Batteryless, Wireless, and Secure SoC for Implantable Strain Sensing

The past few years have witnessed a growing interest in wireless and batteryless implants, due to their potential in long-term biomedical monitoring of in-body conditions, such as internal organ movements, bladder pressure, and gastrointestinal health. Early proposals for batteryless implants relied on inductive near-field coupling and ultrasound harvesting, which require direct contact between the external power source and the human body. To overcome this near-field challenge, recent research has investigated the use of RF backscatter in wireless micro-implants because of its ability to communicate with wireless receivers that are placed at a distance outside the body $(\sim 0.5$ m), allowing a more seamless user experience. Unfortunately, existing far-field backscatter designs remain limited in their functionality: they cannot perform biometric sensing or secure data transmission; they also suffer from degraded harvesting efficiency and backscatter range due to the impact of variations in the surrounding tissues. In this article, we present the design of a batteryless, wireless and secure system-on-chip (SoC) implant for in-body strain sensing. The SoC relies on four features: 1) employing a reconfigurable in-body rectenna which can operate across tissues adapting its backscatter bandwidth and center frequency; 2) designing an energy efficient 1.37 mmHg strain sensing front-end with an efficiency of 5.9 mmHg $\cdot $ nJ/conversion; 3) incorporating an AES-GCM security engine to ensure the authenticity and confidentiality of sensed data while sharing the ADC with the sensor interface for an area-efficient random number generation; 4) implementing an over-the-air closed-loop wireless programming scheme to reprogram the RF front-end to adapt for surrounding tissues and the sensor front-end to achieve faster settling times below 2 s.

environments where they need to adapt to changing conditions, such as the available wireless power, bandwidth, and channel conditions. This article takes a new approach to developing wireless and batteryless implants by co-designing all the subsystems that comprise the end-to-end wireless sensing stack. A key strength of our design is its cross-layer reconfigurability that spans energy-harvesting, wireless communication, security, and analog biometric sensing.
We present the design, implementation, and evaluation of a batteryless wireless and secure SoC for implantable sensing. Our design extends our previously proposed SoC [4] by introducing innovations along two key fronts: 1) a fully integrated security-sensing-communication stack: which shares blocks along the stack for an area and power efficient design; 2) over-the-air closed-loop wireless programming: which leverages the asymmetric wireless link to break the tradeoff between energy-efficiency and latency. Our SoC incorporates a reprogrammable rectenna for integrated and adaptive energy harvesting and backscatter communication, a lowpower security engine, and an efficient sensing front-end for biometric sensing.
We demonstrate our SoC's ability to perform pressure sensing for applications in long-term monitoring of gastrointestinal (GI) motility. Such monitoring can be achieved by sensing the internal pressure along the GI tract [5], and is used to understand digestive system disorders and the development of obesity. The reasons why our design is particularly desirable for this application are multifold. First, our batteryless design eliminates the need for surgical replacement, mitigates the health hazards due to battery depletion risks for in-body implants, and reduces the overall size of the implant (since batteries can consume up to 50% of the volume of GI sensor) [1], [6] while directly integrating the power management with energy harvesting. Second, our built-in security builds on past work [7], [8] to bring privacy, confidentiality, and authentication to sensed in-body biometrics. Third, our programmable RF front-end allows our sensing to adapt to different in-body tissues, which is particularly beneficial for a sensor traveling along the GI tract [6]. Finally, our sensor's ability to communicate with far-field wireless transceivers outside the body makes it more desirable than prior approaches that require direct contact with the human body [3], and paves the way for more seamless future biomedical and bionic systems.

II. IN-BODY STRAIN SENSING
This section describes the design of the strain sensing node, circuit-level implementation, over-the-air programming to reconfigure the chip, and the logic controlling the whole operation while guaranteeing system security.

A. SYSTEM OVERVIEW
The in-body strain sensor, shown in Fig. 1, incorporates a built-in security engine to securely transfer data between an  external reader and the in-body sensors. The SoC is integrated with a flexible antenna and laminated on a batteryless sensor to extract internal biometric data.
The SoC design includes four main subsystems: 1) a power management unit (PMU) with energy harvesting; 2) a transceiver for wireless communication; 3) a sensor front-end for data acquisition; and 4) a security engine for authentication and encryption as illustrated in Fig. 2. The RF energy harvester is designed as a 6-stage rectifier while the PMU has a voltage limiter to protect the internal devices from being overstressed at high RF input power, a comparator which turns on the LDO when the stored voltage is above 0.5 V, and a power-on reset circuit to initialize the internal registers. The transceiver is made up of a receiver chain and a backscatter transmitter. The receiver employs an envelope detector to downconvert the RF input signal, then an integrate-and-dump filter resolves the incoming bits which use pulse interval encoding (PIE). However, the transmitter utilizes a ring oscillator and digitally modulates its frequency to backscatter an FM0-encoded bitstream back to the reader. The sensor interface acquires the strain signal through the analog front-end (AFE) which incorporates a low-noise amplifier (LNA) followed by a programmable gain amplifier (PGA) then a 10-bit ADC. A security engine based on [8] is implemented on-chip to provide authenticity and confidentiality through the use of advanced encryption standard in Galois-Counter Mode (128-bit AES-GCM). In addition, the counter mode initialization is performed via a true random number generator (TRNG) implemented as a switched-capacitor chaos map sharing the ADC with the sensor front-end.

B. CIRCUIT-LEVEL IMPLEMENTATION 1) SENSOR FRONT-END
The sensor AFE consists of an LNA directly interfacing with the pressure sensor, a PGA scaling the signal to the full dynamic range, a SAR ADC to provide 10-bit data samples, and an accumulator-based offset cancelation loop shown in Fig. 3. The input stage of the LNA incorporates a 7-bit binary weighted units of pMOS differential pairs for programmable offset. The input pair sizing is then selected by the dc cancelation loop where a larger effective area produces a smaller offset. Once the cancelation loop settles, the sampled data is fed to the SAR ADC to be processed by the encryption engine, and finally, backscattered to the base station.
Over-the-air programming creates a closed-loop system with the base station where the internal states of the sensor front-end are backscattered through the uplink packet. Therefore, the base station can use the "Reconfigure" command to set the initial state of the accumulator and provide a faster settling time to the dc cancelation loop.

2) RECONFIGURABLE WIRELESS RF FRONT-END
In-body sensing requires placing the SoC with its antenna inside the human body. The main problem, however, is that the antenna impedance and efficiency are affected by the surrounding tissues. An antenna optimized to measure the strain on the stomach is not optimum when used to measure the strain on the bladder. As each tissue has different electrical properties (conductivity and permittivity), then antennas of identical dimensions would have different current density distributions and, subsequently, electromagnetic fields.
To overcome this challenge, we designed a reconfigurable RF front-end where the node can be programmed to operate across a wide range of tissues. Our reconfigurable rectenna is shown in Fig. 4 where a high-Q antenna is conjugatematched with a capacitive 6-stage rectifier. A 6-bit capacitor bank tunes the input impedance of the rectifier to match the antenna across tissues, while another capacitor bank is used as a programmable load coupled to the loop antenna to program its current distribution and, hence, radiation properties.
The antenna design builds on our previous work on programmable rectennas for RF backscatter [6]. In comparison to this previous work, which only tunes the center frequency, the SoC presented here can be optimized in a high-Q highefficiency mode or adapt to scale its bandwidth by a factor of 8 for higher bandwidth datarate. This is based on two design features: first, designing a bandwidth-programmable antenna with programmable matching; second, in contrast to the existing designs, our system can utilize closed-loop over-the-air programming to adapt and reconfigure its RF front-end.
A programmable-bandwidth antenna is implemented by designing a loop antenna which is coupled with an outer loop terminated by a tunable capacitive load. The tunable capacitive load modifies the effective electrical length of the outer loop controlling how much the fields of both loops can add up around the frequency of interest. When the resonance frequencies of each loop with the capacitive rectifier are staggered by tuning them to be slightly different, a wider bandwidth is achieved as both resonances combine to produce almost one effective wider peak.
As adding tunable resistors to the harvesting chip would only increase the losses and ultimately reduce the system's energy harvesting efficiency, we only tune the capacitive part of the harvester's input impedance. As for the real part, the antenna's outer loop is loaded by a tunable capacitor bank which couples with the inner loop tuning the radiation resistance and the bandwidth of the antenna without considerable degradation to the harvesting efficiency. Therefore, the real part is tuned via the antenna's outer tunable capacitor bank (C t ) while the rectifier's input capacitance is programmed through the input matching capacitor bank (C m ).
Since the node is batteryless, the antenna does not know how and when to change its programming. Here, we leverage the asymmetric link by having the reader do a configurationspace search to find the optimum configuration for the required mode of operation. Each configuration is then stored on the reader and sent to the node in the form of a Reconfigure downlink command.

3) POWER MANAGEMENT AND COMMUNICATION
The PMU has a voltage limiter to protect the internal devices from being overstressed at high RF input power, a comparator which turns on the LDO when the stored voltage is above 0.5 V, and the power on reset (PoR) circuit to initialize the internal flip flops. The LDO is implemented with a feedback error amplifier and a Miller-compensated pMOS pass transistor to generate a regulated voltage at different current loads.
The transmitter chain incorporates an FM0 encoder, a continuous phase frequency encoding scheme, to modulate a pair of antenna switches which modulate the node's reflection coefficient between a reflective state and absorptive state. A current-starved ring oscillator generates the clock necessary to frequency modulate the backscattered bitstream.
The receiver extracts the RF envelope through ac-coupling the output of the (N-1)th output of the N-stage rectifier to the input of a cascade of differential amplifiers. The amplified signal is quantized through a Schmitt-trigger comparator then an analog integrate-and-dump decodes the pulse-interval-encoded (PIE) input stream to extract the packet's content.

C. SECURITY AND LOGIC
The core of the security engine consists of an energyefficient hardware accelerator for advanced encryption standard (AES) [9] in Galois counter mode (GCM) [10] for authenticated encryption and decryption. The two main components of the AES-GCM module are the AES block cipher (the forward cipher mode is used for both encryption as well as decryption) and the GHASH hash function (which uses Galois field multiplications for authentication). The AES block cipher is implemented as a 128-bit data-path with a parallel architecture leveraging larger active area to provide an energy-efficient design operating for only short periods of time [8]. The AES implementation consists of 20 parallel instances of the S-Box (substitution function) which is not only the most important nonlinear component of the AES function but also accounts for majority of the area and power consumption. Out of these 20 S-Boxes, 16 are used for the encryption and the remaining four are used for the key schedule computation. The GHASH implementation consists of four Galois multiplication stages which are executed iteratively and provides the perfect a balance between area and energy efficiency [8]. The AES module takes 11 cycle per block encryption and the GHASH module takes 32 cycles per Galois field multiplication. Fig. 5 shows a block diagram of the overall architecture of the AES-GCM engine.
The on-chip AES-GCM security engine ensures node authenticity and data confidentiality. It is employed to authenticate the downlink commands as well as encrypt the uplink biometric data being transmitted via RF backscatter. The AES-GCM engine utilizes a counter-mode encryption  algorithm which requires a unique initialization vector (IV) to be used with each new block of data. A switched-capacitor chaos map is designed as a TRNG to generate the IV for the security engine. Such TRNG comes at a 2% power overhead only as it shares the ADC with the sensor front-end to quantize the TRNG output for each block-cipher.
The SoC finite-state machine is illustrated in Fig. 6 starting with the Charge state where the RF harvester stores the input RF energy on an off-chip storage capacitor. The chip keeps charging till the stored dc voltage V DC is higher than the PMU turn on threshold V ON . Then, the LDO is powered up and regulates the supply for the receiver chain (Receive state) to decode the input bits until the correlator detects a 20-bit preamble signifying the beginning of the packet. The 280-bit downlink packet is stored, with a custom structure illustrated in Fig. 7, and the encrypted command is passed on to the AES-GCM engine along with the IV and authentication tag for authentication and decryption. The AES-GCM engine is then turned on with a 128-bit hardwired preshared key. The chip remains in the Decrypt state till the engine triggers an interrupt signal along with the authentication result. An unauthenticated packet returns the chip to the Receive state while an authenticated one is decrypted into a downlink command to reprogram the chip over-the-air (Reconfig RF or Reconf sense states) or to begin in-body strain sensing in the SENSE state.
For the SENSE state, the chip progresses by turning on the AFE and ADC along with the dc offset cancelation loop. After the front-end settles, the ADC is multiplexed to the TRNG to generate the IV and buffer it for encryption. Finally, in-body strain sensing starts with the data being encrypted, buffered, and backscattered to the reader. The internal accumulator states and gain configurations of the LNA and PGA are concatenated in the first 20 bits of the uplink packets, as illustrated in Fig. 7.
In principle, a nonvolatile memory can be used to store the optimum configurations for the chip. However, nonvolatile memories require a higher voltage for their read and write cycles compared to the subthreshold regime of 0.5 V that the implant utilizes. Over-the-air programming mitigates increasing the node's supply voltage, and power, by taking advantage of the power asymmetry between the reader and the node.

D. ADC-SHARING AND TRNG
The generation of the IV for the AES-GCM engine at a low area and power overhead can be achieved by sharing the SAR ADC between the sensor front-end and the TRNG circuit.
The TRNG is implemented as a discrete-time chaos-map, utilizing a residue amplifier which is similar to a 1.5-bit stage of a pipeline ADC but the residue is fed back to the input [11], [12]. With a random initial sample based on the circuit noise, the output bifurcates into a range of random numbers covering the voltage range of the whole amplifier.  The random discrete output is then sampled by the SAR ADC as it is multiplexed to take its input from the TRNG as shown in Fig. 8 during the TRNG state of the systemlevel finite-state machine before beginning encryption in the Encrypt start state.

III. MEASUREMENT RESULTS
The chip was fabricated using a TSMC 65-nm CMOS process occupying a 2 mm×1 mm active area as shown in the die photo of Fig. 9 which illustrates the main building blocks of the SoC. A commercial strain sensor is interfaced with the sensor node SoC which is mounted on a PCB with a printed microstrip antenna.
The active power consumption of the chip varies from one state of operation to the other. The maximum power is consumed when the security engine is on and draining 4.8 μW contributing to 43.7% of the total active power, as shown in Fig. 10. In contrast, during other states the power is dominated by the PMU which provides a stable supply for the logic, AFE, transmitter, and receiver.

A. WIRELESS HARVESTING AND COMMUNICATION
The strain-sensing SoC is evaluated in a wireless setup where an RF signal generator is configured as a custom base station.
The SoC powers up by harvesting the incoming RF energy through its reconfigurable rectenna, storing it over an offchip capacitor, and generating an LDO-regulated supply of 0.55 V for the remaining baseband and communication circuits. Then, the PMU generates a PoR signal to initialize all internal states of the finite state machine to their default values as shown in Fig. 11.
The downlink packet can be a Reconfigure command used to reprogram the sensor and the RF front-ends, or a "Sense" command used to acquire biometric data. When a Sense packet is received, the SoC turns on the sensor interface to sample the strain input. The time-domain measurements of the uplink operation are shown in Fig. 12 in which the security engine triggers an interrupt signal once an authentication result is ready. An authenticated packet progresses to power up the sensor interface and the ADC starts buffering its samples serially. Then, the AES-GCM engine encrypts the data and the payload is framed with the header to be backscattered to the reader.
The measured uplink bitstream is magnified in Fig. 13, showing the packet composition. The packet header contains a 32-bit node ID to select the sensor as well as a 96-bit IV used by the security engine. Then, the payload carries a variable-length data followed by a 128-bit authentication  tag. The bursts of packets are then buffered, FM0 encoded, and backscattered to the base station.

B. CROSS-TISSUE ADAPTATION
To characterize how the SoC adapts across different tissues, the rectenna is programmed across its range of bandwidths and center frequencies showing its full range of coverage. To tune the bandwidth and center frequency, we varied the tuning capacitance C t and the matching capacitance C m and the antenna's programmable architecture described earlier.

1) BANDWIDTH ADAPTATION
The first adaptation experiment focused on evaluating the bandwidth reconfigurability. To measure the bandwidth, we transmitted an RF carrier frequency from an external source and programmed the SoC to backscatter at different frequencies. The RF signal generator starts by transmitting a constant frequency signal to power up the node. The node then backscatters at different frequencies and a spectrum analyzer is used to measure and record the magnitude of the backscattered power at different frequencies around the RF carrier. Such measurement is then repeated across different configurations for the antenna programming and the optimum configurations for different bandwidths are plotted.
The measured RF bandwidth is shown in Fig. 14 where the rectenna can be reconfigured from a high-Q 10-MHz configuration to an 80-MHz bandwidth according to the available channel conditions.

2) RESONANCE ADAPTATION
The second adaptation experiment focused on evaluating the system's resonance reconfigurability. Similar to the above experiment, we transmitted an RF carrier frequency from an external source and programmed the SoC to backscatter at different frequencies. We measured the backscatter response using a spectrum analyzer. The measured RF bandwidth, plotted in Fig. 15, shows that the system's programmability extends to tuning the center frequency to adapt for different tissue environments where the optimum frequency of operation can be tuned covering a range of 300 MHz demonstrated in the figure.

C. IN-BODY PRESSURE SENSING
The main function of the sensor front-end is to amplify the strain signal to the full dynamic range of the 10-bit ADC while canceling the offset to prevent saturation. The measured time-domain digital output is plotted in Fig. 16 for a continuous pressure measurement setup. The figure demonstrates how the 10-bit samples track the input reference pressure during a fixed-value pressure and as it progresses inside its dynamic range. Additionally, Fig. 17 plots the long VOLUME 3, 2023 41 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.  term accuracy by measuring the dc output for a fixed pressure input over a 4-h period. The measurement is performed at a low-gain configuration (G = 100) achieving a resolution with a standard deviation of σ G,low = 1.37 mmHg, whereas a high-gain configuration (G = 1000) provides a resolution of σ G,high = 0.9 mmHg deviation. The PMU provides a 0.55-V supply to the sensor frontend which consumes 1.65 μW, achieving a conversion efficiency of 4.29 nJ/conv. step. With a 1.37-mmHg resolution, the system achieves a figure of merit (FoM) of 5.9 mmHg·nJ/conv. step, where a lower FoM (FoM = [Power/f s · resolution]) represents higher accuracy and energy efficiency for the design. Our sensor front-end is compared to the state-of-the-art pressure sensor front-ends in Table 1 showing our lower power consumption and, subsequently, FoM.

D. OVER-THE-AIR PROGRAMMING
Closing the loop with the base station takes advantage of the power asymmetry between the node and the reader providing a way to store internal states and program the node for optimum performance.
To evaluate the benefit of our design, we measured the settling time of the sensor front-end under three different settings. Our baseline setting does not incorporate any method for fast settling. The second setting involves turning on the fast settling capability of the design. The third setting incorporates setting the dc-preset to previously determined value, which can be obtained from an earlier settling period and transmitted over the air to the wireless node. Fig. 18 plots the differential output of the front-end as a function of time for each of these three settings. The figure shows that the baseline without fast settling (orange plot, takes about 500 s) to settle. In contrast, with our front-end's fast settling, the voltage settles within 20 s. Finally, our entire design (which would include both fast-settling and over-theair programming with the stored internal state), the settling time drops to only 2 s. This result shows that our design reduces the settling time by two orders of magnitude, which yields very low latency and higher overall energy efficiency by eliminating the settling time overhead of the front-end.

E. EX-VIVO MEASUREMENTS
The full system is evaluated in an ex-vivo setup demonstrated in Fig. 19(a). An RF signal generator is uses a log-periodic antenna with a TX gain of 5-6 dBi in order to provide the RF signal to power up the node. The transmit power is around 10 dBm and the node is at a distance of 15-30 cm from the TX antennas. The reader employs a similar logperiodic antenna connected to a software radio (USRP) as a reader to decode and extract the backscattered biomedical data, while an air pump is used to mimic the flow of fluids in tissues. The chip with the microstrip antenna are mounted on a piezoelectric sensor and laminated on a pork stomach as shown in Fig. 19(b).
To test the system functionality, the measured differential voltage from the sensor front-end is plotted against time as an external pump inflates and deflates the pork stomach. The measured voltage tracks the pressure variations, shown in Fig. 20 showing clear regions of inflation and deflation along the time axis as the fluid is pumped into the stomach. Table 2 compares the SoC to existing batteryless sensing nodes. This work uses far-field to power up and communicate with a distance of up to 30 cm from tissues and consumes a  maximum active power of 11 μW only. Compared to existing nodes, it provides a tunability of 300 MHz in the center frequency with an 8× scaling of BW while guaranteeing security through a 128-bit AES-GCM engine.

IV. DISCUSSION AND CONCLUSION
This article describes an implantable strain sensing node as a fully integrated wireless and batteryless implant enabling far-field with 300-MHz frequency tunability and 8× BW programmability. It incorporates a low-noise sensor frontend achieving a 4.3 nJ/conversion at 1.65-μW active power consumption. Secure sensing and communication is achieved with a low-power TRNG through ADC-sharing between the security engine and the sensor interface.
One of the key advantages of our design is that it relies on far-field energy harvesting and backscatter, allowing it to communicate with a transceiver at a distance from the human body. Such contactless sensing would result in a qualitative improvement in future healthcare monitoring. In particular, in contrast to existing designs that require body contact (e.g., by requiring the user to wear an inductive coil vest or placing an ultrasound probe on their body), our approach would enable a seamless user experience that is similar to how mobile devices connect to nearby wireless access points today. For example, if an external transceiver is placed near a user's desk or bed, it can automatically power up and communicate with our far-field microimplants whenever the user is in their vicinity.
More broadly, the ability to perform wireless, batteryless, and secure in-body sensing opens important applications for long-term monitoring of internal biometrics, may allow for early diagnosis and intervention, and paves the way for closed-loop treatment systems (e.g., drug delivery) that continuously monitor and respond to critical vital signs.