Page 54 - ChipScale_Jan-Feb_2021-digital

Figure 3: Direct imaging for heterogeneous devices: a) Approaches utilizing silicon interposer/EMIB where all dies need a bump; and b) a die-first approach where all
dies are connected by fine RDL.
the pre-measured position data of each die to High-speed compensation exposure position against the original design data,
dynamically generate compensated exposure system dynamically generates the compensated
data without impacting throughput. Using pre- One challenge to date for the mass data for the entire substrate, and executes
measured data and dynamic exposure pattern production of heterogeneous integration the scanned exposure. While exposing,
generation also enables selective patterning, has been the throughput of the lithography subsequent substrates are measured with the
such as selecting pad openings in arbitrary system. Dies on the reconstituted substrate stand-alone tool and compensated exposure
areas, or not exposing die whose displacement have random and independent offsets from data is produced. By utilizing measurement
is too large. mounter accuracy errors and/or movement and exposure processes in parallel,
Dynamic pattern generation supports (expansion/contraction) from the epoxy high throughput, including exposure
unique functions for high-volume mold compound (EMC). These offsets data compensation, is realized. The key
manufacturing as well. One of these is must be measured before exposure and aspects for this approach are high-speed
an auto-numbering function to provide the pattern modified to compensate. This measurement and high overlay accuracy.
traceability as required for critical devices measurement time increases with the The method described above is proven
(automotive, healthcare, etc.). Auto- number of dies on the substrate, further to be very effective at compensating for
numbering functions can simultaneously reducing the throughput of the exposure displacement. For example, we mounted
add unique ID numbers for each die during system. However, we have shown it is 600 dies onto a 515x510mm substrate.
the normal exposure process. Using simple possible to pre-measure the die with a stand- The stand-alone measurement system was
rules, unique ID numbers are automatically alone tool to realize the full advantages of used to measure the displacement of each
generated and applied to individual dies. the direct imaging approach (Figure 2). In die, with results showing displacements
Auto-wiring is another example, beneficial the measurement tool, the exact positions of |Average|+3σ = 81.5µm and 162.0µm in
to advanced packages requiring electrical of all dies are recorded. Twin time delay the X and Y directions, respectively. After
interconnects between multiple dies, each integration (TDI) cameras are used to exposing with the dynamically-generated
with independent displacement. In these scan the substrate and create an image compensation data, overlay results were
packages, die displacement compensation of the entire surface. Our algorithm then |Average|+3σ = 1.9µm and 1.4µm, clearly
alone is not enough—the exposure data for efficiently and rapidly extracts localized showing our system’s ability to flexibly
the interconnect lines will also need to be images, wherein alignment mark matching compensate for die displacement across the
adjusted, and dynamic pattern generation is performed on each die. With this system, entire panel.
with a direct imaging system can do so the measurement time for 5,000 mounted
without impacting throughput. In both the dies on a 515x510 substrate is as low as Summary
advanced packaging and MEMS industry 120s. After measurement, the substrate In this article, we have discussed the
segments, direct imaging’s unique capabilities design and measurement data are loaded flexibility of direct imaging systems
are improving yield and enabling traceability to the direct imaging tool. The direct for advanced packaging and MEMS
of critical devices. imaging tool then evaluates the measured industries. Now, we propose using direct

52
52 Chip Scale Review January • February • 2021 [ChipScaleReview.com]