Who took my chips?

Mark Lapedus of trade newspaper Electronics Buyers News reports: “Taiwan Semiconductor Manufacturing Co Ltd (TSMC) is telling customers that they must sign a three-year lump-sum contract to guarantee capacity….TSMC’s move, which has been greeted coolly by some customers, comes at a time when IC makers are scrambling for more wafer capacity. They may be left with no choice but to strengthen their ties with the Taiwanese company.”

There’s nothing like a shortage to get customers scrambling to tie up long-term deals when previously they were all about just-in-time delivery. However, Lapedus’ report is not from recent weeks, when carmakers have complained to local politicians who in turn complained to the Taiwanese government who promised to get right on that to TSMC management. It’s from July 1995, a distant time when certain chips were as hard to find as a Buzz Lightyear toy and the customers for silicon then were not really big enough to demand a response from their governments. Chips these days are serious business, especially if you’re in a business politicians care about.

However, a little over a year later, analyst firm Advanced Forecasting described how monthly sales had dropped by a quarter from $12bn per month to $9bn. The reason? A dramatic build-up of production capacity driven by the earlier shortages that had suddenly translated into a glut, even as the first internet boom was getting under way. The semiconductor industry never looked back: and not in a good way.

In 1995, unit prices hit their highest point for years, after a long slump that had followed the personal computer booms of the 1980s. After 1995, unit prices dropped. And they kept dropping. It was only last year that the process seemed to have gone properly into reverse with prices rising even amid a slowdown in unit sales and potentially not yet another false ray of hope for chipmakers who keep getting their margins trashed. I found an old piece of mine from spring 2008 that suggested capacity had tightened sufficiently to turn the situation around as long as economic growth continued. Well, we all know what happened next. A year later, governments were throwing stimulus money around to try to soak up the inventory.

Today, some things are different. The TSMC of 1995 was tiny compared to the leading chipmakers of the day. Companies like Xilinx were highly dependent on its output and so were easily encouraged to take part-shares in a new fab. Other companies cut similar deals only to find they had no way to consume the capacity they had pre-booked. But the most meaningful shortages hit other customers, such as thousands electronics equipment builders reliant on Motorola microcontrollers, which focused on its larger customers partly because that made financial sense and partly because it was easier to supply them because a large proportion of their products needed to be loaded with their firmware in the fab.

Before 1995, Microchip Technology was a struggling microcontroller suppler that suddenly found it had a technology that would quickly propel it to number one in 8-bit microcontrollers. They used a memory that could be programmed by the customer instead at little more cost than the Motorola version. The sudden shortage forced those customers to jump ship and many of them did not come back.

This time around, the shortages are most acute in more advanced silicon, which is where TSMC now has a commanding position and that only seems to be becoming more apparent now that Intel is considering outsourcing much more of its production. If you are a carmaker looking for chips for your ADAS computers, this is not good news. Carmakers have a reputation for forcing suppliers into painful contracts. However, it is not as if the other customers of TSMC are exactly generous. Everyone plays the same game, though the companies that book ahead are in the best position.

It takes about three months for a wafer to get through a fab today on the most advanced processes simply because of the sheer complexity of the operations that need to be done on them. Naturally, customers with long-term projections of where and when supplies are needed will be the ones that a foundry such as TSMC will be most comfortable with. This is, seemingly, at odds with the just-in-time supply arrangements carmakers like to use and which is playing a factor now in the problems they have with supplies.

The reality is that though they might have enjoyed JIT-like responses from their chipmaking suppliers, someone has been holding inventory because that is the only way you can make the two types of production work together. And that inventory got squeezed out as car production went back online towards the end of last year. This was not helped by carmakers cancelling orders in 2020 because of Covid-19 slowdowns while IT equipment shipments took up the slack. Result? Shortage for the carmakers.

This, naturally, has led to some anguish about the ability of carmakers to do JIT production in an environment where perhaps half of the cost goes into chips. But as with 1995 it is important not to overreact to the situation and come up with an answer that works for no-one. One issue is that these chips are not all the same. Economics suffers from the 'lump of labour' fallacy; analysis of manufacturing supply chain issues suffers from a fallacy we could call the 'slice of silicon', and you have to pay attention to which kinds of IC are most heavily affected by any shortages.

One of the problems of 1995 was that shortages of specific products stalled production lines. You could probably buy 90 per cent of what you needed but your favoured supplier sold some vital component to its favoured customer. The upshot is that you still could not ship anything. Similarly, today’s vehicle is stuffed full of devices that can be made at numerous fabs around the world on older lines that may suffer short-term issues but which are quickly resolved. But if the primary ADAS controller is on the waiting list, you are not shipping that car.

The response by the car industry is perhaps to be expected, but it probably will not change their approach all that much over time. They may have to start holding more inventories of certain parts but it seems unlikely that this will be the case for the vast majority of the chips they consume. They may enter into more long-term supply agreements with TSMC as it tries to ramp up capacity over the coming year or another player, such as GlobalFoundries or Samsung if they prove able to respond, similar to 1995. We are not likely to see a slump in chip prices come 2022, but the carmakers may have learned a valuable lesson in how to deal with a source of speciality ingredients that take some time to cook.