The project of the month at the small contract house where I worked was the control system for a wafer loader. This machine was to extract a 3" diameter semiconductor wafers full of IC's from a "boat" full of similar wafers, orient the wafer and transfer it into a tester. When the tester finished its job, the loader would accept the wafer back and transfer it to the output "boat".
This thing was about the size of an HO model train layout and about as much fun. There were conveyor belts and elevators, spinning vacuum chucks and swinging arms that would fling wafers across the room like shiny Frisbees if you didn't handle the vacuum right.
Most all of this hardware was working well, but the elevators for the wafer boats were proving a challenge. The boats were plastic troughs, U-shaped in cross-section with slots along their length to hold the wafers. To load the machine, the boat was stood up on end and placed on the elevator which had been previously raised to its top-most position. The control system was then supposed to lower the elevator until it detected that the bottom-most wafer in the boat was in the correct position for the mechanism to engage it and whisk it out of the boat and onward towards the maw of the tester.
It was the "correct position" part of that we were having trouble with. The spec said that boats could be partially filled, so it was up to the loader to locate the bottom most wafer in the boat. The job was assigned to a clever optical sensor that consisted of two IR LEDs flanking a photo-transistor, potted in a single case.
The photo-transistor was tightly coupled to a short piece of glass fiber that extended to the surface of the clear potting compound, so the transistor would not "see" the LEDs directly, but only the light reflected back to it. One of these sensors was placed under the elevator looking up at the wafers descending toward it. The theory was that when the reflected light signal, as sensed by the photo-transistor, reached a level corresponding to the "correct position", the control system would stop the elevator.
Reflectivity to the rescue-or not
The flaw in the theory was that the albedo of the wafers varied wildly. The backs of some wafers were perfect silver mirrors, and some were black as coal. If we set the threshold for the bright ones, we smashed the dark ones when we didn't stop the elevator in time. Setting the threshold for the dark ones left the bright ones high and out of reach.
Running out of time, we were wracking our brains for a solution.
- Detect the wafer from the side? Might work, but the wafers were thin and curved so it wasn't a sure thing, and the mechanical changes would take more time than we had.
- Use a switch? Nope, the spec called for no contact with the wafer.
Finally while studying the sensor data sheet for the nth time, I noticed a spec for "minimum distance of operation." The geometry of the sensor was such that as the sensor approached the reflective surface, the areas illuminated by the LEDs would eventually slip out of the field of view of the photo-transistor.
Watch out for the wall
Picture the situation by imagining yourself walking towards a wall in the dark, holding a flashlight in front of you and looking through a cardboard tube. At a distance, the flashlight illuminates a large section of the wall and you will see the light through your tube. However as you walk towards the wall, the circle of light from the flashlight will get smaller and smaller and eventually won't be visible through the tube anymore.
This happened with our sensor at about 0.1 inch, and since it was set by the geometry of the sensor, it was very repeatable. So, set the sensor threshold to detect the dirtiest wafer, drop the elevator at high speed until the sensor detects the bottom wafer, then go to low speed.
When the signal drops out, STOP! Bingo! The "correct position."
(Mike Hughes is an old farm boy who has spent the last 30 years working with successively smaller computers. He figures that by the time he’s ready to retire, the things will just about reach invisibility