Manufacturing process 18A: Intel starts mass production in Fab 52

Contents

Even if new production processes at Intel have always been associated with certain start-up difficulties, the effects have never been as great as with the disaster surrounding the 10-nanometre generation. These delays led to Intel losing its undisputed supremacy in the PC market, and it has been in a deep crisis for years. To overcome this, former CEO Pat Gelsinger issued the motto "5 nodes in 4 years" in 2021, i.e. to introduce five new manufacturing processes within four years. However, Intel cancelled the fourth stage called 20A in mid-2024 without replacing it. And at the end of 2024, Gelsinger had to take his hat off even before the final stage 18A was ready for series production.

That has now happened: Today, 9 October, Intel is officially opening its Fab 52 semiconductor plant, which has been under construction since 2021 and is located on Intel's Ocotillo campus in Chandler, Arizona. However, the starting signal is not a hard date, especially as silicon wafers spend several weeks travelling through a fab to go through the countless processing steps. The start of mass production began back in July. During a visit to the site, various advanced lithography systems with extreme ultraviolet (EUV) exposure technology were on display, which come from the Dutch supplier ASML and are responsible for 18A. Some visible spaces still offered room for expansion to expose more chips.

Each individual lithography system requires the space of several allotments – and yet they are just one piece of the puzzle among countless other, more compact machines. In view of the enormous precision with which the tiniest transistor structures are created, it seems strange that a fab is not actually finished. There are always construction workers with hard hats and large tools working at some point. All machines are assembled on site, which takes many weeks – while wafers are already being exposed a few metres away.

Intel's Fab 52 is specially designed for huge EUV imagesetters – including upcoming systems with high numerical aperture (High-NA EUV), which Intel intends to use from 14A: the ceiling is higher than in the Fab 42 next door, which is ten years older. Both the cranes running on the ceiling and the floors have higher load-bearing capacities. And yet the principle remains the same: above and below the floor where production takes place, there are further floors for cooling and power supply, for example. Another important part is the air: the entire building is slightly pressurized so that no dust particles can get in from outside. All floors are also covered with perforated panels, except for the areas intended for machines. There is a constant flow of air from top to bottom; the entire volume is renewed six times per minute. Nevertheless, the circulation is quiet and without any noticeable draught.

The best of the best of the best ...

Intel advertises 18A as the first production process in the 2-nanometer class that has reached series maturity. In the best MAGA tradition, the presentation also included the addition that the process and associated factories were researched, developed and built in the USA. I wonder if this has anything to do with the fact that President Donald Trump's US government recently acquired a ten per cent stake in Intel. There's always a bit of cheating involved: As usual, the machines used come from suppliers all over the world, the prominent EUV imagesetters from ASML in the Netherlands, for example.

Either way, the classification alone has no bearing on actual performance, i.e. whether processors manufactured with it achieve higher clock rates or use less energy. In reality, manufacturers rarely opt for one of the two extremes, but optimise in both directions. For the first two chips manufactured with this technology, the Panther Lake notebook CPU (Core Ultra 300) and the Clearwater Forest server processor (Xeon 6+), Intel is currently not specifying any clock frequencies. There are also no thermal design power (TDP) specifications for Panther Lake and only the rough 300 to 500 watts for Clearwater Forest, for which the existing platform and its socket are generally designed.