The production took place on 28th, 29th and 30th
of April 2003 at a contractor of individual Computers. Since the
pictures and especially the video take up quite a bit of web space,
we might remove them shortly. If you like them, please download and
keep them on your hard drive. Use by the press is explicitly allowed
if you mention the source. |
Jeri is a little camera shy :-)
before we even started, I took a
close look at the circuit boards to check if they have been properly
tested. I've had problems before with boards that came as
"electrically tested", but still with a 4-6% failure rate, which is
typical for untested boards. Since there's a lot of valuable parts
on the C-One, I did not want to take any risk, and searched for the
needle prints of the electrical test adapter with a microscope. The
picture shows a section of the FPGA pads, which have a distance of
0.5mm (about 1/50th of an inch). The needle prints are easy to see.
Prior to SMD placement, the SMD
solder paste is "printed" on the board. Think of the solder paste as
viscous solder. It stays in shape after the mask is removed, and
it's (of course) only on the SMD pads. The mask is 0.15mm thin sheet
metal, and the holes are laser-cut into it, which is currently the
most precise method of making such a mask. Many people still think
that etching the mask is state-of-the-art, but laser is a lot more
precise, faster, and about the same price.
The non-vision machine places
all 0805 parts (ferrites and capacitors), Tantalums, PLCC sockets
and SO-ICs. This is a view from the back with all the part feeders.
Precision of this machine is about 0.25mm (1/100th of an inch) due
to possible shifts of the board itself, which cannot be recognized
without a vision system. It would be possible to place 0603 parts
with this thing, but the robot head is equipped with a nozzle that
can also carry heavy parts like PLCC chips and big tantalums. This
nozzle would swallow the small 0603 parts with the vacuum.
This is the robot arm of the
vision machine. The arm uses a nozzle that can be exchanged
automatically, without manual interaction. This way, 0603 parts and
the heavy FPGAs can be placed without interruptions. The camera
module, pickup-nozzle and centering/checking clamps are easy to see.
The centering clamps are also used for an identification of the
part: Resistors and capacitors are checked for the right value "on
the fly", and if the part has the wrong value, it's dropped in the
"bad ID" box. About one out of 1000 resistors has the wrong value,
so this ensures production quality.
The "critical parts" like
QFP208 and TQFP100 are aligned with the fixed vision camera. This is
a picture of the monitor that watches the part from the bottom and
recognizes it's shift/rotate alignment before it is placed on the
board. The board shift itself has been measured before with the
flying vision camera on the robot arm. Precision of the vision
machine is better than 0.05mm.
after the "small machine"
has placed the "uncritical" parts, the half-assembled board is
placed in the vision machine. The picture shows the machine near the
end of the program, where the second FPGA is about to be placed.
After SMD placement, the board
is run through the "pizza street", the reflow-oven. Hot air is used
to solder all parts at the same time. The temperature curve can be
adjusted precisely in five zones, as the last picture shows.
After manual placement of the
through-hole parts (connectors, slots and pin headers), the whole
board is run over the wave solder. Liquid solder is pumped from the
bottom against the board. The soldering process itself only takes a
few seconds, but the board must run over the flux foam and the
pre-heat area of the machine, which takes about a minute. Still a
very good time, given the fact that it's a few hundred solder points
on the C-One! To all of you who are worried about the space for the
second PCI: The drills have been protected against soldering before
the board has been run through the machine, so a second PCI can be
added without special equipment - if this will ever be necessary :-)
Before the board is fired up
the first time, all critical solder joints are inspected under a
microscope. The picture shows the side of the most critical parts,
the FPGAs in QFP208 package. You see a perfect placement and
error-free soldering. This is not always the case, sometimes
solder bridges have to be removed manually.
This picture shows the board
and the two spots that require manual SMD rework. The additional 10K
resistor near the FPGA ensures safe startup of the EP1K100, and the
additional 3k3 resistor above the CPU is responsible for startup of
the whole computer (ATX power-on circuit).
final_board.jpg (caution: 436K!)
Same picture as above, but without the microscope add-ins and the
vision_working.mpg (caution: 3.9M video!)
is a small video of the vision machine working. After I push the
"start" button, the machine starts looking for three vision points
on the board to determine it's position and rotation (You can see
the end of this step when the light on the flying camera goes off).
The video is about three minute long, and does NOT show the full run
of the program (that was nearly 5 minutes). To be honest, it's quite
boring to see the machine run at it's 50-parts-per-minute beat,
especially when you've heard the noises of two machines in the same
room continuously for three days in a row.