A smart knee implant using triboelectric energy harvesters

A schematic for the triboelectric generator is shown in figure 1. The harvester is composed of two major parts separated with rubber springs: an upper aluminum part and a lower PDMS part bound to a bottom aluminum layer. The device operates based on the periodic contact and separation between upper and lower parts. The required separation is small, within a range of less than 700 μm. The upper Al part is a used as a mold with micropatterns. The lower part is made from a thin aluminum substrate glued to a polydimethylsiloxane (PDMS) layer. The PDMS layer is fabricated using the upper aluminum as a mold, and the two layers have reverse micropatterns. This PDMS material is considered biocompatible and used widely for implantable devices applications [29]. When a cyclic axial force is applied to the generator, the rubber springs will be compressed and then the upper and lower parts undergo a continuous contacting and separating motion. This contact and separation mechanism between the two surfaces with micro-patterns spurs charges to be generated. The more force the larger the output will be. This feature enables the triboelectric generator to be an effective energy harvesting mechanism at the knee joint.

Zoom In Zoom Out Reset image size

The fabrication process for the PDMS layer depends on the upper AL mold. This mold is designed and made by Progressive Tools with dimensions of (5.2 × 3.7 × 0.2) cm. The harvester area is 19.24 cm² and is as large as the usable area of a typical tibial tray in knee implants. Our mold is designed with semi-cylindrical grooves, as shown in figure 2(a), to increase the surface contact between the two layers. First, the top Al mold is immersed in distilled water and placed in an ultrasonication machine for 10 min. Then the process is repeated twice with acetone and water, respectively. After that the Al is dried with Nitrogen. A 10:1 mixture of a silicone elastomer base and a curing agent is made to form the PDMS layer. The mixture is placed in a vacuum chamber at −17 PSI until all air bubbles are removed from the mixture. After that, the mold and a square piece of glass are treated with SIH5 841.0 material in a vacuum chamber for half an hour. This treatment helps in peeling the PDMS layer from the Al mold after baking. The PDMS mixture is poured on the glass, and immediately the molds are placed over the PDMS mixture and pressed slowly at an angle to reduce the air bubbles in the final PDMS layer. The glass with the mixture and Al mold is placed on a heater and baked at 80 °C for an hour. After baking, the PDMS layer is peeled from the Al mold yielding a 320 μm thick PDMS layer, with a semi-cylindrical reverse pattern, as shown in figure 2(b). To create the lower electrode, a thin Al layer is spin coated with a thin layer of the PDMS mixture and baked for an hour at 80 °C. Then, the Al layer is plasma treated and bonded to the PDMS layer.

Zoom In Zoom Out Reset image size

The working principle of the triboelectric mechanism is depicted in figure 3. Initially, the two layers are separated and are without charge, as depicted in figure 3(a). When the device is subject to a compression load at the knee, the two layers contact each other, as shown in figure 3(b). The Al layer donates electrons to the PDMS layer. Therefore, on the contact, the Al layer gets positively charged while the PDMS layer gets negatively charged. The PDMS layer induces electrostatic charge at the bottom AL layer. When the mechanical load is released, the mechanical restoring force from the rubber springs separates the two layers. At this stage, if the Al layers are connected, the potential difference between them forces the current to flow from the upper Al layer to the lower Al layer, see figure 3(c). This current keeps flowing until the two layers are fully separated and reach an equilibrium where the triboelectric charges and electrostatic charges are equalized, as illustrated in figure 3(d). With continued cyclic loading, the two layers start to approach each other, and capacitance is charged. The charged capacitance and charged electrode cause a current flow in the opposite direction from the lower Al layer to the upper layer, as shown in figure 3(e). The cyclical reversal of electron flows mean the device produces an alternating current.

Zoom In Zoom Out Reset image size

Figure 4 shows the proposed schematic for the instrumented TKR. TKR consists of the femoral, tibial tray, and the UHMWPE bearing parts. For optimal load measurements, the triboelectric generators are placed between the tibial tray and the UHMWPE bearing. Because the backs of the tibial tray and the UHMWPE bearing are flat, we simplify the analysis by placing our generator between two flat surfaces and test it with an MTS machine to investigate the performance. For future work, the generators will be placed in a TKR implant and be tested in a VIVO (Advanced Mechanical Technology, Inc., Watertown, MA) joint simulator.

Zoom In Zoom Out Reset image size

The goal of this experiment is to test the generator under normal walking activity. Body loads are Fx and Fy, which act in lateral and anterior directions, respectively, and the axial force component, Fz, which always acts in the direction of the implant shaft. For walking and jogging, the highest force component was found to be the axial force, Fz, which is almost equal to the total force [9]. In this test, the axial force profile is approximated as a sine-wave signal at 1 Hz. The experimental setup used to evaluate the performance of the triboelectric generator is shown in figure 5. The setup consists of an MTS 858 Servo Hydraulic Test System with FlexTest SE Controller. The MTS is responsible for conducting a cyclic axial load, while the amplitude is tuned with the FlexTest controller. The MTS has a built-in load cell to measure the applied force on the generator.

Zoom In Zoom Out Reset image size

The generated voltage signal from the triboelectric generator is measured using a Keithley 6514 and an ExceLINX program. The Keithley 6514 has an internal impedance of 200 TΩ to ensure the voltage measurements of the harvester are not compromised by the resistance of the measurement instrument. The experimental test is repeated at different external resistances connected to the harvester to obtain the optimal resistance where a maximum power can be achieved. A variable resistance box is used to control the resistance.

To characterize the electric output of the triboelectric generator, the short-circuit current (I_sc) was measured under different periodic compressive forces of 200, 500, and 1000 N applied by the MTS machine at a constant frequency of 1 Hz as shown in figure 6. By increasing the applied load, a higher current can be achieved. However, our interest in this study is the instantaneous current because of the presence of frontend electronics. To extract the maximum power, we identify the internal resistance of the harvester and match it with the circuit resistance. The internal resistance is referred to as the optimal resistance. To estimate the optimal resistance for the triboelectric energy harvester, the testing was conducted at a specific cyclic load with a force of 100 N with varied external resistance up to 142 MΩ. The variations of the voltage and current outputs with the external resistance load are shown in figure 7. The voltage and current curves intersect at the optimal resistance of R = 58 MΩ. The corresponding average power outputs of the harvester with the external load resistance are shown in figure 8. The optimal resistance from the power curve is obtained at 58 MΩ, which matches the result from the intersection of the voltage and current outputs in figure 7. In addition, a maximum average power of 0.125 μW is generated at this optimal resistance. The load of 100 N is significantly below the typical knee loads, which are more than 2000 N. This testing was only conducted to estimate the internal resistance of the device.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Forcing frequency affects the harvester output. Figure 9 shows the output voltages of the triboelectric generator under three different frequencies, 1, 2, and 3 Hz. The voltages clearly increase with faster frequencies. At high frequencies, because there is more contact in a certain amount of time, more charges accumulate, resulting in higher voltages. The variation of the root mean square of the generated output voltage with a wider range of frequencies is shown in figure 10. The output voltage increases rapidly up to 6 Hz where the increases slow down and stabilize at approximately 11.4 V.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Next, the performance of the generator is explored under different cyclic loading. The main goal of this test is to evaluate the harvester output under equivalent TKR loads to design a suitable load data processing unit (section 4). The generated voltages for 100, 500 and 2300 N are shown in figure 11. The results show the output voltage increases with the applied cyclic load. This increase in the output voltage can be correlated to the larger surface area of contact. With an applied force of 2300 N, which is equivalent to the load from walking of an average weighted person, at the optimal load resistance of 58 MΩ, the harvester generates a 100 V peak-to-peak (18 V rms) and a 6 μW of average power. Based on the results from figure 11, the output voltage is not linearly proportional to the applied force. The applied force is a sinusoidal wave, while the output voltage is not sinusoidal, and this difference is related to the nonlinear mechanism of triboelectric charge generation [21]. However, the root mean square of the voltage is proportional to the load (figure 12). The root mean square of the harvester analog signal can be easily obtained in the frontend electronic circuit and be related to the load amount. Figure 12 shows the rms of the voltage output of the triboelectric generator under different applied cyclic loads up to 3300 N and 1 Hz frequency at the optimal resistance extracted before. The rms voltage outputs are proportional to the load applied, where greater voltage can be generated with more applied force. The reason for this can be related to the surface area of contact between the generator layers. The triboelectric principle relies on the contact and separation of the generator layers (PDMS and Al) both at the macro and micro-scales. Because micro-patterns exist in both layers, higher applied loads cause larger penetration into the PDMS layer leading to an increase in the triboelectric charges generated on the surface and thus higher output voltage. With a maximum applied load of 3300 N, the triboelectric harvester generates an rms voltage of 23 V at the output.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The corresponding average power generated under the same axial loading are shown in figure 13. The maximum average power generated is found to be 9.5 μW at a maximum applied load of 3300 N. This relationship between the applied force and the voltage and power outputs is considered promising for powering load sensors and their telemetry circuitry for TKR applications. Next, we design and evaluate the performance of a frontend electronic circuit that can use the harvested energy to digitize the load data.

Zoom In Zoom Out Reset image size

The electrical model of the triboelectric harvester that provides the required input power, as well as the signal to monitor, is described in this section. The behavior of the harvester is characterized by using equations (1) and (2) [30],

$$\begin{eqnarray}&&V=-\displaystyle \frac{1}{{C}_{M}}Q+{V}_{M}-{Ir},\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&r=R(\displaystyle \frac{{V}_{M}}{V}-1).\end{eqnarray} \tag{ 2 }$$

In equation (1), the three terms on the right-hand side can be modeled using three circuit elements, a voltage source V_M, a variable capacitance C_M, and an internal resistance r. The capacitive term originates from the capacitance between the two electrodes of the harvester, the voltage term arises because of the separation of the polarized tribocharges, and the resistance term models the impedance of the voltage source. The electrical model can be represented by a serial connection of these three circuit elements, as shown in figure 15.

Zoom In Zoom Out Reset image size

The electrical model is connected to an external load resistance R based on equation (2). Increasing the velocity of motion increases the frequency of AC voltage source V_M, which in turn decreases the matched resistance. The tribocharge density does not affect the matched load resistance. From basic thermodynamics theory, V_M and C_M depend only on the moving distance (x) and structural parameters and not on motion parameters such as velocity and acceleration. Using the finite element method and continuous fractional interpolation, a V_M–x and C_M–x relationship for a contact-mode triboelectric harvester has been generated [31–33]

$$\begin{eqnarray}&&{V}_{M}=\displaystyle \frac{\sigma x}{{\epsilon }_{0}},\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}&&{C}_{M}=\displaystyle \frac{{\epsilon }_{0}S}{{d}_{0}+x},\end{eqnarray} \tag{ 4 }$$

where σ is the triboelectric surface charge density, ₀ is the permittivity of free space, S is area of dielectrics in the harvester and d₀ is the effective dielectric thickness. The values of these physical and electrical parameters based on the designed harvester are listed, respectively, in tables 1 and 2.

Table 1.  Physical parameters used to generate the electrical model.

Parameter	Symbol	Value
Dielectric 1	r1	2.4
Thickness of dielectric 1	d1	320 μm
Dielectric 2	r2	1.6
Thickness of dielectric 2	d2	0 μm
Effective dielectric thickness	d0 = d1 / r1 + d2 / r2	133 μm
Width of dielectrics	w	5.2 cm
Length of dielectrics	l	3.7 cm
Area of dielectrics	S = w × l	19.24 cm2
Separation distance	x	0.5–1 mm
Tribo-charge surface density	σ	100 μC m−2
Internal resistance	r	57 MΩ
External load resistance	R	58.3 MΩ

Based on these values, SPICE simulations of the electrical model have been performed. The relationship between output voltage and the separation distance x is depicted in figure 16. The voltage increases with the distance x between the dielectrics.

Zoom In Zoom Out Reset image size

The transient output voltage produced by the electrical model is compared with the measured output voltage of the harvester in figure 17. The output currents are also compared in figure 18. As demonstrated by these figures, the model can represent the harvester with sufficient accuracy. Note that as the distance between the plates of the harvester changes, the harvester impedance also changes. These harvester dynamics are accounted for in the model using a variable capacitance C_M that depends upon the separation distance x. Thus, the input impedance of the model dynamically varies.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

According to these results, a maximum voltage of 17.95 V and a maximum input power of 6.09 μW is generated by the model at the matched optimal resistance, which represents the measured harvested power accurately. Note that the input impedance of the electrical circuitry is approximately 37.2 MΩ at 1 Hz. Since the optimal resistance is 58.3 MΩ, the delivered power to the circuit is reduced from 6.09 μW to 4.46 μW. The input impedance is relatively large due to sufficiently low frequency of 1 Hz. This power is used for the entire electronic frontend circuitry, as described in the following sections.

The harvested AC signal is converted into a DC voltage of approximately 1.2 V through a negative voltage converter rectifier (NVC) [34], as shown in figure 19.

Zoom In Zoom Out Reset image size

Due to the full gate cross-coupled topology, an NVC rectifier performs the rectification process with a high efficiency. During the positive input cycle, when the input voltage is less than the threshold voltage (V_ac < V_th), neither of the transistors conducts and no current flows through the circuit. When the input voltage becomes greater than the threshold voltage (V_ac > V_th), transistors M_P1 and M_P2 provide a current path to the output. The operation is similar during the negative cycle where the transistors M_N1 and M_N2 provide the current path.

The NVC rectifier achieves high power efficiency when the load is purely resistive. The efficiency, however, is reduced when a reactive element is added to the load due to the reverse current induced from output to the input port. This current flows due to the bidirectional MOSFET characteristic and stored charge within the capacitive load, where the output voltage is higher than the input voltage. The NVC rectifier provides a low sensitivity to input voltage variations and is widely used in low voltage, high-efficiency energy harvesters.

A LDO regulator, as shown in figure 20 converts the input DC voltage into a regulated voltage of desired value within a specified tolerance range while compensating for variations in the output current.

Zoom In Zoom Out Reset image size

The LDO consists of a voltage reference circuit for producing the voltage V_ref, an error amplifier EA that produces an output voltage proportional to the difference between the reference and output voltages, a pMOS power transistor M_P to deliver the required current to output, resistive divider to determine voltage conversion ratio, and an output capacitor C_L to ensure stability.

The error amplifier consists of a differential input stage followed with a common source output stage, producing a DC gain of approximately 70 dB. The high gain ensures a high-resolution error signal at the output. The phase margin is 57.81°, ensuring a stable operation.

Reducing the dropout voltage and quiescent current increases the power efficiency of the LDO. Hence, the power transistor is sized sufficiently large to lower the dropout voltage and quiescent current of the regulator, while permitting the flow of the required large current to the load. Furthermore, diode-connected nMOS transistors, M_N1, M_N2, M_N3, and M_N4, are used (instead of resistors) to further enhance the efficiency and reduce physical area. These transistors are sized to achieve the desired voltage conversion ratio of 0.75. Thus, the unregulated input voltage of 1.2 V is converted into a regulated voltage of 0.9 V.

It is important to note that the proposed LDO is free from any external bias voltage, making it applicable to energy harvesting implantable applications.

The output of the triboelectric harvester is used not only as the power source but also as the data signal that is digitized to monitor loads. The delta-sigma ADC is highly suitable for biomedical implants and energy harvesting applications due to low complexity, low power, and small area. Irrespective of the number of bits, a delta-sigma ADC consists of only one integrator and comparator, as shown in figure 21.

Zoom In Zoom Out Reset image size

The analog input voltage V_in is first differentiated from the output of 1 bit digital-to-analog converter (DAC) using a switched capacitor circuit. An integrator then adds the output of the differentiator to the value it has stored from the previous integration step. The output of the integrator is then fed into a comparator, which compares the output with voltage V_mid. It outputs a logic 1 if the integrator output is greater than or equal to V_mid, otherwise, it produces a logic 0. A 1 bit DAC feeds the output of the comparator to the differentiator through a feedback loop. The output of the comparator then goes through a D flip-flop to produce the final digital bit stream [35]. The schematic diagram of the designed ADC is shown in figure 22.

Zoom In Zoom Out Reset image size

The voltages V_H, V_mid, and V_L are, respectively, 0.9 V, 0.6 V, and 0.3 V. The capacitors C₁ and C₂ are kept at 100 fF. The two non-overlapping clocks ϕ 1 and ϕ 2 run at 50 kHz to reduce power consumption and ensure correct functioning of the triboelectric knee implant. The reset and reset2 signals have the same frequency, as determined by

$$\begin{eqnarray}&&{f}_{R}=\displaystyle \frac{{f}_{{clk}}}{{2}^{N}},\end{eqnarray} \tag{ 5 }$$

where f_clk is the clock frequency, f_R is the reset and reset2 frequency and N is the number of bits. In the designed ADC, N is equal to 8, and the reset and reset2 frequencies are 195 Hz. The signals Vϕ2 and $\overline{V}\phi 2$ are produced using a CMOS NAND gate and an inverter. The clock signals, analog input voltage, and digital output voltage obtained from the ADC are shown in figure 23. To check the accuracy of the ADC, the analog and digital figures of merit, represented, respectively, by equations (6) and (7), are calculated. The percentage error between the two should be minimized

$$\begin{eqnarray}{\rm{AnalogFOM}}=\left\{\begin{array}{cl}\displaystyle \frac{{V}_{{in}}-{V}_{L}}{{V}_{H}-{V}_{L}} & \mathrm{if}\ ({V}_{{in}}\lt {V}_{H})\\ \displaystyle \frac{{V}_{H}-{V}_{L}}{{V}_{{in}}-{V}_{L}} & \mathrm{if}\ ({V}_{{in}}\gt {V}_{H})\end{array}\right.,\end{eqnarray} \tag{ 6 }$$

$$\begin{eqnarray}&&{\rm{DigitalFOM}}=\frac{{\rm{Number}}\,{\rm{of}}\,{\rm{1}}{s}\,{\rm{in}}\,{\rm{output}}\,{\rm{data}}}{{{\rm{2}}}^{{N}}}.\end{eqnarray} \tag{ 7 }$$

Zoom In Zoom Out Reset image size

For example, for an input voltage of 0.84 V, the analog FOM is 0.90 and from the digital bit stream, the digital FOM is 0.88. The percentage error between the two is only 2.22%. The figures of merits were calculated for various input voltages (ranging from −57.05 to 58.97 V) and the average percentage error is approximately 5.75%, demonstrating that the designed ADC is sufficiently accurate to digitize the sensed gait signals.

The power efficiencies of the designed NVC rectifier and LDO regulator are listed in table 3.

The overall power consumed by the rectifier, regulator and ADC circuits is 4.46 μW. Thus, it is possible to achieve a self-powered digitization circuitry by relying on the triboelectric energy harvester.