EmbeddedRelated.com
Blogs
Memfault State of IoT Report

Supply Chain Games: What Have We Learned From the Great Semiconductor Shortage of 2021? (Part 4)

Jason SachsDecember 31, 2022

Today we’re going to look at what’s been going on this past year in the chip shortage, particularly in the automotive markets. I’m going to share some recent events and statements that may shed some light on what’s been happening.

In Part Three we went through a deep dive on some aspects of Moore’s Law, the semiconductor foundries, and semiconductor economics, and we looked at the game Supply Chain Idle. We touched on a couple of important points about the economics of foundries:

  • Fully-depreciated equipment has been producing chips for mature technology nodes fairly steadily, even though these now make up a small portion of revenue
  • Smaller ICs like microcontrollers (MCUs) are still being produced on mature nodes, because they’re not economical to design and manufacture on more advanced nodes
  • The prices to make ICs on these mature nodes are low, because they’re fully depreciated, but this poses a dilemma: now that prices are so low, there’s no economic impetus to add more capacity

For Part Four I wanted to take a breather from IC manufacturing economics to look at where things are today, since I published Part One at the very end of 2021.

Hint: it’s not exactly unicorns and roses.

Let’s get those disclaimers out of the way early, so we can get this over with.

Disclaimers

I am not an economist. I am also not directly involved in the semiconductor manufacturing process. So take my “wisdom” with a grain of salt. I have made reasonable attempts to understand some of the nuances of the semiconductor industry that are relevant to the chip shortage, but I expect that understanding is imperfect. At any rate, I would appreciate any feedback to correct errors in this series of articles.

Though I work for Microchip Technology, Inc. as an application engineer, the views and opinions expressed in this article are my own and are not representative of my employer. Furthermore, although from time to time I do have some exposure to internal financial and operations details of a semiconductor manufacturer, that exposure is minimal and outside my normal job responsibilities, and I have taken great care in this article not to reveal what little proprietary business information I do know. Any specific financial or strategic information I mention about Microchip in these articles is specifically cited from public financial statements or press releases.

Furthermore, nothing in this article should be construed as investment advice.

Oh, and there’s no game to play along with in this article. If you really want one, try Industry Idle — somewhere in there is a lesson about supply chains and inventory and transportation, but it doesn’t seem as enlightening as the others.

Que Sera, Sera / Storm Clouds Cantata

As I write this, it’s in the thick of the third-quarter earnings season. I have mixed messages to share, on three issues.

You’ll Take the High Road and I’ll Take the Low Road

One is that the chip shortage situation as of late 2022 has changed. I said in Part One that the chip shortage was not a monolithic phenomenon, that there was actually a group of shortages happening in different market segments. And they’ve now diverged. The leading-edge shortage is essentially over; the trailing-edge shortage continues. Executive board chairman Steve Sanghi gave a concise summary during Microchip’s November 3 2022 earnings call:

So it began a year and a half ago, with capacity being constrained at all the nodes: trailing-edge, leading-edge, and the middle of the way. What has happened in the last few quarters is with the personal computers and cell phones, which are significant consumers of semiconductor, and mostly semiconductors on the bleeding edge of technology: you know, processors and very high-end chips in the cellular phone. With the correction in that market, the leading-bleeding-edge capacity now is really no longer constrained. You have seen dramatic downside guidance by a lot of the very leading-edge people. So today, if you wanted a 7-nanometer, 10-nanometer, 14-nanometer capacity, you can have everything you need.

But the trailing-edge capacity continues to be extremely constrained. On trailing-edge, we built some inside, and we also bought some from the foundries, and we are constrained on both. Inside, we haven’t been able to get all the equipment we wanted. Most equipment that was even scheduled to be delivered got pushed out by many months, sometimes many quarters, because the equipment supplier wasn’t able to get semiconductors for their parts. So the inside products that we’ve done inside remain constrained on many different corridors. And the capacity we buy outside is really very similar. The trailing-edge capacity remains constrained, and we currently believe will remain constrained well into 2023.

Willy Shih covered this in a recent article in Forbes[1] with these section titles that pretty much describe the situation:

  • The high end is seeing softness — demand for PCs/mobile phones has dropped
  • Commodity segments like memory appear to be swimming in inventory — DRAM pricing is falling again
  • But the automotive sector continues to struggle — mature nodes are still supply constrained

Shih goes on to say:

One lesson we can learn from this mess is that when you take an aggregated view of the semiconductor sector, combining the high end, commodity, and mature chip sectors, it can mask what is going on in different segments of the market. I find the same problem when combing through government import/export trade data or Bureau of Labor Statistics stuff. There’s always great value in understanding the detail.

A Recap of 2022

The second issue to note is that there have been a number of “interesting” events affecting the semiconductor industry and the general economy this year. Here are some:

(I apologize if this seems US-centric; feel free to send me other relevant worldwide events to the semiconductor industry from 2022 and I will add to this list.)

Long-term Trends

And the third issue is that certain long-term trends are continuing — we covered some of these in Part Three:

That’s the short answer.

I’ve got a bit more to say, but I need a shot of soothing wholesome fatalism to keep me going… there’s very little we can do to control what’s going on, so why worry? Que sera, sera....

What does it all mean? Well… 2021 was a year of record earnings growth for many semiconductor companies. Lots of demand. Shortages everywhere. Desperate pleas for more capacity, and many companies announced plans in 2021 and 2022 to build new fabrication plants. But this feeds the semiconductor cycle, as we saw in Part Two. So the big question is, when does the clock strike twelve, when is the ball over, when does the magic carriage turn back into a pumpkin, when does Cinderella’s gown turn back into rags — when do we get our semiconductor glut, that we know is coming?

Doom and Gloom From the Analysts

There are a number of semiconductor analysts out there looking with some caution at the industry’s near future. Here is a sense of what they’ve been saying.

  • Bill McClean of IC Insights pointed to Micron’s June 30 2022 fiscal results as a “canary in a coal mine” with future guidance foreshadowing a decline in the business cycle.

  • Bill Jewell of Semiconductor Intelligence warned of “Semiconductor Growth Moderating” as early as February 2022, with a number of similar forecasts — “Semiconductors Weakening in 2022”, “Electronics is Slowing”, and more recently, “Semiconductor Decline in 2023”. This particular summary included a revenue table of top manufacturers, that caught my eye:

    Table © Semiconductor Intelligence, LLC, reproduced with permission.

    Note the split of consumer/computational electronics (flat or downhill) and automotive/industrial electronics (growth).

  • Robert Maire of Semiconductor Advisors posted a somewhat whimsical warning called “The Semiconductor Cycle Snowballs Down the Food Chain — Gravitational Cognizance”, comparing the chip cycle to that moment in the cartoons that Wile E. Coyote hangs in the air for a split-second after running off a cliff until he realizes he will fall:

    This reminds us very much of where the semiconductor industry is today. The industry has been running so fast and focused on speed that it hasn’t yet realized that the basis that supports the industry has gone away, that is that demand has dropped and will see further declines.

    We have been talking about the industry being in a down cycle for months now. Memory prices have dropped (usually one of the first signs) inventories have grown, lead times are down. More importantly, demand for semiconductor rich electronic devices is dropping.

    However, some semiconductor and semiconductor equipment companies are still reporting great earnings, record breaking earnings in some cases. This makes it very difficult to talk about a down cycle when you are still making big bucks.

    The speed at which the industry has been running has driven so much momentum into the industry that gravitational cognizance has been delayed.

  • Malcolm Penn of FutureHorizons has been predicting a semiconductor downturn for some months now. Penn is one of the veterans in the industry, and an astute observer: he’s seen most of the cycles since the beginning, and in terms of market dynamics, there is nothing new under the sun. Here’s some of what he’s been saying this year:

    • In May 2022, FutureHorizons predicted a bearish 22% decline in the semiconductor industry. The metaphor he uses is a roller coaster at the top of its arc:

      Image © FutureHorizons, republished with permission.

    • In an August 2022 article: “Right on cue with our December 2021 forecast, the current semiconductor Super Cycle is finally drawing to a close and the 17th market downturn has now well and truly started.”

    • In a September 13 webinar: (I can’t link directly to this presentation, but here are the highlights on YouTube)

      What should be happening in the downturn is that we invest like crazy, ready for the next upturn. But that was not what will happen. What will happen is: investments will be cut back. And will stop, delay, push out plans. And that will unfortunately feed the next upturn, because we will eventually not have the capacity on place when we actually need it. All of the wisdom says invest countercyclically; all of the financial infrastructure tells you: “You cannot do that.” So, you’ve got this dilemma here. Again, it goes back to the onshoring dilemma here, it’s that the structures in the industry are not conducive to actually taming this industry cycle. It’s just in fact the opposite: it’s conducive to fueling the industry cycle. So we will see capex cut back, we will see a dramatic cutback in capex, it will drop right back down again, it will overshoot just as it has done in the past, just as the sales will overshoot. You’ll see negative capex growth numbers, just like we’re seeing or will see negative semiconductor sales numbers, and then they will delay it. When they suddenly wake up and realize, “Oh we’ve got to start investing again,” they will be at least a year behind the ball, just as they were in 2021, and just as they were in every previous upturn since then.

  • In October, Citigroup (aka “Citi”) cut earnings estimates for Texas Instruments, NXP Semiconductors, and Microchip Technology, with one of its analysts stating the following, according to Seeking Alpha:

    We expect NXP Semiconductor (NXPI) and [Texas Instruments] will both report weakness in bookings and we reiterate our belief that this is just the beginning of the downturn and every company/every end market will feel it.

Bad news (perhaps) for employees and shareholders, good news (maybe) for customers waiting desperately for their orders to come through. All three companies are in the automotive/industrial market, which is the supply-constrained side of the 2022 split in the chip shortage.

Prediction Time

The situation makes me think of Alfred Hitchcock’s 1956 film The Man Who Knew Too Much. We seem to be reaching a kind of slow, dramatic crescendo in the mature-node chip shortage. It can’t last forever — can it? — and the suspense is killing me.

I’m going to state a conjecture and make a prediction. I don’t have any special information, and I’m not going to claim that I have a correct conjecture and prediction, merely that they are a possibility that needs to be considered.

The conjecture is that there is a long-term disturbance in the supply-demand equilibrium for mature nodes, and that the dynamics of the normal semiconductor cycle aren’t going to correct it, without the addition of some deliberate long-term shifts in supply strategy across the industry.

The prediction is that since the length of the semiconductor cycle is somewhere around four years, we shouldn’t plan on this taking less than two complete cycles, or eight years, for the industry to find out a better solution for mature-node equilibrium. Since this shortage was first in the news in December 2020 and became widely recognized by early 2021, that brings us to early 2029. Maybe.

That doesn’t mean that we’ll have to wait eight years for supply to catch up to demand — the analysts seem to be pointing to 2023 for a semiconductor downturn — but rather, I’m not expecting any mature-node glut to last very long, and then whammo! we’ll be back to more shortages. I don’t think we’re out of the woods until we have industry leaders stating confidently that they’re making a coordinated effort to do something different so that there is enough supply. Something on the order of the NTRS/ITRS roadmapping efforts.

Because right now, it seems like the CEOs are just doing what they can, letting us know what they see, and waiting for someone else to make a change so that their company can follow and join in. Most of the big, bold steps are at the leading edge.

Again, I’m just a guy writing a blog in my spare time; what do I know.

Something Is Happening Here, But You Don’t Know What It Is

Let’s take a step back for a moment, and forget that we’re waiting on tenterhooks for something to change in 2022. Forget about 2020 and 2021, even. Sure, car companies lost their place in the chip line when the coronavirus first hit, as they hunkered down for the gathering storm, and when it turned out that what everyone really wanted was game consoles and laptops and cell phones and smart watches for something to do while stuck inside, there were never enough chips for the car companies to catch up — we’ve heard that narrative before, and now it’s time to put it out of your mind.

What’s been happening in the long-term, and how did we really get into this mess?

Abstract Neoclassical Economic Criticism 101

The real issue, as I see it, is that economics doesn’t guarantee a nice supply-demand equilibrium, and yet too many of us labor under the well-intentioned belief that it does — or that we should behave as though it does, and have faith that free enterprise will bring a solution whenever things get out of balance. The truth of the matter is that economic models are not that simple, and even simple economic models can demonstrate turbulence and instability. (Remember M.U.L.E. and the Smithore Gambit in Part Two?) In reality, all we can say is that there are economic forces that encourage equilibrium: when demand is sustained for something, there is a tendency for it to get more expensive, until there is enough of an economic incentive for someone to make more of it. And similar behavior is true in an oversupply: prices tend to decrease and drive the industry towards a self-correcting drop in production. Chip manufacturing is so expensive, and so time-consuming, and so large of an endeavor, though, that unbelievable amounts of supply-demand imbalance can occur before the underlying condition is fixed. Imbalance conditions can be rather “sticky”. We are finding that out the hard way.

To cut to the chase: this semiconductor cycle is different.

That is a big claim, and I’d like to think that it’s wrong, that Malcolm Penn is right, and this is just the 17th market downturn in the semiconductor industry, just another normal cycle. Vanity of vanities, saith the Preacher; the thing that hath been, it is that which shall be; and that which is done, is that which shall be done; and there is no new thing under the sun. To every thing there is a season…

I talked about the semiconductor cycle in Part Two; in case you don’t remember, here is Integrated Circuit Engineering’s description of the semiconductor cycle:

Swings in production growth rate are closely tied to capacity utilization, ASPs of devices and capital spending (Figure 1-2). For the industry as a whole, when capacity utilization is high, ASPs rise and companies are more profitable, which in turn, encourages capital spending. However, with increased spending, capacity constraints loosen and ASPs tend to drop, decreasing company profitability. The decreased profitability (pre-tax income) then reduces the amount of capital available to invest in future needs.

There, like a wheel going round and round, or a pendulum swinging back and forth, with near predictable regularity every four years or so — heavily influenced by DRAM’s lurching capacity increases and price nosedives. The Economist cited IC Insights and Penn’s Future Horizons in January, saying a glut is coming:[2]

IC Insights, a research group, reckons that, across the industry, capital spending rose by 34% in 2021, the most since 2017. That torrent of money is welcome news for the industry’s customers, who have been struggling with shortages for over a year. For the industry itself, it is the latest iteration of a familiar pattern. Bumper revenues, like those reported by Intel on January 26th and Samsung the next day, compel companies to expand capacity. But because demand can change much more quickly than the two or more years needed to build a chip factory, such booms often end in busts. The chip business has swung between over- and undercapacity since it emerged in the 1950s, observes Malcolm Penn of Future Horizons, a firm of analysts (see chart). If history is a guide, then, a glut is in on the way. The only question is when.

And I don’t dispute that. The Preacher knows what he’s talking about. Tech companies are battening down the hatches for rough seas. Bloomberg just reported on Nov 11 that GlobalFoundries announced job cuts and hiring freezes — right after reporting what seems like great increases in revenue and income:[3]

The company based in Malta, New York, and majority-owned by the government of Abu Dhabi, is among semiconductor producers seeking funds from the US government via the \$52 billion CHIPS Act to expand domestic chip manufacturing. In planning layoffs and implementing a hiring freeze, GlobalFoundries is joining many of its peers in the broader technology industries.

Intel Corp. recently said it will undergo cost reductions, which Bloomberg News reported would include a significant number of job cuts. Other chip companies, including Micron Technology Inc., have slowed hiring. Meta Platforms Inc. has started widespread layoffs, while Qualcomm Inc., Twitter Inc., Apple Inc. and Amazon.com Inc. are among those that have paused hiring for many divisions.

On Tuesday, GlobalFoundries said third-quarter revenue jumped 22% and projected sales and profit in the current quarter that topped analysts’ estimates. The company is trying to win share in the market for outsourced chip production and gain enough scale to compete with industry leader Taiwan Semiconductor Manufacturing Co.

Tire Swings

What’s different, and a bit unnerving, is the mature-node situation. If you think of the semiconductor market as a bunch of tire swings, hanging from a big tree, regularly moving back and forth between overcapacity to undercapacity — look, there’s the one for memory and there’s the one for PC/mobile microprocessors, back and forth, there they go…

Photograph "’09" © 2009 kristina sohappy,
used under CC BY 2.0 license, desaturated from the original.

Well, one of the tire swings covers mature nodes, and it’s been pushed so hard it’s gotten stuck on a rock ledge or a tree branch.

It’s not the biggest tire swing, by any measure. Here’s that table of semiconductor manufacturers again:

Table © Semiconductor Intelligence, LLC, reproduced with permission.

The auto/industrial market doesn’t show up until you get to manufacturers #10-14 — and even then, it’s just a portion of their business. Once upon a time this wasn’t the case; TI and Motorola were regularly at the top. In 1988 the top semiconductor companies by revenue, according to Dataquest, were NEC, Toshiba, Hitachi, Motorola, Texas Instruments, Fujitsu, and Intel, in that order.[4] The market has shifted, though, with mobile devices now a big part of the picture.

In other words, this sustained mature-node chip shortage is only a small part of the semiconductor market… yet an important one. The term “golden screw” has been bandied about: when someone is building product X, if you’re missing one tiny “golden screw”, even if you have all the other pieces, you still can’t complete production. I am not sure when this term made its way into the collective consciousness of the electronics industry, but the first instance I could find was in a November 2021 article in Fortune quoting Infineon’s Helmut Gassel:[5]

Infineon, the world’s largest supplier of semiconductors to the automotive industry, reported on Wednesday that the cost of outsourcing its production to chip foundries is set to climb steeply.

Unhappy about sinking money into fresh capacity for outdated processors, these specialized high-volume chip manufacturers want their investment costs covered in exchange for churning out more semiconductors.

“For this fiscal year [to end of September], we have secured somewhat more foundry capacity, but it’s for sure not anywhere close to what we need,” said operations chief Jochen Hanebeck.

For now, Infineon is living from hand to mouth, prioritizing shipments to those customers where the demand is greatest.

The majority of its job, according to marketing chief Helmut Gassel, is taking the chips they can get from the foundries and assigning them based on the greatest economic benefit to their customers: “that which is needed most at a given moment — the ‘golden screw’ so to speak, without which the rest of a product cannot be built.”

Part of the reason for the problem is a general concern in the chip industry — Infineon included — not to chase too much demand and overinvest. The fear of building up excess capacity that in the coming years will erode profits remains part of the collective memory of Infineon from the DRAM glut during the 2000s.

Yet that solves only part of the problem.

Infineon would ideally prefer to invest in its own production of specialized semiconductors like inverters for power electronics, a product segment where foundries are absent. Yet the bottlenecks have gotten so bad that Infineon sees no other alternative than to invest in its own capacity in Dresden to produce certain outdated microcontrollers, something it is typically loath to do for commodity chips.

Hmmm. Not economically desirable, but necessary as a last resort. In Part Two we saw in M.U.L.E. that when goods are scarce, the price goes up, and rational economic actors will take advantage of the situation to add supply. The tire swing is stuck, and we’re all just waiting for it to come down again, but it’s not. Somebody has to do something about it.

TSMC, as I mentioned in Part Three, is doing something about it, by adding more capacity — on 28nm. Meanwhile, we’re still waiting for that 40nm - 90nm tree swing to come down. As Anton Shilov mentioned on AnandTech:

But the cheap wafer prices for these nodes comes from the fact that they were once, long ago, leading-edge nodes themselves, and that their construction costs were paid off by the high prices that a cutting-edge process can fetch. Which is to say that there isn’t the profitability (or even the equipment) to build new capacity for such old nodes.

This is why TSMC’s plan to expand production capacity for mature and specialized nodes by 50% is focused on 28nm-capable fabs. As the final (viable) generation of TSMC’s classic, pre-FinFET manufacturing processes, 28nm is being positioned as the new sweet spot for producing simple, low-cost chips. And, in an effort to consolidate production of these chips around fewer and more widely available/expandable production lines, TSMC would like to get customers using old nodes on to the 28nm generation.

“We are not currently [expanding capacity for] the 40 nm node” said Kevin Zhang, senior vice president of business development at TSMC. “You build a fab, fab will not come online [until] two year or three years from now. So, you really need to think about where the future product is going, not where the product is today.”

Okay, maybe the price hasn’t climbed up enough to motivate more capacity? That tree swing should come down soon. Yep. Any day now....

Scooped by Doug, Kurt, Steve, Ganesh, Willy, and BCG

I’m doing this part-time, so it’s taking me a long time to research all this stuff and write it up. A few others who are involved more closely in the chip shortage have discussed the issue more quickly than I have; they’ve made some astute comments, which I wanted to share.

Fabricated Knowledge, Nov 2021

Doug O’Laughlin at Fabricated Knowledge wins first prize for bringing up the mature-node capacity conundrum in November 2021, in his article The Rising Tide of Semiconductor Cost.[6]

It’s not just the newest, fastest, and most expensive chips that are costing more because of technological problems. The most interesting trend recently is that old chips are starting to drive price increases. The recent inflection in Automotive semiconductors is driving demand, not for the latest and greatest, but older and more mature technologies. The problem is that there never has been meaningful capacity added for older technologies, and most of the time fabs would just become “hand-me-downs” as the leading edge pushed forward and the fab equipment would continue to be used and depreciated. Using fully depreciated fab equipment meaningfully lowered the cost to make a semiconductor, especially after 10+ years.

That’s been an important aspect of pricing. A fab being maintained without much incremental capital, yet still producing chips, is what has driven down the price of older chips so much over time. Often there would even be improvements in yield, which further lowered costs. The thinking went that a leading-edge chip that cost hundreds of dollars in 2000 would cost pennies in 2021 because the fab would be fully depreciated.

But there’s a problem with that. We have a situation that’s never happened before. Historically, demand was clustered toward the leading edge. Today, we have demanded at the lagging edge, too. In fact, the demand for older chips is starting to rise sharply. This is driven by automotive and IoT production, as most of these applications are older, more mature technologies, which have better yield, cost, and, importantly, reliability. And even though demand has spiked, no capacity has been added. It’s very rare to add to lagging-edge production.

Up until very recently, there was ample capacity at the lagging edge. It would have been unheard of to add capacity to the lagging edge. A fab was considered “full” if it was running at 80%. Now, trailing-edge fabs are running at close to 100%. Something must change.

This is one of the drivers of the automotive semiconductor shortage: higher demand and no supply or incentives to add more supply. Most semiconductor firms and fabs are obsessed with the leading edge because they can make higher profits. Now firms are waking up to maintaining the lagging edge. The “old” chips are becoming just as important as new chips, and to add capacity, firms had to start making large capital additions again.

There are obvious pricing problems in this scenario. With a new fab, a company can’t turn a profit selling a lagging-edge chip at the price that was previously dictated by a fully depreciated fab. Prices have to go up.

O’Laughlin then quotes Silicon Labs’ CFO John Hollister in the company’s October 2021 earnings call,[7] talking about the normal routine of fab hand-me-downs:

Hold on just one — just in terms of the cost increases and the durability of that, I think that we are going into kind of a new phase of the semiconductor industry, where we’ve got Moore’s Law and advanced nodes becoming more and more expensive and you’ve got mainstream technology now full. And it used to be that the digital guys would move out and the N minus one, N minus two, N minus three nodes would those would be fully depreciated fabs that you would move into. We have now reached a point where the mix, the ratio between advanced and mainstream is causing fabs TSMC and others to build new mainstream technology, and that means that those fabs are not fully depreciated.

So a large element of the cost increases that the industry is seeing right now is because of the additional Capex that’s having to be put in to build new capacity across the nodes, not just at the advanced nodes, but across. And so if you look at the cost increases that we’re seeing in other — it’s across the industry, there is a certain element of that, that’s durable over time. And so this is a step function in terms of the cost structure of the industry to match the demand that we’re seeing and the increasing content of electronics throughout the economy and the acceleration of demand that we’ve seen through the pandemic has really push that forward and driven us into the supply constraints.

We’ll work through that, but to work through that is requiring a lot of Capex, and that’s got to be recouped. And that’s got to flow upstream from our suppliers to us, to our customers and that’s what you’re seeing right now.

O’Laughlin goes on to state that supply constraints at the lagging edge are likely to continue, something that is “completely new territory for the industry”:

There is really only one solution — adding capacity — but fabs are uncertain they could make a profit adding greenfield lagging edge without raising prices. In turn, they’re concerned about the demand for these “new” lagging-edge chips.

It’s a standoff. Fabs and semiconductor companies are uncertain that this will last, but Auto OEMs and the like seem to have insatiable demand. For a fab, it’s hard to change your behavior in one year against a trend that has lasted decades. Even harder is to go through those difficult changes and expect customers to accept higher prices.

I’m just quoting here, so if you want to read more, I’d suggest reading the original article. It’s now been a year, and things really haven’t changed much. Everybody’s hoping the tire swing will come down and get back to normal.

NXP, September 2022

NXP participated in Citi’s Global Tech Conference in September 2022, with CEO Kurt Sievers engaging in quite a candid discussion with Citi’s Chris Danely. I’ve posted an extended transcript below in the Addenda section, but here are some highlights:

  • The worst constraints are at the foundries in the 28 - 180nm range
  • In addition to working on increases in supply, NXP is engaging in more direct communication with car companies to understand future demand
  • Decrease in demand from PC/consumer/mobile doesn’t affect supply at these constrained nodes
  • NXP purchases finished wafers from the foundries in the 90nm to 5nm range, making up about 60% of their supply — not clear if this is revenue or number of wafers — and is expected to grow to 70-75%.
  • NXP believes it is covering 80% of its true demand (after discounting possible double-orders) by its supply capacity.

There are two areas that I want to cover in a little more detail.

One is on the topic of non-cancelable orders.

Sievers stated that much of what NXP considers as true demand from its customers — and it’s only able to service 80% of this demand — comes from non-cancelable orders, much of which are in the automotive and industrial markets:

A lot of that is these famous NCNR orders, which are the non-cancelable non-reschedulable orders, where in the meantime, the level of those for next year is bigger than what we have for this year. I mean, it’s just important in the current environment where everybody fears about the macro breaking together and whatever, our NCNR order level, which is firm, confirmed orders through the end of next year, are bigger already now than what we have for this year, with a move between— inside the portfolio.

So, we have none of those anymore for mobile, which is probably not surprising to anybody, but we have a significant increase in industrial. Automotive is about the same, which is in line because automotive was the first to escalate. So, a lot of our automotive customers gave us these orders already early for ’21, and they continue to do for ’22.

In industrial, industry has only found out too late, so they were too late to give us these NCNR orders for ’21, for ’22 — so the learning was now to be early to do this for ’23. So, we see a significant uptick, actually, in these NCNR orders for next year. But all together 80% coverage. Now, I don’t know how things move going forward, but that gives you a feel on how robust that still is.

The increase in “commitment” activity in the semiconductor industry (NCNR being one example) is a change that’s come out of the chip shortage. This is involved in both sides of NXP’s supply chain, with commitments also between NXP and its foundry suppliers:

Kurt Sievers — I mean, I can be very open what we have, we have like \$4 billion of supply, or purchase obligations from our perspective, from foundries, which sounds like a big number, but if you know that this is spread more or less evenly over five years, and we do in the quarter more than \$3 billion revenue, then it’s not much at all. It’s actually a pretty small number. And it’s more than out-balanced by NCNR commitments which we have from our customers.

Chris Danely — Yeah. So, it just kind of moves on down the line, right?

Kurt Sievers — Philosophically, I think it is a good concept, because especially in these industrial and automotive markets, design wins and the lifetime of design wins are very lengthy. So, there is enough visibility to enter into these kinds of agreements. And I think it just helps the whole supply chain. So, in principle, I think it’s a good thing. And it goes a bit away from this call-by mentality, which we might have had in the past.

Is it going to cover everything going forward? No, I definitely don’t think so. But for certain deals and certain specific single-source components, I think it’s a good move.

Chris Danely — Has it changed the capital intensity at all, up or down, these upfront payments? Or is it just the same part, you’re just paying a little more earlier?

Kurt Sievers — It’s not even always that we have to pay something upfront. It’s more the obligation that we take it over time. So, but there is all sorts of constructions, I mean, that there is different deals with different foundries, I guess, by all of our competitors doing, everybody is doing a little bit his or her thing?

No, I don’t think overall it changes the capital intensity, but it hopefully forces people to be more thoughtful about what future demands are. And with that, what future investments we should actually take or not take.

And the second topic is price increases, which Sievers summarized as follows:

First of all, the policy we have very transparently taken right from the start is that we will pass on all of the increased input cost to our customers in such a way that it exactly protects our gross margin percentage. Not abusing it to pad gross margin, but also not being the victim in the middle, which is losing profitability on this.

We’ve done this right from the start, I think Bill and I had a very sharp eye to be really precise on this, and we are very open to our customers, this is how it works. And so far, we have found reasonable, I would say acceptance. I mean, raising prices is never easy, but I would say, so far so good.

I do not think this is behind us. At least in the technologies and process capacities which we need. I dare to say it’s— it looks pretty obvious that our input costs will continue to go up next year. So, we are somewhere in the middle of it, but it continues, which also means that NXP will have to raise prices also next year. So, we are not at the end of this. And that has to do with the fact that foundries keep investing into these mature nodes and there continues to be a supply and demand imbalance in these mature nodes.

The topic of price increases is a tricky one. If I put on my customer/consumer hat, when I hear price increases I have one of those visceral reactions — NOOOOO!!!!! — because I know it’s just one more burden for people to bear, and it adds to the inflationary environment. But from an economic perspective, price increases are important to alleviate the supply-demand imbalance. For a company like NXP, if it charges more, then it should incentivize lower demand from its customers. If NXP’s foundry suppliers charge more, then that should help incentivize the creation of more fab supply. Both should help bring supply and demand toward equilibrium.

NXP’s strategy is the “honest shopkeeper” approach to maintain gross margin, passing along what it thinks is a fair and transparent price increase to its customers, to maintain a healthy relationship with those customers. It’s a middle ground between eating the cost increases — which is what would happen with a commodity if one producer’s costs increased but the commodity price stayed constant — and price-gouging. Is that the best choice in this drastic shortage? I don’t know; it’s something to think about.

Microchip Technology, May 2022

Steve and Ganesh are the top executives of Microchip: Steve Sanghi, Executive Chairman of the Board, and Ganesh Moorthy, President and CEO. They’ve been commenting on the chip shortage regularly in earnings calls. Here they are in May 2022:[8]

Christopher Rolland (Analyst, Susquehanna International Group) — Great. Thank you, Ganesh. And then one for either you or Steve. A lot of people have talked about potential over capacity for the industry in ‘23 and ‘24.

Would love to get kind of your views on whether you think that’s a thing or not? And what its effect on industry pricing might be? Thanks.

Ganesh Moorthy (President and Chief Executive Officer) — So I’ll give you my view and then maybe Steve will add to it as well. So it’s a bit of a misnomer when you talk about overcapacity in ‘23 or ‘24. When you look at where is the capex being spent by the industry, right? Industry spent over \$100 billion of capex last year. The vast majority of that capex, over 90% of it, is being spent on the bleeding-edge nodes.

These are the nodes that are 16-nanometer and smaller, so 16, 10, seven, five, three, etc., is where all that is being spent. Where the capacity is not being invested in at the rates that are required, and where, for example, all of the constraints that the industry is fighting through — short, medium, and long term — are on these trailing-edge, specialty technologies. On 300-millimeter, that is typically anything which is 40-nanometer and larger in size: very little capacity investment coming online to be able to help that. On 200-millimeter, eight-inch wafers, there’s almost nothing that is being done, outside of what some of the IDMs have been doing, and which really is still a far cry from what is needed.

So while there is capex spending taking place, of quite significant amount, it is being spent disproportionately on the bleeding-edge technologies, and there are still significant constraints left on the trailing-edge, specialty technologies that we don’t see easing up into ‘23 and ‘24. Steve, do you want to add some more to it?

Steve Sanghi (Executive Chairman) — Yes. Yes, I would. So what has happened historically is that the foundries built a leading-edge fab, depreciated it fully over four years, providing leading-edge chips to the likes of Qualcomm and AMD, and others. And when the leading-edge guys moved to the next node, then they took that capacity, a depreciated fab and repurposed it for microcontroller, mixed-signal, connectivity, and those kind of products.

And that’s how over many, many years, trailing-edge capacity became available. Now what has happened now is that link is broken. The leading-edge lithography has gone to 14-nanometer, 10-nanometer, even seven, five, and three-nanometer, while the microcontrollers and analog, because of functionality needed, are still in the range of 65-nanometer to 180-nanometer. And so, therefore, the trailing-edge capacity no longer easily becomes available, because somebody moved to the next node.

Secondly, starting at 90 nanometer, the wafers became 12-inch, less than 90 nanometer, the wafer’s at eight-inch. And eight-inch is largely aluminum back-end and 12-inch is largely copper back-end, and one is not compatible with the other. So a 12-inch fab becoming available, doesn’t easily give the capacity for an eight-inch product to move to 12-inch. So combination of those factors and the fact that the foundries are adding almost no capacity on the trailing edge, it is quite possible that the trailing-edge capacity is forever constrained.

And that’s why we are making aggressive attempts to add capacity internally to provide that growth to our business and to our customers, and you’re seeing some of our competitors do the same thing.

Wait, what?

it is quite possible that the trailing-edge capacity is forever constrained.

I don’t know that I agree with Sanghi on “forever”, but there is no easy answer here, and it’s going to take time for the industry to come to some kind of agreement. TSMC isn’t rushing in to save the world, and you’re hearing companies in this space, like Microchip and TI, essentially go it alone. I mentioned in Part Three that the only way Moore’s Law keeps on chugging away at the leading edge is a coordinated effort. It lowers risks just enough to make the financing work. Going alone is a tough road.

As for TI: it’s a large enough company that it is pursuing several 300mm capital expenditures at once, something I mentioned in Part Two.

For Microchip, there was some really interesting news in the November earnings call, namely the possibility of a 300mm fab:[9]

Ganesh Moorthy — While we are seeing some loosening of constraints in our supply chain, we continue to have several internal and external capacity corridors that remain very constrained. We are continuing with our carefully calibrated capacity increases, seeking to serve what we believe is a long-term consumption growth. We believe our calibrated increase in capital spending will enable us to capitalize on growth opportunities, serve our customers better, increase our market share, improve our gross margins, and give us more control over our destiny, especially for specialized trailing-edge technologies.

As you may have seen, Microchip has expressed its view that the recently approved CHIPS Act is good for the semiconductor industry and for America, as it enables critical investments which will even the global playing field for U.S. companies, while being strategically important for our economic and national security.

For a very long time, an important component of our business strategy has been to own and operate a substantial portion of our manufacturing resources, including wafer fabrication facilities in the U.S. This strategy enables us to maintain a high level of manufacturing control, resulting in us being one of the lowest-cost producers in the embedded control industry.

In light of this strategy, and potential grant funding from the CHIPS Act, the Investment Tax Credit provision, as well as state and local grants and subsidies, Microchip is in the early stages of considering a 300-millimeter U.S.-based fab for specialized trailing edge technologies.

This fab project, if we decide to pursue it, would be intended to provide competitive growth capacity as well as geographic and geopolitical diversification. The availability of grants, subsidies, and other incentives will all be important considerations in our analysis, and will also help determine the location and timing for the fab.

In the later Q&A session, Moorthy mentioned the potential 300mm fab would not be a quick effort, that it is “something we think about over a 20-year-plus time frame” and

I think on 300-millimeter, where — if we started on a fab tomorrow, it’s 4-plus years away before that fab is starting to ramp. So these are not decisions we make in a single cycle. We think through these across cycles on a long-term secular growth basis and what our position is and what we want our capabilities to be out in time.

Moorthy also handled a question about the mix of process nodes in a 300mm fab:

So firstly, on process technologies, those are still being worked. But largely, we use our 300-millimeter foundries today on process technologies that are 90-nanometer and smaller in size. And the workhorse technologies for trailing-edge tend to be at 40, 65, 90, in that general neighborhood. But those — we wouldn’t limit ourselves just to that. Again, I want you to think of this as— this is a 20, 25-year look at what we would do with the 300-millimeter fab. The reasoning for it is we have — as our business has grown, the portion of our business that we do with 300-millimeter has also grown. And the investment in the trailing-edge part of 300-millimeter technologies has not been there with many of our foundries at the level that we have wanted. And — but it takes a certain scale to get there.

And if you had a full-boat fab that you needed to build, the way in which the breakeven points and the absorption points come about are different from when there is a fab that can be built with government funding and the Investment Tax Credits and whatever local things come in. So clearly, that has changed the equation as to when does it make sense financially. But that’s not the only reason why. We think trailing-edge 300-millimeter technology is going to have constraints for a long time to come and a portion of that being within Microchip would allow us to better serve those markets.

There are those 40-90nm nodes again. The question remains — will there be enough justification to build a 300mm fab, even with CHIPS Act funding? And if not, how is Microchip going to obtain capacity in this range?

Willy Shih

Willy Shih is a supply chain expert and a frequent contributor to Forbes Magazine. He’s written about the semiconductor shortage several times. Here are a few excerpts.

From Aug 3 2022 while commenting on the CHIPS Act:[10]

Chip shortages were caused by changing demand patterns brought on by the pandemic. There was a burst in demand for work-from-home gear like notebook computers, appliances, and other goods, and manufacturers had a hard time shifting production. I explained this a while ago, but it’s important to understand that several issues have gotten conflated here: loss of leadership on the high-end, but an inadequate capacity to meet the surge in demand for trailing-edge technologies (i.e., the older stuff that is no longer leading-edge). Capacity for trailing-edge technologies was tight before the pandemic because making those kinds of chips is not particularly profitable and you want to run your factory pretty fully loaded so you can make some money. If you build a new factory with some of that subsidy money, when it comes on stream in two-plus years it will certainly help with the shortage. But by then, a lot of people are already worrying about a glut, because manufacturers have already been adding capacity like crazy.

From Nov 3 2022, explaining how the market segments have behaved differently and why we still have a mature node shortage:[1]

Most automotive semiconductors are produced on “mature” nodes. A recent McKinsey report stated that most of the demand in this sector was for 90 nm and above. For perspective, the 90 nm node was the bleeding edge of chip technology around 2002, twenty years ago. Partly this was because the types of components used in vehicles didn’t benefit from newer technologies, and there is a long and expensive process to move them to more modern nodes. The factories making these chips use older tools, and since these are not terribly profitable commodity parts, there has been little incentive to invest in growing the capacity. Capacity was already tight before the pandemic, and then during a roughly eight-week period in 2020 when car factories were mostly shut down, most of the OEMs pulled back their orders. Meanwhile an explosion in demand from other sectors filled all that manufacturing capacity, so when the automakers came back to reorder, the lead times had extended way out. They are still catching up.

And most recently, from Nov 20 2022, focusing on the automotive chip shortage itself:[11]

There are several standout features of automotive chips. The first is that they must operate for a long time over wide temperature extremes while subject to lots of shocks and vibrations. Automakers expect an operating lifetime of 15 years and tolerate a failure rate of zero parts per billion during that time. They also want replacement parts to be available for 30 years. Most consumer electronic devices (like your phone) have failure rates measured in parts per million and would be considered obsolete after five years. If your PC encounters an error, reboot and give it another whirl. If your engine controller suddenly fails, you don’t pull over to the side of the road and reboot (although I have heard of something like this happening with an electric vehicle’s infotainment system). The Automotive Electronics Council (established by the Detroit Big Three) maintains a range of qualification standards for chips. For operating temperature, it defines Grades 0, 1, 2, and 3 operating ranges, with Grade 1 covering -40°C to +125°C and Grade 2 from -40°C to +105°C. That has a high-end limit hotter than the temperature of boiling water, by the way. This is a considerably more challenging range than most consumer chips will ever see. The chips need to be reliable, so they must be designed and tested to have a sufficient operating life under extreme conditions.

The second requirement is they must be designed with safety in mind. A lot of this is covered by ISO 26262 – Functional Safety Standards, which covers a range of things beginning with how they are designed to how failures are handled.

Finally the processes for making chips at semiconductor fabs have to be “qualified,” which typically takes six months. The fabs also need to have modifications to their process design kits for high temperature device models, thicker interconnects, and other things that enhance reliability. After that the chips must be extensively tested before they can get built into vehicles. That means accelerated life testing at elevated temperatures and harsh conditions to simulate many years of service. Mainstream automakers have taken as long as 3-5 years to design, test, and validate new chips.

I pointed out earlier that many automotive microcontrollers use 90 nm technology, and it has been difficult to add capacity. The shortages over the past two years have prompted some automotive chip vendors to migrate to 65/55nm nodes, and some have even jumped to 40 nm. But DigiTimes says it will take as much as five years for the new chips built with 40nm processes to clear validation processes and get put into new vehicles, which means the existing technology will be in use for some time to come. And that’s why the auto chip shortage is talking longer than most to alleviate.

I agree with most of Shih’s points, but there’s an assumption here, and that is that 90nm is the problem child and 65/55nm and 40nm are better in terms of availability. As I mentioned in Part Three, we don’t have a lot of visibility to this; TSMC did discuss this in a January 2021 earnings call:[12]

Robert Duncan Cobban Sanders (Deutsche Bank AG, Director, Research Division) — Yes, I just got one question, actually. Just could you please then comment more on the wafer shortage situation and how severe it is at present? Which node are you seeing the shortage most acute? Is it 65-, 90-nanometer, 0.11, 0.13, whatever it is? And how far out are you essentially booked out at some of these nodes? And do you think there’s wafer upside to what you’re pricing at these nodes?

Jeff Su (TSMC, Director of Investor Relations) — Okay. So Robert, your question is on the tightness or shortage in the wafer. He is asking, is it at particular node such as 65-nanometer, 90-nanometer, 0.13, how short it is and how long it will last.

C. C. Wei (TSMC, CEO) — Robert, most of the shortage actually is in the mature node. It’s not in the 3-, not in the 5- or 7-nanometer per se, but in all the mature node, especially in 0.13 micron, in 40-nanometer, in 55-nanometer, in those area.

Kudos to Robert Sanders for asking. But that was nearly two years ago. The analysts have had seven earnings calls since then, and none of them have brought it up again. :-( So we don’t know how things have played out. The closest TSMC has gotten to this topic was in the July 2021 earnings call, in a response to a question from Krish Sankar of Cowen and Company, about the technology nodes used by TSMC’s automotive customers:

C. C. Wei (TSMC, CEO) — Krish, I think I have mentioned that automotive MCU is the biggest one that we have and it’s in 55-, 40- and 28-nanometer with the majority still in 55 and 40. And in the next 2 to 3 years, it will be moved to 28-nanometer. That’s in our current plan, and we are working with our customers on that.

This sounds promising, but I still don’t see that things will magically be solved in the next few years. I pointed out in Part Three that the economics are prohibitive for smaller low-cost chips to move to more advanced technology nodes, except at incredibly high volumes, because of the increase in up-front design and manufacturing costs.

Mobile phones and other consumer electronics are able to take advantage of the cost savings of semiconductor integration by combining functions into one chip: more complex integration on a chip opens the door to using a smaller technology node. In the automotive industry, this is not so easy — notably because cars and trucks are physically large objects. I can say with certainty that there will never be a single car-system-on-chip, because there are so many sensors and actuators and inputs and outputs distributed around an automobile. The industry is trying to get at least some benefit out of integration, through zone controllers to simplify the interconnection between different sections of automotive electronics… but I’d be willing to bet that in 20 years you’ll still see some individual MCUs or voltage regulators scattered around inside a car, and they’ll be using a technology node that makes the most financial sense for each chip.

Boston Consulting Group

Boston Consulting Group put out a report recently (October 2022), titled Automotive Industry Semiconductor Outlook, along with a summary article[13] that starts out ominously:

The semiconductor shortages that have plagued the automotive industry may persist in some form through as late as 2026.

Yikes! Pardon the interruption; that was just a knee-jerk reaction of mine. I’d better let the BCG folks continue.

The semiconductor shortages that have plagued the automotive industry may persist in some form through as late as 2026. Pandemic-induced manufacturing and logistics challenges are easing; consumer chip demand has reached a saturation point and is in a cyclical decline. But continuing growth in automotive semiconductor demand is inevitable, as penetration of semiconductor-intense automotive applications, such as higher levels of advanced driver assistance systems (ADAS) and electrification, increases.

The growing sophistication of vehicles means that, despite recent capacity investments, the imbalance between chip supply and demand will persist—at least for some types of semiconductors—over the next four years. However, the nature of the chip shortages and the device types affected will change over time, requiring automakers to actively manage risks as the situation evolves.

Here’s the lead content from the report itself:

The automotive semiconductor market is expected to grow by more than 9% annually through 2030.

  • The adoption of electric vehicles (EVs) and advanced driver-assistance systems (ADAS) will substantially increase the semiconductor content in vehicles, even as production volume remains steady.
  • Battery electric vehicles (BEVs), which are expected to have the highest market share among EVs by 2026, have twice the semiconductor content of internal combustion engine (ICE) vehicles, owing to the need for discrete-power and analog chips.
  • ADAS Level 2+ is expected to gain the largest market share among assistance systems. Each additional level of sophistication exponentially increases the need for memory and logic computing.

As a result, some semiconductor supply challenges are expected to persist through 2026.

  • Shortages of analog chips and MEMS may persist given limited planned-capacity investments.
  • Discrete-power chips may experience additional demand pressure with the adoption of 800-volt vehicles; there may be insufficient wide-bandgap manufacturing capacity to meet demand.
  • Approximately 50% of future fabrication capacity is planned in mainland China, which will increase risk if the planned capacity does not come online or is inaccessible to Western OEMs and Tier 1 suppliers.
  • Automotive demand growth will be highest for logic chips made on 20nm to 45nm nodes in order to meet the increasing computing needs of centralized electrical/electronic architectures; we expect this to ease demand pressure on mature node sizes larger than 55nm.

There’s a lot to parse here, but the report itself is an easy read, at only 14 pages of content, mostly consisting of graphs and charts and bullet points, like this one on page 13:

This gets into some of the issues I mentioned in Part Three, namely that new mature-node fabs can’t compete economically against existing depreciated fabs at the same technology node. But the “28nm is the current sweet spot for cost-per-transistor” meme is unsubstantiated, and contains unwritten assumptions — at what volume? at what die area? including or excluding design costs? I’m more inclined to believe Scotten Jones’s relative-cost-per-unit graph from Technology and Cost Trends at Advanced Nodes[14] — I included this graph in Part Three, which shows the cost-per-transistor still decreasing down to 5nm.

Slide used by permission of IC Knowledge LLC.

But again, the only catch is that the volumes need to be high enough so that the costs of design and masks don’t negate that cost advantage.

At any rate, the BCG report raises a number of issues, of which you can get a sense just by reading the slide titles:

  • Pandemic-induced manufacturing and logistics challenges are easing, but supply issues will persist
  • The auto industry currently occupies a small share of the semiconductor market, but it’s growing rapidly
  • Increasing semiconductor content per vehicle will promote demand, even as total vehicle production remains steady
  • BEVs are expected to have the largest share of the market and require the most semiconductor content
  • On average, BEV powertrains require more discrete-power chips and analog content than internal combustion engines
  • ADAS Level 2+ will see the highest penetration growth through 2030, increasing demand for logic and memory
  • Analog and MEMS will be the key semiconductor challenges through 2026
  • Automotives’ transition to more advanced logic nodes should ease demand pressure on mature sizes larger than 55 nanometers
  • For nonlogic chips, automotive growth will place stress on analog and MEMS
  • Uncertain access to mainland China’s fabrication capacity may increase risk in the automotive supply chain
  • Outside of mainland China, underinvestment in mature capacity persists owing to a cost penalty for new fabricators and older chips
  • Automotive OEMs are implementing new semiconductor engagement models

I’m not sure I agree with all of the conclusions, but I think most of BCG’s points are worth considering.

Aftertastes: Takeaways After Digesting Five Scoops

There’s a lot of turbulence and uncertainty and signs of continued supply-demand imbalance ahead.

I think the biggest lesson is that we cannot take the dynamics of the semiconductor markets for granted. It’s not always like DRAM uniformly across the industry, with building-new-capacity to the rescue that then causes a glut and leads the way to the next shortage — a roller-coaster ride where there’s just enough uncertainty in timing and magnitude to keep life interesting. There are enough respected voices out there raising reasonable concerns. And we ought to at least keep an eye on things for the long term.

What about the short term? What else has been going on in 2022, and what’s in the tea leaves for 2023?

Clues from Earnings Calls

I’ve been reading transcripts of this years’ quarterly earnings calls from a number of companies. The ones most relevant to the continuing chip shortage, aside from the foundries, are probably the larger industrial and automotive semiconductors manufacturers with their own fabs. These include:

The earnings calls are a chance for company management to present and interpret on what’s happened during each quarter, and for financial analysts to ask questions. I mentioned in Part Two that this is a prime chance to find out something interesting besides the accounting numbers. The analysts are always searching for any forward-looking signals in the wind, like trackers eager to catch a scent of their prey. And there’s a lot of interesting little tidbits that have happened in 2022’s calls. Here are a few themes that have come up in the last twelve months.

(And again, nothing in this article should be construed as investment advice.)

Increasing internal manufacturing capacity

All of these companies mentioned increasing internal manufacturing capacity in at least two earnings calls in the past twelve months.

In this subsection and the following ones, I’m going to show a bunch of tables like the one below, which list a checkmark when I noticed a topic was mentioned in the earnings calls. (Lack of a checkmark doesn’t mean they didn’t mention it; I may have missed some.) The column labels Q1 - Q4 represent the calendar quarter in which the earnings call took place — Q1 = January – March, Q2 = April – June, Q3 = July – September, Q4 = October – December — and refer to the preceding quarter’s results. (Note that some companies use a fiscal year that does not line up with the calendar quarters, like Analog Devices, whose fiscal year ended October 29, and Micron Technology, whose fiscal year ended September 1. I didn’t cover Micron in this article because they’re in the memory segment, which has its own not-so-little market dynamics — see Part Two — that are not really impacting the automotive chip shortage.)

Increasing capacityQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

What’s not available is a sense of how much they are increasing capacity. We can look at numbers about revenue, but unless they’re touting new fab construction or purchases, it’s hard to tell whether “increasing capacity” means they added 1% more manufacturing capacity, or they doubled it. Some highlights:

  • Diodes Incorporated bought ON Semi’s fab in South Portland, Maine, and has been ramping down capacity commitments from its “GFAB”, TI’s former fab in Greenock, Scotland. I don’t know how often this happens, but to smooth out the disruption in production, sometimes a fab sale is accompanied by an agreement to continue production for its former owner. In this case it means that Diodes Inc is still producing a decreasing amount of chips for TI at the Scotland site, but gradually the fab output will ramp up for their own production. From the company’s May 4 call:

    If you look at the GFAB we purchased in year 2019. That facility, we are supporting Texas Instruments, and that support is coming down 10% a year, the loading coming down 10% a year. Therefore, we already start to qualify Diodes’ product to ramp it up in that area – I mean in that fab. So, you can see when Texas Instrument demand or our commitment to the Texas Instruments are down 10% each year, we can go up 10% to support our own demand. And remember their 10% and compared with our number there, then wafer is much more than what we’re talking about 10%. So that is GFAB. Then, obviously, you already know our acquisition for the wafer from onsemi, and that we expect to close next month or end of this month, that’s what we expecting. Okay.

  • TI: increases in capacity in Richardson, Texas, bringing the former Micron fab in Lehi, Utah online, and two new fabs coming up in Sherman Texas — see Part Two, and numerous other recent articles; for example, EE Times covered the company’s analog expansion in October:[15]

    Not that anyone needed more proof that the analog segment in the electronics industry is booming, but Texas Instruments (TI) is serving it up anyway: On Sept. 29, TI announced it began initial chip production at RFAB2, a new 300-mm analog wafer fab in Richardson, TX, that is connected to RFAB1, its 13-year-old 300-mm analog wafer fab. A few years from now, the pair will be able to produce more than 100 million analog chips every day, TI asserts.

Passing on cost increases

As I mentioned with NXP, inflationary cost increases from suppliers are being passed on to customers.

In many cases (Analog Devices, Microchip, NXP), company executives are emphatically stating that they are not trying to take advantage of the situation (think Smithore Gambit) and are only increasing price enough to maintain gross margins.

Others (Diodes Inc, ON Semi) have stated that they intend only to pass on supplier price increases to customers, but it’s not clear whether they are including gross margin in their price increases — in other words, if they sell chips for \$1 each and it costs them 40 cents to make them, but now their cost went up 4 cents (10%) to 44 cents each, will they just pass on the raw cost increase (sell for \$1.04 each) or maintain gross margins (sell for \$1.10 each)?

Infineon and TI have mentioned this idea of “market prices”; here’s Infineon’s Andreas Urschitz on the company’s August 3 call:

We have been historically a bit slower in average to increase prices relative to the one or the other peer. However, we are on track now to very decisively managed to deploy market prices, which allow us to extract a very fair share of the industry profit pools towards our P&L. And when I fair, I mean fair for both, our customers and Infineon. That’s where we stand.

And TI’s Rafael Lizardi on April 26:

On the pricing in general, we are pricing with our customers. Our process on that has not changed. Our process is to price to market. And as prices have moved up over the last two or three quarters, and that certainly did happen in first quarter, we have moved our prices as well, and growth in first quarter did benefit from the pricing tailwind.

Market prices presumably mean they decide on a case by case basis to stay competitive in a general sense....

ST and Renesas have not mentioned any strategy on this topic; Renesas just stated “we are not in the phase of lowering prices” on July 28, and ST has only mentioned the idea of a “favorable pricing” situation in several of their earnings calls.

Pass on cost increase / price to marketQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Backlog and inventory

One theme for 2022 is that the backlog — that is, the amount of customer orders that have been placed but not yet fulfilled — is very high (sometimes characterized as “strong”), or increasing.

Backlog high / increasingQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Some of these companies have emphasized that they would like to see more inventory, or that their inventory is increasing but this is a good thing.

More inventory = good!Q1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Others mentioned an interesting concept, that they prefer in a constrained environment like this to keep inventory outside the “channel”, which is earnings-call-jargon for the distribution channel: that is, instead of sending their manufacturing output to distributors to relieve shortages, they will allocate that output directly to the customers they select.

There’s a potentially unwanted side effect to this strategy: it will probably help larger customers more than small customers, which have to work harder to get the attention and assistance of the semiconductor manufacturers. If Catherine Hepbull from Rooster Novelty Products calls up Kansas Instruments to ask for help getting 5000 microcontrollers to meet production needs for her internet-enabled dog whistle, she’ll probably get sent back to the distributors — whereas if Helen Hornblower, VP at Frob Motor Company, complains to KI that her suppliers are short by 2 million microcontrollers she can’t get to make this year’s Frob Fission and Frob Exploder models, she might have a chance at getting at least some more supply. When revenue talks, suppliers listen.

Prefer inventory outside channelQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Golden Screw

The “golden screw” concept has come up, sometimes directly, sometimes in the guise of “kitting”, which is the process of procuring and gathering parts together for assembly.

golden screw / kitting issuesQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

In some cases there has even been a vicious cycle, where the semiconductor manufacturers can’t get new equipment as fast as they would like, because the equipment manufacturers can’t get the semiconductors they need. SEMI, one of the semiconductor industry organizations,[16] called this issue the SME (semiconductor manufacturing equipment) multiplier effect.[17]

Alarm over this issue traveled all the way to an article in the Wall Street Journal in early May 2022:[18]

The drought in chip availability that has hit auto production, raised electronics prices and stoked supply-chain worries in capitals around the globe has a new pain point: a lack of chips needed for the machines that make chips, industry executives say.

The wait time it takes to get machinery for chip-making — one of the world’s most complex and delicate kinds of manufacturing — has extended over recent months. Early in the pandemic it took months from placing an order to receiving the equipment. That time frame has stretched to two or three years in some cases, according to chip-making and equipment executives. Deliveries of previously placed orders are also coming in late, executives say.

CEO Ganesh Moorthy mentioned Microchip’s prioritization of these orders to help alleviate the problem in the May 9 earnings call:

And over the last 4, 5 quarters, it’s become more challenging to get equipment in on time. So delays are there. It’s a bit of a vicious cycle. Many of the delays are caused by shortages in semiconductor components. Those in turn, delay the equipment, which delays the ability to solve that problem. We have, in fact, taken the initiative to prioritize supply for many of the semiconductor equipment manufacturers, so that we do our part to both help the industry and help ourselves in doing that, and I believe others are doing it as well. But at the moment, the equipment lead times are getting worse, not better.

vicious cycleQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Cancellation / non-cancelable non-returnable orders

The term NCNR (non-cancelable non-returnable orders) has come up often:

NCNRQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Several of these companies take non-cancelable orders with some kind of preferable terms, like the higher priority of Microchip’s Preferred Supply Program (PSP), or perhaps better pricing. (Hotels and airlines do the same thing: you can get a reduced rate if you’re willing to give up the flexibility of cancellation.)

The reason to do this is not just a “gotcha” to get extra money from customers who change their minds; it’s a way to ensure that orders are real. There’s a lot of worry about double-ordering — we’ll explore this more in a future article — so non-cancelable orders are a way to encourage customers only to order what they need, rather than adding a little extra just in case. In a response to the question, “How non-cancelable non-returnable are your NCNR orders really?”, Kurt Sievers of NXP responded (July 26 2022):

Well, indeed. Sorry, I know that most part of an earlier question, I think I missed – I missed speaking to it. Well, it’s like with every contract. There is clearly a legal system in place. But in the end of course reasonable behavior will prevail. What I can tell you, however is that so far every one of them has been – has been fulfilled. I mean, there is no – we don’t see any in automotive industrial. There is no pushout, there is no cancellation. People want to have more products, now, how that looks at some time in the future, when maybe macro continues to change more, could be different. But fundamentally we put that system in place in order to make sure that our customers are serious. And think two times and three times about what level of orders they want to put on us and mind you that in many markets different products from different customers go into one end product.

So we have to make sure that there is a proper allocation between them then otherwise the car company is only suffering in the end. I mean, if there is too much orders from one Tier 1 and the other one orders not enough, then it doesn’t help the car company. That’s another reason why we are really trying to put NCNR in arsenal place in order to hold people to the truth.

This is also an example of companies mentioning that they have seen low cancellation of orders, meaning they expect much of the demand in backlog is real.

low cancellationQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Mature node capacity constraint

And, of course, the lack of foundry capacity investment in mature nodes gets a lot of blame — whether you treat “mature nodes” as 16nm - 90nm or 28nm - 130nm or 40nm - 180nm or something else.

mature node capacity constraintQ1Q2Q3Q4
Analog Devices
Diodes Incorporated
Infineon
Microchip Technology
NXP
onsemi
Renesas
ST Microelectronics
Texas Instruments

Other issues

There are a few other quirky insights that have come up in this years’ earnings calls.

  • All of these companies have foundry suppliers, and some have been prepaying for capacity (which I mentioned in Part Three). In its February 2022 earnings call, NXP actually disclosed that it had \$4 billion in purchasing obligations to foundries:

    Christopher Caso (analyst, Raymond James) — Kurt, I wonder if you could go through a bit about some of the commitments that you’ve had to make to secure that additional capacity. What is the commitment now to the foundry partners? And how has that changed versus prior cycles? And I’m guessing by your comments that your feeling is that that’s backed up by the commitments from your customers to you. And how resilient do you think that will be over time as conditions change? And obviously, this seems like this is something different for the industry.

    Kurt Sievers (NXP CEO) — Yes. Thanks, Chris. Maybe Bill, you want to speak a little bit first about the \$4-plus billion purchasing obligations, which we entered into?

    Bill Betz (NXP CFO) — Yes, that was in my prepared remarks. At the moment, we have \$4 billion. And those are strongly supported by actually greater than our supply commitments, backed by the non-cancelable, nonreturnable orders that we see. So the demand continues to be strong. Our customers are actually coming to us. And unfortunately, we can’t serve them all.

  • Analog Devices and Diodes Inc each mentioned their intent to pivot quickly to respond to changes in supply or demand. Analog Devices claimed (May 18 2022) they have “cross-qualified” fab processes with external foundries, meaning they can quickly move wafer orders between their internal fabs and their foundry suppliers to meet changes in demand effectively. Diodes Inc claimed (Nov 7 2022) they can shift production capacity from the markets that are decreasing demand (consumer / communications) to those that are increasing demand:

    Emily Yang (Diodes Inc Senior VP of Sales & Marketing) — I think the second part of your question is really about our ability to quickly adjust our capacity and support from one market segment to the others, right? I think the Q3 result is a good testament of our ability. So we did actually quickly adjust it from the slow demand markets like the low MPC consumers or the smartphone and to the automotive and industrial customer base, right? So all our factories are automotive qualified. And so that gives us the capability to quickly adjust. So not only the second — the Q3 but also the second quarter, I think we talked about the same thing as well. So I hope that will give you guys the confidence that we do have the capability and the flexibility to quickly adjust our support.

  • Analog Devices’ CFO had a positive spin on cancellation (Nov 22 2022):

    I would say that unlike others in the industry, we are proactively analyzing our backlog and working with customers to remove orders that they no longer want given the rapidly changing environment. So this strategy, for us, is to seek out cancellation. It helps us align our backlog with current demand, and it really gives us better visibility into where the supply needs and what we need to build. So that has increased our confidence in the quality of the backlog we have. It is still — the coverage is out still over a year, but it is down sequentially. And so while we’re always mindful that there can be some continued noise in that backlog, we feel pretty good about both the guide, and as we mentioned, the near term.

  • ON Semiconductor wins the Gobbledygook Award with this noncommittal statement from Feb 2022:

    We have pivoted to align our investments to the high-growth megatrends in the automotive and industrial, while implementing structural changes, capturing value for our differentiated portfolio and optimizing our manufacturing footprint to increase efficiency and improve our cost structure. These initiatives are translating into financial results as we achieved our record annual and quarterly revenue, gross margin, operating margin and cash flow.

In summary....

In summary, the typical automotive / industrial IC manufacturer has said something like this:

Mumble Semiconductor CEO: We are seeing strong demand across all segments, or maybe just across our automotive and industrial segments. Our consumer and computing markets are showing weakness, but our exposure to them is only moderate, and automotive and industrial is where the demand is at. (Except perhaps “legacy industrial”, in other words markets such as professional audiovisual equipment or teleconferencing, which are not consumer, meaning they are not exposed to the “whims” of the average person.) Actually, industrial might be starting to show some weakness. But not automotive, we see it zooming ahead for years to come. At least we hope that is true. The macroeconomic environment is challenging, but if we do see further reduction in demand, we expect a soft landing.

We are sorry for companies with difficulties obtaining that one last component, the so-called Golden Screw. We will try to meet these companies’ needs, especially the ones who have placed non-cancelable non-returnable orders. We would rather be in charge of the scarce supply at this time, rather than our distributors, so we can control who does not receive it. We are working on increasing our inventory, but right now our backlog is growing faster than we can reduce it. It is not our fault that supply has outstripped demand, because the foundries have underinvested in mature node capacity, although sometimes it is our fault, but we are trying to increase internal capacity in these nodes, except that the semiconductor equipment companies are unable to meet our needs in a timely fashion, so it is not our fault. We will be passing along price increases to our customers, to maintain our gross margins, or perhaps to capture “market prices”, but not to take advantage of the situation. Cancellations are nonexistent, or low, or they are a good thing, because it helps us focus on the real backlog, which is very strong, and conservatively speaking, any weaknesses are less than expected.

All our bets are on the automotive market. We’re pumping out as many chips as we can.

Do we have any other questions?

Too cynical? I suppose so. I feel like I’ve learned the most about the state of the industry from NXP and TI and Microchip, but all nine of these companies are pretty reputable, and if they’re saying cancellations are low or the automotive market remains strong, I would tend to believe them.

Take Me Riding in the Car, Car: The Increasing Prevalence of Automotive Electronics and How We Got Here

All this stuff about automotive electronics needs a little context. As I said in Part One, although the automakers canceled orders amid the coronavirus panic of early 2020, and then got caught at the end of the chip manufacturers’ backlogs when demand came rushing back a few months later — this was not the underlying cause of the automotive chip shortage. (The catalyst, not the cause, the catalyst, not the cause....) Various factors have contributed to the present situation over the past decade.

Probably the most significant long-term factor has been the gradual increase in demand in automotive electronics. Although there have been some recent trends like automated driver-assist systems (ADAS) and a trend towards increased sales of electric and hybrid electric vehicles, the electrification of modern automobiles has been a long time coming. Here is a timeline of many electronic automotive systems:[19][20]

Electronics has been gradually replacing mechanical or hydraulic systems for reasons of cost, performance, or safety — or just making things possible that were not before. This comes at a cost; Deloitte estimated that electronic content in new cars increased from about 18% of the total price in 2000 to about 40% in 2020, and still rising. Just the semiconductor content in cars is around \$500 and expected to reach \$600 by 2030.[21]

This trend should not be a surprise. It’s been in the news for many years. I’ve been looking through the research bulletins of IC Insights, and there are numerous bulletins mentioning the rapid growth rate of automotive electronics since 2011:

It’s exciting. And yet automotive electronics is still only a minor contributor with less than 10% of total semiconductor revenue; the two dominant markets are still personal computers and mobile phones. From December 2017’s Automotive and IoT Will Drive IC Growth Through 2021:

From 2012’s Automotive Market Remains a Bright Spot for ICs in 2012:

Yep, exciting times to be part of an industry with the kind of long-term, continued growth potential shown on that graph.

OK, so if automotive ICs have been one of the all-star semiconductor market segments, why did the industry miss in trying to prepare for this growth?

Many blamed the automotive manufacturers; Kurt Sievers mentioned the main argument behind this thought at the 2022 Citi Global Tech Conference:

But there is another action which is equally important, which is a much, much closer collaboration with our customers to understand the future demand, because part of the issue is a lack of understanding and a lack of transparency on what the demand actually looks like — not the next quarter, but the next 12 to 16 quarters.

The car industry — and I say that very bluntly now — had not had an understanding what the manufacturing cycle-times in semiconductors are. So, they thought from a just-in-time mentality perspective, I tell you, I need 5K more products week after next, so you just produce 5K more.

Now, we have to tell them you should have told us four months ago, and then we could maybe have a chance to do this. So, that educational process, especially in direct communication between the OEMs and us, so the car producers and us, has massively increased. I see this as a big, big benefit for us from being, first of all, a better supplier in understanding future demand, but also opening for us a window of opportunity for design wins, because it just created a whole new world of relationships between NXP and the car companies directly.

So, a lot of this is actually going in a triangle relationship with the Tier 1 suppliers, which are our direct customers. But I think the new part of this is the tip of the triangle which is the relationship to the OEMs.

But I think that blaming the automotive manufacturers is kind of a scapegoat excuse. The catch is that forecasting demand is hard. The little “F” on those x-axis years from 2012 - 2015 stands for “forecast”, and it looks rosy, but how do we really know the future’s going to match a prediction?

Forecasting Follies

Hint: we don’t, and it often doesn’t.

Let’s try our hand at some forecasting. I’ve digitized one of the graphs from the brochure for the IC Insights’ 2022 McClean Report, showing Quarterly IC Volume Shipments, across the whole semiconductor industry, in billions of units.

Here’s that data through mid-2018:

This is a really promising future; it just keeps going up, aside from the little wiggles. If we’re smart, we notice the wiggle happens each year, and certain quarters are almost always lower than others, because of seasonal phenomena like the start of the Christmas shopping season (Black Friday) or the Lunar New Year celebration influencing consumption habits. We can correct for the seasonal changes with some least-squares analysis, and fit a quadratic equation on a log scale:

Wow, what a great fit! Look at that, accelerating growth! This is one of those situations that looks too good to be true, and it was; in 2018, the U.S.-China trade tariff war put a damper on the semiconductor industry, and unit sales decreased, recovering somewhat in late 2019. Now a curve fit looks like it’s slowing down, but it’s still a pretty good fit. Perhaps mid-2017 through mid-2018 was just an optimistic anomaly.

Then 2020 hit, and things got even more pessimistic.

About the time that jump in semiconductor shipments happened in the third quarter of 2020, it would be really hard to believe that demand would just zoom forward. But it did, and here we are today, riding what is probably the top of a cycle, on the more-or-less steady exponential growth (a line on a log scale) throughout this whole time.

So forecasting is difficult. There’s a lot of noise and uncertainty in the data. Semiconductor companies have their own data on customer orders at a much finer-grained level than the large-scale (once per quarter) data shown here, but it just goes to show that temporary upticks or downticks can fool you into thinking there’s some long term trend there that isn’t — or maybe it looks like a trend is short-term and has gone away, but it’s really still there all the while.

Yes, Virginia, There Really Is an Automotive Supply Crunch

How do we really know there’s an overload in unfulfilled demand from automotive customers? Maybe this is one big scam, and the auto manufacturers are trying to use the chip shortage as an excuse so they don’t have to explain away poor consumer demand or some other production problem. Well… it’s hard to be 100% certain, but there’s reasonably clear evidence.

Financial Figures from FRED

The US Federal Reserve maintains a database of economic data called FRED (Federal Reserve Economic Data) that includes quite a few statistics on motor vehicle sales and production. The FRED website has graphs but you can download the raw data series if you want to crunch data further. Here’s one example, “Total Vehicle Sales”, over the last 25 years, showing units per month sold in the USA:

The recessions in the United States are highlighted in gray — 2001, 2008-2009, and a very brief one in spring 2020 — and the automotive industry responded very differently to each of them. Hardly any effect in 2001 (tech bubble), a very prolonged impact for several years after 2008 (subprime mortgage crisis), and a very brief direct impact in 2020 (COVID pandemic).

The data above is unadjusted; FRED also includes some seasonally-adjusted data series, where through the magic of statistics, the underlying trend is estimated to exclude seasonal variation. Here’s the seasonally adjusted version of the “Total Vehicle Sales” data, showing sales at an annual run rate:

Some of the 1-month spikes might be quirks of statistical analysis, but anything longer than a few months should be considered as real.

The more interesting graph relative to the chip shortage is the “Auto Inventory/Sales Ratio” dataset (seasonally adjusted):

This represents the ratio of inventory (new motor vehicles ready to be sold; there’s a graph of that too) to monthly sales of motor vehicles, or in other words, represents the number of months of inventory stock, on average — it excludes light truck sales,[22] but I’m not sure whether it counts only new vehicles on dealer lots, or whether autos at the manufacturing plant count too. The average ratio, since the Federal Reserve started keeping track of this statistic in 1993, was around 2.5, and most of the time it stayed between 2.1 and 2.9. A ratio of 2.5 means that there were 2.5 times as many motor vehicles in inventory than there were sold that month, meaning that on average, the car companies have maintained about 2.5 months of inventory and have sold 40% of their ready-made domestic stock in the US each month. Just as an example, here’s some detailed sales and inventory data from August 2015: about 481 thousand motor vehicles assembled in the USA were sold (again, this excludes light truck sales), inventory was 1.245 million units, and the ratio was 2.589.

The low variation most of the time of the inventory-to-sales ratio means that, in aggregate, car companies have done a good job with inventory control and have kept about the right amount of vehicles in stock for their business. The ratio went up to nearly 4.5 months of inventory in 2009, because sales went down sharply at the end of the 2008-2009 financial crisis — perhaps you remember that both Chrysler and General Motors filed for bankruptcy in mid-2009 — but by early 2010 the automotive companies had adjusted their factory production so that the ratio was back to normal. It spiked again briefly in early 2020 (low sales), but by late 2020 stocks of motor vehicles dropped to 2.0 months of inventory, and fell again, precipitously, in early 2021, dropping below one month of inventory in June 2021, and reaching its all-time low of 0.378 months of inventory in February 2022, with inventory consisting of only 64000 units. Since then inventory has recovered somewhat, but is still very low, with inventory-to-sales around 0.5 — basically, cars have been flying off the lots, with only two weeks of inventory.

Other sources of inventory information

An October 2021 New York Times article also mentioned this problem of inventory:[23]

At the end of September, G.M. had 128,757 vehicles in dealer inventories, down from 211,974 at the end of June and more than 334,000 at the end of the first quarter. In years past, the figure was often about 800,000.

Toyota had 37,516 vehicles on dealer lots at the end of the quarter, and 61,208 at ports serving the U.S. market. At the current sales rate, that is enough to last about 18 days.

Cox Automotive tracks US automotive inventory statistics in more detail; here are a couple of glimpses of what’s happened recently. From the company’s September 2021 New Vehicle Inventory Report,[24] it looks like the overall situation dropped to just about four weeks of inventory in autumn 2021, but it’s still much lower than in prior years:

Inventory levels among different car brands varied quite a bit in September 2021, with some Japanese brands having the tightest supply:[25]

Since then, inventory has climbed somewhat, reaching 53 days of supply on average in November 2022:[26]

This data doesn’t conflict with the FRED data, which excludes light trucks (including minivans and SUVs) and shows a smaller inventory; it turns out that small cars are in a much lower inventory situation than light trucks.[27]

There’s a reason for this discrepancy: automakers have chosen to prioritize the limited chip supply they receive towards vehicles with larger profit margin, which happens to be luxury cars, pickup trucks, and SUVs.

To make a long story short: vehicle inventory is better than the beginning of 2022, but it’s still very tight.

If there is a grand conspiracy out there, it’s fooled everybody.

Blame Game

Not even Toyota can overcome force majeure

I said that I think blaming the automotive manufacturers is kind of a scapegoat excuse. Yes, the automotive supply chain has not been well-positioned for long-term resiliency. The manufacturers could (and should) do better. But this is a very complex chip shortage, and it has had different dynamics in the short-term and in the long-term.

Toyota gets credit for having withstood the early stage of the chip shortage, with the carmaker stating in February 2021 that it had stockpiles of chips, up to four months’ worth in some cases:[28]

“For the near term, we do not see any decrease in production volume due to the chip shortage, but we do see risks of a chip shortage,” Chief Financial Officer Kenta Kon said during a briefing.

Kon said Toyota had heard chip shortages globally might continue until the summer, though the situation might resolve itself earlier.

Asked about why the automaker is seeing limited impact compared with competitors, Kon said Toyota has been constantly providing its short-term and long-term production volume plans to suppliers.

Toyota had learned from the 2011 Fukushima earthquake, as a Reuters article mentioned in March 2021:[29]

Toyota may have pioneered the just-in-time manufacturing strategy but when it comes to chips, its decision to stockpile what have become key components in cars goes back a decade to the Fukushima disaster.

After the catastrophe severed Toyota’s supply chains on March 11, 2011, the world’s biggest automaker realised the lead-time for semiconductors was way too long to cope with devastating shocks such as natural disasters.

That’s why Toyota came up with a business continuity plan (BCP) that required suppliers to stockpile anywhere from two to six months’ worth of chips for the Japanese carmaker, depending on the time it takes from order to delivery, four sources said.

And that’s why Toyota has so far been largely unscathed by a global shortage of semiconductors following a surge in demand for electrical goods under coronavirus lockdowns that has forced many rival automakers to suspend production, the sources said.

“Toyota was, as far as we can tell, the only automaker properly equipped to deal with chip shortages,” said a person familiar with Harman International, which specialises in car audio systems, displays and driver assistance technology.

Two of the sources who spoke to Reuters are Toyota engineers and the others are at companies involved in the chip business.

Toyota surprised rivals and investors last month when it said its output would not be disrupted significantly by chip shortages even as Volkswagen, General Motors, Ford, Honda and Stellantis, among others, have been forced to slow or suspend some production.

But not even Toyota, with its stockpiling strategy, was left unscathed by this super-shortage. By August 2021 Toyota had cut production targets,[30] and the company has disclosed intermittent production stoppages in the various months since then.[31]

In the end, there has been nothing that any of the automakers could do, in the short term, to get around the chip shortage. What “short term” means here is a little tricky to define… but for now, let’s think of it as actions taken in 2020 and 2021 that have affected the industry through 2021 and 2022. (Time delays are a big part of what’s going on, and I’ll talk more about what causes them in Parts Five and Six.)

Automotive manufacturers make the best of a bad situation

Right. So Toyota had backup stock and still suffered somewhat, other automotive manufacturers suffered more. Fortune Magazine illustrated this with a bar graph showing estimated production capacity utilization in the first half of 2021:[32]

The prevailing narrative is that automotive manufacturers canceled orders when the pandemic struck in early 2020.[1][23][33][34][35][36] (Apparently the only ones that didn’t were Toyota, Hyundai, and Kia.[37]) And then, by the time it was clear that demand quickly rebounded, and they needed more components, automakers had to start new orders at the back of a very long queue, behind 2020’s demand for stay-at-home consumer electronics. I’m going to repeat part of a quote from NXP CEO Kurt Sievers at the 2022 Citi Global Tech Conference:

The car industry — and I say that very bluntly now — had not had an understanding what the manufacturing cycle-times in semiconductors are. So, they thought from a just-in-time mentality perspective, I tell you, I need 5K more products week after next, so you just produce 5K more.

Now, we have to tell them you should have told us four months ago, and then we could maybe have a chance to do this.

There’s certainly some blame for this short-term “we-don’t-need-it” / “we-need-it-now” fickleness among carmakers, at first glance… but put yourself in the shoes of some automotive executive who gets to make production decisions back in March 2020. (I’m going to steal shamelessly from my own post on SemiWiki in January 2022.[38])

  • You’ve had two bad years already, 2018-2019.
  • It’s early March 2020 and this COVID pandemic is spreading like wildfire, orders from customers are dropping like rocks as lockdowns/fear takes its toll and people aren’t interested in a new car.
  • Your gross margins aren’t great anyway because it’s a cutthroat manufacturing business; Ford’s gross margin was hovering around 15% before 2020 and decreasing.
  • The car rental companies are a few of your big customers, and they’re not ordering anything. They’re making negative capital expenditures (selling chunks of their fleets) because they’re desperate to stay afloat. Some didn’t: Hertz declared bankruptcy in May 2020.
  • You have no idea how long this is going to last.

What do you do?

And what do your suppliers do? Auto companies place orders with Tier 1s who place orders with automotive IC manufacturers, some of whom place orders from foundries like TSMC. Bullwhip effect magnifies the uncertainty through the supply chain. (We’ll cover this more in Part Six.) Do you call up the IC manufacturers and tell them “See, I have this problem, I don’t think I can afford to place as many orders this quarter with my suppliers, and that means that you probably won’t be getting orders from the Tier 1s… but I might need those chips after all, so could you please just kind of set them aside for our future needs? I can’t pay you anything right now but it would be a big favor.”

It’s a big mess, and I don’t doubt that the cost-benefit analyses of choosing “just-in-time” to reduce inventory was a poor choice and did not allow for rare risks. So I guess they could have kept more inventory (like Toyota apparently did) earlier back in 2018-2019, then when COVID hit, reduce new orders and rely on inventory.... or would the proper order be to anticipate the future, that COVID was a brief disruption and keep on cranking out cars even though their customer demand dropped? What happens at the executive staff meetings when the CEO asks them why they’re doing this, they’re going to get hammered at the board meetings and the stock market for building up a bunch of inventory on their balance sheets?

I’m inclined to believe that for the most part they made decisions that they believed were optimal given the information they had at the time. Maybe they made a few minor errors here and there. I don’t know the right answer and I’m sure I’ve made multiple errors in what I’ve just written, but I’m not about to second-guess that the auto industry as a whole all came to the same suboptimal conclusions.

Long-term supply-chain management breaks down

The real reason behind the automotive chip shortage — at least this is my take on it — is a breakdown in long-term supply chain management. This is bigger than just one company’s doing. By the time March 2020 hit, the die had already been cast. Conditions were ripe for this shortage several years ago, with automotive demand increasing and production of mature nodes getting close to capacity. Here’s our “Automotive Fab Four” supply chain from Part Three:

The thing to note is that no matter what the customer end of the chain does — stockpiling or not — the only way for that end to receive product is if all the upstream links are producing enough. The demand signals have to get back to the fab owners — whether they’re foundries or IDMs — who have to act on it in time to make a difference. If automotive demand for semiconductors is growing, the only ways to avoid a mismatch with supply are to convince the fab owners to build more capacity, or to steer a significant fraction of automotive IC designs toward technology nodes with more capacity. Neither is a quick or easy proposition.

Here’s a question to ask yourself: Suppose you’re a downstream consumer, like Frob Motor Company — how much of an order do you need to place in advance to keep parts coming, and what do you have to do to ensure each supplier in the chain does the same? If Toyota made sure that it always had six months of stock on order, and its Tier 1 suppliers always had six months of chips on order from chipmakers, would that have been enough to avoid the effects of the chip shortage? What if it made sure all of the chipmakers placed six months of those chips on order from the foundries? What about twelve months? Twelve months non-cancelable non-returnable, with prepayment to guarantee capacity, and a cherry on top? Or would there still have been enough of a squeeze on mature node foundry capacity that no one would have been able to guarantee the supply?

This is a nagging curiosity of mine — I don’t know if we’ll ever find out the answer, however.

A look forward

But I’m guessing that, as a reader, you’re more interested in the future — when will the chip shortage end? — rather than the cause of blame in the past.

Demand

One bit of good news is that this chip shortage has been a real wake-up call to the automotive industry, and that people are talking with each other about how to do better. Kurt Sievers of NXP recognized this as a chance to build relationships with the automotive manufacturers directly, not just with the Tier 1 automotive suppliers. In addition to the creative partnerships I mentioned in Part Three, which give the impression the automakers are going to get more involved in the IC design process somehow, the mere increase in communication is a good thing. Just having a better picture of future demand is helpful. Wards Intelligence recently interviewed Kristin Dziczek, policy advisor at the Federal Reserve Bank of Chicago; Dziczek commented on the issue this way:[39]

[Dziczek:] Will it make sense for chipmakers to put more money into mature technologies for automakers when the auto industry is so fickle? As soon as times got tough in 2020, they pulled back. The chip industry wants to see that there is a commitment to this for a long period of time before they put the investment into delivering what needs to be delivered. (Automakers face) really long lead times right now and requirements for long-term commitments.

Wards Intelligence: Chip suppliers say that automakers’ transparency about their sales forecast is important to manage the crisis in the short term. Are automakers being more transparent with their suppliers?

Dziczek: Well, that is what they have been aiming to do, but I heard that now that chip supply was more regularly delivered — meaning if you order X, you get X and not 0.8X — some automotive consumers were starting to trim back their orders, and some chipmakers seeing this order of X instead of X plus (are wondering), “Is demand starting to fall off, or is (the market) just normalizing (because automakers) are not in panic mode anymore?” A weakening demand signal could flow back to chipmakers pretty quickly, because they are nervous about automotive pulling back again.

What else we seeing in the news?

I found this recent article on the Jalopnik website, Toyota Sees a Light at the End of the Tunnel,[40] which at first glance sounds very optimistic. Toyota sees an end to the chip shortage? Really? But then if you look closer, the author just quoted an Automotive News story somewhat out of context, and completely missed the point, which is that Toyota executive vice president of sales Jack Hollis sees the chip shortage continuing:[41]

Hollis acknowledged that Toyota hasn’t had much success guessing when restrictions on semiconductors will lift, and said the industry shows no sign of pulling back on its demand for technology.

“It feels like the microchip, the whole semiconductor issue, it feels like it’s going to take forever. Just to be candid, it feels like this problem is here to stay,” he said, explaining that as automakers expand their lineups to include more electronic-intensive electric vehicles and grow their sales, demand for microchips will increase.

“Can we ever get to a point where we’re really producing microchips at a speed that’s that much faster than the industry is growing?” Hollis said rhetorically. “So the problem stays with us for a lot longer than we might have expected originally.”

Both Dziczek and Hollis alluded to automotive semiconductor demand, in different ways. There are headwinds and tailwinds affecting the potential future demand:

  • increasing semiconductor content in cars is a long-term tailwind (more likely to keep demand high): IC Insights has consistently observed automotive semiconductor revenues increasing faster than the rest of the semiconductor market since 2011; in another article, Dziczek cited a Gartner forecast[42] of a compound annual growth rate (CAGR) of 11.7% between 2020 and 2030.

  • uncertain macroeconomic environment (higher interest rates, possible recession) — short-term headwind

  • slowdown in automotive sales since June 2021 despite high demand / low inventory / the continuing chip shortage — probably a tailwind; when there’s a backlog of car supply for so long, some consumers are giving up and putting off purchases for later, or trying to buy used cars instead, but the prices of used vehicles are still high,[43] and meanwhile all the cars out there on the road are getting older.[44] This seems like it should lead to a hidden backlog of people who will need to upgrade to a newer car sooner rather than later, keeping demand high until the supply has caught up. I don’t have a source for this at a macro level, but I have a couple of anecdotes of friends in this situation.

We don’t know what’s going to happen next, but I’m leaning towards the demand staying high for a while, and as the macroeconomic environment gets eventually better, demand’s likely to keep on rising. That is: auto sales likely to return to their historical average, but more chips per car = higher semiconductor demand.

Supply

How will supply keep up? All these new capital expenditure projects at automotive/industrial semiconductor manufacturers will help somewhat, when more capacity comes online, but how much? A large fraction of automotive MCUs appear to be manufactured at TSMC; several articles in early 2021 cited an IHS Markit estimate that 70% of automotive MCUs are made at TSMC fabs. Here’s one of those articles, from Supply Chain Dive:[45]

TSMC manufactures about 70% of all automotive MCUs shipped, according to IHS Markit. This makes it hard for companies to find backup suppliers, because the largest MCU suppliers — Renesas, NXP, Infineon and others — have outsourced much of their manufacturing to TSMC or UMC, IHS Markit noted.

But again: TSMC has not publicized any plans to increase capacity at nodes above 28nm.

The Supply Chain Dive article also included a bar graph showing IHS Markit data on automotive MCU market share:

The top three manufacturers, Renesas, NXP, and Infineon (which acquired Cypress), according to the IHS Markit data in this graph, held a combined 79% market share of automotive MCUs in February 2021. Are they taking major steps to increase internal capacity? Yes… but Renesas’s main CapEx project, the Kofu factory reopening, is for 300mm power devices, and according to a September 2022 presentation, its long-term in-house capacity for MCU products is intended to decrease, shifting more toward analog ICs.[46]

In NXP’s January 2022 earnings call, Kurt Sievers noted a “modest expansion of our internal front-end capabilities” (fabs), with more CapEx aimed at back-end capacity (assembly & test).[47] He also stated in this call:

There is simply anything which is below 90 nanometers, we will not do in-house. And we see no reason to change this.

Infineon’s new capacity at Villach is for power semiconductors, and according to a November 2022 press release, the planned fab in Dresden is for “300-millimeter analog/mixed-signal and power semiconductors”.[48]

So it really depends on the market segment. For analog, it sounds like there’s some capacity improvement coming online in the next few years — including Infineon’s new Dresden fab and TI’s new fabs; TI is the leading analog IC supplier and with all this expansion, it seems intent on staying in that spot. For automotive MCUs, it looks like our fate is primarily in the hands of TSMC and other foundries. Maybe if we all show up at TSMC’s next ceremony and stare really hard at Mark Liu and C.C. Wei, they’ll decide to build more capacity in the 40nm - 90nm range. What do you think?

Let Them Eat Cake

But wait, there’s more scary thoughts. Sean Tucker has written a number of articles on the chip shortage on Kelley Blue Book’s website, with statements indicating that automakers are turning lemons into lemonade, and changing their business model to adapt to this awful situation. Here’s an excerpt from one of them:[49]

For decades, auto industry practice meant keeping a stockpile of new cars available for sale. Dealers routinely kept so much inventory on hand that they discounted most cars to sell them.

Even once chip production recovers, that practice may not return.

“We’ll never go back to the level of inventories that we held pre-pandemic because we’ve learned we can be much more efficient,” GM CEO Mary Barra told reporters last year.

BMW Chief Financial Officer Nicolas Peter told the Financial Times last fall that the automaker plans to “clearly stick with … the way we manage supply to keep our pricing power at the current level.”

Mercedes-Benz parent Daimler AG has the same idea. “We will consciously undersupply demand level,” Daimler’s CFO Harald Wilhelm told FT.

Ford CEO Jim Farley has suggested that the company may move closer to a build-to-order business model, though he recently promised dealerships Ford would not sell cars directly to customers, Tesla-style.

Dealer groups, too, say big inventories and hefty discounts may not come back.

Automakers are financially interested in making as many cars as Americans will buy. But the chip shortage may be teaching them not to make more than that.

This is probably worse for consumers, especially consumers seeking entry-level cars — fewer discounts, and a conscious focus on higher-profit vehicles built to order — but it is one way of staying in business amid a turbulent situation. Yep, bump up the profit margin. We’ll all be paying for it.

Notes

[1] Willy Shih, Why Are Some Companies Still Short Semiconductor Chips When Others Are Swimming In Them?, Forbes, Nov 3 2022.

[2] The Economist, When will the semiconductor cycle peak?, Jan 29 2022.

[3] Mark Gurman and Ian King, Chipmaker GlobalFoundries to Start Job Cuts and Freeze Hiring, Bloomberg Technology, Nov 11 2022.

[4] UPI, Survey: semiconductor industry grew 32.9 percent in 1988, Jan 3 1989.

[5] Christiaan Hetzner, Chipmakers are demanding steep price hikes—and that means more inflation on shelves and the car lot, Fortune, Nob 10 2021.

[6] Doug O’Laughlin, The Rising Tide of Semiconductor Cost, FabricatedKnowledge.com, Nov 22 2021.

[7] Silicon Labs, Q3 2021 Earnings Call, Oct 27 2021.

[8] Microchip Technology, Q4 2022 Earnings Call, May 9 2021. (Microchip’s fiscal year ends March 31, so most of the fiscal 2022 earnings were actually in 2021. Also, I’ve corrected some of the typos in Motley Fool’s transcription)

[9] Microchip Technology, Q2 2023 Earnings Call, Nov 3 2022. (Again, I have made minor corrections to Seeking Alpha’s transcript.)

[10] Willy Shih, Congress Is Giving Billions To The U.S. Semiconductor Industry. Will It Ease Chip Shortages?, Forbes, Aug 3 2022.

[11] Willy Shih, Why Are Automotive Chips Still In Short Supply?, Forbes, Nov 20 2022.

[12] TSMC 2020 Q4 earnings call, Jan 14 2021.

[13] Aakash Arora, Albert Waas, Ramiro Palma, Jimmy Feng, Karl Breidenbach, Harrison Xue, Eric Jesse, Thomas Lopez, and Minhal Dhanjy, Tracking the Next Phase of the Automotive Semiconductor Shortage, Boston Consulting Group, Oct 7 2022.

[14] Scotten Jones, Technology and Cost Trends at Advanced Nodes, IC Knowledge, presented at SEMICON West, Jul 13 2016.

[15] Brett Brune, With RFAB2, TI Bumping Up Analog Chip Production, EE Times, Oct 25 2022.

[16] SEMI was founded in 1970 as the Semiconductor Equipment Manufacturing Institute, and originally focused on the materials and equipment manufacturers (suppliers to the semiconductor industry), but has expanded over the years to include semiconductor manufacturers (eight out of the nine automotive & industrial companies mentioned here, excluding only Diodes Inc), EDA companies, and many others.

[17] Ajit Manocha and Sanjay Malhotra, Chipping in for Equipment Suppliers: The Equipment Multiplier Effect on the Chip Shortage, SEMI, May 2 2022.

[18] Asa Fitch, Global Chip Shortage’s Latest Worry: Too Few Chips for Chip-Making, The Wall Street Journal, May 3 2022.

[19] Charles Murray, The 10 Biggest Milestones in Automotive Electronics History, Design News, Nov 19 2018.

[20] Kevin A. Wilson, Automotive Technology Firsts, Car & Driver, Apr 28 2016.

[21] Deloitte, Semiconductors — the Next Wave, Apr 2019.

[22] Light trucks include SUVs and vans, and overall have had an increasing share of sales in the USA for the past 40 years. For some reason the inventory statistics on FRED only include “regular” motor vehicles (cars) and there aren’t inventory figures for the light truck category.

[23] Neal E. Boudette, Chip Shortage Makes Big Dent in Automakers’ U.S. Sales, New York Times, Oct 1 2021.

[24] Cox Automotive, New-vehicle Inventory as of September 20, 2021, Oct 14 2021.

[25] Cox Automotive, New-Vehicle Inventory Falls Below 1 Million in September, Prices Hit All-Time High, Oct 14 2021.

[26] Michelle Krebs, New-Vehicle Inventory Climbs in November; Prices Stay High, Cox Automotive, Dec 14 2022.

[27] Michelle Krebs, New-Vehicle Inventory Rebounds in October for Most Brands, Cox Automotive, Nov 16 2022.

[28] Eimi Yamamitsu, Toyota hikes profit forecast 54%, shrugs off global chip supply issues, Reuters, Feb 10 2021.

[29] Norihiko Shirouzu, How Toyota thrives when the chips are down, Reuters, Mar 8 2021. This article is short but insightful, and touches on several aspects of Toyota’s strategies, including some historical context. For example:

Years before, it poached engineering talent from the chip industry and opened a semiconductor plant in 1989 to help design and manufacture MCUs used to control Prius powertrain systems.

Toyota designed and manufactured its own MCUs and other chips for three decades until it transferred its chip-making plant to Denso in 2019 to consolidate the supplier’s operations.

The four sources said Toyota’s early drive to develop a deep understanding of semiconductor design and manufacturing processes was a major reason why it has managed to avoid being hit by the shortages, in addition to its continuity contracts.

[30] Maki Shiraki and Ritsuko Ando, Toyota slashes September output amid chip crunch, COVID resurgence, Reuters, Aug 19 2021.

[31] There have been many articles since mid-2021 about cuts in production or shutdowns at Toyota factories; below are a few of them. (Never mind the numerous articles about shutdowns at other auto makers, which have fared worse, in general.)

[32] Eamon Barrett, How Toyota kept making cars when the chips were down, Fortune, Aug 2 2021.

[33] Bill Jewell, Automakers to blame for semiconductor shortage, Semiconductor Intelligence, Apr 22 2021.

[34] Eamon Barrett, This is how the global chip shortage will end, Fortune, May 9 2021.

[35] GlobalData Thematic Research, Semiconductor supply shortage is just the start for squeezed automakers, Verdict, Feb 12 2021.

[36] Jeanne Whalen, Reed Albergotti and David J. Lynch, Biden can’t fix the chip shortage anytime soon. Here’s why., Washington Post, Mar 1 2021.

[37] Joyce Lee, ANALYSIS – Hyundai bought chips when rivals didn’t; its assembly lines are still rolling, Reuters, Feb 25 2021.

[38] Jason Sachs, comment in thread, The Real Reason Behind the Automotive Industry IC Shortage — A Step-Function Surge in Demand!, semiwiki.com, Jan 26 2022.

[39] Maite Bezerra, Q&A: Chicago Fed Automotive Policy Advisor Talks Chip Crisis, Wards Intelligence, Dec 1 2022. Good information with low noise content here.

[40] Steve DaSilva, Toyota Sees a Light at the End of the Tunnel, Jalopnik, Dec 2 2022.

[41] Larry P. Vellequette, Toyota sees U.S. sales recovering to 15M in 2023, Hollis says, Automotive News, Dec 1 2022.

[42] Kristin Dziczek, Why the Automotive Chip Crisis Isn’t Over (Yet), Chicago Fed Letter, Oct 2022.

[43] Manheim Consulting, Used Vehicle Value Index, checked on Dec 26 2021.

[44] S&P Global Mobility, Average Age of Vehicles in the US Increases to 12.2 years, according to S&P Global Mobility, May 3 2022.

[45] Matt Leonard, Why the automotive supply chain is in a semiconductor jam, Supply Chain Dive, Feb 23, 2021.

[46] Shuhei Shinkai, Finance presentation, Renesas Electronics, Sep 28 2022.

[47] NXP, Q4 2021 Earnings Call, Jan 31 2022.

[48] Infineon, After a record 2022 fiscal year, Infineon significantly increases its long-term financial targets, and is planning a major investment in a new factory in Dresden; positive outlook for 2023, press release, Nov 14 2022.

[49] Sean Tucker, Microchip Shortage Still Limiting Car Production, Sep 16 2022.

Wrapup

My take on the economic disturbances of 2019 - 2021 is that the wiggles of the last few years have made it hard to see long-term trends, and we’re still increasing demand for mature nodes behind it all. I’d love to be wrong about that, but here’s the state of things as I see it:

  • Automotive electronics has had a long-term growth trajectory slightly above the overall semiconductor industry for the past dozen years, with clearly-identifiable causes (continued conversion to microcontrollers-in-all-the-things; increased prevalence of electric or hybrid electric vehicles; ADAS)
  • Much of the semiconductor content in automotive is in this “mature node” area
  • All the major fab-lite automotive semiconductor manufacturers are saying they still have strong demand for automotive ICs
  • Other markets for mature-node ICs are also seeing long-term growth (industrial, IoT) although may be weakening in demand
  • Various amounts of incremental capacity are being added at these manufacturers’ fabs (lots at TI; some new fabs built or being built by ON/Infineon/Renesas/ST; statements of new capacity at the others) but a large portion of it seems to be for power electronics or analog, with microcontrollers presumably coming from foundries
  • Aside from some planned 28nm capacity at TSMC’s Kumamoto fab, not much mature-node capacity being added at foundries (see Part Three)
  • Long-term prognosis for mature-node supply availability: anyone’s guess at this point.
  • These are market forces that are not a huge portion of the semiconductor industry’s volume, and not at the leading edge, so the industry doesn’t have a roadmap or a business-as-usual workflow to deal with them. And it’s why I think it will take a couple of cycles before the industry gets the hang of it

We looked at a number of different topics along the way in this article:

  • A quick look at events in 2022 — the chip shortage has bifurcated, with it easing at the leading edge or for consumer electronics, but continuing for the automotive/industrial markets
  • Analyst predictions — gluts are coming
  • My prediction of no good long-term solution to the automotive shortage until around 2029 (just pretend you’re seeing another Doris Day / Jo McKenna scream picture here; I can’t squeeze it in without disrupting the bullet points)
  • A handful of other sources that have raised issues of concern about mature-node supply-demand imbalance over the last year or so
  • A quick look at 2022’s earnings calls of automotive/industrial semiconductor manufacturers
  • Some long-term automotive trends: increases in use of automotive electronics
  • Why the uncertainty/transients in demand may have obscured early warnings about the automotive chip shortage
  • A closer look at automotive inventory trends to double-check whether there really is a production shortage
  • Whose fault was the automotive chip shortage, and does it really matter?
  • A look forward at automotive demand, supply, and possible changes in the automotive maker business model

In the next two parts, we’ll be looking at some of the dynamics behind the chip shortage:

  • Part Five will focus on cycle time, queues, and inventory — in other words, how long does it take a semiconductor fab to manufacture chips, and what are some of the factors that contribute to this time delay
  • Part Six will focus on lead time and the bullwhip effect — how long is the delay between ordering ICs and when you actually get them, why has it gone to crazy levels lately, and why has it stayed so long

Thanks for reading and let’s hope for the best in 2023!

Addenda

Further Reading

Acknowledgements

This article would not be possible without the assistance and encouragement of Brant Ivey and Malcolm Penn.

Changes Among the Semiconductor Market Analysts

The rarefied field of semiconductor market analysts has recently undergone a few changes. TechInsights Inc. recently announced the acquisitions of The McClean Report (after which Bill McClean announced IC Insights would be closing) and IC Knowledge LLC.

I have learned a lot from Bill McClean’s posts on the IC Insights website on changes in the semiconductor markets; I don’t think that I would have been able to write parts of this article without being so confident in the continued growth of automotive electronics.

Scotten Jones of IC Knowledge LLC has likewise provided numerous glimpses of the costs of semiconductor manufacturing, something I was able to utilize in my coverage in Part Three of why MCUs don’t use leading-edge technology nodes. I hope he continues his work at TechInsights and keeps sharing some of his takeaways from the industry.

Both of these sources have been very illuminating to me. Professional analysts can afford to buy high-grade market research and analysis reports, but the rest of us are limited to whatever crumbs of knowledge they decide to throw out for free. I hope that TechInsights decides to make at least some bits of information freely available to fill the void left by these two experts.

NXP’s Kurt Sievers at Citi Global Tech Conference, September 7 2022

Here’s a full transcript of the discussion between Citi semiconductor analyst Chris Danely, and NXP’s CEO Kurt Sievers, up until the audience Q&A. (Danely and Sievers took about 31 minutes of the 40 minute session.) This is based on Seeking Alpha’s transcript, but I have made numerous small corrections, so if the SA folks have any objection to my posting this, please let me know and perhaps we can work something out.

The audio recording itself still seems to be available on Veracast, and it’s worth listening to the source directly.


Chris Danely — Hopefully my mic works. Yes, it does. Alrighty, guys, so I was literally stopped by people in the halls and asked this from friends of mine in the auto industry, about shortages. How do we stand on shortages? Are they getting better? Are they getting worse? What’s NXP doing to hold these off?

Kurt Sievers — Yeah, thanks, Chris. So, I’d say the short version is it’s getting a little bit better, but you probably still have to wait a long time for ordering a new car, which is loaded with electronics. So, we have across the industry, a couple of very persistent shortages, which I also estimate to carry forward easily through next year.

In total, it got a little better, which is a mix of increased supply and some softening in demand on the European side. Now, that softening in demand is way above what we can service. So, it doesn’t mean there is any revenue impact, it’s just that the gap between what is serviceable and what they would like to build, I think has shrunk a little bit, especially in Europe, since some of the demand has come down.

But overall, especially in the famous mature nodes, so say in the 16 to 90 nanometer space, there is persistent shortages, which in the end, it’s all not a miracle, are really a function of a structurally strongly increased demand, which has to do with content increases, especially in — with EVs, and ADAS applications.

Chris Danely — Yeah. And then I guess what is NXP doing to at least try to ensure or minimize shortages going forward for previous cycles? I know we’re always going to have shortages, much to the customer’s chagrin, but —?

Kurt Sievers — Yeah, I think two obvious things. One is clearly we are increasing our supply. I mean, you’ve seen us growing now for more than eight quarters in a row, and that is a result of gradually improved supply, both in-house with our own manufacturing facilities, but also in cooperation with our foundry partners, which are providing about 60% of our wafer supply.

But there is another action which is equally important, which is a much, much closer collaboration with our customers to understand the future demand, because part of the issue is a lack of understanding and a lack of transparency on what the demand actually looks like — not the next quarter, but the next 12 to 16 quarters.

The car industry — and I say that very bluntly now — had not had an understanding what the manufacturing cycle-times in semiconductors are. So, they thought from a just-in-time mentality perspective, I tell you, I need 5K more products week after next, so you just produce 5K more.

Now, we have to tell them you should have told us four months ago, and then we could maybe have a chance to do this. So, that educational process, especially in direct communication between the OEMs and us, so the car producers and us, has massively increased. I see this as a big, big benefit for us from being, first of all, a better supplier in understanding future demand, but also opening for us a window of opportunity for design wins, because it just created a whole new world of relationships between NXP and the car companies directly.

So, a lot of this is actually going in a triangle relationship with the Tier 1 suppliers, which are our direct customers. But I think the new part of this is the tip of the triangle which is the relationship to the OEMs.

Chris Danely — And some semi companies have talked about keeping a higher normalized level of inventory going forward. There’s been some talk about supply chain, whether it’s automotive, industrial, or IoT, or whatever, keeping a higher level of inventory going forward. What are your thoughts on that?

Kurt Sievers — Yeah, inventory definitely is one of the easy means which people see to help the situation. To be very clear, we, as a semiconductor company, have absolutely no interest, and no intention, to hold more inventory for this. If anything, then that needs to happen further down the chain with those who need these parts.

What I can quote here is that some of the car companies actually have the idea that at least they want to have inventory which is covering the manufacturing cycle time. So, say something between three and six months of inventory for critical parts to be safe against any fluctuations in the manufacturing.

Now, this is not the case today. This is an intention; it is a strategic plan by some of the OEMs. From anything I can oversee, at least for our product portfolio, none of our customers has any chance to build this as we speak, because there is simply not enough supply. So, that’s why it is an intention, but it’s not anything which happens practically right now.

Chris Danely — Okay. And then, just to touch on the shortages again, because I get this question a lot as well. Are there, I guess, different parts at different times going into shortage? Or has it been, let’s call, it five, 10, 15 parts that have been consistently in shortage for a while? And is it NXP? Is it your dastardly competitors? Is it a mix? Maybe if you could just shed a little insight on what’s going on?

Kurt Sievers — Yeah, it’s not that much about particular parts, it’s about specific technologies, it’s specific nodes. And most of those are indeed in the bracket between — now, I’d say between 28 and 180 nanometers.

And that was ranging all the way from what you need for microcontrollers to analog mixed-signal. So, it’s not just like — because there was some saying at some time, it’s particularly microcontrollers. No, it’s all across.

And since many of these shortages are originating from the foundries, it is not a NXP only or competitor A, competitor B only thing because it comes from the foundry and the same foundry is supporting all of us.

So, it’s much broader than NXP, and it’s now really playing out more clearly, which are the nodes which are missing. And we try as an industry to build the best mix of parts from these available technology nodes. So, say, let there be a shortage, say, in 55 nanometers, then the whole industry needs parts in 55. And then the art is to make sure that we maximize the parts which can be built with that available 55 nanometer capacity, which is very complicated, because there is thousands of different products by several different semiconductor companies going through 20-plus different tier one companies into 10-plus different OEMs. So, it’s a pretty big puzzle to be sure you optimize the existing capacity, but below the line, it’s still not enough.

Chris Danely — And in terms of supply getting a little bit better, is that simply because, let’s say, foundry X has seen a little bit of decline in demand from, say, PC or handset and they’re able to allocate some of that capacity to auto and industrial or— I know you guys have definitely, sort of, broadened your foundry horizons that a combination of both. Could you maybe talk about what’s—

Kurt Sievers — It’s much more the latter, so the positive impact from the more soft demand in PC consumer and low end mobile, we haven’t seen any positive impact on the wafer supply which we need. Zero. It’s really — unfortunately I have to say it that clearly — absolutely zero.

And the reason is that these technology or process technology buckets are not fungible. It is different technologies which are falling free by the lower demand in — say in mobile, versus what we need in automotive.

Now, I think over time foundries have some flexibility to retool and actually create more capacity out of that, but that takes time. So that’s not a — this is not a quick move.

Now, most of the increased supply is just the straight and direct results of efforts, investments from the foundries over this over two year period now. I mean, I had the first escalations of supply shortages pretty much exactly two years ago. It’s now two years ago that I was asked into the first CEO-level meetings with automotive and industrial customers — actually first automotive, then industrial — on shortages, and since then, I mean, the whole industry has worked manically on increasing capacity and that’s what we start to see now as a positive result. But it’s not a cross-move from mobile or consumer.

Chris Danely — Okay. Can you just maybe update everybody on your manufacturing strategy? How much is internal versus external and then where that’s trending in the future?

Kurt Sievers — Yeah. So, our manufacturing strategy for the back end, I stopped there because that’s simple and straightforward, is to do the majority in-house, so we are about 85%, we have our own factories. And we’ve also been able to scale them to the tune which was needed over this period. I mean, obviously, we also needed much more back-end manufacturing capacity, but that’s relatively quick to add. True, lead-times were long, but still that wasn’t really the source of shortage.

For the front end, we have about 60% from third parties and about 40% from our internal facilities, where the strategy is very straightforward. What we do internally is, is all in eight inch and at or greater than 90 nanometers. And it is very much geared towards proprietary special technologies. Think about gallium nitride, silicon-germanium, high-voltage mixed-signal technologies, which are proprietary and owned by NXP.

That’s what we what we use our own factories for, while all the more standard CMOS 90 nanometers all the way down to volume 16 nanometers, which we ship now in 14, is what we are getting from foundry partners. And again, that makes today about 60%. We think over time, it’s probably going to grow to 70% or even 75%.

Chris Danely — And does the current shortages or what we’ve seen over the last couple of years, has that changed or altered that strategy, it sounds like — do you ever think maybe we should do a bit more internally, or I guess you can’t buy fabs anymore?

Kurt Sievers — It has reinforced that strategy. I mean, it’s very obvious that we have won market share over the last two years now through this period, against all of those people who depend more on in-house supply.

But it’s not because we are magic, it’s simply with the option to work with foundry suppliers, you have just more options to get access to scale and cost. And I believe for CMOS, which gets excessively more expensive, if you go to smaller nodes, that will even be more true, the longer you go forward.

What we are, however, intensifying, and I think that’s a learning out of this period, is that for those technologies, which are proprietary, we have to make sure that we also have good planning on a go-forward basis in-house, which means we also have to expand our in-house manufacturing capacity. Which for a long time, it looked like we had enough, but also that one is now kind of something where we— which is why we are currently spending almost 10% CapEx this year in order to get up to speed there. But the fundamental philosophy of the strategy, I think, has been strongly reinforced by the current situation.

Chris Danely — And by the way, is it the same situation on obtaining equipment, as it is on obtaining semis — the lead-times are 12 months plus? Have you seen any relief there? Can you give us any hope?

Kurt Sievers — I am hearing it’s getting a little better. Again, this is — to a large extent, I say I’m hearing, because a lot of this is via our foundry partners, what they get. That seems to get a little better, because indeed, that’s been part of the bottleneck.

Chris Danely — One quick number you talked about, you’re still — your, sort of, orders are still well above your capacity growth, can you give us a sense of roughly what sort of capacity growth you think you can achieve next year and has that changed and I’d really like to dig through the boundaries?

Kurt Sievers — No, what we said is the following: we have de-risked our backlog. You might have noticed we have never published our backlog, and I say this because some of our competitors spoke about the backlog size. We never published this, because we felt right from the start that it is loaded with double orders and all sort of—

Chris Danely — Customers who never do that.

Kurt Sievers — Not true. Customers are people, human beings; if they don’t get something, they ask for more. So, that’s why— it’s just normal. And it’s not easy to differentiate between what is a true order which is substantiated, versus what is probably just double ordering.

Bill and I, we put a lot of effort in doing this, and we have de-risked it, and the result from what we see now for next year, is that we think we are about 80% supply covered against what we think is true demand.

The backlog is much bigger. I mean, if I gave you the number you would say, this is just crazy. So, after that de-risking, we land at about 80% supply coverage against order levels.

A lot of that is these famous NCNR orders, which are the non-cancelable non-reschedulable orders, where in the meantime, the level of those for next year is bigger than what we have for this year. I mean, it’s just important in the current environment where everybody fears about the macro breaking together and whatever, our NCNR order level, which is firm, confirmed orders through the end of next year, are bigger already now than what we have for this year, with a move between— inside the portfolio.

So, we have none of those anymore for mobile, which is probably not surprising to anybody, but we have a significant increase in industrial. Automotive is about the same, which is in line because automotive was the first to escalate. So, a lot of our automotive customers gave us these orders already early for ’21, and they continue to do for ’22.

In industrial, industry has only found out too late, so they were too late to give us these NCNR orders for ’21, for ’22 — so the learning was now to be early to do this for ’23. So, we see a significant uptick, actually, in these NCNR orders for next year. But all together 80% coverage. Now, I don’t know how things move going forward, but that gives you a feel on how robust that still is.

Now, on the supply side, I will not guide next year. So, I will not tell you how much more we can ship next year. But it keeps gradually improving, just like we’ve had it over the past couple of quarters, which is a combination of our internal expansion, as well as the growing capability of the foundry partners.

Chris Danely — So, do you think that the— I guess the largest shortage problem, which right now appears to be in the automotive space, do you think that that could switch to the industrial space, given those dynamics next year?

Kurt Sievers — Chris, the biggest problem isn’t industrial currently, at least for us. I mean, I can only make that statement for—

Chris Danely — I guess the automotive customers are just the loudest.

Kurt Sievers — They had apparently a better lobby, and a better way to express their needs. I think the bigger gaps we have currently is in industrial. The good news is that from a technology or capacity fungibility perspective between industrial and automotive, it’s quite good. So, if either one of them softens, we can move to the other, which is very different to what I said about mobile and consumer markets before, where we cannot really shift very much. But between those two, we are quite fungible.

Now, it needs time. I mean, it’s still — if you think about cycle times in manufacturing, it’s not a quick action, but fundamentally, it’s possible.

Chris Danely — Okay, let’s talk about your foundry strategy, and I guess what’s changed as far as the number of foundries you work with and what you’ve done over the last couple of years, as far as expanding the foundry base, and how much that’s helped.

Kurt Sievers — Yes, what really has changed is that in the past, we used foundries to kind of be the overflow mechanism. So, that means we have the same technologies with foundries which we could do internally. With what I explained earlier, we are changing this, because now internally is largely dedicated to proprietary technologies and the foundries do the CMOS world.

So, that means the relationship to them goes much more into a partnership, long-term partnership model — versus being just a tactical, kind of, overflow mechanism. Which is also important, we’ve all seen how important supply is, but also technology innovation is super important.

We— I think we talked publicly about the five nanometer design, which we are doing with TSMC as we speak, we just hit a major, major design win on this five nanometer part. So, that is important. What is also changing there is the regionalization of their factory footprints.

So, we are seeing, predominantly in the U.S. and following now also in Europe, we see large customers putting a firm requirement on us, that if they award us a big new design win, then we have to assure that the wafer manufacturing has to be local. So, if it is a U.S. company, then they want that we can assure that the wafer fab which will produce those parts is in the U.S. That is a consequence of the whole supply trouble from the past years and of the geopolitical turmoil.

In order to be able to comply with this, we are actually happy from a manufacturing footprint that we have three facilities in the U.S. and one in Europe. So, for our own stuff, we are actually quite well-positioned. And, as we all see, TSMC, GlobalFoundries, Samsung — they all start to build both in Europe and in the U.S. So, I mean that is following, but we are quite happy to see this.

And that, of course, plays also in the whole discussion around the CHIPS Act in— both in Europe and the U.S., which is supporting those manufacturing strategies.

Chris Danely — Yes, that’s a nice segue; what does the CHIPS Act mean to NXP? Can you benefit? Are you putting in your application? Apparently, there is, you know, strings attached, what’s the— is there any impact?

Kurt Sievers — We would never, never do a strategy which is influenced or built on subsidy potential. We have supply manufacturing strategies, and then we are happily taking the opportunity to help them by subsidies. But we would have never done it because there is the CHIPS Act here or there.

Now, since there is one in the U.S., and there is another one in Europe, which is exactly the places where we need more capacity, we will certainly have our applications in place and hope to benefit in the right way.

I mean, I just learned last night that— I think the number I understood was that a quarter of the U.S. CHIPS Act money, a quarter of the \$52 billion, is actually going to be dedicated to mature nodes, which is a pretty good result, I would say—

Chris Danely — I think one of your foundry partners might have had some influence in that.

Kurt Sievers — We’ve been speaking about this all along, because it was important to understand that a lot of the shortages and that all of the drama here in the U.S. was actually due to mature nodes. So, I felt it was an important move to say that a solid portion of that money is also going to be dedicated to mature nodes. And again, that’s what NXP needs the next eight to 10 years.

Chris Danely — Yep. Now, for you guys, you talked about your non-cancelable orders. I guess on the foundry side, they’re trying to hit you up for some preliminary investments, non-cancelable orders. Do you see this as a wave of the future, or just a manifestation that this is such a powerful upturn, I’d appreciate your thoughts on what you see going forward?

Kurt Sievers — I mean, I can be very open what we have, we have like \$4 billion of supply, or purchase obligations from our perspective, from foundries, which sounds like a big number, but if you know that this is spread more or less evenly over five years, and we do in the quarter more than \$3 billion revenue, then it’s not much at all. It’s actually a pretty small number. And it’s more than out-balanced by NCNR commitments which we have from our customers.

Chris Danely — Yeah. So, it just kind of moves on down the line, right?

Kurt Sievers — Philosophically, I think it is a good concept, because especially in these industrial and automotive markets, design wins and the lifetime of design wins are very lengthy. So, there is enough visibility to enter into these kinds of agreements. And I think it just helps the whole supply chain. So, in principle, I think it’s a good thing. And it goes a bit away from this call-by mentality, which we might have had in the past.

Is it going to cover everything going forward? No, I definitely don’t think so. But for certain deals and certain specific single-source components, I think it’s a good move.

Chris Danely — Has it changed the capital intensity at all, up or down, these upfront payments? Or is it just the same part, you’re just paying a little more earlier?

Kurt Sievers — It’s not even always that we have to pay something upfront. It’s more the obligation that we take it over time. So, but there is all sorts of constructions, I mean, that there is different deals with different foundries, I guess, by all of our competitors doing, everybody is doing a little bit his or her thing?

No, I don’t think overall it changes the capital intensity, but it hopefully forces people to be more thoughtful about what future demands are. And with that, what future investments we should actually take or not take.

Chris Danely — Another one that pops up is inflation and input. Hopefully — my mic’s going in and out — Hopefully, with inflation, how is NXP dealing with increasing input costs? And then, is this something that you’ve had to change your planning strategy going forward?

Kurt Sievers — Clearly, the growth in input cost, I haven’t seen this in 28 years. So, it’s a significant change. Let me give you a couple of perspectives on this. First of all, the policy we have very transparently taken right from the start is that we will pass on all of the increased input cost to our customers in such a way that it exactly protects our gross margin percentage. Not abusing it to pad gross margin, but also not being the victim in the middle, which is losing profitability on this.

We’ve done this right from the start, I think Bill and I had a very sharp eye to be really precise on this, and we are very open to our customers, this is how it works. And so far, we have found reasonable, I would say acceptance. I mean, raising prices is never easy, but I would say, so far so good.

I do not think this is behind us. At least in the technologies and process capacities which we need. I dare to say it’s— it looks pretty obvious that our input costs will continue to go up next year. So, we are somewhere in the middle of it, but it continues, which also means that NXP will have to raise prices also next year. So, we are not at the end of this. And that has to do with the fact that foundries keep investing into these mature nodes and there continues to be a supply and demand imbalance in these mature nodes.

The— this, by the way, is also sitting on inventory sizes, because I hear a lot of debates about growing inventory. It is important for everybody to realize that the good part of that is actually the inflation, which sits on the prices or the valuation of these— of those inventories, that spooks sometimes the ratios which people are looking at.

And the same holds true for revenue growth. We— last year, I think we talked about the number, we had a small single-digit increase in pricing in— so calendar year ’21, which was a tailwind, if you will, to our revenue growth. We will let you know what it will be for this year. At the beginning of next year, it will be bigger. And that also explains some of the— where some people say, how can you ship so much product? I mean, some of it is simply pricing.

And we don’t think in our world — again, I cannot speak for the total semi industry, I can talk for those two markets, auto and industrial, which are the majority of NXP. We don’t think that this is going to fall back down again, but we think we will lift those pricing levels now to a new plateau, which will be reached, I don’t know, maybe sometime next year.

And then we will again enter into what we used to have in the past, which is some small annual ASP erosion, but not a snap back to the levels where we came from.

Chris Danely — Okay. A couple more things related to China. You talked about your increasing use of foundries. A lot of fear, speculation, that there could be a potential invasion of Taiwan by China. Is this something that NXP talks about — that you talk about with the other semiconductor mucky-mucks, and clearly you’ve diversified away from Taiwan somewhat — does this enter into the planning stage or anything?

Kurt Sievers — Well, a couple of perspectives. One is – and that plays into the earlier discussion, you’ve certainly also seen the reports about Chinese foundries starting to see under-loading in the current environment. We really don’t use them much at all, because historically, for automotive and industrial customers, which have very high quality, reliability, and safety requirements, we were not allowed to use them. So, we have to use with the bigger players. In the meantime, indeed, we also don’t necessarily intend to use them, because of the geopolitical risk.

Now, when you speak about Taiwan, you speak about one big company, which of course we use. I am not in a position, Chris, to oversee the dramatic consequence if they were cut off, from a supply perspective, to the rest of the world. I mean, that would be just, I can only say catastrophic, in my view. Certainly, the geographic diversification they are doing is, which we talked about earlier, is a helpful step, but that’s only one step. And what more could be done or should be done, I also can’t oversee. I mean, that’s a big topic.

Chris Danely — Sure. Let’s talk about something a little easier: other geographies, like North America versus Europe. You guys are pretty much based in both. How do you see things sort of evolving there? I mean, clearly, there are some worries in Europe about energy costs and demand there. In North America, it seems like demand is holding up in most areas. So, I’d appreciate some input.

Kurt Sievers — Well, I’m not an economist, but I think it’s easy to see that Europe is in tough waters, given the energy crisis. So, if I had to compare, say, China, Asia, Europe, and the U.S., then I would say probably Europe is in the most difficult situation.

China has actually taken significant incentivizing measures to, for example, electric cars after the Shanghai shutdowns, COVID shutdowns, which seems to take traction, so we see positive dynamic in China. And the U.S. so far, for what we are serving, which again is, of course, largely automotive, is holding up pretty strong. So, that’s why short-term, my biggest concern would be with Europe.

Now, just one other spooky thing. We only have one factory in Europe. So, the whole energy cost burden or threat is comparatively super-small for us. And that factory is in the Netherlands, it’s not in Germany. I point out Germany because Germany sees the biggest energy cost issue in Europe. So, we are, comparably, to some competitors, in a pretty good position, from a supply cost perspective in Europe.

But so much more importantly is, of course, how is the macro developing, and what’s the demand and the consumer sentiment in Europe, which I think is most muted from all the regions in Europe.

Chris Danely — Okay, we do have some time left. This is supposed to be an interactive session, I apologize. If anybody has any questions from the audience, I’m more than happy to survey. Otherwise I’m fine to keep flapping my gums up here. Going once…


Aside from the Danely/Sievers dialogue, this article is © 2022 Jason M. Sachs, all rights reserved.



Memfault State of IoT Report

To post reply to a comment, click on the 'reply' button attached to each comment. To post a new comment (not a reply to a comment) check out the 'Write a Comment' tab at the top of the comments.

Please login (on the right) if you already have an account on this platform.

Otherwise, please use this form to register (free) an join one of the largest online community for Electrical/Embedded/DSP/FPGA/ML engineers: