Intel’s Ohio Investment: Connecting Some Dots

Intel’s $20 billion plan to build two new semiconductor factories or “fabs” in New Albany, Ohio, was described by company CEO Pat Gelsinger at the White House announcement on Friday as the largest single site investment in the state’s history. But this is about much more than alleviating the current chip shortage and supply chain woes, it’s about trying to regain the lead in the industry. I tried to get an interview with the company before the announcement, but to no avail. Nonetheless there are a lot of interesting crumbs out there, so let’s connect some dots.

$20 Billion seems like a lot of money (spent over the multiple years of construction), but in this industry it is not. Intel is already spending that much on two new fabs in Chandler, Arizona, and Taiwan Semiconductor Manufacturing Company (TSMC) is spending $12 billion in nearby North Phoenix. The Ohio investment is for two fabs, so that means around $10 billion each. When you size a fab, there’s a minimum efficient scale which determines your manufacturing capacity. In modern fabs, chips are made on 12 inch silicon wafers, and capacity is typically measured in 12 inch wafer starts per month – how many wafers do you put into the front of the line each month. This is determined by the capacity of your bottleneck step, as well as how much time the overall processes take. Wafers may take as much as two months to complete, depending on the number of steps in a recipe (as many as 700 or more). For advanced technologies, which are referred to by the “nodes” that describe their minimum feature size (e.g., 5 nm node), $10 billion might buy you 20,000 wafer starts per month. 

To put these numbers into perspective, TSMC has four “Gigafabs” in Taiwan, Fab 12A/12B in Hsinchu Science Park, Fabs 14 and 18 in Tainan Science Park, and Fab 15 at Central Taiwan Science Park near Taichung. Each of these has a capacity of around 250,000 wafer starts per month across a range of nodes. TSMC also has a fab in Nanjing, China which processes everything from 7 nm to 28 nm nodes, and it recently announced plans to enlarge that fab from 40,000 to 100,000 wafer starts per month. 28 nm is popular for automotive and consumer electronics chips, and globally TSMC has capacity for 180,000 wafer starts a month at just that node. Overall it has capacity for well over a million wafer starts per month. Volume is critical for semiconductor manufacturing – it is through production learning that you develop the capabilities to produce at scale.

TSMC also recently increased their planned capital spending for this year to around $40 – 44 billion. Meanwhile Samsung spent 40 trillion Korean won ($33.5 billion) in 2021, and this year plans to spend 50 trillion won (almost $42 billion). So while Intel’s investments are substantial, investments being made in other countries are also on a tear, so we should not get deluded into thinking one company’s investment alone will restore America’s lead.

Let’s look at another “dot.” Earlier last week, Intel confirmed that it had ordered the first Twinscan EXE:5200 scanner from ASML in the Netherlands. This is the latest generation of extreme ultraviolet (EUV) lithography tools that play a crucial role in creating the atomic scale patterns on chips. While earlier versions of these machines cost over $150 million each, this particular model costs in excess of $340 million for a single unit. That’s more than the list price of an Airbus A350-900, and airlines usually get big discounts (like as much as 50%). When that litho tool reaches production readiness it should crank out 200 wafers per hour.The scary thing is that fabs usually have to buy more than one of these machines. Since it is the most expensive tool, fabs buy enough of the other process tools so that litho is always the bottleneck. But the other tools aren’t cheap either. So building a $10 billion fab might mean a billion for the building infrastructure and the rest for all the different tools.

The EXE:5200 is so expensive because it has a high numerical aperture (NA). The NA describes the range of angles over which a system can accept or emit light. It is not quite the same as the f-number that is familiar to photographers, which links the size of the maximum opening of the lens to the focal length. Rather the NA determines the smallest thing you can make with an optical lithography system. A measure of this capability is the critical dimension, which is the size of the feature that can be printed at a specific height above the wafer. It is proportional to the wavelength of light used and inversely proportional to the NA. So if you want to make smaller things, you use smaller (shorter) wavelengths of light, or you have to use a higher NA optical system. ASML spent over $10 billion in R&D getting to the EUV light (13.5 nm) in those $150 million plus scanners. Apparently the extra $200 million per scanner gets you a higher NA. That incremental expense should give you a sense of the degree of difficulty of doing this. There is another aspect to this articular dot. There is so much demand for semiconductor manufacturing tools right now that lead times for delivery are extraordinarily long. Equipment delivery times are now the longest tentpole.

Third dot. Gelsinger stated in his White House briefing that Intel was committing $100 million to developing the talent pipeline in the region. This is extremely important and also very telling. For all the money a chip company has to spend on manufacturing tools, people are what enable you to run them and produce products. On the one hand, Ohio does not have much in the way of semiconductor manufacturing, but there are a lot of automotive, aerospace, and other engineering-driven professions as well as great engineering schools in the state and neighboring Indiana, Michigan, and nearby Pittsburgh. Working with local schools will help build a talent pipeline. It signals the increasing competition for talent in cities like Phoenix and Austin, Texas, so going to the Midwest is a smart move. Ohio also isn’t as arid as Arizona or Texas, and fabs use a lot of water. I have questioned before why we like to build fabs in deserts.

When we connect these dots, a fuller picture emerges. First, we should not be confused into thinking that Intel’s investments are going to alleviate the short term chip shortages in the auto industry and elsewhere. These are investments being made for the long term, and they are substantive and meant to keep up with the frontiers of chip manufacturing technology. There will have to be more to follow. Gelsinger already alluded to Ohio becoming a megasite with perhaps $100 billion invested over time. He is clearly bent on climbing back to the top in manufacturing leadership. These moves take courage, especially as a public company CEO in a market where many shareholders would rather have cash returned to them. Let’s hope he and the company are successful.