Wire Bonding: The Ultrasonic Bonding Mechanism Open Access

Deformation in a metal occurs by dislocations moving (slipping) along preferred crystallographic planes within the crystal lattice. Deformation behavior across the spectrum of metals is described by the Homologous Temperature¹  (the % of the melting temperature in Kelvin degrees). Below 40% of the melting temperature metals are rigid, between 40% and 60% they are subject to creep (slow deformation under a load), above 60% of the melting temperature they are unable to support loading. The four main bonding wires are copper, gold, aluminum and silver. Figure 1 shows a graph of the Homologous Temperature. It explains why we bond Al wire at room temperature and gold, copper at approximately 150°C. In each case bonding is at approximately 30% of their respective melting temperatures and the metals have similar deformation properties. Ultrasonic energy provides the additional energy required for deformation and welding.

Early work on the effect of ultrasonic energy on metal deformation was by Langenecker² . Using dog-legged Aluminum Instron tensile bars he mechanically deformed them over both a range of temperature and over a range of ultrasonic energy (room temperature). Figure 2 shows the effect. Langenecker showed that ultrasonic energy was analogous to the effect of heat on deformation, but the temperature did not increase. Langenecker showed that ultrasonic energy increased dislocation density allowing “easy slip” of dislocation planes. A load (force) was still required but it was significantly less than that required for the same deformation at room temperature.

Ultrasonic vibration (amplitude) is transmitted from a stack of piezo-electric crystals (piezo-electric crystals convert an ultrasonic electrical current to a mechanical vibration) through a tapered horn (the taper provides mechanical amplification) to the bond tool.

Early work by Joshi³  using laser interferometry looking for displacement discontinuities from a wedge bond tool tip to the bond interface explained “ultrasonic scrub”. Joshi determined that at the tool/top of wire interface there was a discontinuity. The tool “scrubs” the top surface of the wire. At the wire/substrate interface the motion was different. Almost immediately (<2msec) the two surfaces pinned and there was no net difference in motion. There was no “scrubbing” at the bond interface. Once the surfaces “pin” together Langenecker’s explanation that the two surfaces deform and mix together to form the intermetallic weld is applicable^4,5,6 .

Langenecker¹  provides an equation for acoustic stress induced by an ultrasonic system, Stress is directly proportional with ultrasonic frequency. However, the realities of available ultrasonic systems also generate a significantly smaller ultrasonic amplitude at higher frequency. Ramsey⁷ , et al., found that significantly shorter bonding times could be used for equivalent bond strength. In general, the transition from 60kHz to 120kHz has been implemented. Normal ball bonding times have been reduced from 15–25mS with 60kHz systems to 10–15mS for 120kHz systems. A more important finding is that equivalent bond strength could be achieved with less deformation at the 120kHz system. Where 60–80% deformation was the standard for wedge bonding with 60kHz systems, 20%is now the standard for 120kHz. In ball bonding below 50μm bonded ball diameters 60kHz systems were unable to achieve tightly controlled deformation and achieve high reliability production. Most recent high-speed automatic wire bonders allow selection of 60 or 120kHz ultrasonics for individual bonds. It is possible to select 120kHz for ball bonds and 60kHz for second bonds.

Figure 3 is a block diagram of the ultrasonic system. High frequency electrical signals are generated by the phase-locked loop ultrasonic board. Phase-locking is necessary because as small as a 5⁰ difference in frequency off peak can result in a 95% loss in output. Figure 4 shows transducer Admittance (1/Impedance). Calibration of Impedance is required after every capillary change and if any part of the system is mechanically adjusted. Use of a calibrated torque gage is highly recommended. Designing a transducer to resonate cleanly at a specific frequency (or at frequency multiples such as 60 and 120kHz) without interference from stray modes requires significant FEM modeling.

Today most fine pitch bonding uses 120kHz systems and constant current control mode for driving ultrasonics. Figure 6 compares Constant Current vs Constant Voltage control modes. Constant current has proven to provide much better ball size control for very small ball bonds. Constant voltage is sometimes used for second bonds. They are operator selectable.

Cratering is fracture of the die and bond pad metallization. It can be a deep fracture that lifts a divot of Si die material or with multilayered bond pad metallization⁸  it can be a small fracture several layers below the bond pad. Fractures below the surface are not immediately apparent and through electromigration can lead to device failure by “dark” currents and current leakage. Craters are revealed by etching the bonds from the die (KOH or NaOH for Al bond pads) and examining for cracks or wrinkles in the metallization. Multi-layered bond pads require more detailed procedures or FIB to determine the failure location.

The dominant variables effecting cratering are bond pad metallization⁹  and ultrasonic energy. Aluminum- 1% Si bond pad metallization is not stable. Over time and at elevated temperature Si precipitates out of the matrix as pure nodules. In die that have a thermal SiO₂ layer directly below the bond pad the nodule welds itself to the thermal oxide layer. Any minor ultrasonic-vibration fractures the nodules attachment nucleating a fracture. Fracture growth requires much less energy than nucleation and results in a crater. Koch demonstrated that the use of thin, barrier metal layers like Ti, W, Ti-W between the thermal oxide layer and the Al-Si bond pad resolved the cratering problem. The Si nodules still precipitate but the barrier layer prevented attachment and fracture nucleation.

Of the programmable bonding parameters ultrasonic energy has the largest effect on cratering. Adjusting parameters to minimize ultrasonic energy (smaller bonded ball diameter, higher bond force, higher temperature, longer bond time, thicker bond pads) can help mitigate the effect of ultrasonic energy.

Although market share has been lost to Flip Chip over the past 10 years, wire bonding still represents the largest portion of semiconductor interconnection. Between 75–80% of all packages are wire bonded. It is a complex welding process, joining two metals to form an intermetallic (or monometallic in some cases) weld. Understanding the process requires knowledge of materials, mechanics and statistics. My guidelines for understanding a failure are:

1. Try to determine where within the bond cycle the process is failing.

2. Close examination by SEM with EDS is almost a necessity. Look at both sides of the failure.

3. Look for signs of mechanical damage along the wire loop.

4. Review pull test and shear test data. Pull testing¹⁰  is all resolution of forces. If the pull test hook is placed above the ball, second bond is not adequately tested and may in fact be very weak.

5. Designed experiments are necessary for understanding the process. The principal parameters effecting bonding are: ultrasonics, bond force, temperature and bond time (in order of the strength of their effects). However, many other material, setup and mechanical factors can be critical.