Mobile QR Code QR CODE

  1. (Shenzhen institutes of advanced technology, Chinese academy of science, Shenzhen, China)
  2. (Guangdong power exchange center Co., Ltd., Guangzhou, China)

Manipulator, CMUT, CMOS-MEMS, failure analysis


The number of old people with severe disability and disabled people caused by accident is increasing. Paramedics need to regularly turn the patients over, move them around, and massage these long-term bedridden patients to prevent pressure sores and other diseases. Up to recently, the number of nurses per thousand people in China, the European Union, the United States, Japan, and Norway is 3.6, 8, 9.8, 11.49 and 17.27 (1). Therefore, the study of a dexterous robot manipulator is of great significance to reduce the burden of nurse. The robot manipulator can also help standardize the rehabilitation process for the patients (2). Different from other robot manipulators, the safety and comfort of human limbs are the first consideration in the design of humanoid nursing manipulator. The main parameters affecting the safety and comfort of grasping human limbs are the stress and strain generated by the grasping force inside the human limbs. Therefore, it is necessary to develop robotic fingers with proximity sensors, so that it continuously adjusts the strength to achieve soft touching and grasping. Ultrasonic transducer plays very important role in the proximity sensors.

At present, piezoelectric ultrasonic transducer is commonly used on the market (3). However, it cannot be applied flexibly because of its large size. Due to the small area of robotic fingers, the sensors placed in it are required to be small. Compared with the traditional piezoelectric ultrasonic transducer, the advantages of CMUT fabricated on CMOS-MEMS technology are as follows. Its size is only about 10 mm2. It is compatible with CMOS process and easy to do integration and mass production. Further more it has friendly impedance matching characteristics. Because of these advantages, CMUT becomes the preferred choice for the proximity sensors.

CMUT (Capacitive Micro-machined Ultrasonic Transducer) benefits a lot from the rapid development of MEMS (Micro-Electro-Mechanical System). The receiver and transmitter of CMUT share the same structure and it can be used as a receiver as well as a transmitter, as shown in Fig. 1. In the receiving mode, the incident ultrasonic wave causes the deformation of the membrane, thus affecting the capacitance between the top electrode and bottom one. A suitable circuit is applied to convert the current signal caused by the change of the capacitance into a voltage signal. In other words, the transducer realizes the energy conversion from ultrasonic energy to mechanical energy and then to electrical energy (4). In the transmission mode, a DC voltage is firstly applied between the top and bottom electrode of CMUT, where the membrane is attracted by the electrostatic force. And then an AC signal is applied. The membrane vibrates at a certain frequency, which emits ultrasound wave at the same time. In the transmission mode, CMUT realizes the energy conversion from electrical energy to ultrasonic energy.

The rest of the paper is as follows. In section II, a removable column is ingeniously designed for our CMUT. In section III, the designed CMUT is fabricated and the condition of the device is checked. In section IV, the common failure of CMUT device is studied and the cause of the failure is determined.The last section is the conclution.

Fig. 1. Transmitting mode (left) and receiving mode (right) of CMUT.


II. Title

In this section, the integration between CMOS and MEMS process for CMUT fabrication is discussed. A special CMUT structure with three cantilevers supporting a circular membrane is proposed. 0.18 um CMOS standard process is adopted in our design.

1. The Design of the Air-coupled CMUT

In standard CMOS process, the materials and their distribution of each layer have been determined already (5). Considering the feasibility of MEMS process and the availability of poly layer, SiO2 and Al are taken as the material of CMUT membrane where Al in Metal2 is applied as sacrificial layer.

Considering the dielectric loss in the air, the frequency of air coupled ultrasonic transducer is generally 40-200 kHz. The low-frequency CMUT usually occupies a large area on chip. However, under CMOS standard process, the available area in a wafer is limited. In order to improve the area utilization on chip, we propose a novel CMUT structure with three cantilevers supporting a circular membrane. Fig. 2 shows the design of our CMUT structure.

In CMOS standard process, the thickness of metal2 as sacrificial layer is 0.55 um only. Due to the existence of liquid surface tension in wet etching process, the top electrode and the bottom one often stick together, which leads to device failure. Obviously, the increase of Metal2 thickness can solve the problem. However, the Metal2 thickness is fixed in standard process. Thus, a central column between the top electrode and the bottom one is proposed to resolve the device failure problem. This column will be removed by MEMS process later.

Fig. 2. The proposed CMUT structure.


2. The Structure Design of the Air-coupled CMUT

In this section, the process steps are demonstrated by using the cross-sectional view. In MEMS process, twice dry etching and one wet etching are used to release the device. It is worth noting that the process has self-stopping capability.

After CMOS process, we obtain the die from foundry. The chip cross-sectional view of the CMUT structural layers is shown in Fig. 3(a). M1 is designed as the bottom electrode. M2 is designed as the gap filled with air. M3 is as the top electrode. M4 is as the protection layer of the second dry etching, which acts as a mask. M5 is as the protection layer for the first dry etching, which also acts as a mask. In addition, it is noted that there are two vertical dotted lines and a circle in the middle of the diagram, which is our proposed structure of the central column.

Fig. 3. (a) The cross-sectional view of the CMUT structural layers. (b) After the first ICP etching.


We use ICP (inductively coupled plasma) to do deep etching. It starts etching from top to bottom, with its characteristics of excellent verticality and uniform rate (6). The ICP etching has good selectivity because it can only etch non-metallic compounds with a proper power. It stops etching when encountering metal such as Al. The etching result is shown in Fig. 3(b).

The aim of this process is to corrode the sacrificial layer. Here, we use a corrosive solution mixed acetic acid and nitric acid. In MEMS process, the wet etching of a certain material can only corrode the exposed metal without passing through the protective layer by controlling the concentration of corrosive solution (7). The wet etching result is shown in Fig. 4(a). It can be seen that the central column provides additional support for the membrane. After the wet etching, the samples need to be dried in supercritical dryer.

There are two purposes of the second dry etching. One is to expose the pads so that the device can be connected with an offchip test circuit through wire bonding. On the other hand, the central column needs to be removed so that the membrane can vibrate. The cross section after the second etching is shown in Fig. 4(b).

So far, the complete anti-collapse, self-stopping MEMS processing for CMUT has presented where the reliability and fault tolerance of the whole process are both considered.

Fig. 4. (a) After wet etching, (b) After second ICP etching.



Based on micro-nano scale, MEMS is a kind of precise fabrication technology. Therefore, the whole process should be completed in an ultra-clean room. The process of the fabrication mainly involves dry etching and wet etching. In this section, the MEMS process and test will be presented.

1. MEMS Fabrication

After obtaining the chips from foundry, we take the microscopic picture of the die, as shown in Fig. 5(a). The CMUT array is in the square region in deep orange. These top Layers mixed Si3N4 with SiO2 are above the CMUT structure. Because the thickness of Si3N4 is relatively thin, it is transparent under microscope. This figure corresponds to the section of Fig. 3(a).

After dry etching, the result is shown in Fig. 5(b). SiO2 and Si3N4 on Al are etched away. The yellow is metal Al. Fig. 5(b) is consistent with the results of our model.

Fig. 5. (a) Initial profile of die, (b) After first ICP etching.


In wet etching, the metal Al is corroded away with acid solution. Then the chip is taken out and dried with supercritical drying apparatus. As shown in Fig. 6(a), the dark orange part is SiO2, which meets the expectation.

Because the thickness of the layer between M2 and M3 in second dry etching is thinner than that in first etching, the erosion period needs to be reduced with the same formula. The etching result is shown in Fig. 6(b). After this etching, the pads are exposed. So far, MEMS process in the ultra-clean room has been completed.

Fig. 6. (a) Wet etching, (b) Second ICP etching.


2. Device Testing

After observing the specific profile of CMUT under SEM (Scanning Electron Microscope), the CMUT device is wire bounded to a PCB, and then connected to the impedance analyzer for testing.

A. SEM Observation

The SEM used in the test is Zeiss Sigma 300. As shown in Fig. 7(a), the surface of CMUT is smooth and the structure of the disc and beam is complete without any damage. As we can see in Fig. 7(b), the outer diameter of CMUT is 240 um, the inner diameter is 178 um, and the diameter of central small disc is 14 um, which all meet the design specifications.

Fig. 7. (a) CMUT structure in SEM, (b) CMUT parameters in SEM.


B. Impedance Analysis

As shown in Fig. 8, the device is wire bounded to a PCB test board. Then, it is connected to an impedance analyzer which is Keysight 4990A. In the test, the scanning scale is from 1 kHz to 1 MHz. Unfortunately, we cannot get any signal during the impedance test. Some MEMS failures show in the test. We will perform failure analysis in the following section.

Fig. 8. The device is wire bonded to a PCB.



The goal of this section is to find out the cause of the CMUT failure, which is very important for the design, fabrication, and production yield. The failure analysis of the device is carried out in the two stages of CMUT processing and its final released device.

1. Study on the Fabrication Process

In the step of dry etching, due to the top-down etching process of ICP etching, the processing conditions of the device surface can be observed directly through the microscope, which seems in good condition. The purpose of the wet etching stage is to form a cavity which can help CMUT vibrate freely. The cavity is under the membrane and cannot be observed directly. Therefore, the process quality of wet etching in MEMS processing is our focus.

Here, we use high power ultrasonic cleaning equipment with similar frequency to peel off the top layers of CMUT. The detailed operation is as follows. The CMUT is processed as the same procedure as the previous one. It is placed in a small closed bottle filled with deionized water. The bottle is placed in an ultrasonic cleaner for 60 min. As shown in Fig. 9(a), CMUT top electrode layers were successfully peeled off, and a pit is exposed. We use a surface profilometer to test the peeled CMUT. The profilometer measures the depth of the groove along the red dotted line, as shown in Fig. 9(a). The measured result is shown in Fig. 9(b).

In the process of wet etching, there are two kinds of corrosion effects. Insufficient or too long corrosion time leads to rough surface and little bumps appear in the detected curve. If the corrosion time is appropriate, the plane will be smooth and the probe curve will be smooth as well. Due to the edge jitter effect of the profilometer, the double peak structure of the bump appears, which is circled in Fig. 9(b). In addition, the protrusion in the middle is caused by our proposed central column. The curves of the rest parts are smooth, which indicates that the sacrificial layer has been corroded enough in the wet etching.

Fig. 9. (a) CMUT after being peeled off the top electrode layers, (b) Test result using the profilometer.


2. Analysis on Released Device

This analysis is mainly aimed at the devices after the completion of MEMS processing. From the types of actual physical failure, device failure can be divided into the 4 types.

(a) Fracture failure. Fracture failure means that the device breaks when the load applied on the movable structure of the device exceeds the maximum value it can bear without significant damage precursor.

(b) Adhesion failure. When two smooth surfaces touch each other, they tend to bond together due to some adhesive forces. For MEMS devices, the ratio of the movable surface area to the cavity height is relatively large, which makes the capillary force and van der Waals force relatively enhanced in the micro scale. They are the causes of adhesion failure between micro structures.

(c) Wear failure. It usually occurs in two solid materials in contact. Under the mechanical operation of polishing, the wear failure caused by the relative mechanical movement between the surfaces is usually accompanied by the removal or thinning of the surface material.

(d) Fatigue failure. Fatigue failure refers to the failure of the device for a long-term operation under the periodic excitation of the load lower than its yield strength. If the periodic load exists in the device for a long time, the material strength of the device structure will be weakened gradually, resulting in small cracks on the surface. Most MEMS micro actuators have movable structures. The movable parts often work under periodic loads. Therefore, the fatigue failure can be delayed but cannot be avoided.

According to Fig. 7(a), the surface of CMUT disc is flat and the cantilever is not fractured. Therefore, fracture failure can be ruled out. From the review of fabrication process at the beginning of this section, the top layer of the electrode is successfully peeled off. That means there is no any adhesion. Thus, the adhesion failure can be ruled out. In the process of MEMS, the device is not polished so that the wear failure is not the cause. In conclusion, fatigue failure is most likely the cause.


In this section, fatigue analysis on CMUT structure is carried out with COMSOL stress life section. The structural fatigue life analysis method is based on the following assumption (8). If the material of the structure is consistent, and the stress concentration factor and load spectrum are the same, then the fatigue life is also consistent. The analysis process is shown in Fig. 10.

The optimal operating point of CMUT is that the center displacement occupies about 1/3 of the cavity height (9). With the steady-state simulation of COMSOL, it is found that the static voltage applied at the center position is 12.3 V. In pre-stress analysis, we need to set multi-physical field to electro-mechanical mode. The fixed constraint is set at the edge of the beam structure. The stress map is calculated. As shown in Fig. 11, the maximum stress is at the pining place of cantilever.

Fig. 10. Fatigue model analysis using COMSOL.


Fig. 11. The max stress in CMUT.


In order to make CMUT structure reciprocating under stress, cosine function is defined at the boundary load, as shown in Fig. 12(a). The amplitude is the cosine vibration with the maximum value of boundary load. Then, the spring foundation is set to replace the fixed constraint in solid mechanics, and the boundary conditions are consistent with the fixed constraints in steady-state simulation. In addition, in order to calculate the fatigue life, we need to input a suitable S-N (Stress-Life) curve. The curve takes the fatigue strength of material standard specimen as ordinate and logarithm of fatigue life as abscissa, which represents the relationship between fatigue strength and fatigue life of standard specimen under certain cyclic characteristics, also known as stress life curve. According to the article (10), the S-N curve of CMUT thin films based on CMOS-MEMS technology can be obtained, as shown in Fig. 12(b).

In COMSOL, the required parameters of fatigue simulation have been prepared, and the calculation is carried out. The operation results, the fatigue life, and stress intensity distribution of CMUT structure are shown in Fig. 13. It can be seen from the figure that the minimum fatigue life is 6027 times, which is located at the junction of arm end and disc of CMUT. It shows that the interface between arm end and disc is the most vulnerable place. In our experiments, the excitation signal is a 200 kHz square wave. The test signal is transmitted three cycles per second and the number of excitation in each cycle is 8, according to the excitation mode of commercial air-coupled transducer. Therefore there are 24 vibration cycles per second. The CMUT vibrates more than 6027 times in only 252 seconds. The CMUT structure is at the end of its stress life time.

In conclusion, the reason that the expected signal of the CMUT cannot be detected is because of fatigue damage.

Fig. 12. (a) load-displacement curve, (b) S-N curve.


Fig. 13. (a) Results of fatigue life simulation, (b) The detailed display.



This paper has proposed a novel CMOS-MEMS compatible method to fabircate CMUT. The CMUT array has been designed and fabricated using the standard 0.18um CMOS process and post CMOS process. The adhesion of the upper and lower electrodes, which often appears in the CMUT fabrication, has been described. A central column is proposed to resolve the problem, by supporting the upper and lower electrodes in the wet etchings. It can be ingeniously removed without affecting the device.

Our CMUT device is fabricated in line with the expectation. The geometry parameters and surface morphology of CMUT are consistent with our design as well. The failure of CMUT device has been studied. The cause of the failure has been determined to be the fatigue damage.


This work was supported in part by National key R&D program of China (Grant number 2016YFC0105002, 61674162, 2018YFB2100904, U1913601, 2018YFF010 12500), Shenzhen Key Lab for RF Integrated Circuits, Shenzhen Shared Technology Service Center for Internet of Things, Shenzhen government funds (Grant numbers JCYJ20180305164616316).


Cohen B., Hessels A., Kelly A., Chen L., Larson E., 2019, Impact of Exposures and Outbreaks on Staff Nurse and Infection Preventionist Workload, Am. J. Infect. Control, Vol. 47, No. 6, pp. s7DOI
Liu B., Wang L., Liu M., 2019, Lifelong federated reinforcement learning: a learning architecture for navigation in cloud robotic systems, IEEE Robot. Autom. Lett., Vol. 4, No. 4, pp. 4555-4562DOI
Kiihamaki J., Dekker J., Pekko P., 2004, ‘Plug-up’—a new concept for fabricating SOI MEMS devices, Microsyst. Technol., Vol. 10, No. 5, pp. 346-350DOI
Savoia A. S., Caliano G., Pappalardo M., 2012, A CMUT probe for medical ultrasonography: From microfabrication to system integration, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, Vol. 59, No. 6, pp. 1127-1138DOI
Khanghah M. M., Sadeghipour K. D., 2014, A 0.5 V offset cancelled latch comparator in standard 0.18 um CMOS process, Analog Integr. Circuits Signal Process., Vol. 79, No. 1DOI
Zhang R., Ren Z., Yu S., 2008, Loss-Reduction in Flexibly Vertical Coupled Ring Lasers Through Asymmetric Double Shallow Ridge and ICP / ICP Cascade Etching, Vol. 20, No. 22, pp. 1821-1823DOI
Zhang Y., Zhu Q. S., Itoh T., Maeda R., Toda \A., 2013, High-resolution wet etching technology of thick electroless nickel alloy film for MEMS devices and packaging, Proc.2013 14th Int. Conf. Electron. Packag. Technol. ICEPT 2013, pp. 396-399DOI
Arora S., Arora A., George P. J., 2012, Design of MEMS based microcantilever using comsol multihysics, Int. J. Appl. Eng. Res., Vol. 7, No. 11, pp. 1-3Google Search
Degertekin F. L., 2013, Harmonic cmut devices and fabrication methods., J. Acoust. Soc. Am., Vol. 130, No. 3, pp. 1781Google Search
Zhou W., Xu P., 2013, Theoretical investigation on the dynamic performance of CMUT for design optimization, Acta Mech. Solida Sin., No. 1, pp. 99-110DOI


Jiajun Liu

Jiajun Liu earned his MS degree in Shenzhen Key Laboratory for RF Integrated Circuits at Shenzhen Institutes of Advanced Technology, Chinese academy of science.

He is currently with Guangdong power exchange center Co., Ltd.

His R&D interests are in the fields of sensors and information management system.

Benxian Peng

Benxian Peng earned his Master degree of mircoelectroinc engineering in 2008, from Shanghai Jiao Tong University, He currently foucus on the research of the device and IC design of ultrasound chip.

Fengqi Yu

Fengqi Yu earned his Ph.D. degree in Integrated Circuits and Systems Lab (ICSL) at UCLA.

Before joining Shenzhen Institute of Advanced Technology in June 2006 as a full professor, he worked at Rockwell Science Center (USA) as a design engineer, Intel (USA) as an analog circuit design engineer, Teradyne (USA) as a Sr.

IC design engineer, Valence (USA) as a Sr. principal engineer, and Suzhou CAS IC Design Center (China) as VP and RF department director.

His R&D interests are in the fields of low power communication networks, CMOS integrated circuits, CMOS sensors, wireless sensor networks, and RFID.