Lock-in thermal IR imaging using a solid immersion lens

Transcription

1 Microelectronics Reliability 46 (2006) Lock-in thermal IR imaging using a solid immersion lens O. Breitenstein a *, F. Altmann b, T. Riediger b, D. Karg c, V. Gottschalk d a Max Planck Institute of Microstructure Physics, Weinberg 2, D Halle, Germany b Fraunhofer Institute for Mechanics of Materials, Heideallee 19, D Halle, Germany c Thermosensorik GmbH, Am Weichselgarten 7, D Erlangen, Germany d ELMOS Semiconductor AG, Heinrich-Hertz-Str. 1, D-44227, Dortmund, Germany Abstract A hemispherical silicon solid immersion lens (SIL) was used to improve the spatial resolution of front-side thermal IR imaging in lock-in mode. The bottom of the SIL was cone-shaped to reduce the footprint of the SIL to the size of the imaged region. Caused by the lock-in operation mode, the detection limit improves by 2-3 orders of magnitude, and scattered light does not limit the image contrast. By using this SIL in combination with an IR camera working in the 3-5 µm wavelength range, a spatial resolution of 1.4 µm was obtained for thermal IR imaging. An automatic SIL positioning facility was constructed to place the SIL exactly in the center of the imaged region and to easily remove it after the detailed investigation. 1. Introduction Solid Immersion Lenses (SILs) made from silicon material are increasingly used to improve the spatial resolution of optical failure analysis techniques like light emission or laser microscopy [1]. In the infrared (IR) spectral range above 1.1 µm wavelength, where silicon is transparent, silicon is an optimum material for SILs because of its high refraction index of n If the object under investigation is "immersed" by the material of the immersion lens, the wavelength of the light is smaller in this region by a factor of n, which is the final reason for the improvement of the diffractionlimited spatial resolution by this factor. It is shown in this contribution that a SIL can also be used to improve the spatial resolution of infrared (IR) thermography, which is most effective in the lock-in mode SIL as an add-on In most previous SIL applications, the SIL was an integral part of the microscope objective. Here we are using the alternative approach to use the SIL as an add-on to an existing conventional microscope objective [2]. If a SIL should be used in this way, it acts like an additional magnifying glass, producing an additional optical magnification factor M. If the IR microscope objective was already designed to give a resolution close to the diffraction limit, the magnification factor M of the SIL has to be at least as large as its refraction index n, in order to fully exploit the theoretical resolution improvement capability of the SIL. A hemispherically shaped SIL, having a height H equal to its radius R, just shows a magnification factor of M = n. In addition, such a hemispherical SIL has a focal plane just in the object plane, which makes it * Corresponding author: breiten@mpi-halle.mpg.de Tel: ; Fax:

2 easier to change between working with and without SIL. However, such a SIL is not optimum with respect to optical aberrations. A so-called "Super-SIL" or "Hyper-SIL" is characterized by a larger height of H = R + R/n, and it shows no spherical aberration and minimum coma or astigmatism [1]. A Hyper-SIL made from silicon shows an optical magnification factor of about M = 8. For such a lens the focal plane is lying well below the object plane. 2. Experimental 2.1. Lock-in thermography The first attempt to image local heat sources in ICs by IR-based lock-in thermography (LIT) was published already in 1996 [3]. However, the spatial and the thermal resolution of this system were not sufficient yet for general use. After the introduction of high-resolution lock-in IR-thermography and lock-in FMI to IC failure analysis [4], several other approaches to use the lock-in principle in failure analysis have appeared. So the socalled "Stabilized" thermography techniques by FA Instruments [5] as well as the Lock-in OBIRCH technique by Hamamatsu [6] are basically single-phase LIT approaches. Note that OBIRCH and thermography are investigating the same type of defects: In both cases only a resistance-limited current can be detected [2]. Lock-in thermography (LIT) means that the power dissipated by the faults is amplitude-modulated, e.g. by modulating the supply voltage applied to the device, and that the thermal signal is processed and averaged on-line to yield an image of the periodic temperature modulation. Advantages of LIT compared to steadystate thermography are an improvement of the detection limit by 2-3 orders of magnitude, a suppression of the lateral heat diffusion, which is responsible for thermal blurring, and (only for twophase LIT) the availability of display modes like the phase image, which show inherent IR emissivity correction [4, 7]. Two-phase LIT images can be displayed in different ways, as the T-modulation amplitude image, which is the standard representation, the in-phase (0 ) image, the phase image, and the 0 /-90 image [7, 8]. The 0 image shows the best possible spatial resolution, but still contains the emissivity contrast. The phase image is inherently emissivity-corrected and shows local heat sources of different power in a comparable brightness, hence it allows best to see weak heat sources besides strong ones. However, its spatial resolution is worse than that of the 0 image. The 0 /-90 image combines the advantages of the 0 and the phase image [8]. Usually, one of these images is superimposed to the topography image to show the exact location of a fault. All investigations shown here have been made by using the TDL 384 M 'Lock-in' thermography system by Thermosensorik Erlangen, Germany [9], which is equipped with a 384x288 pixel mercury-cadmiumtelluride focal plane array detector with a pixel pitch of 24 µm working in a spectral range of 3-5 µm. The use of a 5x microscope objective without SIL leads to a pixel resolution of 4.8 µm. Due to the high numerical aperture of the objective of 0.7, the diffraction-limit of the 5x microscope objective is below 4.8 µm for the spectral range 3-5µm. Therefore, the spatial resolution is limited by the pixel resolution. Alternatively, for making overview images of the whole chip, a standard 28 mm focal length objective was used, the magnification factor of which could be increased by using additional distance rings The optimum SIL shape We have tried both hemispherical and Hyper-SILs with a radius of R = 1 mm and 3 mm, respectively. Before a LIT image can be made, the operator always has to orientate oneself on the surface of the chip. Since in this phase the sample is essentially at constant temperature, this orientation is performed basically according to the emissivity contrast of the uppermost metallization layer(s) of the chip. Note that for microscopic investigations the sample is very close to the IR camera, which is internally cooled. Metallized layers show a very weak IR emissivity, they essentially reflect the IR light coming from the surrounding, which is very cold in this case. Hence, there is very little amount of IR light reflected in the region of metallization layers. Therefore, in IR microscopy, metallized layers appear dark compared to bare silicon regions because of this emissivity contrast. Until now, we did only succeed to get meaningful IR topography images by using a hemispherical SIL, but not by using a Hyper-SIL. We believe that the reason of this is the high magnification factor of the Hyper-SIL. As for any other microscope objective, the contrast of the image decreases with 1/M 2. Hence, for the Hyper-SIL, the image contrast is much weaker than for the hemispherical SIL. Therefore, for thermal IR imaging, a hemispherical SIL is obviously better appropriate than a Hyper-SIL, in spite of its higher aberrations. For front-side investigations, it is not useful to use a complete hemisphere as a SIL because of its large footprint. Note that the SIL has to touch the surface of the IC. Since there are usually bonding wires at the

3 edge of the die, for using a complete hemisphere, the investigated region cannot be closer than the SIL footprint radius to the bonding wires. This would exclude an essential part of the die from being investigated. In fact, only the central part of the circular base plane of the SIL is used for imaging. Therefore, we have ground our SILs cone-shaped according to Fig. 1 with the cone angle of +/- 45, being adapted to the light acceptance angle of our IR objective. Fig. 1: SIL shape used 2.3. Automatic SIL positioning facility a which allows an easy change between working with and without SIL and always positions the SIL exactly in the middle of the image field. This facility is connected to the bottom plane of our 5x microscope objective and allows to retract the SIL and to position it outside of the light cone of the objective. It reduces the working distance from 10 to 6 mm. Note that the SIL is only freely hanging in this facility. When it is set down to the die surface, it is freely standing there at the surface of the die, so that the z-movement of the objective, which is necessary for focusing, is not leading to any harmful touching of the surface. Fig. 2 shows the lower part of the microscope objective with the SIL positioned outside (a) and in the light path (b). Changing between these two states takes less than 10 s and leaves the image position unchanged. 3. Results The following examples came from failure analysis work on ASICs for automotive application. For some defective ASICs, the failure sites could not be located by emission of visible light or other imaging techniques like liquid crystal microscopy and OBIRCH. These ASICS were manufactured in a 0.8 µm-high-voltage- CMOS technology with 3 metal layers. The defective devices are showing significantly increased current consumption values. The identification of local heat sources, combined with other information from testing and probing, allowed to find the origin of a fault High resolution LIT using a SIL with R = 1 mm b Fig. 2: SIL positioning facility, with the SIL (see arrows) retracted (a) and in working position (b) Our first experiments were made by manually positioning the SIL on top of the die by using a tweezers [2]. However, it has turned out that in this way it is very hard to place the SIL in the correct position just in the middle of the image field. After the SIL was removed, it was nearly impossible to meet the same position once again. Moreover, there was the danger to destroy the bond wires accidentally. Therefore we have constructed an automatic SIL positioning facility, We have used two SILs, both shaped according to Fig. 1, a small one having a radius of R = 1 mm and a larger one having a radius of R = 3 mm. Without SIL, our objective allows to image a region of 1.84x1.38 mm 2 (384x288 pixels, 4.8 µm pixel resolution). By using the SIL, the pixel resolution improves to 4.8µm/3.42 = 1.4 µm, hence the image region reduces to 538x403 µm 2. Fig. 3 shows a topography image (a) of a region in the logic field of a µ-controller imaged through the smaller SIL with R = 1 mm. In all cases, a usual optical immersion oil was used to fill the gap between the surface of the die and the SIL. We see that in this image the uppermost metallization layer becomes visible in the center of the image. However, already at a certain distance to the center, the image becomes increasingly bright, and also some image distortions appear. The bright contrast towards the edge is due to the reflection

4 of thermal radiation from the outside into the camera via the upper surface of the SIL, which becomes increasingly efficient with increasing distance to the center [5]. Only in the middle of the SIL, where the surface is horizontally oriented, the reflected light stems mostly from the cold interior of the IR camera, hence it has a low intensity. The dark regions in the corners of Fig. 3 are showing the original surface of the die outside of the SIL. 50 µm Fig. 4: 128x128 pixel section from Fig. 3 (topography) 100 µm Fig. 3: Full frame (384x288 pixel) topography image of a region imaged through a SIL with R = 1 mm Because of this strong brightening of the topography image towards the edge, and because of the increasing image distortions towards the edge, only the innermost part of the image can be used for a SIL with R = 1 mm. Indeed, if only the innermost 128x128 pixel areas of the originally 384x384 pixel sited images are displayed (see Figs. 4 and 5), the spatial resolution obtained is impressive. Only in the edge region of the topography image (Fig. 4) some distortions become visible. The edge definition of the metal lines, hence the real optical resolution of the image, is indeed in the order of 1 pixel, having a size of 1.4 µm here. In the LIT amplitude image in Fig. 5, three different heat sources can be distinguished. The pattern of metallization lines could be compared with layout data for the identification of the cell, which generates the thermal radiation. Since this amplitude image still contains the emissivity contrast of the metallization lines, the positions of the heat sources can be revealed. For a clear identification of the location in the cell circuit a more closer view is needed, e.g. by probing. With these information a further qualification can be made and the physical root cause can be worked out by probing. Fig. 5: Lock-in thermography (T-amplitude) image of the region of Fig. 4, imaged through a SIL with R =1 mm 3.2. Results using a SIL with R = 3 mm 50 µm Since the SIL with R = 1 mm obviously does not allow to use the whole image area of the camera, a three times larger SIL with R = 3 mm was fabricated according to Fig. 1. Only this larger SIL is able to be used together with the automatic SIL positioning facility shown in Fig. 2. Fig. 6 shows an overview image of another µ- Controller (topography + phase image) taken by using the standard 28 mm objective together with a lens extender (distance) ring. Here the phase image was

5 used for showing both weak and strong heat sources with a similar brightness. The image shows different heat sources, which are both due to the regular operation of the circuit and due to faults. The arrows are pointing to two fault positions, which were investigated in detail. Fig. 7 (0 + topography image) shows both defects by using the microscope objective without the SIL. We see that the two defects (arrows) show a very different power, leading to a different brightness in the 0 image. The exact location of these faults relative to the layout of the metallization cannot be concluded from this image yet. could be defined in the layout (see Fig. 10). This will be the presupposition for demonstrating the physical root cause. 1 mm 100 µm Fig. 8: Microscopic image with 3 mm SIL (topography + 0 ) of the left defect in Figs. 6 and 7 Fig. 6: Overview image (phase + topography) 100 µm 500 µm Fig. 7: Microscopic image without SIL (topography + 0 ) of the defect region in Fig. 6 (arrows pointing to faults) The gain in resolution which could be reached with this 3 mm SIL is illustrated in the next images. Only by using the SIL with R = 3 mm (Figs. 8 and 9, both 0 + topography images) the exact locations of the faults relative to the metallization lines can be revealed. Note that the usable image field for the R = 3 mm SIL indeed fills out nearly the whole full frame field of view of the camera. In this analysis example the verification of test results could be done by identifying the source cell of the high intensity hot spot in Figs The source of thermal radiation in the cell circuit could be identified in one transistor and the measuring points for probing Fig. 9: Microscopic image with 3 mm SIL (topography + 0 ) of the right defect in Figs. 6 and 7 4. Conclusions and outlook We have shown that IR-based 2-phase lock-in thermography (LIT) performed by using a silicon solid immersion lens (SIL) greatly expands the application field of thermal IR microscopy in failure analysis. The use of LIT alone leads to an improvement of the detection limit compared to conventional (steady-state) IR microscopy by 2-3 orders of magnitude [4, 7]. The variability of the display options of 2-phase LIT provides a high flexibility to choose e.g. between high resolution imaging (0 image) or imaging of heat sources of different magnitude with inherent emissivity correction (phase image). Without using a SIL, for submicron technologies even a high NA microscope

6 objective having a pixel resolution of 5 µm, which is close to the diffraction limit, does not allow to exactly localize the heat sources relative to the layout structure. The limitation is not the LIT image itself, which may be blurred anyway to a certain degree e.g. due a certain depth position of the heat source, but the orientation on the surface in the topography image, which is the raw IR image of the camera. By using a SIL, the resolution of this image can be improved to about 1.4 µm, which is sufficient to see the metallization lines even for sub-µm technologies, as shown here for a 0.8 µm-hv-cmos technology with 3 Al-metallization layers. The resolution, which can be reached by this setup, is high enough to give the needed information for successful localisation of failure sites and demonstration of physical root causes. All facilities used for this contribution are commercially available [9]. Fig. 10: Comparing the images from Figs. 6 9 with the layout lead to defect location in a cell, identified from test, and allowed the localization of a transistor in this cell dissipating the heat. It should also be possible to use this technique for backside inspection, if the substrate is not too highly doped. Backside LIT investigations without SIL have already been demonstrated [4]. For backside inspection, the height of the SIL has to be reduced by the residual thickness of the wafer. Fortunately, for thermal IR investigations, the demands on the planarity of the opened backside of the die are not as high as for light emission microscopy because of the considerably larger wavelength used here. The efficiency of our SILs will be further increased by covering them with an anti-reflection (AR) coating. Also the spatial resolution of the SIL imaging can further be improved, if an IR camera for a lower wavelength range (e.g. 2-3 µm) together with a higher magnifying objective is used. The considerably lower thermal photon yield in this spectral range can at least partly be compensated by investigating the samples at an elevated temperature of, say, 100 C. Following this way, even sub-micron spatial resolution should be attainable for IR-based LIT in future. Acknowledgements The authors are grateful to J. McDonald (QFI, San Diego) for some discussions about SIL application. References [1] J. McDonald: "Optical and infrared FA microscopy", Tutorial at ISTFA 2005, see: and_infrared_fa_microscopy.pdf [2] O. Breitenstein, F. Altmann, T. Riediger, D. Karg: Lockin Infrared Microscopy with 1.4 µm Resolution Using a Solid Immersion Lens, Electronic Device Failure Analysis 8 (2006) 4-13 [3] D. Wu, J. Rantala, W. Karpen, G. Zenzinger, B. Schönbach, W. Rippel, R. Stegmüller, L. Diener, G. Busse: "Application of Lock-in Thermography Methods", Review of Progress in Quantitative Nondestructive Evaluation, ed. D.O. Thompson, D.E. Chimenti, Plenum Press, New York, 1996, Vol. 15, pp [4] O. Breitenstein, J.P. Rakotoniaina, F. Altmann, J. Schulz, G. Linse: "Fault localization and functional testing of ICs by lock-in thermography", Int. Symp. Test & Failure Analysis (ISTFA), Nov. 2002, pp [5] J.B. Colvin: Moiré Stabilized Thermal Imaging, Proc. 12th Int. Symposium on the Physical and Failure Analysis of Integrated Circuits (IPFA), 2006, pp [6] M. de la Bardonnie, R. Ross, K. Ly, F. Lorut, M. Lamy, C. Wyon, L.F.Tz. Kwakman, Y. Hiruma, J. Roux: The Effectiveness of OBIRCH Based Fault Isolation for Sub- 90nm CMOS Technologies, Proc. 31th Int. Symposium for Testing and Failure Analysis (ISTFA), 2005, pp [7] O. Breitenstein, M. Langenkamp: Lock-in Thermography - Basics and Use for Functional Diagnostics of Electronic Components, Springer, Heidelberg / New York, 2003, ISBN [8] O. Breitenstein, J.P. Rakotoniaina, M.H. Al Rifai, M. Gradhand, F. Altmann, T. Riediger, New Developments in IR Lock-in IR Thermography, Proc. 30th Int. Symposium for Testing and Failure Analysis (ISTFA), Nov. 2004, pp [9]