Page 41 - ChipScale_Jul-Aug_2020-Digital

2. A dual-headed picker that can keep the number of nozzles. For just a few nozzles, Separation of pick and place
12 place nozzles supplied with dies. die transport and dip fluxing dominate As with most modern FC bonders, the
3. For each bond head, a die distributor the cycle time and adding a few more pick and place processes are separated to
and a set of two intermediary die nozzles dramatically improves throughput. allow them to occur in parallel. Through
shelves; the pickers and distributors With more and more nozzles, however, the use of die distributors and intermediary
populate die onto one set of these the necessarily sequential die placement die shelves, we took the separation of these
shelves, while the bond heads can processes eventually dominate the cycle processes one step further. Each bond
pick die from the other set of shelves. time, and adding more nozzles results in head has a dedicated picker, distributor and
4. Two fluxers, one for each head, for diminishing returns. For a 200ms place two die shelves. The sequence is that the
simultaneous dipping of all six dies time, going from two to four nozzles dual-headed picker hands the die over to a
on a bond head. delivers a 61% improvement in throughput, distributor, which then places it onto a shelf.
5. An up-looking die alignment vision but from 8 to 16 nozzles, the improvement The spacing of the die on this shelf matches
system that uses a strobing light is only 29%. Between 12 and 16 nozzles, that of the nozzles of the place head. Once a
source to capture images on the fly the improvement is merely 9% for a 33% shelf is populated with the 6 dies, it moves
(i.e., the bond head does not stop increase in the number of nozzles. backwards ready for pick up by the place
over the camera). Clearly, the throughput advantages of head with simultaneous transfer of all six
adding more nozzles need to be balanced dies. At the same time, the distributors
The most common type of FC bonders in against the added cost and complexity of start populating the second shelf. The same
the market only have two bond heads, each designing in and operating more nozzles. sequence happens in parallel for the other
with a single nozzle. With this architecture, From a machine design perspective, the bond head. This system allows the bond
dies are taken individually through the challenge here is mainly associated with the heads to run essentially independently of
sequence of pick up, transport to the specific requirement for co-planarity of the nozzles, each other and also provides a small buffer
site on the substrate, alignment, placement, as well as maintaining placement accuracy of dies should the pickers lag the placers
and return to the pick up site. Clearly, a for all nozzles without the need to dial in during some sequences of the process. Of
multi-head bonder is faster as it can take accuracy independently for each nozzle. course, these die shelves also enable the
several dies simultaneously through most of From a user’s perspective, adding more use of only two pickers for 12 nozzles. A
these steps. Table 2 summarizes which steps nozzles increases the cost of application- direct handover between picker and placer
in the bonding sequence are performed on all specific tooling, especially for the place would necessitate 12 pickers, which would
of the dies in unison, and which are carried tools themselves. be prohibitively expensive, or a sequential
At this point, it is worth highlighting handover, which would be far too slow.
another aspect of a multi-nozzle FC bonder
that process and production engineers Ease of use
should take to heart: the throughput reward A multi-nozzle system requires detailed
for reducing place time is much larger attention to ease of use, especially when
than for a single nozzle per bond head it comes to teaching or changing over
applications. With Katalyst™, we realized a
system that allows the operator to teach the
placement with a single nozzle and for that
teaching to be copied to the other 11 place
heads. Of particular importance is that the
accuracy concept of the machine allows for
Table 2: Parallel processing steps per bond head. attaining accuracy without dialing in offset
for each place head individually. This is
out sequentially for each die on a bond head. accomplished by the use of an automatic
Of course, each of the two bond heads work calibration system that determines any
in parallel for all process steps. Aligning the differentials in the placement process of
die to the specific bonding site and actually Figure 2: Throughput as a function of the number of the individual placers. While production
placing the die must, necessarily, be carried nozzles for 100, 200 and 300ms place times. engineers will, generally, still verify
out in sequence as the pitch of the die on the placement accuracy of each of the nozzles,
bond head cannot match the pitch of the bond system. Figure 2 illustrates this point. The a single x,y placement offset is required for
locations on the substrate. In Katalyst™ curves for 100ms and 300ms place times all nozzles.
we have realized a system in which die strongly diverge as the number of nozzles The bonder also has fully-automated
placement is the only portion of the sequence increases. For a 12-nozzle system, the pick and place tool exchangers complete
that is not carried out in parallel. As will 100ms process delivers a 42% throughput with the use of ID codes to verify the use
be discussed later, parallel fluxing required improvement over the 300ms process! In of the correct tooling. The bonder can
challenging design innovations. the case of two nozzles (one per bond head), hold a full set of tools for five applications
the improvement is only 14%, because the at a time. This along with automated
Why 2 x 6 nozzles? majority of the time is used on die handling, or semi-automated verifications allows
Figure 2 shows the throughput of a two- alignment and fluxing, not placing the die. application changeovers in 20 to 30min, as
bond head bonder as a function of the total summarized in Table 3.

39
Chip Scale Review July • August • 2020 [ChipScaleReview.com] 39