Packaging and Interconnection of a Back-Lit CCD Sensor Device by Joshua V. Bennett B.S. Materials Science and Engineering, Massachusetts Institute of Technology 1996 SUBMITTED TO THE DEPARTMENT OF MATERIALS SCIENCE AND ENGINEERING IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY MASSACHUSETTS INSTITUTE OF TECHNOLOGY June, 1997 JUN 1 7 2003 ©Joshua V. Bennett LIBRARIES Signature of Author ____ yr · Department of Materials Science and Engineering May 9, 1997 Approved by " William Robbins Technical Supervisor, Draper Laboratory Certified by Yet-Ming Chag, Kyocera Professr of Ceramics Thesis Supervisor ( / , Accepted by Linn Hobbs, John F. Elliott Professor of Materials Chairman, Departmental Graduate Committee SCI'NCE Packaging and Interconnection of a Back-Lit CCD Sensor Device by Joshua V. Bennett Submitted to the Department of Materials Science and Engineering on May 9, 1997 in Partial Fulfillment of the Requirements for the Degree of Master of Science ABSTRACT: This paper describes the engineering and development of a process to electrically and mechanically bond a large area, backlit charge-coupled device (CCD) sensor to a metallized substrate, using a materials system compatible with a high vacuum tube configuration. The CCD sensor is used in light sensor and digital camera devices. Currently, CCD devices are mechanically bonded to substrates using a variety of polymeric materials. For maximum resolution, however, the device must operate at high vacuum. Polymers release monomer molecules when exposed to a vacuum, since the polymerization reaction is far from equilibrium. Polymers also release retained solvent and binder molecules into a vacuum. This release of gas pollutes the high vacuum environment and degrades the tube perfomance. The CCD chip used in this research is extremely large, compared with standard semiconductor devices. The CCD chip measures 0.4 x 0.5 inches. At this size scale, thermal expansion mismatch stresses between die and carrier become an important design consideration, particularly for a die thinned to 10 micrometers. The proposed process results in a mechanical bond that is compatible with the high vacuum requirement, and is applicable to substrates of four possible materials: silicon, aluminum nitride, Pyrex, and LZS (a zirconia/ silicon nitride composite). The process also allows for two possible electrical connection techniques. Thesis Supervisor: Yet-Ming Chiang Title: Kyocera Professor of Ceramics ACKNOWLEDGEMENT May 22, 1997 This thesis was prepared at The Charles Stark Draper Laboratory, Inc. Publication of this thesis does not constitute approval by Draper or the sponsoring agency of the findings or conclusions contained herein. It is published for the exchange and stimulation of ideas. I hereby assign my copyright of this thesis to The Charles Stark Draper Laboratory, Inc., Cambridge, Massachusetts. Permission is hereby granted by The Charles Stark Iaper Laboratory, Inc., to the Massachusetts Institute of Technology to reproduce any or all of this thesis. Table of Contents 1 Acknowledgements Page 4 2 Glossary of Terms 5 3 List of Figures 6 4 Introduction 7 5 Literature Review 9 6 Device and Process Concerns 15 7 Materials Selection 17 8 Electrical Attachment Process Development 24 9 Mechanical Attachment Process Development 27 10 Results: The Packaging and Interconnection Process 38 11 Testing and Reliability of Bonded Devices 39 12 Conclusions 43 13 Future Research 44 14 Appendices: A: Phase Diagrams 46 B: Mask Design 47 15 References 48 1 Acknowledgements First and foremost, I would like to express my gratitude to Bill Robbins and Yet-Ming Chiang for their constant guidance. This research would have been impossible without their support and encouragement. My thanks also goes to DARPA and the Charles Stark Draper Laboratory, who provided the funding to make this thesis possible. I also would like to thank Dominic Fulginiti for his technical knowledge and all he has done to aid this research. Ed Ayers, Sol Kafel, Craig Dement-Myers, and Matthew Lozow were also instrumental in the progress of this thesis. Words cannot express my gratitude to my friends, who made my time at MIT bearable. I especially would like to thank Keith Breinlinger, Anthony Rocco Nole, Jang Don Kim, Neil Jenkins, Howard Fine, and Luis Ortiz. Most of all, I would like to thank my loving fiancee, Penny Pliskin. It is impossible to express how much her constant support and understanding have meant to me in my pursuit of this degree. 2 Glossary of Terms CCD: Charge Coupled Device. A CCD receives electrical or photonic current over an area, and converts this into a usable signal for displays. Cordierite: A ceramic material, 2MgO-A12 0 3-5SiO 2 Device: In this research, device refers to the electrically and mechanically bonded die and substrate arrangement. Die: The silicon chip that contains the actual CCD device. P-Eucryptite: A ceramic material with a slightly negative thermal expansion coefficient, Li20-A120 3-2SiO 2 LZS: A composite material consisting of roughly 90% Si 3N4 and 10% ZrO2 . LZS is manufactured by Ceramic Process Systems, Inc, Chartley, MA. 3 List of Figures 1 Schematic of an EBCCD Apparatus Page 7 2 Optimal Glass Profile at Die Edge 15 3 Aluminum-Silicon Phase Diagram 16 4 Physical Derivation of the Young-LaPlace Equation 19 5 Creep Map of an FCC Metal 26 6 Furnace Cycle 6 30 7a 2760 Glass on a Pyrex Substrate 34 7b 2760 Glass on a Silicon Substrate 34 8 Anneal Cycle for Bonded Devices 37 4 Introduction Large area CCD-based digital cameras have a huge number of possible applications. Because the CCD sensor apparatus can be designed to detect a wider range of wavelengths than conventional light cameras, they can be used in new applications that are not possible with conventional cameras. Electronic CCD cameras are currently used to create digital photographs, as well as in night-vision equipment. It is also possible to use infrared CCD cameras in medical applications to create a detailed picture based on the heat produced in different organs or tissue in the body. This is extremely useful in oncology, since cancerous cells commonly produce a different heat signature than healthy cells. As the CCD cameras become more complex over time, one desirable quality is lowlevel light detection. One way to increase the quality of images detected at low light levels is to decrease the pixel size and increase the total sensor area. To understand the physical effects of changing these variables, one must comprehend the basic workings of a Electron Bombarded CCD device (EBCCD). A schematic of an EBCCD camera device is drawn below in Figure 1. Photon I A GaAs Imaging Equipment I I I . CCDn Figure 1: Schematic of an EBCCD sensor apparatus As shown in figure one, photons hit the photocathode material. The photon excites an electron from its ground state. The space between the CCD die and the photocathode is held under vacuum, in the range of 10-11 torr in this research. The electrical potential across this space accelerates the electron towards the die, which acts as the anode. The electron impinges on the die active area and activates a pixel. The number of electrons impinging on a pixel determines the relative intensity of light, or brightness of that portion of the image. The devices in this research are backside illuminated. This means that the electrons travel through the back of the die, and impinge on the underside of the active area. This allows all electrical connections to be removed from the space between the die and photocathode. To accomplish this physical arrangement, the die is thinned to a thickness of about 10 micrometers. This allows electrons to travel through the silicon die without a loss of resolution. The standard bond between a die and substrate in this type of thinned assembly employs polymeric materials. These bonds can be carried out at temperatures at or near room temperatures, using epoxy resins or thermoplastic materials. These materials can be filled to match their thermal expansion coefficients to silicon, so any thermal stresses that occur during the operation of the device can be minimized. In addition, because the moduli of these materials are often low, they can deform in compliance with the thermal stresses without failing due to fracture or permanent deformation. The assembly considered in this research, however, must be compatible with a high vacuum environment. Polymeric materials are not compatible with high vacuums. Polymers are the result of monomer combination reactions, and when the final polymer is exposed to a vacuum the kinetics of the polymerization reaction will favor a breakdown into monomer 8 molecules. In addition, polymers commonly contain residual solvents or other organic artifacts of polymerization and processing. Therefore polymeric materials can not be used to create the bond between the CCD die and substrate in this research. Lead oxide-based electronic sealing glasses are proposed as a substitute material to achieve the bond in this research. Glasses are compatible with the high vacuum environment required for this sensor, and should adhere to the materials proposed for use in this packaging process. However, glasses must be processed at higher temperatures than polymeric bond materials, and they commonly have thermal expansion coefficients that are much larger than silicon. Glasses also have much higher moduli than polymers, and are brittle under tensile stresses. It is proposed that these concerns can be overcome to produce a die-substrate bond that is both durable and practical, and is compatible with the high vacuum environment of the chosen CCD device. It is theorized that stresses due to thermal expansion coefficient mismatches can be minimized in a number of ways: minimizing processing temperatures, adding low thermal expansion coefficient filler materials to the electronic glasses, and constraining a thin bond glass layer between thicker die and substrate layers. 5 Literature Review The bonding of large area dies with electronic glass has not been discussed in detail in any known literature. There is much discussion of polymer-based packaging of large area sensors, however [1]. 5.1 Requirements for bond There are many mechanisms that can result in a bond between two dissimilar materials [2,3,4]. First, the bond can be purely mechanical, or a physical interaction of surface roughness that serves to interlock the materials. Because of the high surface finish of the materials in this research, and the fact that the bond glass is liquid in the bonding cycle, mechanical bonding can not be relied on to produce a permanent bond in this materials system. A bond can also be produced by the interdiffusion of two dissimilar materials. Because this research proposes firing oxide glasses in contact with passivated silicon, it is expected that some interdiffusion of these similar materials occurs. This likely contributes to the bond between the die and glass, as well as the glass and possible substrates such as Pyrex and silicon. A chemical bond at the interface between the two materials is also possible. There may be some possible network bridging when the oxide glass contacts silicon atoms in this research. This would contribute to the interfacial bonds. A fourth mechanism for a bond between dissimilar materials is an adsorption and wetting model. This model necessitates adsorption via molecular or van Der Waals bonding at the interface. This type of bond will occur if the interfacial energy of the two materials is less than the surface energy of the two separate materials in the bonding atmosphere [4]. 5.2 Glass Bond Literature The bonding of silicon wafers with borosilicate glass has been discussed [5]. Boron was deposited on a silicon surface using a physical vapor deposition process, and an oxide layer was thermally grown. Bonding was carried out at 450'C in an 02 atmosphere. To investigate 10 bond strength, the bonded silicon wafers were scored and broken. If fracture occurred in all layers without glass peeling away from the silicon surfaces, the bond was considered good. They report no good bonds using a spun-on glass technique, with a wide variety of borosilicate glass compositions and process variables. This spin-on glass application technique is closer to the application techniques used in this research than the PVD/ oxide growth method. Success in die bonding with silver-filled, conductive PbO-ZnO-B 20 3 glass has been reported [6]. This glass is used to provide electrical interconnects, as well as mechanical bonding between substrate and die. This type of glass has produced void and crack free bonds after sintering at a temperature of roughly 4000 C. However, the experiments used a glass paste with a binder. Because of the high vacuum environment of the sensor in this research, a glass without binders is desirable. The glass bond in this research must also be non-conductive. The lowest temperature, non-conductive bond reported occurs at 450 0 C [6], with the optimal firing temperature for a lead borate alumina glass at 900 0 C. The materials system in this research requires glass that can be fired without binders at a temperature near 400'C. The relationship between voids in IC/substrate bond layers and resultant damage of chips due to stress concentrations has been investigated [7]. This signifies the need for a continuous bond glass layer that is free of voids, thereby minimizing any possible sources of damage to the active die surface. 5.3 Stress Calculations The calculation of stresses due to thermal processing is vital to the choice of materials for this application. Thermal expansion stresses cause a variety of problems in the sandwich type processing. The most important stress component in the device is the amount of stress in 11 the glass layer, both before and after die attachment. This is due to the relative thickness of the bond layer, and the low yield and rupture stress values characteristic of glasses. Bowing of the die, and residual stresses in the thinned die are also concerns. The tensile stress in the glass can be approximated before die attachment by the Stoney formula. This formula is derived in The Materials Science of Thin Films [8]. This formula is derived from basic thermoelasticity theory in a biaxial frame of reference, and gives the stress is small compared to the component in the radial direction, ao. The Stoney formula assumes a layer thickness that is much smaller than the substrate thickness. The formula is expressed below: 1 a2 a61 E,E (a, -a,)AT Eas s +Ea +EEazaa(4aa +6a a +4a eq. 1 Es and Ef are the Young's moduli of the substrate and film, respectively, a is the thermal expansion coefficient, and as and af are the thicknesses of the substrate and film. AT is the difference in temperature from the bond conditions. Using the final glass composition in this research, one mil of glass fired on an LZS substrate results in a tensile glaze stress of 23 ksi. Another estimate of glaze stress can be found in Introduction to Ceramics [9]. This formula takes into account the relative thickness of the substrate and glass, but is more accurate at very small glass thicknesses, compared to the substrate thickness. The calculated stresses from the two equations converge when the difference between the Young's moduli of the glass and the substrate is small, and when the substrate is significantly thicker than the glass layer. Because of the less rigorous derivation, the Kingery equation is expected to be a rougher 12 approximation of the value of stress than the Stoney formula. The Kingery formula is shown below: ogiaz= E,(To -T')(a -a) 1-3 a+ + 6 f as a. eq. 2 Values are defined identically to those in the Stoney formula. With a structure consisting of one mil of glass on an LZS substrate, equation 2 gives a value of 108 ksi in tension. This stress level is much different than the value calculated from equation 1. While both equations express the value of stress with the same considerations, the derivation of the Stoney formula is more vigorous. This is due to the fact that the Stoney derivation begins with the consideration of materials with similar thicknesses, and then considers the special case where one material is significantly thicker than the other. The Kingery equation is derived by disregarding any constraint on the thicker material due to the thinner material. Both equations predict similar stress levels, as long as the Young's moduli of the substrate and glass are similar. As these properties become dissimilar, the Stoney formula is expected to give a better approximation of the actual stress levels. In addition, This research also investigated the use of substrates with dissimilar expansion coefficients that the silicon die. A method of calculating the stress in the bond layer between two planar materials with dissimilar thermal expansion coefficients has been proposed [10]. Because the bond layer is constrained by the two significantly thicker layers, the thermoelastic properties of the bond glass are not considered. The shear modulus of the glass has a significant effect on the stress level in the glass, so this property is included in the 13 derivation. This formula gives the maximum value of shear stress at the surfaces of the glass, and does not allow for bowing of the sandwich. Because bowing is not allowed, the value of stress calculated will be somewhat higher than the true value of stress in a similar physical setup. Some bowing will occur in any real physical arrangement. The formula is given below: (a1 - a 2 )A TG sinh(pr) ,8frcosh(fl) with: eq. 3 77 Elt i E2t2 eq. 4 G and q are the shear modulus and thickness of the glass. r is the radial distance from the center of the bond. E and a have their usual significance, and t is the thickness of the sandwiched layers. 1is the full radius of the assembly. In stress calculations in the Appendix, the die is defined as layer 1, and the substrate is layer 2. Using equation 3, the maximum shear stress occurs at the extreme edge of the assembly. At this point, r equals 1,and the hyperbolic sine and cosine terms become a single hyperbolic tangent function. For all but very small values of/ and r, this hyperbolic tangent function essentially equals one. For a 1 mil layer of glass on a Pyrex substrate, the hyperbolic tangent function equals 0.99 at 0.4 millimeters. This would imply that the shear stress at the die comers is independent of geometry, as long as the die face diagonal is less than 0.8 millimeters. Care must be taken when applying this formula to a real situation. 6 Device and Processing Design Concerns Steps in the device fabrication process dictate certain limits on the materials and processing of the die attach process. Most importantly, the die thinning process is extremely sensitive to the die bond condition. The silicon die is thinned using a HF-based acid etch process. This acid solution will damage the active side of the die if contact is made. Therefore, the bond glass must provide a seal to the edge of the die. The non-passivated die edge, however, shows excellent wetting compatibility with all glasses used in this research. For the acid etch process to produce a uniform thickness across the profile of the die, no glass can adhere to the die edges. Figure 2 below shows the optimal glass profile at the die edge to provide a seal without impeding the thinning process. die glass substrate not enough glass die die glass glass substrate optimal substrate too much glass Figure 2: Optimal glass profile at die edges. The CCD die used in this process has conductive lines of aluminum. Aluminum and pure silicon have a eutectic point at 5770 C. This eutectic point is depressed to with the addition of boron or other doping elements. The die is composed of P-type silicon. The Al-Si phase diagram is shown below. WEIGHT PER CENT SILICON 10 05 700 20 30 1 1401 1 601 50 70 1 1 90 1 wt-*/.si 10 15 600 20 1 5770 (Af) 500 1t500 159 ~14120 (1.65) I I I 1400 400 0.16(0.17) 0 80 300 (AL+ Si) 7Z 7Z - - 200 a- 1300 7 200 1 000 rL 0.5 1.0 At-%i 1.5 /S 100 - 900 800 700 6600 600 5770- 1 Si500 A ^^ 0 Al 10 20 E 30 40 50 60 70 80 ATOMIC PER CENT SILICON 90 100 SI Figure 3: The aluminum-silicon phase diagram [8] If this eutectic temperature is exceeded, the liquid will change the charge-carrying paths and produce shorts in the die. This would destroy the chip's performance. This temperature limits the choice of materials used in the bond. Interdiffusion of aluminum and silicon is also a concern in this physical arrangement. To lessen the effect of possible interdiffusion, the aluminum used in conductive lines on the die contains silicon at the solubility limit at room temperature. There is also a passivation layer under the conductive aluminum layer. Oxidation of materials in the die is also a concern. As charge-carrying materials in the die oxidize, their electrical resistivity changes. This could degrade the performance of the die. 7 Materials Selection The substrate holds the die in place and acts as the mechanical support for the die, and The substrate also provides electrical interconnects between the die and subsequent packaging and pins. The substrate must have a Young's modulus high enough to provide mechanical support to the die, and must have a thermal expansion coefficient that is close to silicon to minimize thermal expansion mismatch stresses. The substrate must be electrically insulating to isolate contact bondpads. It is also desirable to have high thermal conductivity to dissipate heat from the sensor operation. Since phonon conductivity and electrical conductivity are usually proportional properties, choosing a material with these qualities is somewhat difficult. Based on the above criteria, research was limited to four possible substrate materials: passivated silicon, aluminum nitride, Pyrex, and LZS. LZS is a dispersed composite material, made up of roughly 91% Si 3N4 and 9% ZrO 2 . Table 1 below shows a comparison of some thermal and mechanical properties of the four material. Table 1: Material properties of substrate materials Density (p) Young's CTE (a) Material Termal Conductivity modulus (E) Si 3N4 2.25 x10 -6 280 GPa 3.44 g/cm' 30 W/mK ZrO 2 8.0 x 10-' 180 GPa 5.8 g/cm 3 2.09 W/mK LZS* 3.3 x 10- 286 GPa 3.79 g/cm 3 30 W/mK Silicon 3.3 x 10- 170 GPa 2.33 g/cm 3 150 W/mK A1N 4.2 x 10 300 GPa 3.26 g/cm' 180 W/mK Pyrex 2.9 x 10-' 63 GPa -2 g/cm 3 1.1 W/mK Based on the above criteria, LZS is the most likely candidate for the substrate material. Experiments were carried out on all four candidate materials, however. It is foreseeable that other material properties may prove to be more desirable. LZS seems to maximize the desirable properties in this research. 7.1 Bond Glass The bond glass connects the die to the substrate. The glass layer must be mechanically strong to hold the die in place. Because the device operates in a high vacuum environment, the bond glass must not release gas into the sealed device. This undesirable process is commonly called outgassing. Because of this outgassing concern, no polymeric materials or organic binders can be used in the packaging process. Because of the temperature limits of the die, the glass processing temperature must be low. Based on comparisons between experiments and the manufacturer's recommendations, the glass viscosity must be in the range of 10-4 Poise for necessary glass flow in the bonding process. It is also desirable for the glass to be vitreous, or not experience crystallization and devitrify after one firing cycle. The process used in this research relies on multiple firing steps, and the glass must be capable of viscous flow during each heating cycle. To provide a good bond with both the substrate and the die, the bond glass must wet and adhere to both surfaces. This is both a physical and chemical process, requiring both intimate mechanical contact and adhesion between the two materials at the interface. Adherence is a complex phenomenon, and can not be easily measured. One approach to quantify adherence is to determine the relative interfacial energies of the materials under 18 investigation. The surface energies between the glass, substrate, and atmosphere can be related using the Young-LaPlace equation with a droplet of glass on a substrate: Yls Yxx Y= sv - Ylv COS 0 represent the surface energies of the liquid-solid, solid-atmosphere, and liquid-atmosphere interfaces. 0 is the angle observed at the edge of the liquid droplet. Figure 4 below shows the geometric derivation of this equation. YI Ysv 7 1s Figure 4: Physical derivation of the Young-LaPlace equation. This equation is derived from investigations of liquid droplets on solid surfaces If the angle is less than 900, the liquid wets the solid surface. An angle less than 900 indicates that the system can reduce its energy state by covering the solid with the liquid, since the interfacial energy between the liquid and solid is lower than the solid-atmosphere energy. This arrangement will allow the liquid to flow and coat the solid surface. It is desirable for the bond glass to have a thermal expansion coefficient that is as close to silicon as possible. This is extremely significant, as the bond glass is in direct contact with the active side of the die. Even if the liquid glass wets the substrate and die surfaces, differences in thermal expansion coefficient can cause spalling of the bond glass layer. In addition, it is desirable to have a glass with a fracture toughness and shear modulus high enough to resist deformation under any thermal stresses. 19 7.2 Bond Glass Choices Based on the thermal processing limits of the die, lead oxide-based electronic sealing glasses were chosen for experimentation. Because initial glass layers are fired on the substrate, it is permissible to approach the Al-Si eutectic temperature. However, because it is usually desirable to have some deformation throughout the bond glass layer during the die bond process, all glasses used in the mechanical bond must have processing temperatures near or below this eutectic temperature. Table 2 below shows the glasses used in these experiments, with the manufacturer's recommended firing temperatures for each. Table 2: Glass Characteristics Firing Temperature Glass Thermal Expansion Coefficient [ x 10-6] EG 2760 (Ferro) 380 0 C 11.35 EG 2008 (Ferro) 410 0 C 7.0 EG 7585 (Ferro) 415 0 C 6.8 EG 2805 (Ferro) 580 0 C 3.37 LS 0111M (Nippon El.) 4600 C 5.0 In all cases except 2805 glass, it is necessary to exceed the recommended firing temperature of the glass to produce a fully dense layer. Processing at the recommended firing temperature in all other cases produces a glass layer that is porous. Table 3 below shows the compositions of the glasses used in this research. All values shown are the compositions reported by the manufacturers. 2760 is a single-phase lead oxide20 based glass. 2008, 7585, and 2805 are mixtures of 2760 and fillers. The inert fillers serve to lower the thermal expansion coefficients of the glasses. 0111M is also a filled glass. One concern with the 2805 glass, as described by the manufacturer, is that the P-eucryptite fillers can cause the glass to crystallize after multiple firings. Table 3: Glass compositions Composition Glass EG-2760 <90% PbO, <15% B20 3 , <10% ZnO, <5% SiO 2, <5% A12 0 3 EG-2008 2760 composition + 2MgO-A120 3-5SiO 2 (cordierite) EG-7585 <70% PbO, <15% SiO 2 , <15% B20 3, <10% A120 3, <5% ZnO, <5% MgO, <1%BaO, <1%SnO 2 EG-2805 <60% PbO, <20% SiO 2, <15% A120 3, <10% B2 0 3 , <5% ZnO, <5% Li 20, <5% MgO <1% BaO or, 2760 composition + cordierite and Li20-A12 0 3 -2SiO2 (P-eucryptite) Ferro glasses are available in a variety of powder sizes. Table 4 below shows the power sizes used in this research: Table 4: Ferro glass mesh sizes Average diameter (microns) Powder Type Maximum Diameter (microns) VEG 6.0-10.0 44 MVG 4.5 25 SRRG 1.5 10 The VEG grade of all glasses were used, as well as SRRG grade 2760. Visible observations have shown that the glass powder is composed of flakes. The filled glasses are physical mixtures of glass flake powder and separate crystalline filler particles. The process proposed in this paper is designed to be compatible with the VEG grade of all glasses. 21 Adhesion between the glass and the substrates was investigated in relative terms only. Two types of tests were performed; a visual wetting angle observation, and a scratch test. To visually inspect the wetting angle, a small amount of glass powder was placed on each substrate, and the substrate was placed on a hotplate. The assembly was heated until the glass consolidated, and exhibited flow. The samples were then cooled, and visually inspected under a microscope. Measurements were made by observing droplets 1-2 mm in diameter. Because the angles were only visually estimated, the values below only show relative wetting among the candidate materials. The s in many of the cells signifies that the glass layer spalled from the substrate. In these cases, the glass did not adhere to the substrate, and no actual wetting angle could be observed Table 5: Wetting angles for glasses and substrate materials Substrate w Glass Pyrex Silicon A1N LZS Passivated silicon (die) 2760 s 2008 450 7585 0111M s s s s 500 550 500 450 450 450 400 300 300 330 Adhesion of the candidate glasses was also investigated relatively using a scratch test. Glass layers were applied to all candidate substrates by depositing an alcohol/glass slurry using an airbrush. The samples were then fired at temperatures sufficient to form fully dense layers. The surfaces were then scratched with a blade, with enough force to remove the glass under 22 the point of the knife. The scratched surface was then observed. Relative adhesion strengths were qualitatively compared by observing the relative force required to draw the blade across the surface. This could also only be estimated in relative terms, since this experiment was performed manually. This scratch test revealed that the differences in thermal expansion coefficients affected the strength of the bond. As a result of these experiments, 2008 was chosen as the bond glass for LZS and Pyrex substrates, and 0111M was chosen for use with AIN and silicon. The manufacturer discontinued the 0111M glass during this research, so 7585 was chosen as a substitute for A1N and silicon. The filler particles in the filled glasses remain solid through the firing process. Therefore, they do not contribute to the glass-die bond if they are in direct contact with the die face during the bonding process. For this reason, all bonding procedures use a layer of 2760 directly under the die. 7.3 Metallization In addition to the substrate and bond materials, the mechanical characteristics of the proposed electrical interconnects must also be investigated. The substrates must be metallized to provide interconnects between the die and the subsequent packaging. This metallization must be compatible with both possible die attach mechanisms. For most interconnects in microelectronics, gold is the desired material. Because of its low chemical activity, gold provides a surface free of oxides. Surface oxides increase the resistivity of the contacts. For this reason, the top layer metallization on the substrates is gold. There are additional layers to increase the metallization adherance on the substrates. Because most of the substrates have an oxide on the surface, the metallization must be mechanically and chemically compatible with this oxide layer. The first metallization layer on the substrate is titanium, because titanium dioxide has roughly the same density as titanium. This means that as the metal in contact with the substrate oxidizes, it does not expand. This allows a stress free oxide-oxide bond. A layer of platinum is deposited between the gold and titanium to act as a diffusion barrier between the two layers. 8 Electrical Attachment Process Development The die has metallized bond pads of deposited aluminum. The electrical attachment process must provide an electrical connection between these aluminum bondpads and the gold metallization on the substrate. There are two proposed methods for connecting the die to the substrate, a wirebonding procedure and a collapsed ballbond procedure. 8.1 Wirebonding In the proposed wirebond process, the wirebond is connected to the underside of the aluminum bondpad on the die. The backside of the bondpad is exposed using a selective etching process. After the standard thinning etch procedure, the back side of the die active area is masked, and the silicon under the aluminum bondpads at the edges of the die is etched away. The etching process can be controlled so that only the metal bondpad is left. The glass bond under the die supports this thin layer of metal. The exposed bondpad can be wirebonded to the substrate metallization at room temperature, after the mechanical bond is accomplished. This wirebond can be created using gold wire, using standard wirebonding procedures. 8.2 Ballbonding The second attachment method is more complex. It consists of a ball bond, with a metal ball deposited on the aluminum bondpads on the die. The standard ballbond is made using gold wire, leaving a four mil diameter ball attached to the aluminum bondpads. The die is then aligned with the substrate, and put under a load, and then heated to the glass firing temperature. The ball collapses, and provides an electrical interconnect between the die and the substrate. The attachment is loaded using arrangements of 2 or 4 lb. spring clips. The yield stress of gold is 207 MPa in tension. One two pound clip, assuming contact with all 30 ballbonds, corresponds to a compressive force of 177 MPa. Any additional clips will cause plastic deformation at room temperature. This will lead to deformation and prevent a permanent connection. It is desirable for the bonding process to be carried out with a creep mechanism, allowing for interdiffusion and mechanical connection between the bondpad and the gold balls. The ballbond attachment is carried out at a homologous temperature of roughly 0.54, based on the 1337K melting point of gold. With the same arrangement of one two pound clip, this corresponds to low temperature power-law creep. The mechanism for this deformation is dislocation motion, assisted by diffusion through the dislocation core. Figure 5 below is a creep deformation map for a FCC metal, showing deformation regimes in terms of homologous temperature (T/Tm) and normalized shear (o/p). All pure FCC metals will show the same mode of deformation at similar homologous temperatures and normalized shear stresses. For this reason, the deformation map for gold and silver will be essentially the same. A Compressional stress state can be resolved onto pure shear axes by dividing the stress value by 25 1.73. Figure 5 shows the deformation behavior in a pure shear mode. Because only order-ofmagnitude approximations are necessary in these calculations, deformation in compression can be roughly equated with deformation under shear. TEMPERATURE (C) S4 -- O C 200 400 00 Oc b' E z in- LdJ Ix W _m LJ0 0I 0 Ln W Ix a: (I) 0 w 4-J Cr W w I O Z U, HOMOLOGOUS TEMPERATURE, T/T. Figure 5: Creep map of an FCC metal. In this case, the metal is silver. With the addition of glass, the contact area is raised significantly. Assuming a glass pad of 0.4 in. x 0.5 in, the overall stress in the bond, using one four pound clip, is now 287KPa. This implies that diffusional creep controls deformation in the gold balls. These calculations show the large variation in gold deformation mechanisms based on the contact area of the glass. To ensure consistent bonding mechanics, the glass must be flat, and roughly the same height as the glass. This can be accomplished by carefully hand lapping of the ballbonded die and the glass bond layer. Lapping the die for about two seconds on a 26 3000 mesh diamond lap produces balls that are uniformly 1.2 to 1.7 mils high. 9 Mechanical Attachment Process Development To mechanically attach the die to the substrate, a bond glass layer is first deposited on the substrate. This glass layer is built up through a series of depositions and firings. The glass may undergo a surface preparation between subsequent layers. After the bond glass layer is complete, a die is aligned with the substrate, and the whole assembly is fired under a load. The whole assembly will then go through a compaction and stress anneal cycle. 9.1 Glass Deposition Powdered glass is applied to the substrate using an airbrush. All glasses used in the final process are manufactured by the Ferro Company, and are available in various mesh sizes. Experiments were performed using the VEG and SRRG mesh sizes, and the VEG mesh was selected for processing. This selection was based on ease of handling in the airbrush process. The powdered glass is mixed with isopropyl alcohol to form a slurry. Based on initial experiments with spraying procedure, a solution of 7.5 g of glass in 20 mL of alchohol gives an even coat of glass without clogging the airbrush. Solutions with less alcohol gave a glass layer that was less uniform after firing, so 20 mL was chosen as the optimal airbrush solution. Constant agitation is necessary throughout the spraying process. To ensure a consistent glass mixture, the solution is made in 20 mL batches and mixed fully before each spraying. Then a small volume of the solution is poured into the airbrush fluid receptacle. This provides enough solution to coat roughly ten substrates. The substrates are coated using nitrogen as the propellant, at 20 psi. The spraying distance is roughly 8 inches, and the spraying time is 27 roughly one second per 0.4 x 0.5 square inches of substrate area. After coating, the excess solution is returned to the 20 mL container. The alcohol evaporates from the substrate in a few seconds, leaving an even layer of the powdered glass on the substrate. After firing, the above procedure produces a glass layer that is roughly 0.5 mils thick. 9.2 Glass Firing All glass layers were fired in belt furnaces. To guard against die degradation due to oxidation, die bonding cycles were all carried out under a nitrogen atmosphere. There is some concern that a reducing atmosphere may cause reduction of the lead oxide in the bond glass[7], but this did not appear to be significant enough to have an effect in this research. This concern suggests that the total number of firing cycles should be kept to a minimum to lessen the amount of time that the glass is exposed to a reducing environment. The table below shows the belt furnace profiles used in this research. Cycle numbers 2 and 3 are brazing profiles for standard production, and all glass firing cycles are derived from these two basic profiles. Because all the filled glasses in this research require higher firing temperatures than the 2760 glass, all filled glasses are fired in cycles 1 or 2. The four zones identified in table 6 are equidistant through the length of the furnace. The percentages in each zone refer to the power to the zone heaters. The furnaces are controlled through feedback from two thermocouples, located at two of the four zones. Furnace 1 is controlled by tuning to the temperature of zones one and four, and furnace 2 by the temperature of zones two and three. Table 6: Firing Furnace Cycles Cycle Furnace Zone 1 Zone 2 Zone 3 Zone 4 Atm. Number 1 1 1 500 0C 2 100% 4 2 100% 5 2 90% 6 2 100% 7 2 90% 8 2 90% 9 3000 70% 70% 4700 100% 3 Notes Speed 100% 2 Belt 2 90% 70% 4200 4300 80% 80% 4000 4100 80% 80% 3850 4050 60% 65% 3850 3650 90% 65% 4050 3650 100% 65% 4200 3850 100% 65% 4200 3850 100% 65% 3 in/min 100% 3000 70% Air Filled glass firing cycle Air 3 in/min Default N2 4 in/min Default N2 4 in/min Non-intermixing 100% 100% 100% attempt Air 2 in/min 100% Die bond attempt Air 2 in/min 65% Die bond attempt Air 2 in/min 65% Die bond attempt N2 4 in/min 2760 firing cycle N2 2 in/min Die bond cycle 65% 65% Oxide glasses have a large gradient in its thermal expansion coefficient near their glass transition temperature. The manufacturer suggests that any residual stresses arising from a cooling cycle through this temperature can be removed using an anneal cycle at a temperature of 3150 . The transition temperature, which the manufacturer identifies as the softening temperature, occurs at 350'. To minimize residual stresses in the glass during firing cycles, a slow cool through the transition temperature is desired. It is also desirable to fire the glass layers at as low temperature as possible, to further lessen residual stresses. Cycle 3 also had a slower ramp profile, with a fast cool. This is undesirable for glass firing. Further cycles attempted to produce a profile with a slow cool. Figure 6 below shows a profile obtained by running a thermocouple through firing cycle six. This is believed to be the slowest cooling profile obtained. Figure 6: Cycle 6. The cooling profile obtained in the belt furnaces is too fast to provide for any significant stress relief, so an anneal cycle was added. The development of this anneal cycle will be discussed in section 9.7, Anneal Cycle. 9.3 Masking The substrate is metallized prior to the glass deposition steps. The metal runs must be free of glass to ensure electrical contact between the carrier, substrate, and die. A masking fixture was designed to shield the metalized area of the substrate during the glass spraying process steps. The mask consists of an aluminum plate with recessed cavities for the substrates, and a mylar sheet with an opening that matches the dimensions of the desired glass bond area. The design of this mask assembly is included as Appendix B. There are two possible mask sizes, which correspond to the two possible electrical attachment techniques. The ballbond attachment technique uses a masked area that is smaller than the die, to allow the ballbonds at the edge of the die to contact the metallized areas of the substrate. The final bonded die assembly must have glass at the edge of the die to prevent damage to the active side during thinning. This arrangement relies on surface tension to cause the glass to wick to the die edge after bonding. With the second electrical attachment technique, the masked area is slightly larger than the die. The wirebonds can be attached to the metallized area beyond the glass, and not necessarily directly under the metallized pad on the die. This technique does not rely on glass wicking to produce a fully sealed die edge. After the glass is applied with the airbrush, the substrate is visually inspected with a microscope. Any overspray is removed with a vacuum pencil. Experiments with the SRRG grade of 2760 glass show more residual overspray than the VEG grade. This is due to the difficulty of removing the overspray of the extremely fine SRRG grade particles. This vacuum process is tedious, and likely to be a bottleneck in a production-scale process. Therefore, other masking techniques have been proposed. One is to remove glass using a photoetch process. Using this method, no masking apparatus is necessary, and the entire substrate can be coated with glass. After the final glass deposition step, the glass above the metallized area can be etched away. This technique would also eliminate the need for glass movement in the ballbond-type arrangement. 9.4 Optimal Layer Thicknesses The total bond glass layer thickness must be thick enough to produce a continuous bond between the substrate and die. In the ballbond-attached arrangement, there must be enough glass to allow for flow to the die edges in the time duration of the bonding furnace cycle. The glass layer also must have a flat surface, with a surface that is uniform enough to provide contact with the die over the entire die surface. Flow occurs in the vitreous glass phases in the glass layer, but the filler particles remain solid throughout all furnace cycles. During initial research, spalling and substrate adhesion concerns led to a necessary 3:1 filled glass: 2760 glass ratio. To obtain the necessary surface to provide for complete bonding over the die surface, glass layers of 3-4 mils were necessary. This deviated from the original concepts of using very thin layers, but glass surface irregularities were determined to be the dominant factor in interfacial bonding. Using the stress models proposed in section 4.3, this gives a level of stress of 23 ksi, for a 4 mil layer of glass on an LZS substrate. It was discovered that hand lapping can greatly improve the surface of the glass, without the need for a 3-4 mil glass layer. Lapping will remove the non-vitreous material along with the glass, and produce a very flat layer. Substrates are now lapped after deposition 32 of the initial filled glass, before the application of the 2760 glass layers. Because the added 2670 glass is completely vitreous, no lapping is required after 2760 application. The filled glass, 2008 or 7585, is applied as before, and a layer of 2.5-3 mils is built up. The substrate is then lapped until the filled glass layer is uniformly 1 mil thick. This ensures a continuous layer after lapping. A 0.5 mil layer of 2760 glass is then applied on top of the filled glass. At the firing temperature, the matrix glass in the filled glass and the applied 2760 both flow. The result is a 1.5 mil layer of glass with a uniform flat surface of vitreous glass. A 1.5 mil layer of glass on a LZS substrate is calculated to have a residual stress level of 23 ksi (164 MPa). 9.5 Crazing General lead oxide based glass can support a tensile stress of 12 ksi [9]. Because all the glasses in this research are lead oxide based, it is assumed that they will have an ultimate tensile stress of about 12 ksi. Based on a Stoney formula approximation, the tensile stress in glass deposited on an AIN substrate is 23 ksi. Since AIN has the highest thermal expansion coefficient of all the substrate materials (4.2 x 10-6), this arrangement should have the lowest possible residual stress. Because all glass/ substrate samples exceed the ultimate tensile stress of the glass, crazing is evident on the glass surface. Crazing is a network of cracks over the surface, which relieve the residual stress. Crazing could be detrimental to die performance in a number of reasons. The small cracks could retain atmosphere when the device is operated under a vacuum. The gas would pollute the high vacuum environment and degrade the resolution of the CCD. Also, if the cracking occurs after the die is bonded to the glass, the sudden strain could damage the active surface of the die. Figures 7a and 7b below show crazing of 2760 glass on Pyrex and silicon. Crazing grain sizes scale roughly with the calculated stress values. The crazing grain size of the glass on Pyrex, Figure 7a. is slightly larger than the grain size of the glass on silicon, Figure 7b. top,%Ij Figure 7a: 2760 glass on Pyrex Figure 7b: 2760 glass on silicon The severity of crazing is greatly reduced with the addition of the filled glass layer. and an anneal treatment. Samples using the final glass composition and process have been etched to the point where only the passivation layer is left on top of the glass, and cracks are not observed in the passivation layer. Crazing is still evident in the glass, however. 9.6 Die Attach The die is attached to the substrate by placing the die against the bond glass on the substrate, and heating the arrangement until the glass becomes vitreous. A load is applied to the arrangement, so the bond glass layer can deform and provide intimate contact with the die surface. The bond glass surface must be extremely flat before bonding can occur. A template jig was designed to align the die ballbonds and the metallized pads on the substrate. Because the chip is bonded in a flip-chip arrangement and all substrate materials (except Pyrex) are opaque, this can not be done using visual alignment. The jig has an opening to allow for the application of a spring clip. The spring clip holds the arrangement together during the initial bonding cycle. The amount of force on the unbonded arrangement is critical for device performance. The arrangement can be clamped using from one to four clamps, and the clamps are available in either two or four pound loads. Careful control of glass layer and ballbond height is critical in the bonding operation is critical, as the amount of stress on the glass and ballbonds is dependant on the contact area of the arrangement. If the ballbonds are higher than the glass, then all the stress from the clip is distributed among the ballbonds, and not carried by the glass. If the ballbond height is not uniform, and not all the ballbonds contact the substrate, the stress levels in the higher ballbonds can be extremely high. Preliminary experiments have demonstrated that stresses in the ballbonds can exceed the compressional yield stress of silicon, and cause chips in the die directly under the ballbond. Ballbond lapping produces ballbonds of uniform height, and alleviates this problem. To ensure electrical connection, it is desirable to have ballbonds that are slightly higher (0.2 mils) than the glass pad. This ensures that the ballbond will connect with the substrate metallization. 35 After clamping the die/substrate arrangement in the jig, it is heated in cycle 9. This seals the die, but does not allow for enough glass deformation to produce a good bond. After this initial bond, the die can be removed with the application of a small shear stress. However, this bond cycle provides a seal of the active surface of the die in a nitrogen atmosphere. This is believed to ensure against oxidation degredation of the die. 9.7 Anneal Cycle After the initial seal cycle, the arrangement is bonded in a small (440 cm 3 ) box furnace, in air. Because the ballbonds and glass are now in contact at both interfaces, a larger bonding force can be applied without damage to the silicon surface. In addition, flat pieces of alumina are placed on both sides of the bonded arrangement to distribute the stress equally over the surfaces. This furnace cycle heats the arrangement to a temperature where the glass is again viscous, and then cools to anneal the glass at 315 0C. This anneal temperature is below the softening temperature of 2760 glass. According to the manufacturer, an anneal at this temperature for ten minutes will remove the residual stress from bonding. This furnace cycle is diagrammed below: 415C 400 T time Figure 8: Anneal cycle for bonded devices This cycle produces a bond that is uniform, and can withstand manual applied shear stresses. If the glass surface is sufficiently uniform before bonding, the bond will have no visible porosity. Some crazing is still evident under the die, although to a much lesser degree than samples solely bonded in a belt furnace. If the sample is electrically bonded using the ballbond method, the glass must flow outwards past the connected ballbonds to seal the entire active area of the die. Many experiments were performed to determine what process variables would allow consistent flow of glass to the die edge during the final bond/anneal cycle. However, it seems too difficult to produce repeatable results on a laboratory scale. Substituting the mask with the larger opening for wirebonding, with a vacuuming step to remove the glass from the metallization runs before firing, should produce more repeatible results. The new photoetching process should eliminate the need for any mask apparatus. 10 Results: The Packaging and Interconnection Process This development suggests a process that will produce an optimal bond between the die and substrate, using the proposed materials systems. 10.1 Glass Layer Firing The first step in the packaging process is the choice of substrate material. This process is compatible with four substrate materials. To minimize extraneous variables in the process, aluminum nitride and silicon substrates follow an identical filled glass procedure; and Pyrex and LZS substrates follow a common procedure. 2008 glass is used as the first layer on Pyrex and LZS substrates, and 7585 glass is used on silicon and aluminum nitride. This glass is deposited using an airbrush procedure, described in section 8.1, Glass Deposition. Successive glass layers are deposited and fired until the total glass thickness reaches roughly 3.5 mils. This ensures a continuous layer after lapping. The glass is then hand lapped to a thickness of 1 mil. It is necessary at this point to have a continuous layer of glass. After the sample has been lapped to a glass layer thickness of 1 mil, 2760 glass layers are deposited and fired using the firing cycle 8. The glass layer thickness is measured after each firing, and the glass is built up to a total glass layer thickness of 1.5 mils. 10.2 Bond Cycles If the die is to be electrically connected to the substrate using the ballbond procedure, the ballbonded die must be face lapped. A hand lap of 2-3 seconds on a 3000 mesh lap will ensure uniform ballbond height. After a 1.5 mil glass layer is deposited and fired on the 38 substrate, it is aligned with a die in the alignment jig. One four pound clip is placed on the diesubstrate assembly, and the arrangement is fired in cycle 9, under nitrogen. This will seal the assembly. The assembly is then sandwiched between two flat pieces of alumina to distribute the bonding load. Four four-pound clips are then attached to the sandwich, and the whole assembly is fired in the anneal/ bond cycle described in section 8.7, Anneal Cycle. The device is then ready to be thinned and connected to the CCD sensor apparatus. 11 Testing and Reliability of Bonded Devices The reliability of these devices can not be fully tested until a whole assembly is placed in operation. However, there are numerous ways to quantify the success of the mechanical and electrical bonds. While these tests provide some measure of how well a bonded assembly will perform, only field tests of the whole CCD device under operating conditions will give a true representation of possible performance. 11.1 Manual Mechanical Bond Test After a die is bonded to a substrate, adhesion is investigated in a purely qualitative basis by applying a small amount of manual force to the die. This only ensures that the die will not separate from the substrate with regular handling during subsequent testing, and that some contact is made between the die and the glass. 11.2 Thermal Cycling It is desirable for the mechanical bond to be able to withstand the cyclic stresses involved in repeated thermal cycling. To test cyclic thermal exposure resistance, bonded samples using all four possible substrate materials were put through two separate thermal exposure profiles. The tests were carried out in air, and attempted to emulate the range of temperatures that a finished device would be exposed to during operation. The first set of samples were exposed to 21 cycles from temperatures of -250 C to 750 C. The cycles were somewhat sinusoidal, and had a frequency of 1.3 cycles per hour. The second set of samples were cycled from -300 C to 750C with a frequency of 0.8 cycles per hour for a total of 12 cycles. No degradation of the bonds was observed after these thermal cycles. It is assumed that the thermal cycles used in this experiment are less severe than any possible use conditions, so degradation of the bond due to temperature fluctuations during use should not be a major concern. 11.3 Die Connectivity To test the success of electrical bonding attempts, all bonded devices were tested for electrical connectivity. This was accomplished by contacting the metallized runs on the substrate with electrical point probes, and measuring electrical characteristics across circuits on the die. This gave a discrete method to determine electrical connection status at each ballbond. This test also attempted to determine allowable thermal exposure of the dies by comparing impedance changes with duration of thermal exposure. No changes in electrical behavior were observed, however. Table 7 shows the results of electrical connectivity tests for devices. The column labeled 'bonds connected' gives the number of bonds that showed a standard electrical response for diodes in the known circuits. This column also reports the total number of bonds that were probed. In most cases, all 31 bonds were not probed. This table shows the increase in connected bonds chronologically, with sample numbers. The increase in connected bonds through subsequent experiments should be noted. It should also be noted that all samples from number 43 onwards show full electrical connectivity of all probed bonds. This demonstrates that the electrical and mechanical bonding procedures are compatible. Sample numbers 28-42 are the results of experiments with the 2805 glass. These experiments were discontinued, as the 2805 does not exhibit the required viscoelastic characteristics in the range of temperatures required for bonding. This table also presents the results of some experiments with 2004 glass. This glass is another 2760 derivative that has recently become available. 2004 glass has good viscous properties at the bonding temperature, and shows promise to replace the 2008 glass in the process described in this research. Table 7: Results of electrical connectivity tests notes thermal history bonds Test sample connected number 6 13/25 final treatment in 4 mils of glass cycle 2 rebonded- one side of die connected 6 13/23 rebonded again, same result 7 11/26 one bond only works with pressure on die 9 25/25 bond cycle only no glass 11 4/16 bond cycle only no glass, substrate is broken, some metallization gone 13-27 no glass- ballbond experiments 28-42 experiments with 2805 glass 43 29/29 before final anneal glass lapped- final thickness 1.2 mils die with polyimide coating 43 29/29 43 29/29 44 25/25 44 25/25 bonded with 8 lb load after die thinning full process probe tested experiments with 2004 glass 44-56 57 after final anneal 25/25 SiC coating on die, 2004 glass 12 Conclusions This research demonstrates that it is possible to package silicon CCD dies with materials that are compatible with extremely high vacuum environments. Bonds using materials with dissimilar thermal expansion coefficients can be accomplished, as long as the bond layer is physically constrained to lessen the effects of this dissimilar thermal expansion coefficient. This research suggests that crazing can occur in the bond glass layer without causing extensive damage to the operation of the device. However, the extent of this crazing should be minimized through further research. The results of die thinning experiments have shown that the level of physical constraint drastically changes when the die is thinned to 10 pým. This suggests that physical constraint alone is not enough to remove the effects of thermal expansion coefficient mismatch. However, the proposed process demonstrates that the glass layer can be engineered to minimize these effects on the active area of the die. The manual and thermal cycling test results demonstrate that the assembled device should be able to withstand any mechanical stress during final device assembly and use. The results reported in Table 7 demonstrate that the process can produce a mechanically bonded assembly that is fully electrically connected. Therefore this process can be used in this application, with the expectation of a fully connected and working device. It can be observed that all samples from sample number 43 onwards demonstrate dies that are fully electrically connected. This reproducibility demonstrates the robustness of this process. Therefore, slight variations in process variables will not cause failures in bonded devices, and it is likely that the process can be scaled to an industrial level. This research suggests that while it may not be possible to fully eliminate the effects of thermal expansion coefficient mismatch stress, it is possible to design a process to minimize these effects to a point where they do not effect the performance of the CCD sensor. 13 Future Research The initial successes of the 2004 glass show promise. Dies have been attached to Pyrex substrates using this glass, and no crazing has been exhibited after bonding. This is significant, since the thermal expansion coefficient of this glass is roughly the same as the 2008 and 7585 glass. There are many problems yet to be addressed considering this glass, however. A continuous 2004 glass layer has not been achieved, so a glass layer free of both porosity and crazing has not been observed. This is significant, as porosity is theorized to remove some of the stress that would cause crazing in a continuous layer. There are also concerns that porosity could cause pollution of the high vaccum environment in service. To lessen the extent of porosity in glass layers, proposed experiments will investigate glass porosity in glass layers fired under vacuum. Theoretically, the low pressure environment will discourage the formation of pores in the viscous molten glass. Photographs of thinned dies show no surface effects of the glass porosity when bonded using 2004 glass. This suggests that a continuous glass layer may not be necessary for optimal die performance. These results also suggest that porosity may be more desirable than crazing. Experiments are proposed to investigate the effects of coatings on the die surface. It is proposed that any possible damage to the die surface due to crazing could be lessened with the application of a barrier coating. Polyimide, spun-on glass, and silicon carbide coatings have been proposed. Preliminary experiments using the polyimide coatings suggest that damage to 44 the die surface can be prevented. However, this polyimide is an organic material, and not compatible with the high vacuum environment of the sensor assembly. Experimentation must be performed using inorganic coatings, such as silicon carbide or the spun-on glass. Li20-A1203-SiO2 S10, 14 Appendices 14.1 Phase Diagrams PbO-SiO2 m• 2050* Rustum Roy and E. F. Osborn, J. Am. Ceram. Soc., 71 [6] 2086 (1949); slightly modified by M. Krishna Murthy and F. A. Hummel, ibid.. 37 [1117 (1954). PbO-B 20 3 -SiOz R. F. Geller, A. S. Creamer, and E. N. Bunting, J. Research Natl. Bur. Standards, 13 [2] 243 (1934); RP 705. 40 Of i PbO-AlsOr-SiOj; high-PbO portion. R. F. Geller and E. N. Bunting, J. Research Natl. Bur. Standards 31, 264 (1943); RP 1564. 46 R. F. Geller and E. N. Bunting, J. Research Natl. Bur. Standards, 23 18] 279 (1939); RP 123. 14.2 Mask Design [-V - 0.375" 0.5" SII '0'I I 2.750" pin offset 3.25" 0.75" 0.5" O Pins: 0.122" dia. O Pin holes in cover: 0.125" dia. 3.25" sq. Material: Aluminum, 6061T6 2.750" Cover - 0.250" thick O O Cavity 1: 0.65" x 0.55" on center Depth: 0.032" 12.750" hole offset O O 2.50" sq. 47 3.25" Cavity 2: 0.65" x 0.55" on an (+0.5", +0.75") offset Depth: 0.040" Cavity 3: 0.65" x 0.55" on an (-0.5", -0.75") offset Depth: 0.015" Josh Bennett x2984, rm 4251 15 References 1 Beebe, D. J. and Denton, D. D. A Flexible Polyimide-BasedPackagefor Sensors. Sensors and Actuators. Vol. A44. 1994. pp 5 7 -6 4 . 2 Loehman, R. E. and Tomsia, A. P. Joining of Ceramics. Ceramic Bulletin. Vol. 67, n. 2. pp 375-380. 3 Pask, J. A. From Technology to the Science of Glass/Metaland Ceramic/MetalSealing. Ceramic Bulletin. Vol. 66, n. 11. 1987. pp 1587-1592. 4 Shanahan, M. E. R. Adhesion and Wetting: Similarities and Differences. Rubber World. Vol. 205, n 1. pp 28-36. 5 Field, L. A. and Muller, R. S. Fusing Silicon Wafers with Low Melting Temperature Glass. Sensors and Actuators. Vol. A23. 1990. pp 935-938. 6 Geiger, G. Glass in ElectronicPackagingApplications. Ceramic Bulletin. Vol. 69, n. 7. 1990. pp 1131-1136. 7 Shukla, R. K. and Mencinger, N. P. A CriticalReview of VLSI Die-Attachments in High Reliability Applications. Solid State Technology, July 1985. pp 67-74. 8 Ohring, M. The Materials Science of Thin Films. Academic Press, Inc. San Diego, 1992. 9 Kingery, W. D. Introduction to Ceramics. Wiley. New York, 1976. 10 Chen, W. T. and Nelson, C. W. Thermal Stresses in Bonded Joints. IBM Journal of Research Development. Vol. 23, n. 2. March, 1979. pp 179-188. 11 Frost, H. J. and Ashby, M. F. Deformation-Mechanism Maps. Pergamon Press. Oxford, 1982. p 25.