**3. Results and Discussion**

Figure 7a illustrates the released normal force sensor. The interference fingers in the diaphragm and supporting fingers indicated that these structures were temporarily bonded to the substrate due to the stiction effect. The stiction strength during the stiction process was estimated using the longest unattached cantilever to be higher than 7.06 kPa and lower than 22.31 kPa, as shown in Figure 7b.

Figure 8 depicts the integrated normal force sensor, the size of which is approximately equal to those of LMV358, 1070 µm × 640 µm × 525 µm. The maximum alignment error of eutectic bonding was measured to be approximately 1.5 µm, sufficient for the proposed integration process. A Dage Series 4000 bond tester (Nordson DAGE, UK) was used to test the shear strength of Au–Si eutectic bonding. The shear strength of the bonded test structure was approximately 30.74 MPa. The serial resistance of the Au–Si eutectic bonding area is lower than 2 Ω [32], which is much lower than that of polysilicon piezoresistors and can be neglected.

The stress in the transferred normal force sensor caused by the Au–Si eutectic flip-chip bonding process was estimated by comparing the output voltages of the Wheatstone bridge before and after the transfer. The change in the output voltage was in the range of −7.76 to +7.25 mV. Therefore, the stress produced by the Au–Si eutectic flip-chip bonding process was calculated to be in the range of −9.95 MPa to +9.30 MPa, which can be neglected.

**Figure 7.** Optical images of released structures; (**a**) diaphragm supported by four beams; (**b**) cantilever array. Cantilevers longer than 80 µm are colorful due to the interference patterns of uneven gaps, while those shorter than 60 µm exhibit uniform color, which indicates that all cantilevers longer than 80 µm adhered to the substrate. Stiction strength is estimated using the longest unstuck cantilever and shortest stuck cantilever.

**Figure 8.** (**a**) Scanning electron micrographs of the integrated normal force sensor; (**b**) close-up view of the transferred diaphragm.

The integrated normal force sensor was measured using a set of homemade copper wire weights [24]. The source voltage of the Wheatstone bridge was set to 5 V using Agilent E3631A (Agilent, Santa Clara, CA, USA), and the corresponding output voltage of the instrumentation amplifier was recorded using Agilent 34401A (Agilent, USA), when different masses of the beam-shaped copper wire weights were placed on the diaphragm of the transferred normal force sensor under the microscope; Figure 9 presents the measurement results. The sensitivity was calculated to be 93.5 µV/µN/V. The sensitivity of the piezoresistive Wheatstone bridge was calculated to be 18.8 µV/µN/V at an amplification of 4.96. Nonlinearity was approximately 4%, which was quite large and mainly caused by the uncertainty of the point-of-force application. Five sensors were measured. The deviation of sensitivity was less than 20%. System noise was measured at approximately 200 µV.

**Figure 9.** Measurement results of the integrated normal force sensor. Sensitivity was calculated to be 93.5 µV/µN/V.

Shear force sensors can also be developed with flat diaphragms [33] by modifying piezoresistive Wheatstone bridge routing. When shear force in the x direction is applied on the center of the diaphragm, the left side of the diaphragm moves down while the right side moves up, as shown in Figure 10a. The stress of piezoresistor R<sup>1</sup> in Figure 10b is tensile, while that of R<sup>3</sup> is compressive. When the piezoresistors are connected, as in the Wheatstone bridge in Figure 10c, the output is sensitive to shear force and insensitive to normal force. The bumps in the center of the diaphragm can be used as a mesa to improve shear force sensitivity.

**Figure 10.** Shear force sensors developed by modifying routing of piezoresistive Wheatstone bridges. (**a**) When shear force in the x direction is applied on the center of the diaphragm, the left side of the diaphragm moves down while the right side moves up. (**b**) Released shear force sensor. (**c**) Wheatstone bridge of the shear force sensor.

### **4. Conclusions**

Although the integration of sensors with readout circuits is a crucial technology required for large-area, high-resolution tactile sensing, to date, few integrated tactile sensors have been identified, owing to strict thermal budgets and process compatibility. Tactile sensors are typically integrated with the readout circuit in system levels [1–13]. In this study, an integrated piezoresistive normal force sensor was presented. The surface micromachined normal force sensor was transferred to the readout circuit chip, with a temporary stiction effect handling process.

The piezoresistive normal force sensor was manufactured using surface micromachining. The readout circuit chip comprised two CMOS operational amplifiers, which were redistributed to form an instrumentation amplifier. The SETH process was used to transfer the released sensor to the readout circuit chip. Because the MEMS structure and readout circuits were manufactured separately, they were optimized independently. Because the MEMS structures were released before transfer, readout circuits did not undergo etching for release and could be manufactured using a normal IC foundry. These processes feature excellent compatibility with IC chips.

The normal force sensor was designed for pulse diagnosis instruments. The size of the transferred normal force sensor was 180 µm × 180 µm × 1.2 µm. The maximum misalignment in the flip-chip bonding process was approximately 1.5 µm. The sensitivity was measured to be 93.5 µV/µN/V. The routing of the piezoresistive Wheatstone bridge can be modified to develop shear force sensors; hence, this technique can be used to develop tactile sensors, capable of sensing both normal and shear forces.

The size of the integrated normal force sensors is approximately equal to those of the readout circuit chips, because the sensors are significantly smaller and sit on top of the readout circuit chips. In our experiments, the wafers of a very-old-version operational amplifier (LMV358) were employed to verify the technology, owing to ease of accessibility. The LMV358 chip size is approximately 1070 µm × 640 µm × 525 µm, and of approximately 0.5 µm minimum line width. Hence, extremely compact sensors can be developed with modern instrumentational amplifiers.

**Author Contributions:** N.L. contributed to the design and testing of the device; P.Z. contributed to the fabrication and part of the testing of the device; K.S. contributed to the test idea; C.Z. helped implement part of the fabrication; Y.Z. contributed to the research guidance for the work; H.Y. contributed to the research idea and provided guidance for the work. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the Shanghai Municipal Science and Technology Commission (Project No. 18dz1100600).

**Institutional Review Board Statement:** The study was conducted in accordance with the Declaration of Helsinki, and approved by the ethical committee of Longhua Hospital (2019LCSY035, 2020.08.27), Shanghai University of Traditional Chinese Medicine.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors appreciate financial support from the Shanghai Municipal Science and Technology Commission (project 18dz1100600). The authors also appreciate the assistance of engineers at the State Key Laboratory of Transducer Technology and the staff of Longhua Hospital, Shanghai University of Traditional Chinese Medicine.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**

