Early semiconductor designs were made by hand cutting and peeling patterns in a Rubylith film made up of clear and red sheets that were reduced to make masks. How that evolved to today’s computer-generated patterns written with e-beam exposure tools is well known. Ever wonder what economic impact mask technology innovation has had? I had an interesting discourse on this topic with Jim Wiley when he announced his retirement. Jim started in the Rubylith era and finished just as EUV was coming to production. So I asked, “Any idea of how long it would take to cut a Rubylith for a 5nm design destined for EUV? Also, how big would the Rubylith have to be? Just, back-of-the-envelope.”
Jim’s answer, “Very interesting questions! First, how big would the Rubylith be? If we keep the patterns on the Rubylith the same size as in 1970, about 4 millimeters, a Rubylith for a 5nm design would be about 4 kilometers by 5 kilometers. It would take about 2700 years to cut and peel. Good thing we abandoned Rubylith!”
Now that’s just for one mask level. If we assume a fully loaded labor cost of $100K per year, the labor cost for what we would call the tape out would be $270M-per-mask level. Assuming an extreme design of 100 mask levels, the cost for the entire set would be an industry bankrupting $27B-per-design in 2018 dollars, which is 6 times the size of today’s mask-making industry. There were 271 <8nm designs in 2018, which would work out to $7.3T, which is roughly the combined size of the German, French, and Italian economies.
Photo Source: Jim Wiley’s private collection.