Appendix 2
The Mask-Making Process
Today an engineer sits in front of the display of his workstation and assembles a hierarchy of logic circuits, assisted by a powerful design software that goes by the name of EDA (electronic design automation). When the engineer is finished with the logic design described by using a high-level language, the software takes over, and, employing a rich library of predesigned and pre-characterized circuit layouts, compiles the rest of the chip design and then proceeds to design the masks automatically.
In 1970, each step of the way was essentially done manually. The composite layout was the key document that translated the logical-circuit description of the electronic system into the silicon topology obtained by overlaying several layers of different materials, the pattern of which was determined by the masks.
Unlike today, the composite was hand-drawn on a reclining drawing table with a ruler and colored pencils at 400 to 500 times the actual scale, using a mylar sheet instead of paper to maintain greater dimensional control, especially with changes of humidity and temperature. The name composite indicated that all the mask layers were drawn in the same document, allowing for the visualization of their alignment.
To create the masks, it was necessary to generate a separate document—called rubylith or ruby—for each mask. The ruby was a sheet of transparent mylar coated with a thin red semitransparent mylar film that could be cut on a cutting table and peeled
away. Under the ruby was placed the composite that served as a guide for the process of cutting and peeling, showing the areas to be removed.
The ruby cutting was a long and error-prone process that required careful checking before sending the ruby to the mask-making service that produced the working plates. Since each ruby represented only one mask of the chip, it was necessary to superimpose two or more rubies in various sequences over a light table to check the alignment and integrity of the patterns. With this method, we could create as many shades of red as there were superimposed layers, facilitating the recognition of possible mistakes.
When the rubies were ready, many people were recruited from the lab to help uncover all the mistakes and imperfections.
The ruby was then hung on a wall and photographed by a giant camera to be reduced to a 10x image called a “reticle.” The reticle was then mounted on a step-and-repeat machine to produce the master plate. The master was a 2.5 x 2.5-inch photographic plate containing as many rows of chip patterns at actual scale as would fit in 2.5 inches. The repeated pattern was obtained by photo-reduction of the reticle to 1x (actual scale) and successive exposures of the chip pattern on the master plate to complete a row of patterns. Then a new row would be done the same way until the entire surface of the plate would be filled with as many copies of the same pattern as would fit.
The master plate was then used to create sub-masters by contact photography and each sub-master originated several working plates that represented the actual tooling to be used for wafer production. The working plates were then mounted in the mask-aligner, a machine that was in the cleanroom where the wafers were processed. The working plates could only be used a small number of times because they were invariably damaged by microscopic scratches due to the contact between the wafer and the mask.
The masking process consisted of covering the wafer with a thin layer of photoresist that behaved like a photographic emulsion sensitive to ultraviolet (UV) light. The transparent areas of the mask let the light through causing the exposed photoresist to harden. Underneath the opaque areas of the mask, the photoresist would remain unaffected. The photoresist would then be “developed” like a photographic film by chemically removing it from the areas that were not exposed to UV light, whereas the hardened photoresist protected the wafer from the chemical attack, of the now exposed areas, that would follow.
After the chemical attack, the hardened photoresist was stripped completely off and then the wafer was subjected to a variety of thermal and/or chemical processes depending on the process specified in the run sheet.
The entire cycle from the photoresist to masking and thermo-chemical processing was then repeated as many times as there were masks.