|

Wafer
Identification System Helps AMD Prevent Misprocessing on Metal
Deposition Tool
Advanced
Micro Devices (AMD; Austin, TX) recently retrofitted a 15-year-old
Anelva 1015 sputtering system with new optical character recognition
(OCR) sensor technology. The retrofit was done to demonstrate
the feasibility of integrating a stand-alone, in situ wafer identification
system on existing 6-in. (150-mm) wafer processing equipment.
|
|
|
Retrofitting
an older sputtering tool with an automatic wafer ID system
greatly reduces misprocessing.
ILLUSTRATION
BY JAMES SCHLESINGER
|
The
Anelva, like most older wafer processing equipment, was not previously
equipped with an automatic wafer ID capability, posing a line-yield
risk. "We wanted to automatically read the lot number on the wafers
and stop processing if the actual lot number read did not match
the expected lot number," explains Mark Bradford, a senior engineer
at AMD. "The wafers are transported from operation to operation
in boxes labeled with their lot number. On occasion, someone will
inadvertently mix wafers between boxes. If this is not realized
at the next operation, the wafers could receive the wrong processing.
In the case of the Anelva metal deposition process, once you put
the wrong metal on the wafers, they're trash.
"By
adding wafer reading to the tool, we now have a way to stop the
tool if the wrong wafers are loaded, so scrap due to this type
of misprocessing is [all but] eliminated."
Bradford
evaluated other wafer ID products and ultimately selected an In-Sight
1700 wafer-reading sensor from Cognex (Natick, MA). The
self-contained unit combines advanced recognition software, image
processing, and LED-based image formation in a handheld enclosure.
He chose the sensor primarily because of its ability to read OCR
scribes, regardless of their appearance. "Sometimes there will
be edge-bead removal lines that run right through the middle of
the scribe, and the other systems we tried couldn't handle that
very well," the AMD engineer explained. "The others also had trouble
reading very light scribes that sometimes occur when the laser
scribe is not set up properly."
Before
wafers are transported by belt into the sputtering station on
the tool, they are first positioned onto a flat finding station
where each wafer is mechanically aligned. The wafer reader is
mounted directly above the station, roughly 10 mm above the wafer
surface, and is triggered immediately after alignment. In less
than a second, the sensor illuminates the wafer using built-in
LEDs, captures an image of the scribe, and analyzes the image
using recognition algorithms. An automatic image-tuning feature
dynamically balances illumination and image processing based on
the specific visual characteristics of the scribe being read.
This feature eliminates the need for filters and other image preprocessing
tools often used to handle image variations.
"The
1700 can read anything we throw at it," claims Bradford. "It also
provides us with a large read area, which is important because
the flat align does not have to be precise."
As
each cassette on the sputtering tool starts, the expected lot
number for that cassette is transmitted to the wafer ID system
by a custom Visual Basic (VB) program that Bradford developed.
Once the scribe is read, the sensor unit compares the expected
lot number and the actual lot number. The result of this comparison,
as well as the read information and the score of the read, are
passed back to the VB application. If the comparison fails, the
program automatically stops processing on the tool.
AMD
saw the ID unit's Ethernet connectivity as another important feature.
The entire fab runs on an Ethernet network, so it was advantageous
to find a wafer reader that could link directly to the network.
The sensor can be set up, monitored, and modified using any Ethernet-connected
PC in the facility, and the link between real-time ID information
with work-in-process control software is seamless. The sensor
also provides a familiar Web browser interface that allows fab
workers to adjust reading parameters using menu-driven dialogue
boxes as well as a live image display of the wafers being read.
Another
key feature of the sensor is its ability to send the data read
to a file, and thereby directly track the order in which the wafers
are being processed through the equipment. The factory tracks
this information indirectly by reading the wafer IDs at a separate
workstation. The fab typically runs 48-wafer lots, with two cassettes
of 24 wafers each. The current process calls for the operator
to read the left cassette first, then the right cassette. The
lot is then supposed to be processed in the same way (left first,
then right). The assumption is then made that the order in which
the wafers were processed through the equipment is the same order
in which they were read.
"There
is always some uncertainty about whether the left cassette gets
processed before the right one, or whether wafers get mixed up
between the wafer read step and the actual process step," according
to Bradford. "Knowing the exact order that specific wafers are
processed through a given tool is critical for troubleshooting
when there are process or equipment problems."
After
successfully retrofitting the sputtering tool with the wafer sensor
unit, AMD is looking into upgrading a wide variety of other fab
equipment with automatic ID for improved wafer line yield and
traceability.

MicroHome |
Search | Current Issue | MicroArchives
Buyers Guide | Media Kit
Questions/comments about MICRO Magazine? E-mail us at cheynman@gmail.com.
© 2007 Tom Cheyney
All rights reserved.
|