Open Localization in 3D Package with TSV Daisy Chain Using Magnetic Field Imaging and High-Resolution Three-Dimensional X-ray Microscopy

: With the development of 3D integrated packaging technology, failure analysis is facing more and more challenges. Defect localization in a 3D package is a key step of failure analysis. The complex structure and materials of 3D package devices demand non-destructive defect localization technology for full packages. Magnetic ﬁeld imaging and three-dimensional X-ray technology are not affected by package material or form. They are effective methods to realize defect localization on 3D packages. In this paper, magnetic ﬁeld imaging and high-resolution three-dimensional Xray microscopy were used to localize the open defect in a 3D package with a TSV daisy chain. A two-probe RF method in magnetic ﬁeld imaging was performed to resolve isolation of the defect difﬁculties resulting from many different branches of TSV daisy chains. Additionally, a linear decay method was used to target sub-micron resolution at a long working distance. Multiple partition scans from a high-resolution 3D X-ray microscopy with a two-stage magniﬁcation structure were used to achieve sub-micron resolution. The open location identiﬁed by magnetic ﬁeld imaging was consistent with that identiﬁed by a three-dimensional X-ray microscope. The opening was located on the top metal in the proximity of the ﬁfth via. Physical failure analysis revealed the presence of a crack in the top metal at the opening location.


Introduction
Recently, with the rapid development of the integrated circuit industry, two-dimensional packaging technology has been unable to meet the increasing requirements for chip performance and packaging density [1]. Therefore, in order to continue Moore's Law, the development of two-dimensional integrated circuits to three-dimensional integrated circuits is inevitable [2]. TSV interconnect technology is recognized as one of the most promising technologies for 3D packaging [3]. TSV is connected to micro-bumps through a redistribution line (RDL) [4]. Reliable interconnection is the prerequisite and guarantee for high-quality 3D integration [5]. The tiny scale and layered structure of 3D TSV packages make defect detection and failure analysis extremely complex and challenging [6][7][8][9].
Magnetic field imaging and three-dimensional X-RAY technology are not affected by package material or form [10,11], and they are effective in realizing defect localization on a 3D package [12,13].
Magnetic field imaging (MFI) technology is a non-destructive, non-contact defect localization technology [14]. Different from thermal and optical technology, magnetic field 3D X-ray microscopy with a two-stage magnification structure were used to achieve submicron resolution too. This paper is organized as follows: Section 2 describes the test sample. Section 3 demonstrates defect localization on the TSV sample after thermal cycling test by magnetic field imaging and high-resolution three-dimensional X-ray technology. Section 4 verifies the open location by microscopic physical analysis. Finally, a conclusion is wrapped up in Section 5.

Test Sample
The test sample in this paper contains thirteen groups of TSV daisy chains. Each via is 100 µm deep and 10 µm in diameter, which is filled with copper and with a Ti barrier layer 50 nm thick and a SiO2 dielectric layer 250 nm thick. The TSV connects two metal levels of damascene redistribution copper at the top with one metal level of damascene redistribution copper at the bottom. The vertical structure schematic of each TSV daisy chain unit is as shown in Figure 1. In application, the current is input from the pad of the upper surface metal layer (TOP_M2), travels through the Top_M1 layer, then passes the holes of TSV, down to the lower surface, through the lower surface BTM_M1 layer, then passes the holes of TSV back to the upper surface; it repeats many times, and is finally output from the upper surface. After running a thermal cycle experiment from −55 °C to 125 °C on the sample for 1000 cycles, the DC resistance of three groups of TSV daisy chains was increasing significantly. The changes in DC resistance for thirteen groups of TSV daisy chains are shown in Table 1. Channel 6, with the largest resistance change, was selected for analysis.  After running a thermal cycle experiment from −55 • C to 125 • C on the sample for 1000 cycles, the DC resistance of three groups of TSV daisy chains was increasing significantly. The changes in DC resistance for thirteen groups of TSV daisy chains are shown in Table 1. Channel 6, with the largest resistance change, was selected for analysis.

Magnetic Field Imaging
To locate the defect, both fail and reference samples were scanned with the SQUID sensor for an overview scan first and then the GMR sensor for a zoomed-in scan of the die area by Neocera Magma EFI HiRes, as shown in Figures 2 and 3. AC and RF scans were acquired for the reference sample. The AC and RF scans of the reference sample helped us understand the way the current is supposed to propagate in the sample when there is no failure. Only RF scans were acquired for the failing sample. We used a two-probe RF technique for the failing sample to try to isolate the failure channel from the whole grid.

Magnetic Field Imaging
To locate the defect, both fail and reference samples were scanned with the SQUID sensor for an overview scan first and then the GMR sensor for a zoomed-in scan of the die area by Neocera Magma EFI HiRes, as shown in Figures 2 and 3. AC and RF scans were acquired for the reference sample. The AC and RF scans of the reference sample helped us understand the way the current is supposed to propagate in the sample when there is no failure. Only RF scans were acquired for the failing sample. We used a two-probe RF technique for the failing sample to try to isolate the failure channel from the whole grid.  For the AC scan in channel 6, the current input from the pad of upper surface J12 travels through the TOP_M2 layer to the TOP_M1 layer, then passes the TSV down to the lower surface, travels through the BTM_M1 layer, then passes the TSV back to the upper surface, it repeated many times, and finally output from the pad of the upper surface J3. The current image is shown in Figure 4.
The main circuit path, as shown in the AC scan, shows a weaker signal. The relative intensity of the signal in the fail is lower compared to the reference. The fact that we see a weaker signal in the fail sample points to a possible failure in that path.
The RF signal followed the whole circuit rather than the single line connecting J12 to J3 in the AC case. For some reason, the RF signal propagated more strongly in the top part of the circuit for both the reference and the fail. This complicated the analysis. The RF scan image from pin J12 on reference and fail sample is shown in Figure 5. The same behavior can be observed from pin J3.
For this reason, a two probe RF technique on pins J3 and J12 was performed. This can help the RF signal to propagate through the path of interest.

Magnetic Field Imaging
To locate the defect, both fail and reference samples were scanned with the SQUID sensor for an overview scan first and then the GMR sensor for a zoomed-in scan of the die area by Neocera Magma EFI HiRes, as shown in Figures 2 and 3. AC and RF scans were acquired for the reference sample. The AC and RF scans of the reference sample helped us understand the way the current is supposed to propagate in the sample when there is no failure. Only RF scans were acquired for the failing sample. We used a two-probe RF technique for the failing sample to try to isolate the failure channel from the whole grid.  For the AC scan in channel 6, the current input from the pad of upper surface J12 travels through the TOP_M2 layer to the TOP_M1 layer, then passes the TSV down to the lower surface, travels through the BTM_M1 layer, then passes the TSV back to the upper surface, it repeated many times, and finally output from the pad of the upper surface J3. The current image is shown in Figure 4.
The main circuit path, as shown in the AC scan, shows a weaker signal. The relative intensity of the signal in the fail is lower compared to the reference. The fact that we see a weaker signal in the fail sample points to a possible failure in that path.
The RF signal followed the whole circuit rather than the single line connecting J12 to J3 in the AC case. For some reason, the RF signal propagated more strongly in the top part of the circuit for both the reference and the fail. This complicated the analysis. The RF scan image from pin J12 on reference and fail sample is shown in Figure 5. The same behavior can be observed from pin J3.
For this reason, a two probe RF technique on pins J3 and J12 was performed. This can help the RF signal to propagate through the path of interest. For the AC scan in channel 6, the current input from the pad of upper surface J12 travels through the TOP_M2 layer to the TOP_M1 layer, then passes the TSV down to the lower surface, travels through the BTM_M1 layer, then passes the TSV back to the upper surface, it repeated many times, and finally output from the pad of the upper surface J3. The current image is shown in Figure 4.  The main circuit path, as shown in the AC scan, shows a weaker signal. The relative intensity of the signal in the fail is lower compared to the reference. The fact that we see a weaker signal in the fail sample points to a possible failure in that path.
The RF signal followed the whole circuit rather than the single line connecting J12 to J3 in the AC case. For some reason, the RF signal propagated more strongly in the top part of the circuit for both the reference and the fail. This complicated the analysis. The RF scan image from pin J12 on reference and fail sample is shown in Figure 5. The same behavior can be observed from pin J3. Two RF probes were connected to pins J3 and J12 (Channel 6) at the same time. Some unusual behavior in the frequency sweep for the failing sample was noticed in comparison with the frequency sweep for the reference sample, as shown in Figure 6. The reference sample sweep has only one voltage dip, where the RF signal drops significantly. The failed sample sweep shows two dips at two different frequencies. The typical behavior is to only have a single dip. Scans are performed at different frequencies. The reference sample current density data were taken at 43.0 MHz, with one probe on the left side and one probe on the right side. As shown in Figure 7, the signal shows no significant decay along with the green cursor at all three modes. For this reason, a two probe RF technique on pins J3 and J12 was performed. This can help the RF signal to propagate through the path of interest.
Two RF probes were connected to pins J3 and J12 (Channel 6) at the same time. Some unusual behavior in the frequency sweep for the failing sample was noticed in comparison with the frequency sweep for the reference sample, as shown in Figure 6. The reference sample sweep has only one voltage dip, where the RF signal drops significantly. The failed sample sweep shows two dips at two different frequencies. The typical behavior is to only have a single dip. Scans are performed at different frequencies.
(a) (b) Two RF probes were connected to pins J3 and J12 (Channel 6) at the same time. Some unusual behavior in the frequency sweep for the failing sample was noticed in comparison with the frequency sweep for the reference sample, as shown in Figure 6. The reference sample sweep has only one voltage dip, where the RF signal drops significantly. The failed sample sweep shows two dips at two different frequencies. The typical behavior is to only have a single dip. Scans are performed at different frequencies.
(a) (b) The reference sample current density data were taken at 43.0 MHz, with one probe on the left side and one probe on the right side. As shown in Figure 7, the signal shows no significant decay along with the green cursor at all three modes. The reference sample current density data were taken at 43.0 MHz, with one probe on the left side and one probe on the right side. As shown in Figure 7, the signal shows no significant decay along with the green cursor at all three modes. Failed sample current density data were taken with both probes at the same time with two different frequencies, as shown in Figure 8. The signal decays in the same general area for both datasets along the green cursor. The red arrows point to that area of signal decay.  Failed sample current density data were taken with both probes at the same time with two different frequencies, as shown in Figure 8. The signal decays in the same general area for both datasets along the green cursor. The red arrows point to that area of signal decay. Failed sample current density data were taken with both probes at the same time with two different frequencies, as shown in Figure 8. The signal decays in the same general area for both datasets along the green cursor. The red arrows point to that area of signal decay.  Figure 10). Given the proximity of the location to the via, the fifth via was viewed as the likely opening location.

Three-Dimensional X-ray Microscopy
To verify the failure location, the above 3D package with TSV daisy chain was  Figures 9 and 10. The linear decay shows us more precisely where the current signal disappears. The linear decay works best with the signals that propagate in one plane. This particular signal seems to changes planes at the same time as it is decaying. We may experience some overshooting from the actual open location. The possible location of the opening was on the right side of the sample in the proximity of the fifth via (see blue arrow in Figure 10). Given the proximity of the location to the via, the fifth via was viewed as the likely opening location. Failed sample current density data were taken with both probes at the same time with two different frequencies, as shown in Figure 8. The signal decays in the same general area for both datasets along the green cursor. The red arrows point to that area of signal decay.

Three-Dimensional X-ray Microscopy
To verify the failure location, the above 3D package with TSV daisy chain was scanned with a high-resolution 3D X-ray microscopy, Zeiss Xradia 510 Versa. The three- Failed sample current density data were taken with both probes at the same time with two different frequencies, as shown in Figure 8. The signal decays in the same general area for both datasets along the green cursor. The red arrows point to that area of signal decay.

Three-Dimensional X-ray Microscopy
To verify the failure location, the above 3D package with TSV daisy chain was scanned with a high-resolution 3D X-ray microscopy, Zeiss Xradia 510 Versa. The three-

Three-Dimensional X-ray Microscopy
To verify the failure location, the above 3D package with TSV daisy chain was scanned with a high-resolution 3D X-ray microscopy, Zeiss Xradia 510 Versa. The three-dimensional characterization completely displayed the internal structure of the sample, overcame the limitations of the two-dimensional characterization, and combined the high-resolution objective lens to capture the details of the internal structure of the sample to the maximum extent so that the structural characteristics of the TSV array inside the sample could be seen. Two-dimensional X-ray projection image and 3D X-ray reconstruction image of the TSV sample under the overall scan are shown in Figure 11. As shown in this figure, due to the limitation of the resolution, the failure site cannot be observed under the overall scan.
Appl. Sci. 2021, 11, x FOR PEER REVIEW 7 of 11 dimensional characterization completely displayed the internal structure of the sample, overcame the limitations of the two-dimensional characterization, and combined the highresolution objective lens to capture the details of the internal structure of the sample to the maximum extent so that the structural characteristics of the TSV array inside the sample could be seen. Two-dimensional X-ray projection image and 3D X-ray reconstruction image of the TSV sample under the overall scan are shown in Figure 11. As shown in this figure, due to the limitation of the resolution, the failure site cannot be observed under the overall scan.- In order to achieve sub-micron resolution, according to the electrical test results, local scanning should be conducted with channel 6 as the center. Due to the long channel, multiple partition scanning is required, followed by image reconstruction and observation, which requires a longer test time. Under the test voltage of 150 kV and the test current of 15 µA, the three-dimensional stereogram, XY section, YZ section and XZ section morphology of the sample in the local magnification scanning are shown in Figure 12. It can be seen that TSV copper columns are arranged neatly and uniformly in thickness, with no voids or cracks. Defects were observed in the Top_M1 layer connecting the two copper pillars. The fracture morphology of the metal layer observed in the XY section is shown in Figure 13. The defect location identified by magnetic field imaging is consistent with three-dimensional X-ray microscopy. In order to achieve sub-micron resolution, according to the electrical test results, local scanning should be conducted with channel 6 as the center. Due to the long channel, multiple partition scanning is required, followed by image reconstruction and observation, which requires a longer test time. Under the test voltage of 150 kV and the test current of 15 µA, the three-dimensional stereogram, XY section, YZ section and XZ section morphology of the sample in the local magnification scanning are shown in Figure 12. It can be seen that TSV copper columns are arranged neatly and uniformly in thickness, with no voids or cracks. Defects were observed in the Top_M1 layer connecting the two copper pillars. The fracture morphology of the metal layer observed in the XY section is shown in Figure 13. The defect location identified by magnetic field imaging is consistent with three-dimensional X-ray microscopy.
Appl. Sci. 2021, 11, x FOR PEER REVIEW 7 of 11 dimensional characterization completely displayed the internal structure of the sample, overcame the limitations of the two-dimensional characterization, and combined the highresolution objective lens to capture the details of the internal structure of the sample to the maximum extent so that the structural characteristics of the TSV array inside the sample could be seen. Two-dimensional X-ray projection image and 3D X-ray reconstruction image of the TSV sample under the overall scan are shown in Figure 11. As shown in this In order to achieve sub-micron resolution, according to the electrical test results, local scanning should be conducted with channel 6 as the center. Due to the long channel, multiple partition scanning is required, followed by image reconstruction and observation, which requires a longer test time. Under the test voltage of 150 kV and the test current of 15 µA, the three-dimensional stereogram, XY section, YZ section and XZ section morphology of the sample in the local magnification scanning are shown in Figure 12. It can be seen that TSV copper columns are arranged neatly and uniformly in thickness, with no voids or cracks. Defects were observed in the Top_M1 layer connecting the two copper pillars. The fracture morphology of the metal layer observed in the XY section is shown in Figure 13. The defect location identified by magnetic field imaging is consistent with three-dimensional X-ray microscopy.  Figure 14 is the 3D stained image rendered by 3D Viewer software, and the defect structure such as cracks in the sample can be more intuitively observed.

Physical Analysis
To further verify the failure location, decapsulation was performed on the sample, and the surface passivation layer was removed by a reactive ion etcher. No obvious abnormality was observed by metallographic microscopy and scanning electron microscope (SEM) at the position of the open defect localized by magnetic field imaging and 3D XRAY microscopy, as shown in Figures 15 and 16.    Figure 14 is the 3D stained image rendered by 3D Viewer software, and the defect structure such as cracks in the sample can be more intuitively observed.

Physical Analysis
To further verify the failure location, decapsulation was performed on the sample, and the surface passivation layer was removed by a reactive ion etcher. No obvious abnormality was observed by metallographic microscopy and scanning electron microscope (SEM) at the position of the open defect localized by magnetic field imaging and 3D XRAY microscopy, as shown in Figures 15 and 16.

Physical Analysis
To further verify the failure location, decapsulation was performed on the sample, and the surface passivation layer was removed by a reactive ion etcher. No obvious abnormality was observed by metallographic microscopy and scanning electron microscope (SEM) at the position of the open defect localized by magnetic field imaging and 3D XRAY microscopy, as shown in Figures 15 and 16.  Figure 14 is the 3D stained image rendered by 3D Viewer software, and the defect structure such as cracks in the sample can be more intuitively observed.

Physical Analysis
To further verify the failure location, decapsulation was performed on the sample, and the surface passivation layer was removed by a reactive ion etcher. No obvious abnormality was observed by metallographic microscopy and scanning electron microscope (SEM) at the position of the open defect localized by magnetic field imaging and 3D XRAY microscopy, as shown in Figures 15 and 16.  After the surface of the sample was brushed with acid and observed under scanning electron microscopy, the crack defect could be observed at the positions where it is open localized by magnetic field imaging and 3D XRAY microscopy, as shown in Figure 17.

Conclusions
State-of-the-art non-destructive defect localization technology, magnetic field imaging and high-resolution three-dimensional X-RAY microscopy were utilized to localize the open defect in a 3D package with a TSV daisy chain in this paper. Open defects were intentionally induced by running a thermal cycle experiment from −55 °C to 125 °C for 1000 cycles. For magnetic field imaging, in order to locate a possible open failure location, it is very important to identify the circuit involved. The AC and RF scans of the reference sample helped us to understand the way that the current is supposed to propagate in the sample when there is no failure. RF signals may show intensity attenuation as they propagate along the circuit and couple to other circuits that are not part of the circuit under study. In this particular case, there were a lot of different branches that proved very difficult to isolate and complicated identification of the circuit. Therefore, a two-probe RF method in magnetic field imaging was performed to resolve isolation of the defect difficulties resulting from many different branches of TSV daisy chains. Moreover, a linear decay method was used to target sub-micron resolution at a long working distance. For three-dimensional X-RAY microscopy, to achieve sub-micron resolution, a high-resolution 3D X-ray microscopy with a two-stage magnification structure was used for scanning. In order to further improve the resolution, multiple partition scans are required, followed by image reconstruction and observation, which requires a long test time. The defect location identified by magnetic field imaging was consistent with that identified by threedimensional X-ray microscopy. Additionally, the physical failure analysis revealed the presence of a crack in the Top_M1. From the results presented in this paper, it can be concluded that by specific methods, both magnetic field imaging and high-resolution three-dimensional X-RAY imaging can achieve sub-micron resolution in defect location at After the surface of the sample was brushed with acid and observed under scanning electron microscopy, the crack defect could be observed at the positions where it is open localized by magnetic field imaging and 3D XRAY microscopy, as shown in Figure 17.

Conclusions
State-of-the-art non-destructive defect localization technology, magnetic field imaging and high-resolution three-dimensional X-RAY microscopy were utilized to localize the open defect in a 3D package with a TSV daisy chain in this paper. Open defects were intentionally induced by running a thermal cycle experiment from −55 °C to 125 °C for 1000 cycles. For magnetic field imaging, in order to locate a possible open failure location, it is very important to identify the circuit involved. The AC and RF scans of the reference sample helped us to understand the way that the current is supposed to propagate in the sample when there is no failure. RF signals may show intensity attenuation as they propagate along the circuit and couple to other circuits that are not part of the circuit under study. In this particular case, there were a lot of different branches that proved very difficult to isolate and complicated identification of the circuit. Therefore, a two-probe RF method in magnetic field imaging was performed to resolve isolation of the defect difficulties resulting from many different branches of TSV daisy chains. Moreover, a linear decay method was used to target sub-micron resolution at a long working distance. For three-dimensional X-RAY microscopy, to achieve sub-micron resolution, a high-resolution 3D X-ray microscopy with a two-stage magnification structure was used for scanning. In order to further improve the resolution, multiple partition scans are required, followed by image reconstruction and observation, which requires a long test time. The defect location identified by magnetic field imaging was consistent with that identified by threedimensional X-ray microscopy. Additionally, the physical failure analysis revealed the presence of a crack in the Top_M1. From the results presented in this paper, it can be concluded that by specific methods, both magnetic field imaging and high-resolution three-dimensional X-RAY imaging can achieve sub-micron resolution in defect location at

Conclusions
State-of-the-art non-destructive defect localization technology, magnetic field imaging and high-resolution three-dimensional X-RAY microscopy were utilized to localize the open defect in a 3D package with a TSV daisy chain in this paper. Open defects were intentionally induced by running a thermal cycle experiment from −55 • C to 125 • C for 1000 cycles. For magnetic field imaging, in order to locate a possible open failure location, it is very important to identify the circuit involved. The AC and RF scans of the reference sample helped us to understand the way that the current is supposed to propagate in the sample when there is no failure. RF signals may show intensity attenuation as they propagate along the circuit and couple to other circuits that are not part of the circuit under study. In this particular case, there were a lot of different branches that proved very difficult to isolate and complicated identification of the circuit. Therefore, a two-probe RF method in magnetic field imaging was performed to resolve isolation of the defect difficulties resulting from many different branches of TSV daisy chains. Moreover, a linear decay method was used to target sub-micron resolution at a long working distance. For three-dimensional X-RAY microscopy, to achieve sub-micron resolution, a high-resolution 3D X-ray microscopy with a two-stage magnification structure was used for scanning. In order to further improve the resolution, multiple partition scans are required, followed by image reconstruction and observation, which requires a long test time. The defect location identified by magnetic field imaging was consistent with that identified by three-dimensional X-ray microscopy. Additionally, the physical failure analysis revealed the presence of a crack in the Top_M1. From the results presented in this paper, it can be concluded that by specific methods, both magnetic field imaging and high-resolution three-dimensional X-RAY imaging can achieve sub-micron resolution in defect location at long working distances without decapsulation or cutting samples. Both technologies are promising for future 3D-architecture-related quality assessment and failure analysis. Data Availability Statement: The data supporting this study can be accessed by contacting with the first author via: chenyuan@ceprei.com.