Constant Length Wirebonding for Microwave Multichip Modules

Transcription

1 Intl. Journal of Microcircuits and Electronic Packaging Constant Length Wirebonding for Microwave Multichip Modules S. John Lehtonen and Craig R. Moore The Johns Hopkins University Applied Physics Laboratory Johns Hopkins Road Laurel, Maryland Phone: Fax: s: and Abstract This paper examines a method to automatically make constant length wirebonds in microwave Multichip Modules. In general, the goal in microwave packaging is to have short interconnects for control of impedance and to ensure low gain ripple, low noise figure, or low loss in transmitter power paths. At ever- increasing frequencies, interconnects between different monolithic microwave integrated circuit (MMIC) chips or to substrate traces will behave as strong circuit discontinuities, which require a compensation network. Making the interconnect length constant allows the circuit designer to use the theoretical model of the wirebond as an element of a flexible compensation network. The technique of constant length wirebonding allows the actual wire length to remain the same even though the distance between the first and second bond pad changes due to tolerances in chip size and placement accuracy. The constant length wire is achieved with a programmable automatic wirebonder with special software that allows the user to precisely control the movement of the bonding tool (the wire is spooled out through a hole at the bottom of the tool) throughout the entire bonding cycle. Sample electrical components were wirebonded with the standard looping program and with the constant length wire looping program. Electrical test data show that the samples bonded with constant wire lengths had improved electrical performance. The constant length wirebonding operation is also described in this work. Key words: Wirebond, Electrical Interconnect, Constant Length Wirebonding, Microwave, MCM, and MMIC. 1. Introduction In recent years, great efforts have been undertaken to develop assembly processes for packaging components into microwave Multichip Modules (MCMs), which now are being introduced for communications and radar systems. Constant length wirebonding can improve MCM manufacturability for many of the applications shown in Table 1. From the cellular phone frequency range of MHz to military radars at 10 GHz and beyond, the parasitic inductance of an uncontrolled length wirebond used to electrically connect the various ICs and chip components can cause significant signal degradation. A typical wirebond can lead to loss of power and introduction of high noise levels that will significantly reduce the range and performance of the device. Table 1. The wireless radio bands of current commercial interest are concentrated below 6 GHz. As these frequency bands are occupied, future applications will push operation to higher frequencies. Service Frequencies (MHz) Description Cellular GSM DCS PCS U-ISM U-NII HIPERLAN U-ISM LMDS ,100 31,500 Mobile Communications (N. America) Global System for Mobile Communications (Europe) Digital Communications System (Europe) Personal Communications Systems Unlicensed Industrial Scientific Medical (N. America) Unlicensed National Information Infrastructure Wireless Local Area Network Unlicensed Industrial Scientific Medical (N. America) Local Multipoint Distribution Service This paper presents data from test samples that were wirebonded with both the standard uncontrolled length wire pro- 110 International Microelectronics And Packaging Society

2 Constant Length Wirebonding for Microwave Multichip Modules gram and the constant length wire program. The components on the sample devices had varying gaps between them to simulate conditions found in a typical MCM. This work demonstrates that manufacturability of microwave MCMs can be improved due to the relaxed tolerances for component placement, and enhanced electrical performance can be obtained when the constant length wirebonding technique is incorporated into the microwave MCM manufacturing process. 3. Compensation Network A compensation network that approximates a 50-ohm transmission line consists of two capacitors and an inductor between them as shown in Figure 1. Computer simulation and electrical modeling can be used to calculate the required capacitance and inductance values for the compensation network to improve the circuit performance over a range of frequencies Applications For efficient, point-to-point transmission of power and information from an electro-magnetic wave source, the source energy must be directed or guided 1. Typically in MCMs, microstrip transmission lines are used to guide the signal 2. Microwave circuit designers have selected 50-ohm characteristic impedance as a standard design specification for transmission lines, and most circuits are fabricated with the goal of having the entire microwave signal path be at this characteristic impedance. The signal path on a substrate can easily be made to match the 50-ohm requirement by depositing a calculated trace width on a known substrate material. Many processes currently exist for depositing sufficiently accurate trace widths on various substrate materials. In a typical microwave MCM, electrical interconnects must be made between multiple monolithic microwave integrated circuits (MMICs) and other circuit elements. The problem is that a random length of wire (suspended in air) does not have the characteristic impedance of 50 ohms. As operating frequencies increase, the wires to MMIC chips to substrates, or to package feedthroughs behave increasingly as strong discontinuities 3,4. These discontinuities cause the degradation in the electrical signal transfer. Ideally, all components could be placed side by side with no gaps between them and no variations in component thicknesses to make the interconnect lengths zero. But in the real world, variations due to manufacturing tolerances must be considered. Some gap will always remain due to the variations in both chip sizes and irregular edges. Thermal expansion effects require that the interconnection also have a small loop for stress relief. In addition, the placement accuracy capability of the assembly equipment and the possibility that the electrically conductive attachment medium will squeeze up between the components can also increase this gap. As it becomes larger, the length of the wire between the components must increase. When compensation is not used, the signal degradation worsens as the wire length increases. Eventually, the circuit performance becomes unacceptable and a compensation network is required. Figure 1. Example of typical microwave MCM with interconnects at MMIC chips and 50-ohm substrates and equivalent electrical model of the compensation network. each interconnection between the components will have a unique wirebond length that is determined with computer simulation. The use of a compensation capacitor does not add any steps to the existing substrate metallization process since it is made at same time as the 50-ohm microstrip line is deposited on the substrate. The capacitor, located at the end of the trace, is formed by intentionally widening the microstrip line by a calculated amount. Widening the trace has the effect of making two parallel plates separated by a dielectric, the substrate. This forms the standard structure of a capacitor since the substrate backside is also metallized and usually connected to ground in the MCM package. The widened trace on the substrate also serves as an ideal wirebonding pad. The compensation capacitor on the MMIC (or other chip component) is the parasitic capacitance of its wirebonding pad. The wirebonding pad forms a capacitor to ground because it behaves like one side of a parallel plate separated by the device body, also a dielectric. The inductor is made by the length of the interconnect wire. The inductance of the wire is a function of its length and diameter. For a fixed wire diameter, the inductance increases as the wire length increases. For a flexible compensation network to function properly, the interconnect wire length must be controlled with a great degree of accuracy to obtain the calculated required inductance value. A constant length wire is required to make the interconnection across a variable gap. International Microelectronics And Packaging Society 111

3 Intl. Journal of Microcircuits and Electronic Packaging 4. Electrical Model In the electrical model of a typical MCM interconnect at MMIC chips and 50-ohm substrates (Figure 1), each interconnect between components must be modeled separately resulting in unique wirebond lengths for each case. As stated previously, the inductance value in the model is determined (to first order) by the wirebond length, and the capacitance value is determined by the dimensions of the bond pads. The impedance of the lumped element approximation of a transmission line is proportional to the inductance to capacitance ratio, (L/C). However, the maximum frequency of acceptable performance of this LC circuit is proportional to 1/ (LC). Thus, at higher frequencies, one is concerned with the nominal length of the wire as well as with its variation in length. Placing wirebonds in parallel can reduce the inductance, provided they are separated by more than 10-wire diameters 6 or are angled 7. This is true since the mutual inductance between the wires adds to the inductance, but is inversely proportional to the wire separation and proportional to the cosine of the included angle between the wires. Controlling the placement and length of the wirebonds, the interconnection parasitic can be compensated for in a predictable manner. 5. Wirebonding A typical wirebond, as shown in Figure 2, is defined to include the entire length of wire used to make the electrical interconnection from the one component pad to another. The wire is attached to the component pad metallization using thermosonic bonding, which is the local application of ultrasonic energy and heat to the two materials being joined. A wirebond consists of three segments that are commonly known as the first bond, the second bond, and the wire loop. The test samples for this paper were prepared using a wedgebonding machine and thus wedgebonds will be described here. A ball bonding machine with similar loop control software, as described later, should also provide similar results. 112 Figure 2. Typical wedge-bonded wirebond found in a microwave MCM. When the constant length wirebonding program is used, the total amount of wire spooled out by the machine is fixed. The wire loop height will be high for short bonding distances and lower for longer bonding distances. This allows for wider acceptable tolerances for component placement accuracy. As the term suggests, the first bond is the first part of the wire that is bonded to the component pad. The wire will have a flattened (deformed) portion that will be approximately the length of the foot of the bonding tool chosen, and a short tail will be present at the starting end of the wire. The tail is caused by a small amount of excess wire under the bonding tool. The second bond is the part of the wire that is joined to the other component pad. The wire will also have a flatted area that will be the length of the bonding tool except that no tail will remain at this end. The wire loop is the wire joined between the first and second bond. The wirebonding parameters of gram force, amplitude of the ultrasonic energy applied, and time duration of ultrasonic energy application must be adjusted carefully to obtain strong bonds without excessive wire deformation, which will lead to weak or damaged bonds. 6. Test Samples Three test samples were assembled with varying gaps between a substrate and a 50-ohm resistor chip as shown in Figures 3 through Figure 8. The two components were assembled on a flat metal carrier using silver-filled, electrically conductive epoxy paste. The compensation capacitor on the substrate can be seen at the end of the 50-ohm microstrip transmission line. The distances between the wirebonding pads for the three samples were 350 µm (Figure 3), 430 µm (Figure 5), and 520 µm (Figure 7), respectively. The wire used was a 17.8 µm diameter gold wire. Computer simulation showed that a wire length of 685 ± 50 µm would provide acceptable compensation. Wires were initially bonded to the three samples using the constant length looping parameters. After electrical test, the wires were removed and new wires were bonded with uncontrolled length wire looping parameters. Note that the same test pieces were intentionally International Microelectronics And Packaging Society

4 Constant Length Wirebonding for Microwave Multichip Modules used for both cases to avoid introducing other variables that could cause errors in the electrical measurements. Figure 3. Small gap sample with uncontrolled length wires. No loop in wirebond. Figure 6. Nominal gap sample with constant length wires. Slightly lower loop than Figure 4. Figure 7. Large gap sample with uncontrolled length wires. Very slight loop seen. Figure 4. Small gap sample with constant length wires. High loop in wirebond. Figure 8. Large gap sample with constant length wires. Small loop still present. The thin-film resistor chip was mounted such that the gap between it and the substrate is different on each sample. The Figure 5. Nominal gap sample uncontrolled length wires. distance between the bonding pads was 350 µm (13.5 mils) Almost straight wire. on the smallest size gap sample (Fig. 3 and 4), 430 µm (17.0 mils) on the nominal size gap sample (Fig. 5 and 6), and 520 µm (20.5 mils) on the largest size gap sample (Fig. 7 and 8). The wires in Figures 3, 5, and 7 all have different lengths and vary such that the small gap sample has the shortest wires and the wide gap sample has the longest wires. The wires in Figures 4, 6, and 8 were all bonded at a constant length of 685 µm (27.0 mils). A small loop is desirable in the International Microelectronics And Packaging Society 113

5 Intl. Journal of Microcircuits and Electronic Packaging wirebond for mechanical stress relief and for performing nondestructive wirebond pull testing. 7. Constant Length Bonding The wirebonding was performed with a Palomar model V programmable automatic wedgebonder using the loop mode 3 constant length software option. This option allows the user to precisely control the wire length by controlling the movement of the bonding tool tip as shown in Figure 9. In theory, a constant length wire will be bonded because the machine is designed to spool out the same amount of wire each time for each programmed wire in the MCM. After the first bond is made, the bonding tool will move up and over (a user-programmed amount) before the clamp closes. Once the clamp is closed, no additional wire spools out as the bonding tool continues to travel to the second bonding surface as shown in Figure 10. The minimum and maximum acceptable distances between first bond and second bond will vary depending on the desired wirebond length. Table 2 shows typical distance variations that are acceptable for a fixed wire length. Wirebonds that were made outside this range had inconsistencies in length or breakage of the wire. When the distance is too small compared to the total wire length, the loop may become unusually tall and some of the wire will be forced back into the bonding tool during loop formation. Also, if the maximum distance begins to approach the desired wire length, the wire will be too flat and is likely to tear off at the heel of the first bond. Figure 10. A typical wirebonding tool at the portion of the automatic bonding cycle where the wire clamp has already closed and the tool is descending to the desired 2nd bond location. The clamp operates by pushing the wire against the back of the wirebonding tool, thereby, limiting the motion of the wire. In the open position, the clamp mechanism moves away from the tool forming a gap through which the wire can flow with minimal drag. Table 2. Recommended distances between component bonding pads for a proper wire loop formation. Wire Length 380 (15 ) 500 (20) 620 (25) 750 (30) 880 (35) 1000 (40) Nominal Distance 265 (10.5) 365 (14.4) 435 (17.1) 540 (21.2) 665 (26.2) 795 (31.3) Distance Tolerance ± 65 (±2.5) ± 85 (±3.3) ± 135 (±5.3) ± 160 (±6.3) ± 165 (±6.5) ± 165 (±6.5) It is difficult to measure the actual wire length of the wirebond. However in most cases, it is not as important to know the exact wire length as it is to be able to repeatedly get a certain length. Several iterations are usually needed to get the desired circuit performance. Wires can be bonded and the circuit electrically Figure 9. The bonding tool path on the Palomar 2470-V tested. The first wires can be removed and then new wires of wedgebonder can be precisely controlled using the loop mode different lengths can be bonded and again the circuit electrically 3 optional software program. Parameter A is the initial tested. This process can be repeated until the desired electrical distance that the tool moves up after making the 1st 0bond. circuit performance has been achieved and the wirebonding looping Parameter D is the angle at which the tool moves up. parameters have been developed. Using these determined Parameter B is a distance the tool moves back to form a parameters, additional modules can assembled with a high degree samll bend in the wire. Parameter C is the distance the tool of confidence that electrical performance will be as expected. will move toward the 2nd bond location before the clamp When a compensation network is used, the best results are obtained closes. During movements A, B, and C, wire have been by achieving a high loop for short bonding distances and allowed to freely spool out of the bottom of the tool. After low loops for longer bonding distances to maintain constant length the clamp is closed, however, no additional wire is fed out as wires. the tool continues through the bonding cycle to make the 2nd bond. 114 International Microelectronics And Packaging Society

6 Constant Length Wirebonding for Microwave Multichip Modules 8. Standard Wirebonding Standard wirebond looping programs do not give the user the same level of control that the constant length program provides. In standard looping mode, the loop height parameter is a percentage of the distance between the bonding pads. This results in low loops for short wires and tall loops for long wires, which is fine (and desirable) for low-frequency applications. To overcome this variation in loop heights, many modern microwave MCMs are assembled with a requirement for very tight tolerances for component placement accuracy. If the components are consistently attached with the same size gaps in every module the end result will be constant length wirebonds. This places an extra burden on the assembly process and adds extra inspection steps to insure placement accuracy better than ± 50 µm. The accumulated manufacturing dimensional tolerances of all the circuit components including the package, substrates, and MMICs make it very difficult to maintain such tight tolerances and adds cost due to lower assembly yields, and the resulting rework to remove/re-install components and re-inspecting the module. 9. Electrical Test Results Figure 11. Measured voltage reflection coefficient and percent difference from modeled performance of uncontrolled length wires. The modeled wire lengths were 685 m (27.0 mils). The uncontrolled length wires were bonded using a standard wire looping program. Figure 12 shows the effect of constant length wires on the measured voltage reflection coefficient and the difference from the modeled inductance value. The reflection coefficient was within 10% of the modeled figure for most of the frequency range. Again, due to limitations on test equipment, it was not possible to test at higher frequencies. However, these results do demonstrate that the flexible compensation network behavior can be very closely predicted. In order to evaluate the test samples, the voltage reflection coefficient was measured. This reflection coefficient is a quality measure of a microwave interconnection and ranges in value from 0 to 1. Zero reflection coefficient means that all of the microwave energy presented to the interface is transferred across. A reflection coefficient of 1 means that all the energy presented is reflected and none is transferred. A good reflection coefficient is considered less than 0.1. Since voltage squared is proportional to power, this means that 1% of the source power is reflected and 99% transferred. The test samples were installed in a coaxial test fixture and connected to a vector network analyzer to measure the voltage reflection coefficient over a frequency range of 0.2 to 6 GHz. The coaxial connector is included in the measured data and also in the model. Figure 11 shows the effect of uncontrolled length wires on the measured voltage reflection coefficient and the difference from the model, which assumed the wire length is equal to the longest wirebond length. As expected, these three samples varied considerably from the modeled performance. Two of the samples varied more than 25% over a large portion of the frequency range. It would have been desirable to test the samples at higher frequencies because these differences will become more pronounced as frequencies increase. However, due to limitations on test fixturing, the higher frequency testing could not be performed at this time. Figure 12. Measured voltage reflection coefficient and percent difference from modeled performance of constant length wires. The modeled wire lengths of 685 m (27.0 mils) were bonded between a microstrip transmission line with end compensation and a 50-ohm thin film chip resistor connected to substrate ground. Wires of this length were implemented using the constant length wirebonding software option on the same samples as used in Figure Implications for Future Antenna Systems The control of parasitic inductance through constant length wire bonding will have a significant impact on future wireless International Microelectronics And Packaging Society 115

7 Intl. Journal of Microcircuits and Electronic Packaging technology. For example, an active phased array antenna makes use of a number of antenna elements with a small transmitter and receiver immediately adjacent to the antenna element. Controlling the electrical delay, or phase shift, through the active elements with a phase shift control circuit in each element, one can create an electrical delay pattern across the antenna face 8. This delay pattern enhances the performance of the antenna in a desired direction while suppressing the performance in all other directions. Changing the delay pattern by electronically controlling the phase shifter in each element instantly changes the direction of the antenna. In addition to military radar and satellite communications applications, this technology can greatly enhance commercial communications systems. For instance, a cellular telephone operator could rapidly adapt the performance of the tower antenna in response to traffic increases and decreases in different directions. This performance optimization would permit higher traffic density without additional towers. The implementation of this technology in commercial systems awaits reductions in the cost of manufacturing an antenna with tens to hundreds of active elements. The researchers recently fabricated a number of MMIC transmit and receive elements and assembled them into a small array to demonstrate the feasibility of a low-cost active array for a microwave communication system. The transmit modules are shown in Figure 13. An important performance parameter in active array antennas is the variation in amplitude and electrical delay, or insertion phase, from unit to unit. Any variation between array elements must be measured and an appropriate correction made that is unique for each element when its phase shifter is set. Any amplitude variations must be corrected with an attenuation control circuit in the module. Failure to do so will result in inferior antenna performance. The authors utilized the constant length feature for all wire bonds in the RF path of these modules that are shown in Figure 14. The resulting uniformity of these modules is shown in Figure 15. The root mean square (RMS), or 1 sigma, gain and phase variation in degrees between 11 transmit modules, measured at each of 401 frequency points is plotted. The insertion phase variation was nominally 15 degrees RMS over the 4 to 6 GHz frequency range. The accompanying amplitude variation was approximately 0.75 db RMS. Similar uniformity was realized for the receive modules. Figure 14. Gaps betweem MMIC die can vary when constant length wirebonding program is used. Figure 15. Electrical test results of 11 transmit modules designed and assembled at the John Hopkins University Applied Physics Laboratory. The RMS gain and phase variations were measured to be well below the values needed for acceptable active antenna array performance. Therefore, an individual attenuation control circuit in each module is not needed and phase correction for each antenna element was not necessary. These electrical variations are well below the values needed for acceptable active phased array antenna performance. The implication for cost savings is significant in that an attenuation control circuit in each module is not needed. Further, the tedious tasks of measuring the amplitude and phase variation of each module and the bookkeeping of the resulting correction factor can be eliminated. Also, the added memory and computation associated with element correction in the antenna controller would not be needed. The use of constant length wire bonds was directly responsible for the improvement in performance uniformity of these modules and thus has the potential to significantly lower the cost of future active phased array antenna systems. 11. Conclusions Figure 13. A microwave transmit module fabricated to The data presented in this publication demonstrates that with demonstrate feasibility of a low-cost active array for a the use of constant length wirebonding techniques, a flexible compensation network can be designed into a microwave MCM to shipboard microwave communication system. 116 International Microelectronics And Packaging Society

8 Constant Length Wirebonding for Microwave Multichip Modules improve the quality of the signal transfer across a typical interconnection. A wirebond can be treated as a fixed value inductor in the electrical model used for computer simulation of the interconnection. Furthermore, significant cost savings can be realized in the fabrication of modules due to less complicated circuitry in module, higher assembly yields due to relaxed component placement accuracy, and less electrical test data manipulation. 12. Future Work in Microwave Wirebonding Additional work in the microwave wirebonding is planned in the coming year particularly in the area of higher frequency effects. The researchers will further analyze wirebond geometry changes and repeatability of the constant length wirebonding process. The microwave probe station and the spectrum analyzer are being upgraded to enable testing through 26 GHz in the near term. (eventually plan to have test capability through 50 GHz.) Acknowledgment The authors would like to acknowledge the support of Katherine J. Mach of APL s Electronic Services Group for sample preparation and David M. Verven of APL s Advanced Systems Development Group for electrical test. References 5. F. Alimneti, et al., Quasi Static Analysis of Microstrip Bondwire Interconnects, Proceedings of the 1995 IEEE-Microwave Theory and Techniques Digest, pp , Steve Nelson, Optimum Microstrip Interconnects, Proceedings of the 1991 IEEE Microwave Theory and Techniques Digest, pp , Sang-Ki Yun, Parasitic Impedance Analysis of Double Bonding Wires for High Frequency Integrated Circuit Packaging, IEEE Microwave and Guided Wave Letters, Vol. 5, No. 9, pp , September Bruce Kopp, et al, Transmit/Receive Module Packaging: Electrical Design, Johns Hopkins Technical Digest, Vol. 20, No. 1, pp , About the authors S. John Lehtonen received a BSEE from Florida Atlantic University in He has 15 years of experience working in the microelectronics packaging field. Mr. Lehtonen is a Senior Staff Engineer with the Johns Hopkins University Applied Physics Laboratory where he is involved with the design and fabrication of high-reliability microelectronics for military and space applications. His current interests are assembly process development for a variety of electronic miniaturization efforts which include microwave MCMs, chip-on-board, indium bump bonding, and flip chip technology. His address is john.lehtonen@jhuapl.edi Craig Moore received the B.E.E. and M.S. Degrees from Cornell University and has over 35 years experience in microwave circuit and system design and measurement. Currently, Mr. Moore is a principal staff engineer with Johns Hopkins University Applied Physics Laboratory, where he is involved with new technology for future military radar and communications systems. His address is craig.moore@jhuapl.ed 1. David K. Cheng, Field and Wave Electromagnetics, Addison-Wesley Publishing Company, Reading, Massachusetts, Chapter 9, pg. 370, Wolfgang Menzel, Interconnects and Packaging of Millimeter Wave Circuits, IEEE Microwave Theory and Techniques Newsletter, No. 149, pp , Summer G. Baumann, et al., 51 GHz Front-end with Flip Chip and Wire Bond Interconnections from GaAs MMICs to a Planar Patch Antenna, Proceedings of the 1995 IEEE Microwave Theory and Techniques Digest, pp , T. Krems, et al., Millimeter-Wave Performance of Chip Interconnections Using Wire Bonding and Flip Chip, Proceedings of the 1996 IEEE Microwave Theory and Techniques Digest, pp , International Microelectronics And Packaging Society 117