Figure 1(a) schematically shows the scenario of the AI microrobot patrolling the human body under external control. The sensing part at the head of the microrobot harvested the energy from the electromagnetic field generated by the RF coil of the imaging device and modulated the local RF magnetic field to change the signals at the corresponding position in the image.26,27 The dielectric characteristics of the surrounding environment, including the dielectric constant and conductivity, determine the electrical characteristics of the sensing circuit and further affect the enhancement effect on the local RF magnetic field. Diseases can change the dielectric properties of the microenvironment, and therefore, when the microrobot reaches the vicinity of the lesion, the enhanced signal can be changed, suggesting the occurrence of possible disease.28,29 Fig. 1(b) shows a microrobot patterned on the substrate and an as-fabricated free-standing one. The designed device is manufactured step by step following the processes shown in Fig. 1(c). Firstly, a 30 nm Al2O3 layer was deposited as the sacrificial layer on a pre-prepared Si substrate covered with 300 nm SiO2. Then, a double-layer tail containing a magnetic (Fe) and a pre-stress memberane (SixNy) was deposited. The head platform of the microrobot was made of 100 nm SiO2, which acted as the substrate to support the sensor. The three-layer structure of the sensor was sequentially fabricated on the head of the microrobot. The middle-interdigitated electrodes (< 10 µm thickness) occupied approximately 25% of the head area and were surrounded by annular electrodes, which were connected with the bottom electrode through two via holes to form a closed resistance-inductance-capacitance (RLC) loop as shown in Fig. 2(a). The size of the RF coil ranged from ~ 500 µm to ~ 50 µm, and the number of interdigital electrode pairs can be adjusted according to the actual requirements, as shown in Supplementary Fig.S2. Finally, the chip was soaked into NaOH solution to etch the sacrificial Al2O3 layer, and the stress in the SixNy layer was released, leading it to roll up into a helical tail. A more detailed fabrication process is shown in Supplementary Fig.S1.
Figure 2(b) demonstrates the principle of the RLC passive sensor, which is composed of the inductance Ls introduced by the coil, capacitance Cs, and resistance Rs coming from the interdigitated electrodes. The driving coil of the imaging equipment first generated the RF magnetic field, and the external loop-coil of the sensor harvested the energy from the RF magnetic field and induced the current of the sensor circuit. The corresponding current can excite an additional magnetic field, which enhances the local RF magnetic field of the sensor. Considering the distance between the excited coil and the sensor is far enough, the coupling coefficient is usually very small, resulting in the influence of the magnetic field generated by the sensor on the excited coil being negligible. Therefore, the excited magnetic field around the sensor can be considered a constant as well as the voltage generated by the sensor coil. As a result, the current and magnetic field of the sensor merely depends on the impedance of the circuit, and the sensor circuit can obtain the maximum current and the most obvious enhancement effect on the local RF magnetic field at the resonance frequency. The resonance frequency (\({f}_{s}\)) and quality factor (\({Q}_{s}\)) can be calculated as equations:
$$\begin{gathered} {f_s}=\frac{1}{{2\pi \sqrt {{L_s}{C_s}} }} \hfill \\ {Q_s}=\frac{1}{{{R_s}}}\sqrt {\frac{{{L_s}}}{{{C_s}}}} \hfill \\ \end{gathered}$$
1
in which \({L}_{s}\) is a constant, and \({C}_{s}\) and \({R}_{s}\), introduced by interdigitated electrodes, can change with the surrounding environment as shown in Fig. 2(c) (top). When the dielectric and conductive properties of the surrounding solution changed, the equivalent resistance and capacitance of interdigital electrodes responded accordingly. In the equivalent circuit in Fig. 2(c) (bottom), \({R}_{l}\) and \({C}_{l}\) represent the resistance and capacitance when the electric field passes through the solution, respectively, while the \({C}_{sub}\)stands for the substrate capacitance when the electric field passes through the substrate. The total capacitance \({C}_{IDE}\) is the sum of \({C}_{l}\) and \({ C}_{sub}\). The equivalent impedance Z can be calculated using the following formulas:
$$Z=\frac{1}{j\omega C}//\text{ R=}\frac{1}{j\omega C+1/R}=\frac{1/R-j\omega C}{{\omega }^{2}{C}^{2}+1/{R}^{2}}$$
4
and \({R}_{s}\) and \({C}_{s}\) in the sensor circuit can be calculated as follows:
$${R}_{s}={Re}({R}_{l}//{C}_{IDE})=\frac{1/R}{1/{R}^{2}+{\omega }^{2}{C}^{2}}=\frac{1}{1/R+{\omega }^{2}{C}^{2}R}$$
5
$$\omega {C}_{s}=-\frac{1}{{Im}({R}_{l}//{C}_{IDE})}=\frac{1/{R}^{2}+{\omega }^{2}{C}^{2}}{\omega C}=\frac{1}{{R}^{2}\omega C}+\omega C$$
6
in which \({R}_{s}\) decreases with increasing \(\omega C\) and \(\omega {C}_{s}\) decreases with the increased \(R\). \({R}_{s}\) and \(\omega {C}_{s}\) obtain the maximum and minimum values at\(R=\frac{1}{\omega C}\) if \(\omega C\) and \(R\) are fixed, respectively. When \(R>\frac{1}{\omega C}\), the increase in \(R\) leads to a decrease in \({R}_{s}\) and the increase in \(\omega C\) leads to the increase of \(\omega {C}_{s}\). In contrast, when \(R<\frac{1}{\omega C}\), the increase in \(R\) induces to the increase in \({R}_{s}\) and the increase in \(\omega C\) results in the decrease in \(\omega {C}_{s}\). Especially, when \(R\gg \frac{1}{\omega C}\):
$${R}_{s}=\frac{1}{1/R+{\omega }^{2}{C}^{2}R}\approx \frac{1}{{\omega }^{2}{C}^{2}R}$$
7
$$\hspace{0.33em}\hspace{0.33em}\omega {C}_{s}=\frac{1}{{R}^{2}\omega C}+\omega C\approx \omega C,\hspace{0.33em}\hspace{0.33em}Cs\approx C$$
8
According to the structural model, the electrical characteristics of interdigital electrodes were emulated by the High Frequency Structure Simulator (HFSS). The interdigital electrode model was extracted from the microrobot with a head size of 500 µm and 11 interdigital electrode pairs. With the frequency varying from 0.5 GHz to 2 GHz, amplitudes of the real and imaginary components of the impedance decrease obviously in Supplementary Fig.S3. The increase in environmental conductivity will lead to an increase in the real impedance and a decrease in the imaginary impedance. It should be noted that when the conductivity increases from 0.001 S/m to 0.1 S/m, the imaginary impedance is almost unchanged, which accords with the theory of \(R\gg \frac{1}{\omega C}\). The simulation results are consistent with the theoretical analysis results at \(R>\frac{1}{\omega C}\). Therefore, the increased environmental conductivity can lead the loop resistance to increase at the resonance frequency and weaken the enhancement effect.
The local electromagnetic field enhancement ability of the sandwich structure sensor, including the intermediate dielectric layer, and top and bottom electrodes, can be simulated by the HFSS as shown in Fig. 3(a), in which an excited port is used to generate the RF magnetic field and the sensor surrounded by the dielectric as the environmental substitution is placed below the excited port. The substrate without the sensor is set for the control group and the deionized (DI) water (\({\epsilon }_{r}=81, \sigma \text{=0S/m}\)) is used as the environmental dielectric. The local RF magnetic field strength can be evaluated by the magnetic field strength integration. In Fig. 3(b), the\(500\times 500 {\mu }\text{m}\)plane centered on the upper electrode is used to calculate the magnetic field strength. With the frequency increasing, the strength integration of only the substrate increases gradually. In contrast, some definite peaks appear for the sensor, illustrating the enhanced local RF magnetic field effect. According to the theoretical analysis, the greatest enhancement effect occurs at the resonance frequency, the first of which is ~ 1.24 GHz. The magnetic field strength integral for the substrate is ~ 4.8E-7 T∙m2 at 1.24 GHz, while the value can amplify ~ 45 times to ~ 2.2E-5 T∙m2 for the senor. In addition, the enhancement amplitude reaches ~ 560 times at the second resonance point (f = 4.94 GHz). Moreover, the presence of the sensor greatly changed the local magnetic field distribution, especially in the area close to the sensor. The local RF magnetic field distribution of the sensor at 1.24 GHz is shown in Fig. 3(c) and (d). The maximum electromagnetic field strength can reach ~ 300 A/m at 1.24 GHz. With increasing distance from the sensor, the corresponding strength decreases gradually. However, the maximum magnetic field strength is only ~ 3 A/m without the sensor as shown in Fig. 3(e) and (f). The magnetic field in the central area around the substrate is uniform and less than 1 A/m, while the magnetic field strength becomes larger at the boundary and the area closer to the excited port.
The magnetic field enhancement coefficient (MFEC), the ratio of the magnetic field strength integral between with and without the sensor, can be utilized to evaluate the magnetic field enhancement effect at different planes as shown in Eq. (9).
$$MFEC=\frac{{\iint }_{D}\left|{H}_{sensor}\right|dS}{{\iint }_{D}\left|{H}_{substrate}\right|dS}$$
9
The size of the plane is \(500\times 500\mu m\), which is similar to the sensor at the center, and the height gradually increases from the upper substrate plane (\(D= 0\) µm), where the MFEC is ~ 45.45. As the height increases, the relevant value of the MEFC decreases accordingly. When \(D= 200\) µm, the MFEC changes to ~ 1.19. While D increases further to ~ 300 µm, the MFEC reduces to ~ 1.03, but the ability to change the magnetic field distribution remains strong according to the magnetic field distribution in Supplementary Fig.S4.
The structure of the sensor, including the size of the sensor and the number of interdigital electrode pairs, can affect the magnetic field strength as shown in Supplementary Fig.S5. We compared the MFEC at \(D=0\) µm with different sensor structures. The whole microrobot swam in the DI water. According to the simulation results, the resonance frequency peak increased rapidly from ~ 1 GHz to ~ 30 GHz as the size of the sensor decreased from 500 µm to 50 µm. Moreover, the frequency peak has a left shift with increased numbers of interdigital electrode pairs, which can promote the capacitance and reduce the resonance frequency. The statistical result of the resonance peak shifting with the sensor structure changing is summarized in Fig.S6. Overall, the small size of the sensor led to a high resonance frequency, resulting from the limitation of the placement space led to smaller values of inductance and capacitance.
The surrounding environment can affect the local magnetic field enhancement performance of the sensor by changing the permittivity or conductivity, respectively. With the conductivity increasing exponentially from 0.001 S/m to 1 S/m under the fixed permittivity (\({\epsilon }_{r}=81\)) resonance peaks are concentrated at approximately 1.24 GHz when the swept frequency ranged from 0.1 to 2 GHz as shown in Fig. 4(a). The data at \(D=0\) µm were used to calculate the MFEC. However, when the environmental conductivity is 10 S/m, the large resistance of the sensor circuit causes a small current, resulting in the local magnetic field remaining nearly unchanged. The relationship between the resonance frequency, the MEFC, and the conductivity is plotted in Fig. 4(b) based on the results from Fig. 4(a). The MFEC curve was extracted at the frequency of 1.24 GHz (the red line in Fig. 4(a), near the resonance peak). The stable resonance frequency with conductivity varying from 0.001 S/m to 0.1 S/m is mainly due to the small conductivity, where the solution impedance of the interdigital electrode \({R}_{l}\) is much larger than \(1/\left(\omega {C}_{IDE}\right)\). Correspondingly, the capacitance of the whole circuit \({C}_{s}\) and the resonance frequency \({f}_{s}\) remain unchanged. When the conductivity increased to 1 S/m, the corresponding resonance frequency decreased to 1.22 GHz. Moreover, the resonance frequency dropped rapidly to 0.24 GHz when the conductivity reached 10 S/m. Overall, the MFEC is more sensitive to small conductivity changes, demonstrating a monotonic downward trend with increasing conductivity at the frequency of ~ 1.24 GHz. It decreased from 103.6 at 0.001 S/m to 2.4 at 1 S/m, but was only 0.98 at 10 S/m, indicating there was no enhancement effect.
Figure 4(c) shows the simulation results of permittivity from 1 to 81 under constant conductivity (σ = 0 S/m) with frequencies ranging from 0.1 to 4 GHz. With the permittivity increasing, the resonance peak revealed a red shift from 3.7 GHz to 1.24 GHz. Meanwhile, the MFEC was obviously improved from 0.1 to over 100 at the resonance point. The detailed corresponding relationship between resonance frequency and permittivity is plotted in Fig. 4(d). The decrease rate of the resonance frequency gradually slowed down with increasing permittivity. When the permittivity changed from 1 to 11, the resonance frequency dropped from 3.7 GHz to 2.72 GHz. However, the frequency decreased by only 0.14 GHz (from 1.38 GHz to 1.24 GHz) under permittivity varying from 71 to 81, suggesting that the tuned sensitivity of the resonance frequency provided high detection accuracy and response under conditions with low permittivity. To cooperate with the electromagnetic imaging system, we extracted MFEC results with the different dielectrics at 400 MHz as shown in Fig. 4(d) (right), corresponding to the magnetic resonance imaging (MRI) with an operating magnetic field strength (B0) of 9.4 T. Compared with the value near the resonance peak, MEFCs at 400 MHz are all less than 1, which means the senor weakens the local RF magnetic field instead, resulting from the induced magnetic field generated by the coil reducing the excited magnetic field. However, the induced magnetic field is large enough to amplitude the field strength near the resonance peak. When the permittivity changes from 1 to 81, the MFEC increases from 0.11 to 0.96 linearly. The MEFC of less than 1 can make the sensor appear as a dark spot in the electromagnetic imaging system, while the surrounding environment presents as a bright background. Therefore, it is necessary to match the resonance peak with the working frequency of electromagnetic imaging to realize variation and easy observation for in vivo detection.
On account of the theoretical results, we fabricated a suitable structure sensor on the microrobot and connected the device to the prepared external measuring equipment, as shown in Fig. 5(a). A square coil with a side length of 1 mm was used as an external readout coil, which was coupled with the coil surrounding the perceptive interdigital electrodes. The external coil was connected to the network analyzer through the Sub-Miniature-A (SMA) connector for the frequency sweeping test and the measured circuit was made on a printed circuit board (PCB). The chip carrying the device was fixed on the PCB by epoxy resin to prevent the solution from affecting the readout coil from the back. As shown in Fig. 5(b), the solution can immerse the lower end of the PCB and contact the sensor through the front hole during the measurement. The reflection coefficient S11, which reflects the sensor response to the surrounding environment, can be calculated by Eq. (10):
$$\begin{gathered} {S_{11}}=20{\log _{10}}(\left| {\frac{{{Z_{eq}} - {Z_0}}}{{{Z_{eq}}+{Z_0}}}} \right|) \hfill \\ {Z_{eq}}={R_r}+j\omega {L_r}+\frac{{{\omega ^2}{M^2}}}{{{R_s}+j\omega {L_s}+\frac{1}{{j\omega {C_s}}}}} \hfill \\ \end{gathered}$$
10
\({Z}_{eq}\) and Z0 are equivalent impedances of the readout coil and the system (normally 50 Ω), respectively. M is the coupling coefficient of the readout and the sensor coil. Rr and Lr are the resistance and inductance of the measured circuit, respectively. S11 reaches the minimum value near the resonance frequency when Qs ≫ 1.30 We prepared HCl solutions with different conductivities by changing the concentration, demonstrating the relevant pH value. When the device was placed in the HCl solution at pH = 1 and T = 25℃, the resonance frequency of the sensor centered at ~ 1.12 GHz, while the resonance frequency is ~ 1.85 GHz for the substrate only, as shown in Fig. 5(c). As the pH increases from 1 to 4 in Fig. 5(d), the curves shift to higher frequencies. The relationship between the resonance frequency and the pH is plotted in Fig. 5(e) based on the result from Fig. 5(d). When the pH value decreases to 1, the resonance frequency rises to ~ 1.120 GHz. The free H+ in the solution forms the diffusion capacitance, which attenuates the external electric field, resulting in the sensor capacitance decreasing instead of increasing.31 However, the resonance frequency remains nearly unchanged at ~ 1.110 GHz when the pH is close to 7 because of the low concentration of the solution, which leads to a larger resistance than the impedance generated by the capacitance. Nevertheless, the S11 at 1.110 GHz (the red line in Fig. 5(d)) as a function of pH in Fig. 5(e) monotonically increases from − 3.35 dB to -3.17 dB when the pH increases from 1 to 4, indicating the sensitivity of the device as the pH sensor.
Moreover, the sensor can respond to ambient temperature changes. Solutions with different temperatures were obtained by heating the DI water and the permittivity gradually decreased from ~ 80 at room temperature (25℃) to less than 70 at 60℃.32,33 The resonance frequency moved from ~ 1.90 GHz to ~ 1.12 GHz when the device was fabricated on the substrate, which was surrounded by the DI water at T = 35℃ as shown in Fig. 5(f). When the temperature increased from 35℃ to 55℃, both the resonance frequency and the S11 of the device increased. Figure 5(h) demonstrates the relationship between the resonance frequency and the temperature derived from the measured result in Fig. 5(g). The resonance frequency increases from 1.125 GHz to 1.160 GHz linearly with increasing temperature due to the decreased permittivity while S11 at 1.124 GHz monotonically changes from − 7.99 dB to -6.11 dB, which is much larger than the change caused by the varying pH values.
Therefore, the sensor integrated into the microrobot can modulate the local magnetic field in the electromagnetic imaging process depending on the surrounding environment and affect the signals at the position in the image. Apparently, the AI microrobot can be guided to the prescribed destination and monitor the local microenvironment under the actuation magnetic field. We fabricated a series of microrobots integrated sensors with different sizes (50 µm to 500 µm) and released them into the DI water as shown in Supplementary Fig.S8. Smaller microrobots were beneficial to the movement in narrow channels and complex areas of the body, but their resonance frequency would be higher, causing difficulties in cooperation with electromagnetic imaging equipment.
The locomotion control and the posture adjustment of our integrated AI microrobots were carried out to attain the motion characteristics of AI microrobots, as shown in Fig. 6 and Supplementary Movie 1. The actuation magnetic field was generated by three pairs of Helmholtz coils perpendicular to each other, and the intensity and direction of the resultant rotational field can be adjusted according to the motion requirements. The tail was rotated by the external rotating magnetic field, and the rotation was converted by the helical structure into a translational one so that the microrobot could move forward and backward according to the direction of rotation of the field. The DI water was utilized as the work environment. During the whole driving experiment, both linear and circular locomotion were achieved from 0 s to 7 s and from 7 s to 23 s, respectively. In the first linear locomotion stage, a rotating uniform magnetic field in the yz plane was applied, enabling the AI microrobot obviously can spin and crawl like an active paddle accompanied by the external magnetic field stimulus. The magnetic field is as follows,
$$\overrightarrow{{B}_{\text{yoz}}}={B}_{0}{sin}(2\pi {f}_{1}t)\overrightarrow{j}+{B}_{0}{cos}(2\pi {f}_{1}t)\overrightarrow{k}$$
11
where B0 is the magnetic induction intensity and B0 = 6mT, f1 is rotation frequency, and f1 = 1Hz. By comparing the position of the microrobot at 3 s and 7 s, the linear locomotion speed was up to 133.32 µm/s. Such large speed was mainly attributed to a tumbling motion based on a rotation axis at the interface line between the micro-sensor and substrate, leading to the 100 µm length of the micro-sensor becoming the final rotation radius. Then, to achieve the circular locomotion, a gradually increasing yaw angle with the increment speed at 24 °/s in the xy plane was applied to the initial Byoz field. The total external magnetic field is as follows,
$$\mathop {{B_{{\text{total}}}}}\limits^{ \to } ={B_0}\sin (2\pi {f_1}t)\sin (\omega t)\mathop i\limits^{ \to } +{B_0}\sin (2\pi {f_1}t)\cos (\omega t)\mathop j\limits^{ \to } +{B_0}\cos (2\pi {f_1}t)\mathop k\limits^{ \to }$$
12
where ω is the yaw angular speed. The results showed that a circular path started at 7 s and finished at about 22 s. Thus, the average angular speed can be calculated as ω = 2π/15 = 0.419 rad/s. In addition, our integrated microrobot with larger microsensor (300 µm length) were also successfully actuated by same external magnetic field, as shown in the Supplementary Movie 2. The results showed that the optimized helical tails with larger geometry dimension can offer a larger driving force to guarantee the tumbling motion of larger microsensor, due to the larger magnetic material volume integrated on the helical tails. This phenomenon indicated that our manufacturing process was suitable for the integrated microrobots with various scales. In conclusion, our integrated microrobot exhibited fast locomotion speed and flexible path planning. Then, we focus on the formation mechanism of tumbling of our integrated AI microrobot. In our integrated platform, a continued and stable driven moment reliably formed due to the magnetic torque always needed to align the magnetization of the microrobot with the applied field. However, the unsymmetrical geometric dimensions between the micro-sensor and helical tails lead to the position of the applied force was always located at the interfacial line between the micro-sensor and the Si substrate. With the help of the rotating magnetic field, the periodic friction force at the interface between the micro-sensor and the Si substrate enabled the AI microrobot to transform the rolling motion into the linear motion in the plane, finally forming a paddling motion. This transformation motion characteristic is similar to the tumbling motion of peanut microrobots in the previous report [20]. More importantly, due to the rotation of micro-sensors, the posture adjustment of our integrated AI microrobots also can be realized, enabling the precise control of the relative rotation angle between the integrated coil and the external coil. This effective regulation behavior will promote the effective coupling between the coils and ensure the reliability of wireless signal extraction. Therefore, our integrated AI microrobots provide an effective integration strategy for future wireless monitoring systems with a controllable movement mode. In the future, the optimized AI microrobot architectures need to design to achieve the conversion of helical rotational motion into linear motion away from the substrate effect.
Our research strongly implied that the designed sensor integrated on the microrobot could realize controllable mobile sensing in vivo for disease diagnosis. The magnetic layer at the tail of the microrobot can generate force to actuate the microrobot under the excitation of an external magnetic field and the wireless sensor can modulate the local RF magnetic field without providing external power. When the functional microrobot worked in the electromagnetic imaging system, the position of the device appeared as an artifact in the image, especially bright spots when the resonance frequency of the sensor matched the working frequency of the electromagnetic imaging. The conductivity and permittivity of the environment changed the modulation effect of the local RF magnetic field. Compared with in vitro detection, the abnormality around the lesion is more prominent, especially in the early stage of the disease Therefore, the designed device sent to the vicinity of the lesion can improve the detection accuracy of the disease. The AI microrobot is tiny enough to enter the human body noninvasively and can realize the detection in complex positions in the body. Owing to the advanced micro-/nanofabrication technology, large-scale parallel manufacturing can reduce the cost and produce good consistency. In addition, the AI microrobot can cooperate with the electromagnetic imaging equipment to transmit the detection signal wirelessly, so the sensor itself only needs to locally interact with the RF magnetic field and modulate it. According to the previous study, the maximum working frequency of the MRI scanner for the human body is ~ 500 MHz under an operating magnetic field strength (B0) of 11.7 T.34 To reduce the resonance frequency of the AI microrobot effectively to match the MRI, we can increase the number of interdigital electrodes or add the thickness of the upper electrode as shown in Supplementary Fig.S7. Moreover, the refined structural design, including optimization of the inductance coil and loading of fixed capacitance and resistance, can be considered to facilitate cooperation with the MRI.
Even though the designed sensor was sensitive to the conductivity and permittivity change of the surrounding environment, which was usually related to the occurrence of diseases, it is not specific for different influence factors. Next, we will functionalize the upper electrode of the sensor by coating it with selective materials to specifically detect the concentrations of various ions, proteins, glucose, oxygen, and so on.35–40 After the functionalized sensor is combined with the marker, the permittivity and conductivity above the sensor will change. Therefore, the scene in which many microrobots patrol the body under external control are navigated by MRI and actively find abnormal parts according to the sensing signal for in-situ diagnosis and even treatment will be realized in the near future.