm Cu layer is performed over the entire die surface to form the electrical connections between the top surfaces of the PZT transducers and the input pads for the 10 temperature sensor chips on the die (fig. S1H). These monolithically integrated sensor chips are then separated with the dicing saw along the chip edges to create individual sensing motes (fig. S1I). Last, an 8-μm-thick parylene layer is conformally deposited all over these sensing motes to provide biocompatibility for in vivo applications.
The image seen in Fig. 1B
was taken on an FEI Nova NanoSEM 450 SEM. The sample was prepared by mounting a sensing mote to a glass coverslip (25 mm by 25 mm by 0.2 mm) with double-sided Kapton tape. To prevent charging of the surface, a sputter coater (Cressington 108) was used to conformally deposit a 6-nm layer of Au. This was then imaged in the SEM at a working distance of 4.9 mm from the surface, with an acceleration voltage of 5 kV. The spot size was 18 μm, and the image was taken in secondary electron imaging mode, using the Everhartt-Thornley Detector. The scanning dwell time was set to 1 μs.
Preparation of ultrasound gel
The ultrasound coupling gel (Aquasonic 100, Parker Laboratories Inc., Fairfield, NJ) was degassed and warmed. To remove the air bubbles inside the gel, which heavily attenuate ultrasound, gel was centrifuged at 3000 rpm for 1 hour at 22°C. In addition, the degassed gel was microwaved for 8 s to warm it up to the body temperature of mice before use.
A mote was embedded in chicken or pork muscle tissue loaded in a three-dimensional (3D)– printed case or a cylindrical glass container, respectively. The case/container was filled with degassed water for ultrasound transmission and the L12-3v ultrasound probe was positioned directly above the mote. An ultrasound image was first formed to indicate the mote location (see Fig. 2A
and fig. S10A). A customized ultrasound signal was configured to contain duty-cycled pulses separated by 1 ms, where each pulse consists of four cycles of ultrasound at the 8.3-MHz operating frequency. This signal with an ISPTA
of ~0.044 mW/mm2
was used to power the embedded mote in the chicken tissue at a ~3.5-mm implantation depth, while an ISPTA
of ~1.85 mW/mm2
was used to provide power to the mote in pork tissue at a ~2-cm depth. Transmission was done by phasing 128 elements in the probe to focus the energy at the mote location. Backscattered data were picked up by the single element of the probe with the highest backscattered signal return or SNR (see Fig. 2C
and fig. S10B).
Incidence angle measurements
A mote was mounted in a 3D-printed case filled with degassed water for ultrasound transmission. The L12-3v probe was controlled by a rotation stage and placed ~2.2 cm above the mote to sweep from −20° to 20° with respect to the central axis of the mote, while the minimum energy to activate the mote and the corresponding SNR for each measured angle were recorded.
The Institutional Animal Care and Use Committee (IACUC) reviews and approves protocols for Columbia University’s programs for the humane care and use of animals and inspects the animal facilities and the research laboratories. Evaluation of the implanted motes was performed in compliance with IACUC regulations under the approved protocol of AC-AAAZ0451. Mice were obtained from the Jackson laboratory and housed in the Institute of Comparative Medicine facility of Columbia University. Surgeries were performed in animals 3 to 8 weeks old.
Mice were anesthetized with urethane (1 g/kg) administered intraperitoneally. While head-fixed by means of a Kopf stereotaxic apparatus, fur was removed with hair-removal cream, the scalp was cut open, and a 1.5 mm by 1.5 mm cranial window was drilled using 0.5-mm bits (fig. S6A). A single mote fixed to a 1 mm by 1 mm substrate was then placed on the exposed brain and fixed to the cranial window (fig. S6B). The scalp was then closed, covering the implanted device (fig. S6C).
Mice were anesthetized with urethane (1 g/kg) administered intraperitoneally. The hindlimb was fixed to an acrylic base by means of cyanoacrylate glue to minimize movement caused by breathing. The skin and underlying tissue were cut open with fine scissors, and the mote fixed on a 1 mm by 1 mm substrate was placed between the skin and the muscle, without further fixation. The skin was closed, covering the device (see fig. S7).
Sciatic nerve implantation
Mice were anesthetized with isoflurane (1 to 5%, v/v). The hindlimb was fixed with glue to a plastic base. An incision was made right under the femur, and the muscle was cut open to expose the sciatic nerve. Once the sciatic nerve was exposed, it was gently lifted using curved tweezers, and a PI substrate (~1 mm by 3 mm) with two motes attached was carefully placed under the nerve. A cartoon illustration of the strategy for such implantation is shown in the inset of Fig. 4C
. The motes were fixed with glue to the two edges of the substrate, leaving a space in the middle to accommodate the sciatic nerve. In such a way, each mote was conveniently located at each side of the nerve (Fig. 4C
) to record temperature changes right at the nerve during FUS stimulation (Fig. 4E
). In addition, a saline solution was applied to the incision to prevent the nerve from drying out during the experiment.
The mouse hindlimb was placed on top of a custom 3D-printed FUS coupling cone filled with degassed water (Fig. 4, A and E
). The cone was designed for a 3.57-MHz therapeutic ultrasound transducer (SU-107; SonicConcepts, Bothell, WA) where the focus, indicated by cross-hairs on the cone itself (Fig. 4D
), was aligned with the sciatic nerves inside the lower hindlimb. The transducer was connected to a function generator (33120A; Agilent, Santa Clara, CA) and RF amplifier (A150; E&I, Rochester, NY) for generating ultrasound pulses with 3.57-MHz operating frequency and 1-ms pulse duration. Four stimulation intensities (957.99, 1348.02, 1608.21, and 1745.83 mW/cm2
) were used in accordance with parameters found successful in a previous neuromodulation study (43
) to stimulate the sciatic nerves at intensities proven safe under histological examination. Neuromodulation outputs were recorded using a needle EMG electrode (Biopac, CA), and the corresponding temperature changes were recorded by the implanted motes at the sciatic nerve (Fig. 4, B and F
Intramuscular injection of the mote
Mice were anesthetized with urethane (1 g/kg) administered intraperitoneally. The hindlimb was fixed to an acrylic base by means of cyanoacrylate glue to minimize movement caused by breathing. A 1-ml syringe was loaded with saline solution containing seven motes (Fig. 1E
). The injection was performed right below the femur to target the sciatic nerve (fig. S9A) using an 18-G needle (inner diameter, 0.84 mm; outer diameter, 1.28 mm; Fig. 1D
). Following the injection, the skin and the underlying muscle were cut open with fine scissors to confirm the motes’ location right next to the nerve (fig. S9B).
Device fabrication was performed in part at the Columbia Nano Initiative (CNI) cleanroom and the City University of New York Advanced Science Research Center (ASRC) Nanofabrication Facility. Funding: This work was supported in part by a grant from the W. M. Keck Foundation and by the Defense Advanced Research Projects Agency (DARPA) under Contract HR0011-15-2-0054 and Cooperative Agreement D20AC00004. Author contributions: C.S. and K.L.S. conceptualized the study and performed data analysis. C.S. designed and tested the circuits, fabricated the fully integrated motes, conducted ultrasound characterization, performed experimentation, and wrote the paper. V.A.-P. and S.A.L. assisted in in vivo experiments. T.C. and J.E. assisted in device fabrication, ultrasound characterization, and in vitro experiments. S.A.L. and E.E.K. provided advice for FUS-based neuromodulation and guidance for in vivo experiments. K.L.S. provided overall supervision and guidance. All the authors provided comments and edited the manuscript. Competing interests: C.S. and K.L.S. are listed as inventors on a patent filed by Columbia University (patent no. US 10,898,168 B2, published 26 January 2021). The other authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials.