Extraction of Thermal Packaging Stress via Intrinsic Silicon on Glass Stress Measurement

EXTRACTION OF THERMAL PACKAGING STRESS VIA INTRINSIC SILICON ON GLASS STRESS MEASUREMENT Ahmet Arif Aslan1, and Erdinc Tatar1, 2 1Department of Electrical and Electronics Engineering, Bilkent University, Ankara, TURKEY and 2National Nanotechnology Research Center (UNAM), Bilkent University, Ankara, TURKEY ABSTRACT Prediction and measurement of MEMS packaging stress are challenging due to the complex material stack. We decompose the thermal packaging stress of Silicon-onGlass (SoG) sensors by referencing regularly mounted dies to a no die-attach sensor in this work, for the first time. We employ a 3x3 capacitive strain gauge array and measure the intrinsic stress of an SoG sensor over a 60°C temperature range. We show that the intrinsic SoG normal strain is tensile (~0.2-0.25µStrain/°C) since Si CTE is lower than glass. The die-attaches (silver-filled epoxy and stiff AuSn solder) increase the shear and normal strain up to 10X and 2X, respectively, and cause strain nonuniformity and hysteresis over temperature. KEYWORDS Stress sensing, Silicon-on-Glass (SoG) process, Thermal stress, MEMS die-attach. INTRODUCTION MEMS device packaging includes multiple layers, i.e., device, die-attach, package, solder, and PCB. Considerably different mechanical properties of the packaging stack induce mechanical stress on the MEMS device with temperature and aging [1]. Packaging plays a fundamental role on the long-term performance of the MEMS sensors. [2] showed that the long-term drift of a MEMS gyroscope can be calibrated with on-chip stress sensing even though obtaining the calibration coefficients is challenging. The effects of packaging stress on sensors are device dependent and are not straightforward to comprehend since stress is a multidimensional tensor. Die-attach, directly connecting the MEMS to the external world, has a major effect on packaging stress and hence the MEMS performance. [3] showed that the dieattach film stress dominates the MEMS packaging stress through optical profiler measurements. Epoxy die-attaches were characterized for various preparation methods in [4] and exhibit viscoelastic material properties with not welldefined mechanical properties. Electronics packaging stress was monitored with piezoresistive stress sensors from a survivability point of view in [5]. In the previous work, we characterized the packaging stress for siliconebased and silver-filled epoxies over temperature with a stress array [6]. We also demonstrated the efficient compensation of MEMS gyroscope bias against thermal packaging stress with on-chip stress sensors [7]. To better understand the packaging stress, Figure 1 shows our sensor cross-section consisting of different materials, typical in electronics packaging. Table 1 summarizes the mechanical material properties, coefficient of thermal expansion (CTE) and Young’s Modulus (E) [8]. The sensor materials glass and silicon (Si) have low CTE; however, the surrounding materials have notably different CTE and E. In addition, the mechanical properties of the die attach epoxies significantly vary over time and temperature [1, 4]. Packaging materials expand, contract and age differently resulting in undesired but inevitable stress on the mechanical sensor. So, we expect a bending with temperature leading to MEMS anchor displacements affecting the dynamics of the sensor [9]. Unlike previous work focusing on the final stress of the packaged die, we pursue a different approach and unveil the thermal packaging stress by comparing the inherent Silicon-onGlass (SoG) stress with regularly packaged device stress. Figure 1: Cross-section of our sensor. Table 1: Material properties of the packaging stack [8]. Material CTE (ppm) E (GPa) MEMS (<111> Si) 2.6 170 MEMS (glass) 3.2 70 Die - attach (Ag fill) α 1 =30, α 2 =150 E 1 =0.2, E 2 =0.005 Die - attach (AuSn) 16 68 Packg. ceramic 7 270 Solder (lead - free) 22 26.2 PCB (FR4) 17 24 CAPACITIVE STRESS SENSOR ARRAY AND THE TEST SETUP Figure 3 shows the bridge-type capacitive stress sensor (strain gauge) with its front-end which is the building block of the stress sensor array. The imbalanced bridge connections at the top and the bottom convert and amplify the strain induced displacements in orthogonal directions [10]. The mechanical gain (1/tan(α)=6.7) is set by the imbalance angle (α=8.5°). The bridge length (Lbridge=200µm) determines the strain to displacement gain. We employ capacitive stress sensing since it can be easily integrated with capacitive inertial sensors and does not consume DC power unlike piezoresistive stress sensors [5]. The capacitance variations (ΔC=C+ - C-, C+/- represent the sensor in Fig.3) due to substrate strain are read by applying differential clocks (Vmod+/-) and using a charge amplifier. Die attach Solder Glass Ceramic Package PCB Silicon with temp. 979-8-3315-1381-8/25/$31.00 ©2025 IEEE 1744 Transducers 2025 Orlando, FLORIDA 29 June - 3 July 2025 2025 23rd International Conference on Solid-State Sensors, Actuators and Microsystems (Transducers) | 979-8-3315-1381-8/25/$31.00 ©2025 IEEE | DOI: 10.1109/TRANSDUCERS61432.2025.11111343 Authorized licensed use limited to: ULAKBIM UASL - Bilkent University. Downloaded on November 05,2025 at 13:17:49 UTC from IEEE Xplore. Restrictions apply. Figure 2: SEM image of the stress sensor array consisting of 3x3 stress rosettes, 6 rosettes are shown. Each rosette measures normal x and y strain (ε x , ε y ), and shear strain (γ xy ). We applied a sinusoidal clock at a frequency of 10kHz, far away from the lowest mechanical resonant mode (>300kHz). The charge amplifier output voltage is 𝑉 =|𝑉|(𝐶− 𝐶)𝐶 ⁄, where |𝑉| represents the clock amplitude. V out is then demodulated for the stress information. Figure 3: Bridge-type capacitive stress sensor and the front-end. The imbalanced bridge connection converts and amplifies the strain in orthogonal directions. The individual strain gauges are arranged for a stress rosette. Each rosette measures x and y normal strain (ε x , ε y ) and shear strain (γ xy ) with two +/-45° (diagonal) oriented gauges. The diagonal orientation and the clock polarity of the gauges cancels the normal strain and enables pure shear strain measurement [6]. Figure 2 presents the SEM image of the stress sensor array. The array consists of 3x3 stress rosettes, 6 rosettes are shown in Figure 2. A total of 27 distributed individual stress components can be measured from each die. The sensors were fabricated using a wafer-level vacuum packaged Silicon-on-Glass (SoG) process [11]. Figure 4 illustrates the test PCB photo. MEMS die is wirebonded to a 44pin ceramic leadless chip carrier which is soldered to a daughter board. The daughter board houses an on-PCB heater and underneath proportional to absolute temperature (PTAT) sensor. This configuration allows heating of the MEMS chip independent of electronics. The strain gauges share the modulation clocks (V mod+/- ) and have individual charge amplifiers. The amplified outputs are multiplexed and finally demodulated by a digital lockin amplifier (Zurich Instruments, HF2LI). The lock-in amplifier, controlled by a python code, generates the multiplexer selection bits and saves the data. We cycled the temperature three times with a slow rate of 0.5°C/min (ΔT ~ 60 °C) and 30 min. soaks at maximum and minimum temperatures to ensure an isothermal package. There is also an Aluminum cap on the ceramic package (not shown) to improve the thermal stability. Figure 4: Photo of the test PCB, MEMS die is soldered to a daughter board housing an on-PCB heater and underneath temperature sensor. We cycled the temperature three times. EXPERIMENTAL RESULTS We assembled a special die without a die-attach to measure the intrinsic stress of the SoG process. We first drilled a hole at the center of the ceramic package and held the MEMS chip in-place with vacuum during wirebonding. 1745 Authorized licensed use limited to: ULAKBIM UASL - Bilkent University. Downloaded on November 05,2025 at 13:17:49 UTC from IEEE Xplore. Restrictions apply. We call this chip as “no die-attach die” and handled it carefully during the experiments. Figure 5 reports the intrinsic normal and shear strain measurements of the SoG process for the first time. We report strain (ε) as it is more universal than stress (σ=εE). We present the strain variation in our plots. All distributed stress components are plotted at the same figure to observe the uniformity. We note that the stress sensors respond only to the expansion/contraction mismatch between the sensor Si structure and its anchors. For example, if the CTEs of the MEMS device and surroundings are identical, then the strain gauges outputs do not change. Figure 5: Normal (X, Y) and shear (S) strain of the no dieattach sensor, representing the intrinsic strain of the SoG process. Tensile normal stress without hysteresis is measured since Si has the lowest CTE. The sensor was held with vacuum during wirebonding. Figure 6: Glass and Si CTE and their CTE difference over temperature [9]. The measured strain on the no die-attach experiment is due to the glass and Si CTE difference. Figure 6 presents the CTE of glass and Si and their CTE difference over temperature (20°C-100°C) [9]. While the glass CTE is stable around 3.15ppm/°C, Si CTE varies by 20% from 20°C-100°C. The glass CTE is larger than Si CTE, and we expect to see tensile stress in Si with increasing temperature. Our stress sensors generate positive output for tensile strain, and we observe tensile normal X and Y strain in the range of 12µε - 16 µε (0.2-0.25 µε/°C). The average glass-Si CTE difference (ΔCTE) is 0.4 ppm/°C, so the maximum expected strain is Δε=ΔT×ΔCTE = 60°C×0.4ppm/°C = 24 µε, if everything could expand freely. Since Si and glass are bonded, the actual strain is smaller and depends on the layer thicknesses. ΔCTE decreases over temperature leading to a lowering slope in the measured normal stress in Fig. 5. The measured shear strain is small compared to the normal strain and is in the 2µε - 6µε range in the CCW direction. The shear is an indicator of stress uniformity and cannot be predicted easily. The intrinsic strains have negligible hysteresis. We think the stress nonuniformity might be due to the Si cap holding the device only at the edges or any possible temperature gradients because of the non-ideal thermal contact of the sensor and package. Figure 7: Strain measurements for the 1 st cycle of the silver-filled epoxy. A large shear hysteresis (+60-80µε) exists which we attribute to the epoxy humidity absorption. Figure 8: Strain measurements for the 2 nd -3 rd cycle of the epoxy. Smaller hysteresis persists with 2X increase in normal stress and variability in the strain w.r.t. intrinsic. We can extract the die-attach stress with the intrinsic SoG stress knowledge. Figure 7 and Figure 8 present the strain measurements of a die mounted with conductive silver-filled epoxy for the 1 st and 2 nd – 3 rd cycles, respectively. The silver filled epoxy causes a large shear hysteresis (+60µε-80µε) in the first temperature cycle which we attribute to the humidity absorption [6]. The humidity evaporates in the first cycle, and we obtain results in Figure 8 for the 2 nd and 3 rd cycle. Strain measurements of the die soldered with gold-tin (AuSn) die attach are shown in Figure 9. Referenced to intrinsic strain, AuSn induces ~10X shear strain, and its hysteresis is smaller than the epoxy. The peak normal strain is ~2X of the intrinsic for both cases. The tensile strain increases with die attach since the die-attach connects the sensor to higher CTE materials, see Table 1. 1746 Authorized licensed use limited to: ULAKBIM UASL - Bilkent University. Downloaded on November 05,2025 at 13:17:49 UTC from IEEE Xplore. Restrictions apply. More significantly, comparing Fig. 5 with Figs. 7, 8, and 9, both die-attaches introduce strain nonuniformity and hysteresis, adversely affecting the sensor performance. Figure 10 presents the finite element simulation showing the total surface displacement of the package and the PCB for a standard material set as in Fig. 1 with a 60°C temperature rise. Figure 10 visualizes the expected strain over temperature. The bending profile is complex, and convex on the MEMS side as the sensor materials have the lowest CTE. The presented strains reflect individual regions on the MEMS surface. Figure 9: Strain measurements for the AuSn die-attach for three cycles. 10X increase in shear strain w.r.t. intrinsic and smaller hysteresis than the epoxy are observed. Figure 10: Finite element simulation showing the total surface displacement of the package and the PCB for a 60°C temperature rise. CONCLUSIONS Typical packaged MEMS sensor includes multiple materials with vastly different mechanical properties. It is commonly acknowledged that the MEMS packaging stress significantly affects the sensor performance. This work focuses on distinguishing the intrinsic device stress from the final packaging stress for Silicon-on-Glass (SoG) process. Stress measurements from a no-die attach die revealed the intrinsic SoG stress due to the Si-glass CTE difference. The intrinsic SoG normal stress is tensile in the range of 0.2µε-0.25µε/°C, consistent with the theoretical expectations since Si CTE is lower than the glass CTE. Comparing the intrinsic SoG stress with the epoxy and AuSn solder mounted devices show that the normal stress increases to 2X of the intrinsic. However, the shear stress increases up to 10X, a major concern for device performance. This study shows that stress nonuniformity and hysteresis occur due to die-attach. Excluding the 1 st cycle large hysteresis of the silver-filled epoxy, thermal packaging strain is <0.8µε/°C which is <0.14MPa/°C for E=170GPa (111 Silicon). ACKNOWLEDGEMENTS This work was supported by the European Union’s European Research Council (ERC) Starting Grant under the grant agreement 101116162 – 0-drift – ERC-2023STG. Views and opinions expressed here solely belong to the authors. The authors would like to thank Derin Erkan and Tolga Veske for the PCB design and setup automation. REFERENCES [1] X. Zhang, S.B. Park, R. Navarro, and M.W. Judy, “Accurate assessment of packaging stress effects on MEMS devices,” Proc. ITHERM2006, San Diego, CA, USA, May 30-June 02, 2006, pp. 1336-1342. [2] B.E. Uzunoglu, D. Erkan, and E. Tatar, “A ring gyroscope with on-chip capacitive stress compensation,” J. Microelectromech. Syst., vol. 31, no. 5, pp. 741-752, Oct. 2022. [3] W. Mayer et al., “Investigating the effects of stress on die deformation and cross-axis offset drift in modesplit MEMS gyroscopes,” IEEE Sensors Letters, vol. 8, no. 1, pp. 1-4, Jan. 2024. [4] A. Misrak et al., “Impact of die attach sample preparation on its measured mechanical properties for MEMS sensor applications,” J. Microelectronics and Electronic Packaging, 18(2021), pp. 21-28. [5] J.C. Suhling, and R.C. Jaeger, “Silicon piezoresistive stress sensors and their application in electronic packaging,” IEEE Sensors Journal, vol. 1, no. 1, pp. 14-30, June 2001. [6] T. Veske, D. Erkan, and E. Tatar, “Characterization of packaging stress with a capacitive stress sensor array,” Proc. IEEE MEMS2023, Munich, Germany, Jan. 1519, 2023, pp 909-912. [7] D. Erkan, and E. Tatar, “Improving the temperature stability of MEMS gyroscope bias with on-chip stress sensors,” Proc. IEEE INERTIAL2024, Hiroshima, Japan, March 25-28, 2024, pp. 1-4. [8] Physical Constants of IC Package Materials, intel.com, available online April 2025. [9] M. Hosseini-Pishrobat, and E. Tatar, “Modeling temperature effects in a MEMS ring gyroscope,” J. Microelectromech. Syst., vol. 34, no. 2, pp. 150-163, April 2025. [10] H.-W. Ma, S.-M. Yao, L.-Q. Wang, and Z. Zhong, “Analysis of the displacement amplification ratio of bridge-type flexure hinge,” Sensors and Actuators A, 132(2006), pp.730-736. [11] M.M. Torunbalci, S.E. Alper, and T. Akin, “Advanced MEMS process for wafer level hermetic encapsulation of MEMS devices using SOI cap wafers with vertical feedthroughs,” J. Microelectromech. Syst., vol. 24, no. 3, pp. 556-564, June 2015. CONTACT E. Tatar, tel: +90-312-2903193; [email protected] 1747 Authorized licensed use limited to: ULAKBIM UASL - Bilkent University. Downloaded on November 05,2025 at 13:17:49 UTC from IEEE Xplore. Restrictions apply.