inputs and outputs often interact, making the overall global
alignment a moving target.
For previous alignment automation technologies, this would
dictate a time-consuming sequential loop
of aligning this and then aligning that, returning and repeating to undo the impact
of each successive alignment on its predecessors across the device, and then doing it
all over again if drift should occur. Many
of these alignments are sensitive to angle
as well, requiring multiple optimizations
across degrees of freedom as well as across
inputs and outputs and channels.
In 2015, to meet these challenges, PI
demonstrated the ability to perform align-
ments across multiple inputs, multiple outputs, and multiple degrees of freedom all
at once, in parallel. This eliminated the
successive loops of sequential alignments
back and forth across the device and its
channels. Instead, the global consensus
alignment (including tracking if desired)
is achieved in one fast step.
This alignment step alone can multiply
the throughput from 60 devices per hour by
two orders of magnitude, at least regarding the alignment steps.
By eliminating the repetitive loops of sequential alignments, ad-
ditional orders of magnitude of process throughput improve-
ment can be readily achieved. This all means that alignment
ceases to be the pacing item for overall device manufacturing.
Parallelism: The key to process economics
The importance of this to photonomics cannot be understated—
this parallelism is an enabler for process economics at multiple
nodes in the production chain, from verifying the functionality
of devices while still on the wafer, to performing each assembly
step, to validating the devices’ health every step of the way, ensuring that bad or damaged products do not proceed through
further costly packaging steps (see Fig. 2).
Besides the parallel optimization capability—which is
founded on a novel revision of the classical digital gradient
search—there is the need to achieve first-light in the first
place and then to localize the main mode of each coupling
to ensure against spurious lock-ons. Otherwise, the optimization risks locking onto one of the local optima rather than
the global optimum.
This optimization can be efficiently achieved by perform-
ing an areal scan of the target area, and thereby mapping the
coupling cross-section across its full extent. This was formerly
performed in software and required significant time to accom-
plish, both because of communications latencies and settling
dwells necessary at each end of each scan line in convention-
al raster and serpentine scans. Worse, most approaches have
been based on pointwise move-and-acquire sequences, adding
time-consuming settling and communications dwells at each
data point in the scan.
FIGURE 2. A fast 18-axis parallel photonics alignment system is based on two hexapod six-degree-of-freedom nanomanipulators and two high-speed piezo XYZ scanners for double-sided wafer probing of waveguide devices. Fully digital and closed-loop, it implements
exclusion-zone capability, an internal data recorder, nanoscale-stable position-hold, and
firmware-based fast scanning, modeling, and multichannel N × M gradient search capability.
An application video can be found at https://youtu.be/X7zkUWo-lI8. (Courtesy of PI)