Gold Ribbon Bonding for Microelectronic Interconnection: A Designed Experiment to Evaluate Process Opportunities

Transcription

1 Gold Ribbon Bonding for Microelectronic Interconnection: A Designed Experiment to Evaluate Process Christina M. Conway and Nicole L. Cavanah Rockwell Collins, Inc. 400 Collins Road NE, M/S 08-0 Cedar Rapids, Iowa Phone: /342 Fax: /6060 s: cmconway@collins.rockwell.com and nlcavana@collins.rockwell.com Abstract This paper presents results from a Design of Experiments (DOE) study undertaken to evaluate the feasibility of, and limits of applicability for, making gold ribbon bonds using a manual wirebonder. The objectives of this study included the following: Establishing the viability of using a manual wire bonder to make gold ribbon bonds. Determining preliminary equipment process parameters for making robust and consistent ribbon bonds. Investigating the effect of different ribbon types on these process parameters and the resulting bonds. Establishing limits on the resultant interconnection geometries that can be achieved using this process. Determining whether aging affects ribbon bond pull strength. All bonds were made on gold-metallized alumina substrates. The variables evaluated were ribbon size, ribbon supplier/alloy/elongation, bond length, and equipment settings for force, time, temperature, and power. Results show that gold ribbon bonds with good pull strengths can be achieved using the manual wirebonder. Ribbon type affected the pull strength consistency of the resultant bonds. Aging the bonds for 6 hours at 25 C did not degrade pull strengths. Some bond-to-bond variation in loop height was observed and is reported. Since the degree of ribbon bond variation that can be tolerated depends on the specific interconnection requirements of the intended product, the applicability of this process must be considered on a case-by-case basis. Key words:. Introduction Gold Ribbon Bonds, Manual Wirebonding, Design of Experiments, Bond Pull Strengths, and Metallized Alumina Substrates. Wirebonding remains the microelectronics industry s predominant method for accomplishing electrical interconnection between an integrated circuit and the first level package. Ribbon bonding, a subset of wirebonding that utilizes wire with a rectangular cross section as the interconnection medium, has been championed for a number of microelectronic applications. From an electrical performance perspective, ribbon bonds can offer lower impedance and greater current carrying capacity as compared to round wirebonds of similar size,2. Mechanical advantages of bonding with ribbon wire compared to round wire include: lower incidence of heel cracking, the ability to span longer distances with minimal sagging, and better reliability after high tempera- The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN ) International Microelectronics And Packaging Society 435

2 Intl. Journal of Microcircuits and Electronic Packaging ture storage 3, 4. As a result, ribbon bonding is the preferred interconnection method for many microwave and power hybrid devices. Process, material, and equipment limitations govern the geometry, strength, and reliability of ribbon bonds. Since product performance, yield, and cost are directly related to the interconnect process, this study was undertaken to establish the practical limits of using a manual wirebonder to generate consistent-geometry gold ribbon bonds with good pull strengths. 2. Investigation A Design of Experiments (DOE) was used to evaluate the Ni feasibility of, and limits of applicability for, making gold ribbon Pd bonds using a manual wirebonder. The objectives of this study Si <6 --- are listed as follows, Ag <26 8 Establish the viability of using a manual wirebonder to make gold ribbon bonds. Table 2. Comparison of mechanical properties for 99.99% Determine preliminary equipment process parameters for gold ribbons. making robust and consistent ribbon bonds. Investigate the effect of different ribbon types on these process parameters and the resulting bonds. % g % g Spec d Spec d Tested Tested Manufacturer/ Elong. Strength Elong. Strength Size µm x µm Müller Establish limits on the resultant interconnection geometries Feindraht/ that can be achieved using this process. 27 x Determine whether aging affects ribbon bond pull strengths. Williams/ 3 max 45 min x 2.7 A primary goal of the study was to evaluate ribbon bond configurations that are representative of anticipated product appli- Feindraht/ * Müller cations. Seven controllable process factors were identified for 254 x 25.4 Williams/ investigation: ribbon size, ribbon type (identified by manufacturer 254 x max 200 min and elongation), ribbon bond length, and equipment pa- rameters for force, time, power, and temperature. Two different ribbon sizes were selected for evaluation: a fineand *5g over upper spec limit a heavy-gage ribbon. The fine-gage ribbon measured 27 Prior to bonding, the input temperature was set at the workholder. Test coupon surface temperatures were slightly lower µm x 2.7 µm (0.005 in. x in.) and the heavy gage ribbon measured 254 µm x 25.4 µm (0.00 in. x 0.00 in.). than work-holder input temperatures as follows: work-holder The gold ribbon used in this study was obtained from two input temperatures of 20 C, 50 C, and 90 C corresponded to different manufacturers: Müller Feindraht AG and Williams test coupon surface temperatures of 3 C, 36 C, and 7 C, Advanced Materials. All ribbon was specified as 99.99% gold respectively. Table 3 lists the specific two level input variable (Au). The original intent was to evaluate two sizes of the same settings for force, power, time, and temperature that were evaluated in this DOE. kind of ribbon (99.99% Au) from two different manufacturers. However, although identical ribbon was requested from each Table 3 also indicates the ribbon bond lengths that were evaluated in this study. Prior to the DOE, a preliminary evaluation manufacturer, the supplied ribbon contained different alloying elements and/or amounts of alloying elements and was processed was performed to establish what bond lengths the equipment could by methods specific to the individual manufacturer. As a result, repeatedly produce. It was determined that, for bonding from each manufacturer s ribbon behaved differently. Alloy compositions and mechanical properties for the ribbon, as provided by one pad on a substrate to a second pad on the same substrate (both pads at the same elevation), the shortest bond that could be the manufacturers, are presented in Tables and 2, respectively. attained with reasonable consistency measured approximately 508 The bonder used to produce all of the ribbon bonds for this µm (0.020 in.). When one attempted to produce bonds shorter study was a Kulicke & Soffa (K&S) Model 4526 Auto-Stepback than this limit, the bonder routinely lost its ability to control loop Wedge bonder. The bonder was configured with a 90 wire feed height. Rather than producing loops that measured approximately angle kit for deep access bonding and outfitted with the appropriate software to execute table tear wire termination. Settings µm ( in.), the bonder generated looping arches greater than 254 µm (0.00 in.) high. for force, power, and time could be controlled independently for The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN ) 436 the bond (first and second) at each end of the ribbon. In this evaluation, all of the first and second bond settings for force, power, and time were dialed in identically; that is, if the first bond power was set at.5, the second bond power was also set at.5. Table. Comparison of alloy content (in ppm) for 99.99% gold ribbons. Alloying Element Müller Feindraht Williams Al Be <2 3 Ca < --- Cu <50 7 Fe <5 --- Mg <5 5 International Microelectronics And Packaging Society

3 Table 3. Factors evaluated in the two-level designed experiment. Ribbon Size µm (in.) Manufacturer/ % Elongation Force [dial setting] Time [dial setting] Power [dial setting] Temperature C Length µm (in.) 27 X 2.7 (0.005 x ) Müller Feindraht/ X 25.4 (0.00 x 0.00) Williams/< (0.020) 380 (0.50) Since short bonds with low loop heights were desired for microwave interconnect applications, 508µm (0.020 in.) was established as the minimum bond length for this study. Even with this established limit, while making multiple consecutive 508- µm (0.020 in.) ribbon bonds, the bonder occasionally produced a bond with a significantly higher loop. Since this was readily apparent to the operator and occured relatively infrequently, it was deemed not a problem for the purposes of this DOE. The longest ribbon bond that could be produced consistently was 380 µm (0.50 in.). This limit resulted directly from the equipment design. The total stage travel was restricted to approximately 4200 µm (0.65 in.) by the equipment manufacturer 5. All ribbon bonds were made on test coupons fabricated from 635-µm (0.025 in.) thick, as-fired alumina substrates patterned with metal traces. The metallization scheme consisted of angstroms of titanium tungsten (TiW) in contact with the alumina. The TiW was overplated with angstroms of palladium (Pd). A.25-µm (50 min.) thick layer of sputtered gold (Au) over the Pd formed the top layer of metallization. Metal traces on the test coupon were 254 µm (0.00 in.) wide and spaced in a staggered sequence so that the center-to-center distance between the traces was 508, 260, and 380 µm (0.020, 0.085, and 0.50 in.), respectively. Throughs were cut between and parallel to the 508-µm (0.020-in.) spaced traces to provide adequate clearance for positioning a test hook underneath the short, lower-loop ribbons during subsequent bond pull strength tests. The test coupon configuration is shown in Figure. A one-half fraction of a full factorial design was selected for this study. In this experiment, ribbon bond length was confounded with the fifth order interaction among the remaining main effect variables. This DOE required 2 7- = 64 combinations of input variables. In addition to the 64 combinations of the one-half fraction design, 2 mid-point combinations were included to provide information regarding the degree of linearity of the results. Twenty ribbon bonds were made using each combination of variables, for a total of 520 [(64 + 2) x 20] ribbon bonds. The measured output response for this study was ribbon bond pull strength. Bond pull strength testing was accomplished in accordance with MIL-STD-883E, method 20.7, test condition D. Per MIL-STD-883E, the minimum bond pull strength limits for the 27 µm x 2.7 µm (0.005 in. x in.) and the 254 µm x 25.4 µm(0.00 in. x 0.00 in.) gold ribbon bonds are 6 and 20 grams-force, respectively. Execution of the DOE proceeded as follows. Prior to ribbon bonding, each test coupon was ultraviolet ozone (UVO) cleaned for 0 minutes to remove any organic material that might inhibit bond formation. After UVO cleaning, ribbon bonds were made in accordance with the specified input variable combinations per the designed experiment. The design combinations were executed in random order to the extent practical for the process. All ribbon bonds for the DOE were made within eight hours of beginning the experiment. Confirmation runs of promising combinations were made at a later date. A total of twenty ribbon bonds were fabricated for each design combination. Ten bonds were tested in the as-bonded condition and the pull strengths recorded. The remaining bonds were aged, by baking at 25 C for 6 hours, and then pull tested. Among combinations that yielded bonds with non-zero pull strength values, the failure mode was first bond heel breaks with only two exceptions. Both exceptions were second bond heel breaks. 3. Results Table 4 summarizes the data resulting from the one-half fraction DOE. Only data from input variable combinations for which every bond yielded a non-zero pull strength are reported. The data are sorted first by ribbon size, then, by manufacturer/elongation. Within these groupings, the data are ranked, in descending order, by the product of the minimum recorded pull strength of as-bonded and post-aged bonds for each input variable combination ( Min. Min. in Table 4). Figure. Schematic depiction of ribbon bond test coupon. Dimensions are in millimeters [inches]. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN ) International Microelectronics And Packaging Society 437

4 Intl. Journal of Microcircuits and Electronic Packaging Table 4. Data matrix for the one-half fraction DOE. Order # Ribbon Size Manuf./ % Elong. Input Variables Force Tim e Pow er Temp C Bond Length Pull Strength Data (grams) As-bonded Post-aged Avg Min. Max. Avg. Min. Max. Min Min MF/ MF/ MF/ MF/ MF/ MF/ MF/ MF/ MF/ W/< W/< W/< W/< W/< W/< W/< W/< W/< W/< W/< W/< MF/ MF/ MF/ MF/ MF/ The results of an analysis of variance are presented in Figure 2. This Figure depicts a comparison of the resultant mean pull strengths for all bonds made using the half-fraction DOE combinations. The as-bonded condition is referred to as Bef Bake and the post-aged condition is referred to as Aft Bake. The error bars indicate ninety-five percent confidence intervals. No significant difference can be discerned between the mean pull strengths of ribbon bonds in the as-bonded condition and those in the post-aged condition. Figure 2, in combination with other statistical analyses, supports the conclusion that aging the ribbon bonds for 6 hours at 25 C resulted in no significant change in the bond pull strengths. Therefore, all subsequent work in this study was performed using bonds in the as-bonded condition. The half-fraction DOE identified a number of input variable combinations for 254 µm x 25.4 µm (0.00 in. x 0.00 in.) ribbons that can produce bonds with satisfactory minimum pull strengths. Top candidates from each manufacturer/elongation category were re-run to confirm the results of the DOE. Some bond length restrictions were imposed. Only 260 µm (0.085 in.) and 380 µm (0.50 in.) lengths were used for the confirmation runs. To provide visibility to the potential process window for the various combinations, each candidate input combination was bonded at both the 260 µm (0.085 in.) and 380 µm (0.50 in.) lengths. 438 Mean Mean Means Plot for Bef Bake with 95.0% Confidence Intervals Means Plot for Aft Bake International Microelectronics And Packaging Society StdOrd StdOrd with 95.0% Confidence Intervals Figure 2. Plots of mean pull strengths for ribbon bonds in the as-bonded and post-aged conditions. The results of the 254-µm (0.00 in.) ribbon confirmation runs are presented in Figure 3. Figure 3 is a plot of the mean pull strengths with ± 3 standard deviation limits for seven input variable combinations. The combinations are identified by the order number (given in Table 4) followed by the bond length in micrometers; that is, is order number 74 bonded at 380 The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN )

5 µm. The pair of combinations highlighted by the arrows represents the best overall combination, providing the highest average pull strength with the lowest standard deviation for both bond lengths [260 µm (0.085 in.) and 380 µm (0.50 in.)]. Grams-force Figure 3. Mean strengths with 6-sigma variations for 254- micrometer ribbon confirmation runs. Run 4. Fine-Gage Ribbon Bond DOE Of the thirty-eight combinations tested for each ribbon size, twenty-one of the 254-µm (0.00-in.) ribbon combinations met the minimum pull strength requirement of 20 g, while only five of the 27-µm (0.005-in.) ribbon combinations had a measured value of at least 6 g. Only ribbon from Müller Feindraht that spanned 508 µm (0.020 in.) produced bonds that consistently yielded pull strength values greater than zero. To optimize parameters for the fine-gage ribbon bonding process, a second, threelevel DOE was run. A central composite design was chosen for evaluating the 27- µm (0.005-in.) ribbon process. Building on results from the halffraction DOE, the central composite design fixes the ribbon type and bond length at Müller Feindraht/-4% elongation and 508 µm (0.020 in.), respectively. All procedural aspects of the experiment were conducted as described for the half-fraction DOE. Table 5 presents the input variable combinations for the central composite DOE and the data resulting from the experiment. Ten ribbon bonds were tested at each combination. A statistical analysis of the data provided the following conclusions; The input variable combination highlighted in Table 5 (F = 2, t = 3.5, P = 2.75, and T = 50) is the best combination from the experiment for both mean and standard deviation. From studying interaction and cube plots, the recommended best settings for the confirmation runs are F = 2, t =, P = 2.75 and T = 40. Table 5. Central composite DOE matrix. Pull Strength (grams-force) Std Min Dev International Microelectronics And Packaging Society 439 Force 2 7 Time Power Temp ( C) Avg Max The two combinations identified from the statistical analysis were re-run to confirm the results of the DOE. Ten ribbon bonds were produced for each combination and tested. The confirmation run results are presented in Table 6. Pull test failure modes were predominantly first bond heel breaks with a few second bond heel breaks (2/0 and /0 for the respective combinations in Table 6). Table 6. Central composite DOE confirmation runs data. Force Time Power Temp ( C) Avg Pull Strength (grams-force) Std Min Max Dev Both combinations in Table 6 yielded good average bond pull strengths. Applying ±3 standard deviation limits implies that either combination would consistently yield bonds with pull strengths greater than 6 g in production. However, it should be noted that the measured average pull strength for the combination F = 2, t = 3.5, P = 2.75, and T = 50 was significantly larger The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN )

6 Intl. Journal of Microcircuits and Electronic Packaging (to a 99% probability level) during the confirmation run than during the initial tests (Table 5). The difference is suspected to be attributable to overall bond loop height variations between the runs. More evaluation is needed to confirm this hypothesis and to ensure that a robust, consistent process has truly been identified for the 27-µm (0.005-in.) ribbon. 5. Conclusions A study was performed to determine the feasibility of, and limits of applicability for, making gold ribbon bonds using a manual wire bonder. Two designed experiments, a half-fraction of a full factorial and a central composite design, were used to evaluate preliminary process parameters for making 254 µm X 25.4 µm (0.00 in. X 0.00 in.) and 27 µm x 2.7 µm (0.005 in. x in.) ribbon bonds, respectively. The following conclusions were reached; The 254-µm (0.00-in.) ribbon lent itself to a more robust and consistent process, Aging the ribbon bonds for 6 hours at 25 C did not degrade resultant pull strengths, Of two candidate 99.99% Au ribbon types, the Müller Feindraht/-4% elongation yielded more bonds with higher overall pull strengths under the constraints of these experiments, Ribbon bond lengths are restricted to a minimum of 508 µm (0.020 in.) by bond loop considerations and to a maximum of 380 µm (0.50 in.) by equipment design, Loop height showed significant variation, particularly in short ribbon bonds. In application, the required degree of bond loop control may ultimately determine the applicability of these ribbon bonding processes for use in specific products, and Further evaluation of the preliminary 27-µm (0.005-in.) ribbon bond process is warranted to ensure that it is repeatable and robust. Acknowledgments References. George G. Harman, Wirebonding in Microelectronics: Materials, Processes, Reliability, and Yield, McGraw-Hill Publishers, Second Edition, New York, Chapter 2, pp. 3-32, Donald J. Beck, The Case for Ribbon Bonding of Large Packages to PCBs, Surface Mount Technology Magazine, Vol. 8, No. 4, pp , April Bradley K. Benton, Automated Ribbon Bonding, Advanced Packaging Magazine, Vol. 8, No. 3, pp , March David C. Guidici, Ribbon Wire versus Round Wire Reliability for Hybrid Microcircuits, IEEE Transactions on Parts, Hybrids, and Packaging, Vol. PHP-, pp.59-62, June Private telephone conversation with Mr. Joe Martin, K&S, August 6, 999. About the authors Christina M. Conway is a Senior Materials and Process Engineer at Rockwell Collins in Cedar Rapids, Iowa. Ms. Conway earned her Master of Engineering Degree in Metallurgical Engineering and Materials Science from Carnegie-Mellon University in 986, and her Bachelor of Science Degree in the same discipline from the University of Notre Dame in 984. Previously, she held positions at USBI in Huntsville, Alabama and LSI Logic Corp. in Fremont, California. Ms. Conway joined Rockwell Collins in 996. She conducts research and development activities related to advanced packaging and interconnect technologies. Her current research interests involve interconnect and packaging of microwave and millimeter-wave devices. Ms. Conway has authored or co-authored several published technical papers and is a member of IMAPS. The authors gratefully acknowledge the contributions of the Nicole L. Cavanah is a Materials and following individuals: Mrs. Patricia Feller who made, measured, Process Development Engineer in the and pull tested all of the ribbon bonds that were used in this Advanced Technology Center of study, Mr. Melvin Stanard who performed the statistical analyses reflected in this paper, and Mr. Joe Martin and Mr. Lee Levine, communication and aviation electron- Rockwell Collins, a manufacturer of of Kulicke & Soffa, who provided guidance and input regarding ics, in Cedar Rapids, Iowa. Ms. the wirebonder and ribbon wire. Cavanah graduated from Iowa State University with a Bachelor of Science degree in Ceramic Engineering in 993 and is currently pursuing a Masters degree in Material Science and Engineering. She joined Rockwell Collins in May 993 as an Advanced The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN ) 440 International Microelectronics And Packaging Society

7 Operations Process Engineer responsible for microelectronic processes. In her present assignment, Ms. Cavanah is responsible for development of new interconnect and assembly processes, including material evaluation, training, analysis, and process verification. She serves as a technical expert on call for ceramic materials, microelectronics assembly, and cleanrooms. Ms. Cavanah was recognized for her achievements as a Rockwell International DaVinci Engineer of the Year for 999, Rockwell Collins YWCA Tribute to Women in Industry Honoree for 998, and Rockwell Collins DaVinci Engineer of the Year for 996. A patent was awarded in 999 for High Strength Au Wire for Microelectronics. She is a member of the Iowa State Material Science & Engineering Industrial Advisory Council, Mercy Women Center Community Advisory Board, and the International Microelectronics And Packaging Society. As a member of IMAPS, she has served on the technical paper selection committee, presided as a session chair, and presented several papers. The International Journal of Microcircuits and Electronic Packaging, Volume 23, Number 4, Fourth Quarter, 2000 (ISSN ) International Microelectronics And Packaging Society 44