2010 IEEE International Conference on Microelectronic Test Structures, March 22-25, Hiroshima, Japan.

4.4

Analysis of the Performance of a Micromechanical Test Structure to Measure Stress in Thick Electroplated Metal Films
S. Smith , N.L. Brockie , J. Murray , C.J. Wilson , A.B. Horsfall , J.G. Terry , J.T.M. Stevenson , A.R. Mount and A.J. Walton
Institute for Integrated Micro and Nano Systems (part of the Joint Research Institute for Integrated Sytems) School of Engineering, Scottish Microelectronics Centre, The University of Edinburgh, Edinburgh, EH9 3JF, UK Email: Stewart.Smith@ed.ac.uk

School of Electrical, Electronic and Computer Engineering, Merz Court, Newcastle University, Newcastle upon Tyne, NE1 7RU, UK

IMEC, Kapeldreef 75, B-3001 Leuven, Belgium

School of Chemistry, Joseph Black Building, The University of Edinburgh, Edinburgh, EH9 3JJ, UK II. T EST S TRUCTURES The microrotating stress sensor [1][5] used in this study is fabricated in the layer being characterised and consists of three elements; two stress or expansion arms which anchor a third pointer arm, as shown in Fig. 1. An underlying layer is etched away to release the suspended structure while at the same time releasing any residual stress in the expansion arms. These arms expand or contract causing the pointer arm to rotate. In the structure in Fig. 1, a clockwise rotation indicates tensile stress while an anticlockwise rotation indicates compressive stress.

Abstract Previously reported suspended microrotating test structures designed to measure the stress in thick layers of electroplated Permalloy (NiFe alloy) have been analysed using nite element modelling and compared with experimental measurements. These results have been used to optimise a stress sensor test structure and design a new mask, with an array of test structures specically designed to wafer map the stress of thick nickel and Permalloy lms. This is the rst time these structures have been employed for determining spatial variation in lm stress and the results of this characterisation are reported for nickel.

I. I NTRODUCTION Characterisation and control of the intrinsic stress in deposited lms is important for many microfabrication processes and is becoming increasingly critical for many MEMS applications, especially those that require thick layers. The micromechanical stress test structures used in the present work have been previously implemented for evaluating the stress in Permalloy [1] (a generic term for nickel iron magnetic alloy). In this work, the Permalloy was deposited by an electroplating process and, as with many nickel based materials deposited in this way, signicant stresses can be induced, especially when thick layers are required as is often the case for MEMS devices. The stresses in blanket deposited layers are typically assessed through measurement of wafer bow but, as reported in [1], the stress sensor structure designed in this paper has the potential to map the stress across a wafer with high resolution. This paper initially undertakes the rst simulations of these thick lm structures to theoretically underpin their performance and compares these with data from measurements made on fabricated structures. These results are then used to optimise the design of the stress monitoring test structure, which has subsequently been fabricated in an array to determine the spatial variation of stress in a patterned nickel lm. 978-1-4244-6915-4/10/$26.00 2010 IEEE
80
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 24,2011 at 13:56:51 UTC from IEEE Xplore. Restrictions apply.

Fig. 1.

Permalloy stress test structure showing key dimensions

III. S IMULATIONS Stress test structures with the dimensions reported in [1] were modelled using the large strain, nite element (FE) model discussed in [6]. This model does not rely on elastic parameters and the solution is based solely on thermal strain. As the elastic properties of Permalloy at the dimensions used

Fig. 2.

FE simulation procedure for the stress sensor shown in Fig. 1

for these structures are unknown, only the strain can be calculated reliably. This can be arbitrarily converted into stress upon solution using elastic moduli from the literature or from other experiments. Thus the elastic modulus is not a tting parameter, but is still required to determine the stress. The ideal strain value is determined by comparing the simulated displacement with experimental results in an iterative loop as illustrated in Fig. 2. Fig. 3 presents the simulated pointer rotation as a function of the ratio of extension arm separation (Y) to feature width (W) for increasing values of strain (). Fig. 4 shows results from Permalloy test structures that have been tted to the model based on the assumption that the stress is equal for all geometrical variations (=0.083% strain). These show a good t with the general trend, suggesting that, within the variability of the experimental results, the model can be used to predict the sensor operation at different dimensions.
1

6 FE L=650m FE L=750m FE L=850m L=650m L=750m L=850m

Rotation ()

0 0 2 4 6 8

Arm Separation Ratio (Y/W)

(a)
6 FE L=650m FE L=750m FE L=850m L=650m L=750m L=850m

Increasing Strain Normalised Pointer Rotation (a.u.)

0.8 4

Rotation ()

0.6

2 0.4 1

0.2 0 0 2 4 6 8

Arm Separation Ratio (Y/W)

0 0 1 2 3 4

(b) Fig. 4. Results from Permalloy structures of arm width (a) W=8m and (b) W=10m tted with results from FE simulations

Arm separation ratio (Y/W)

Fig. 3. Simulated pointer rotation as a function of arm separation ratio (Y/W) for increasing values of strain

Assuming only a small variation in stress, the optimisation of the structure dimensions W and L can be considered. Fig. 5 shows the predicted maximum pointer rotation as a function of feature width for different arm lengths using the tting stress from Fig. 4. This demonstrates that the sensor must be optimised to increase the sensitivity, as the small variations in geometrical parameters do not give a sufciently large difference. The feature width is the most sensitive parameter to increase rotation [6], following a negative power relationship with stress. However, depending on the process it may be easier to increase the arm length L, which also increases the rotational sensitivity to stress, following a power relationship.

IV. FABRICATION Based on the results published in [1] and the simulations presented above, a new test mask was designed that consisted of an array of stress structures with L=850m, W=8m and Y/W=1.5, 1.75 and 2. The test die is shown in gure 6 (a) while the complete test mask layout for a 200mm (8") wafer is presented in gure 6 (b). Each die in the test mask design has 32 test structures and there are 384 die giving a total of 12288 structures. In each die there are seven structures with Y/W=1.5, seven with Y/W=2, and fourteen with Y/W=1.75. In addition, there are four structures where the expansion arms narrow to 5m where they attach to the pointer

81
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 24,2011 at 13:56:51 UTC from IEEE Xplore. Restrictions apply.

arm.
14

12

Increasing expansion arm length (L)

Pointer Arm Rotation ()

10

L=25 L=150 L=400 L=800 L=1400 L=2300 L=3500 L=5000

0 0 2 4 6 8 10 12 14 16 18

Expansion Arm Width (m)

Fig. 5. Pointer rotation plotted against feature width for different expansion arm lengths

before the excess seed layer was removed with a wet etch process. The structures were fabricated on 200mm silicon substrates coated with a sacricial layer of thermal SiO2 0.5m thick. The stress structures were released from the substrate by a dry HF vapour release in a Memsstar SVR XF tool. After release the stress structures were free to rotate, but there appeared to be little or no movement of the pointer arms on any structures which would indicate low levels of stress in the deposited Ni lm. Subsequent processing and packaging will potentially involve multiple heating steps, for example to cure photoresist or glues used in packaging. To assess the effects of this treatment the wafer was subjected to a brief heating step at 150C for 10 minutes on a hot plate, which mimicked these processes and would identify any changes in the metal layer. After the heat cycle all of the structures showed rotation indicating the development of tensile stress in the nickel lm. Fig. 7 shows two released test structures before and after heating.

(a)

(a)

(b) Fig. 7. Optical microscope images of the pointer arm tip of stress sensors (a) before and (b) after heating to 150C for 10 minutes

V. R ESULTS AND D ISCUSSION

(b) Fig. 6. Schematic layouts of stress sensor chip (a) and wafer (b)

The mask is intended to be used to pattern an array of these structures across a 200mm wafer in order to measure any variation of stress with position. The rst batch of test structures fabricated using this design were in electroplated nickel 5m thick. The structures were plated from a copper seed layer using an 8m thick mould of SPR220-7 photoresist. After plating the resist was removed by rinsing with acetone

Initial measurements of the pointer arm rotation were made by capturing images of the released structures with a digital camera attached to an upright optical microscope and extracting the angle from the image. Two complete sets of structures were measured, one in the centre of the wafer and one near the edge. Each test die contains seven structures with Y/W=1.5, seven with Y/W=2, and fourteen structures with Y/W=1.75. The results are presented in Fig. 8. Each data point in Fig. 8 represents the average rotation for the set of structures with a given value of Y/W, while the

82
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 24,2011 at 13:56:51 UTC from IEEE Xplore. Restrictions apply.

Centre Die Edge Die

4.5

3.5

2.5

2 1.5 1.6 1.7 1.8 1.9 2

Arm Separation Ratio Y/W

software to extract the rotation for each structure. Fig. 9 shows the results from the wafer of nickel stress structures. Where the structures are damaged and could not be measured a value of 0 is used which shows up as black. There is a clear pattern to the data with greater rotations near to the centre but the variability is quite large and there are also structures with large rotations around the edge. The poor yield also leads to many empty spaces in the map which can make it difcult to see the pattern. The problems with yield are mainly caused by the wet etch process used to remove the copper seed layer before release as this also tends to attack nickel. Optimisation of this process to limit the overetching of the nickel structures, as well as the permalloy structures which will be fabricated in future, should help to increase the yield.

Angle of Rotation ()

Fig. 8.

Pointer rotation plotted against Y/W for nickel test structures

error bars show the standard deviation. Statistical tests were performed using the data analysis programme, RS/1, to determine if the differences between structures at the centre and the edge of the wafer were signicant. These tests conrmed, with a 95% condence, that there is a statistically signicant difference between the two sets with greater rotation in the centre of the wafer. These results suggest a difference in stress between the centre and the edge of the wafer, provided that there is no signicant difference in the geometry of the test structures, particularly in terms of the expansion arm width. Each of the test dice also includes a set of eighteen cross-bridge electrical linewidth test structures [7]. In order to investigate any dimensional differences between the structures, one set at the centre and one at the edge of the wafer have been measured and the results are presented in table I.
TABLE I E LECTRICAL LINEWIDTH RESULTS FOR NICKEL TEST WAFER Position Centre Edge Mean Linewidth (m) 5.11 4.98 (Linewidth) (m) 0.13 0.19

Fig. 9. Wafer map of pointer rotation for a 200mm wafer with nickel stress test structures

The difference between the mean values of linewidth at the centre and edge is around 2%, with wider structures in the centre. If the stress structures show similar dimensional bias then it would be expected that the rotation of the pointer arms, for a given value of stress in the lm, should be reduced for the wider expansion arms. In addition, and referring back to Fig. 5, simulations suggest that the observed change in linewidth will lead to a change in rotation of less than 0.1 . All of this is suggestive of a difference in the level of stress in the nickel between the centre and the edge of the wafer. In order to investigate this in more detail and to obtain wafer maps of the distribution of the stress a new system has been developed which uses a PC running LabView to control a 200mm semi-automatic wafer prober with a CCD camera mounted on the microscope. This allows the test structures to be quickly scanned and analysed using image processing

As previously stated, there are 32 test structures in each of the 384 die on the wafer. By taking an average value of rotation for each die it is possible to reduce the granularity of the wafer map and get a more useful picture of the variation. This large scale map is presented in Fig. 10. The average rotation is greater in the centre of the wafer at around 3.5 while it is closer to 2.5 at the edge. The large rotations near the edge in the upper left quadrant of the wafer are a result of measuring one or two structures in each die which show very large rotations. The automated measurement is not able to determine if a structure is broken so these structures may not represent actual variation in stress. It is clear from Fig. 9 that the yield is particularly low in this area. Fig. 10 shows the average for all the structures in each die, but as was mentioned previously there are a number of different structures with different values of expansion arm separation ratio. The average rotation for each value of Y /W has been calculated for each die and is plotted in Fig. 11. In addition the averages for a block of 618 dice in the centre of the wafer and the remaining structures around the edge have also been calculated and are plotted on the same graph. Each point represents the mean value while the error bars show the

83
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 24,2011 at 13:56:51 UTC from IEEE Xplore. Restrictions apply.

Pointer Rotation (degrees)

the standard test structures while Fig. 12 presents the results for the structures where the expansion arms narrow to 5m where they meet the pointer arm.
3.5 4

1 100 80 60 Vertical Position (mm) 40 20 0 -20 -40 -60 -80

1.5

2.5

Fig. 12. Average pointer arm rotations plotted against expansion arm separation ratio for narrowed structures

-100 -100 -80 -60 -40 -20 0 20 40 Horizontal Position (mm)

60

80 100

Fig. 10. Wafer map of the average rotation for each die on the 200mm nickel test wafer

standard deviation (1). The total number of structures that were measured to produce each mean value varies due to the yield issues mentioned previously but there are generally a few hundred results behind each point.

These results show very similar, statistically signicant, differences between the centre and the edge of the wafer and similar variabilities in the rotation for each value of Y /W . There is more variation in rotation with expansion arm separation but this is insignicant in comparison to the overall variability. There is no signicant difference in rotation between the standard structures and those with the narrowing of the expansion arms. In fact the average is slightly lower for the narrowed structures which is the opposite from what might have been expected from theory. Simulations of the strain/stress as a function of a test structure with W=5, L=850 can be seen in Fig. 13. A typical value of elastic modulus of 200GPa has been used to convert the simulated strain into a stress value. Even if the change in rotation between the centre and edge of the wafer is as low as 0.5 that equates to a change in stress of around 20MPa.

Fig. 11. Average pointer arm rotations plotted against expansion arm separation ratio

Although the standard deviations are quite large the differences between the structures near the centre and those around the edge have been shown to be statistically signicant, with a 95% condence, by standard tests in RS/1. Interestingly there is no obvious trend in rotation with Y /W , unlike the initial results. Either this is very small and is being masked by the variation across the wafer or there is no dependence on Y /W , which suggests the structures are operating around the peak of the curve in Fig. 3. Fig. 11 shows the results for

Fig. 13. Simulated stress and strain gures plotted against maximum pointer rotation angle

VI. C ONCLUSIONS AND F UTURE W ORK Finite element modelling has been used to simulate stress test structures and t the results to measurements made on

84
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 24,2011 at 13:56:51 UTC from IEEE Xplore. Restrictions apply.

structures fabricated in thick electroplated Permalloy. These results have been used to optimise the test structure dimensions and design a new test mask, which replicates the stress sensors across a 200mm silicon wafer. This mask has been used to fabricate stress sensor structures in a 5m thick layer of electroplated nickel, which initially showed no obvious rotation after release. However, after heating to 150C the structures rotated to indicate tensile stress. A full set of structures at the edge and at the centre of the wafer were measured and the results showed a signicant difference in rotation. This cannot be accounted for by differences in the widths of the expansion arms in these structures and so the results indicate a difference in stress with position. A semi-automatic wafer prober and image analysis software has been used to measure every structure on the nickel plated test wafer. Wafer maps of the results show a clear pattern with higher rotations in the middle of the wafer than at the edge. The full wafer data has been broken down by test structure geometry and also by position on the wafer. This conrmed that there is a signcant difference in sensor pointer arm rotation between the middle and the edge of the wafer. Overall the results suggest that there is greater tensile stress in the plated nickel near the middle of the wafer than there is nearer to the edge. This is the rst time this spatial variation has been observed in this manner in a patterned, electroplated lm. Further work is required to analyse the spatial stress distribution in electroplated nickel, Permalloy and copper lms by comparing measurements from stress sensor structures with optical measurements of blanket coated layers. Patterned and blanket coated test wafers with nickel and the other materials are in the process of being fabricated but the results obtained so far indicate the capability of this test structure for measuring the cross wafer variation of stress in patterned lms. Future efforts will focus on extending the technique to investigate and characterise other materials and processes.

ACKNOWLEDGEMENTS The authors would like to acknowledge the support of the Edinburgh Research Partnership in Engineering and Mathematics. They would also like to thank Dr. Anthony OHara and Graeme Pringle of Memsstar for help with processing. Jeremy Murray would like to acknowledge the support of the SFC funded SPIRIT studentship scheme. R EFERENCES
[1] S. Smith, N. L. Brockie, J. G. Terry, N. Wang, A. B. Horsfall, and A. J. Walton, Application of a Micromechanical Test Structure to the Measurement of Stress in an Electroplated Permalloy Film, in Proceedings of the IEEE International Conference on Microelectronic Test Structures (ICMTS09), Oxnard, CA, US, March-April 2009, pp. 75 80. [2] X. Zhang, T. Y. Zhang, and Y. Zohar, Measurement of Residual Stresses in Thin Films Using Micro-Rotating Structures, Thin Solid Films, vol. 335, no. 1, pp. 97105, 1998. [3] J. G. Terry, S. Smith, A. J. Walton, A. M. Gundlach, J. T. M. Stevenson, A. B. Horsfall, K. Wang, J. M. M. dos Santos, S. M. Soare, N. G. Wright, A. G. O. Neill, and S. J. Bull, Test Chip for the Development and Evaluation of Sensors for Measuring Stress in Metal Interconnect, IEEE Transactions on Semiconductor Manufacturing, vol. 18, no. 2, pp. 225 261, May 2005. [4] C. J. Wilson, A. B. Horsfall, A. G. ONeill, N. G. Wright, S. J. Bull, J. G. Terry, J. T. M. Stevenson, and A. J. Walton, Direct Measurement of Electromigration-Induced Stress in Interconnect Structures, IEEE Transactions on Device and Materials Reliability, vol. 7, no. 2, pp. 356 362, June 2005. [5] J. M. M. dos Santos, K. Wang, A. B. Horsfall, J. C. P. Pina, N. G. Wright, A. G. ONeill, S. Soare, S. J. Bull, J. G. Terry, A. J. Walton, A. M. Gundlach, and J. T. M. Stevenson, Calibration of MEMS based test structures for predicting thermo-mechanical stress in integrated circuit interconnect structures, IEEE Transactions on Device and Materials Reliability, vol. 5, pp. 713719, 2005. [6] C. J. Wilson, Stress Measurements in Deep Sub-micron Damascene Copper Interconnects, Ph.D. dissertation, Newcastle University, 2009. [7] M. G. Buehler, S. D. Grant, and W. R. Thurber, Bridge and van der Pauw Sheet Resistors For Characterizing the Line Width of Conducting Layers, J. Electrochemical Soc - Solid State Technology, vol. 125, no. 4, pp. 650654, April 1978.

85
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 24,2011 at 13:56:51 UTC from IEEE Xplore. Restrictions apply.