The vast majority of chips are intended for connection with thermosonic bonds: all other methods require some modification to the wafer. As early as 1972, Jordan described three gang-bonding methods: spider-lead bonding, flip-chip bonding and beam lead bonding. Of these, it is the second which has attracted most continuous development, and is the subject of a separate topic note.
The chip used in spider-bonding is of conventional fabrication, but the aluminium interconnecting leads are preformed as a frame stamped out of aluminium foil. Each chip is placed face down at the bonding station over the frame and positioned with the aid of a key on the edge of the chip. All the connections are then ultrasonically welded in one operation.
The outer part of the frame is cropped to leave just the leads, and the device then mounted face up on the substrate and the leads bonded to the substrate conductors, again in a single welding operation. The requirements for this technology are a high degree of accuracy, consistency in bonding pads, and a tool able to produce accurate multiple welds.
However, the process with greatest potential was perceived in the 1970s to be beam lead bonding. Here gold ‘beams’ protruding beyond the chip edges by about 150µm form an integral part of the device, which is mounted face down on the substrate. Welds to the beams provide both electrical and mechanical attachment. Although the processing of beam lead wafers is more expensive than for conventional devices, claimed advantages were:
As with spider-bonding and TAB, one of the problems with beam lead bonding is the difficulty of generating the correct conditions for bonding each of the leads, so there is much to be said for doing this individually, a practice which negates some of the claimed advantages.
More significantly, the (real) improvements in reliability came not from the construction itself so much as from the need for beam lead processing to include a silicon nitride passivation layer. This approach was later adopted for conventional designs, which had the advantage of needing many fewer processes and a much narrower ‘street’ between dice. In consequence, beam leads are relatively little used, except in special applications such as beam lead microwave diodes.
Unpackaged beam lead diodes have extremely low parasitic capacitance and perform well at frequencies above 10GHz, as well as giving improved performance over their packaged counterparts at lower frequencies. Figure 1 shows a beam lead PIN diode with 15fF capacitance: the device is small, 0.7mm end to end, and mounting or bonding into the circuit requires care. A ‘strong beam’ has a typical pull strength of 10 gm. The diode construction is shown in Figure 2: unlike the usual planar configuration, the cathode is in the same plane as the anode rather than at the base of the chip. This eliminates the die attach process, and the beam lead device is normally bonded upside down.
Three methods can be used for beam lead attachment:
Epoxy bonding does not require high temperature or pressure and is suitable for all substrate types. Also the type of metal and finish of the conductor are not critical, and adequate bond strength can be achieved easily. However, epoxy bonding is less reliable, and temperature cycling over wide temperature extremes can weaken the bond.
Thermocompression bonding is carried out with a bonder similar in style to that used for gold wire ball bonding. Both stage and tool are heated, typical temperatures being stage 225ºC, tool 280ºC. A small (100x50µm) bonding wedge is fitted, with 15g per square .0001 inch applied for 500ms. The optimum parameters have to be established by experimentation, and vary considerably with the bond tool used: a flat bottomed tool is generally better than a concave one.
Because of the relatively high temperature and pressure required, thermocompression bonding is not suitable for soft substrates (such as the PTFE boards much favoured by microwave designers) as deformation may occur, putting excessive strain on the diode. Thermocompression bonding is also sensitive to the type of metals being bonded, and metallisations such copper, or gold less than 2µm thick, are more difficult to bond to than thick gold.
Variations on thermocompression bonding are:
Compliant bonding of beam leaded chips uses an intermediate or compliant member, usually of soft aluminium, between the bonding tool and the beams to be bonded. The function of this is to flow around or ‘conform’ to the beams of the chip in order to achieve a uniform bond with minimal controlled ‘squash out’ (deformation of the lead).
Since there is no ‘neck down’ or other major change in cross section of the bond, the compliant bond is inherently stronger than those made by other techniques. Also a flat heated ram can be used instead of a specific tool for each size of the chip to be bonded, since the thickness of the compliance member is greater than the thickness of the chip.
Compliant bonding has the additional advantage of being able to accommodate process variations in chip manufacture, both from metallurgical and dimensional standpoint. It accomplishes this by presenting a fresh tool face with each bond, and it is the flow stress property of the compliant member that becomes the controlling factor in transmitting pressure to the beam lead.
The material used is annealed aluminium foil which has relatively good thermal conductivity and builds up an oxide which will not stick to the gold during the bonding operation. The thickness of the compliant member must be greater than the silicon chip, but also thick enough to prevent lateral flow as it yields around beam leads, since this could cause distortion of the beam leads during bonding as well as some ‘sticking’ due to the rupturing of the oxide surface at the bond interfaces. A typical tape is manufactured from 125µm thick sheet which is slit into reels and punched to create both chip openings and holes for drive pins.
Whilst laboratory scale compliant bonding can be carried out using aluminium preforms, production equipment uses a continuous reel of aluminium, which is advanced so as to present a clean face for each operation, with the tape held flat against the tool face.
The yield stress for the aluminium compliant member at around 300ºC is approximately 5000psi. To allow for friction and lateral distortion on the load, a pressure of 15,000psi is applied through the compliant member to bond the leads, and a typical force setting is 20-30lbs for a device with up to 50 beam leads.
With compliant bonding, there is minimal deformation of the beam, and reduced squash out means that bugging or vertical movement of the chip due to inward extrusion of the beam material is never excessive.
With these high bonding forces, the work station has to be designed so that the devices are held down against the fixed ground work surface. Referencing on the base of the package may cause slight variations in the height of the bonding plane, and a vertical viewing optical system is therefore required.
All these processes are restricted to hard substrates, as the pressure required is sufficient to deform soft ones.
Resistance welding applies controlled heat and force, with the temperature generated by resistive heating. The term ‘welding’ is actually somewhat misleading, as the mechanism involved is really a thermocompression bond, formed by the diffusion of surface atoms from one metal to another. This provides a bond which is stronger than the beam leaded device itself, and the lower input energy results in lower stress than straight thermocompression.
There are three methods of resistance welding:
Of these three methods, the parallel gap method is the easiest to implement and the least sensitive to electrode size. Electrodes for fixed and parallel gap welding have a rectangular cross-section and should be high in conductivity and hardness and low in contact resistance. For beam lead diodes, tungsten, copper-tungsten or molybdenum electrodes are suggested, although for parallel step gap welding, the lower contact resistance of copper-tungsten helps compensate for the unequal force distribution on the electrodes.
For beam lead diodes, the optimum weld energy is 0.1–0.15J, and best results are obtained with pulse duration below 15ms, as this reduces energy loss due to thermal conduction. This is more serious on ceramic substrates or copper conductors because of the higher thermal conductivity of these materials.
In the process of bonding of beam lead devices, a certain amount of ‘bugging’ occurs (Figure 9). This term refers to the chip lifting away from the substrate during the bonding process due to the deformation of the beam by the bonding tool. This effect is beneficial as it provides stress relief for the device during thermal cycling of the substrate. The coefficient of expansion of some substrate materials, especially softer polymeric boards, is such that some bugging is essential if the circuit is to operate over a wide temperature range.
[ back to top ]