Deloitte: The end of the semiconductor shortage is near

Semiconductor chips have been in short supply for the past couple of years, as the supply chain has been whipsawed by the pandemic.

The shortage has impacted the entire electronics food chain of products, as chips are the basic building blocks of all things electronic.

The shortage was complicated because complex devices such as game consoles can require thousands of components. If a single $1 chip isn’t available, it could hold up the shipment and sale of a device, appliance, or vehicle worth much more, Deloitte said.

As recently as December, Deloitte Consulting predicted that the shortage would last through all of 2023. But now the company’s chip analysts believe that the end is within sight, as the chip industry has boosted capital spending and much-needed factories are starting to come on line after lengthy construction projects. Demand is also starting to change as a slowdown in the world economy is expected. Fears around inflation have driven stocks down and concerns about a recession are leading buyers and consumers to be more cautious.

Meanwhile, chip manufacturers are waiting for the U.S. Congress to approve the Bipartisan Innovation Act (CHIPS Act) to provide subsidies for factories to be built in the United States. Intel recently delayed a chip factory groundbreaking on a $20 billion facility in Ohio due to the legislation uncertainty. (I couldn’t get Deloitte to comment on pending legislation).

I talked to Chris Richard and Brandon Kulik, both principals in the semiconductor consulting business at Deloitte. They have decades of experience in the industry.

Here’s an edited transcript of our interview.

semiconductor chips in a pile
Semiconductor chips.

VentureBeat: Did you have a headline in mind when you first reached out to talk about this? It sounds like the semiconductor shortage continuing would not be a headline of the moment, but if there’s some change to that, that’s something to talk about.

Brandon Kulik: To me, if I was writing a headline, I’d say something like, “The end is not here, but it’s closer.” That’s how I’d represent it.

VentureBeat: What makes you reach that conclusion? What do you see?

Kulik: If you’re a semiconductor company trying to forecast, you look at a lot of things. Obviously device demand–there are indicators that device demand is going down, even though services demand continues to be strong. Samsung is a major retail and consumer device manufacturer, and they pushed some orders out. But on the whole, we’re not seeing that trickle down to pulling back on manufacturing capacity or production. I haven’t heard, and Chris can add to this, that there are major demands or realization that the shortage is coming to an end. “Customer demand is dropping so let’s pull back on inventory.” We’re not seeing that.

Chris Richard: The way I look at it, I don’t know all the drivers, but back in 2019, even before COVID really started to hit the press, the shipments in the industry were starting to trend out. It’s very cyclical as you look back over the last 20 or 30 years. We were in a natural drop in 2019. You get into the beginning of 2020 with the lockdown. The drop stayed, and then all of a sudden people realized they had to work from home. They needed new computers, new laptops, new tablets, new phones. Toward the end of 2020 the demand started to go through the roof. We’d already been cutting back, and then it surged.

That’s been the cycle, and as you look back on that period where we were under average for almost eight quarters, we have to make that back up, assuming that demand actually never went away. I think we’re well into next year, end of next year–unless there’s another huge issue. The big one on the horizon is recession. If people start cutting everything back that could change the dynamics. But just recovering from where we undershipped for those eight quarters, we’re looking at well into next year.

Kulik: A lot of it depends on the market you’re talking about. I don’t see shifts in the data center market. They need to process data. That isn’t going away, even in a recession. The need for data services to support enterprises and consumers–social media isn’t going anywhere. Automotive, there’s years of inventory that they still need that they’re behind on. Automotive will continue to push demand, even if we see things like phone sales starting to slow, or PC and laptop sales starting to slow.

VentureBeat: Can you take me back to the onset of the pandemic and how much demand increased? Why did that produce a shortage? How much did demand go up in that short time? And then how long did chip makers need to respond, in order to beef up factories and invest more into capacity? At some point, if their reactions are starting to take effect–how long does it take for that cycle of deciding to invest in something to get to the point where you’re producing?

Richard: To the second part of the question, I can answer that in two different ways. First, the short term, and second, the long term capacity. In the short term, what a lot of people–I’ve probably done about 15 interviews like this over the last couple of years. Consistently what I’ve found is there’s this general perception among the public that because this is a microchip, it must take a micro amount of time to make. It can’t possibly take that long, right?

The reality is, for the most advanced chips, like you would have in your computer or your phone right now, that could be a 26-week lead time from when you start the production in the wafer fabs to the point where it’s in the warehouse and shipping to the downstream customer. Then it could be another month after that to get into a PC or a phone. You’re looking at six or seven months from when that manufacturer starts production to when it’s actually usable by a consumer. That’s the first thing to keep in mind.

If you’re an automotive company, you might think your demand is going to get cut by 20 percent over the next two years. You cut that volume of orders. Then six months later you realize you need those orders. From the time that you cut back to the time they start getting product again, that’s a year. Six months of not ordering and then six months to the pipeline surging again. That’s one way to think about it. That’s within the existing capacity. That’s just the supply chain dynamics of people having orders, placing new orders, and having to wait for them to arrive.

As to the second part, which is more directly the answer you were looking for – I just wanted to give you the full context – to order new equipment and get it in a factory is a minimum of three months, and more likely six to 12 months under normal circumstances. Normal circumstances meaning all your competitors aren’t ordering the same tools at the same time. But at this stage, because lots of companies have gone in and tried to order new equipment for their factories–we’ve heard from a lot of capital equipment suppliers that it’s at least a 12-month lead time from when they place an order to get those tools.

Deloitte's capacity predictions for chips.
Deloittes capacity predictions for chips.

If you place an order now, you get the tool 12 months from now. It can take another quarter to get that tool installed, certified, up to production, passing all the quality standards. You’re really looking at more like 18 months on average, typically, from when you start to add capacity to when that gets installed and able to run a real production. That assumes you have space in your facilities. If you have to build a new factory, add two more years.

Kulik: If you think about how long it takes to add a factory–it’s years. It isn’t like they build it all at once. A lot of these facilities are campuses. They progress and add buildings. The way I see it, they’re picking locations. There are several dozen new fabs that are in process, and several dozen more that haven’t broken ground yet. But they’re going to build, and as they build, demand is going to fluctuate. They’ll speed up or slow down the addition of space as that demand fluctuates. It may be a five-building location, and they may build one or two of the buildings now. They may decide exactly what devices and equipment to build there. It’s a much longer-term process.

It’s not just, “We’re adding a bunch of fabs. Company A is spending $20 billion to do this,” and in a couple of months’ time it’s there. The demand and supply is going to even out a lot more before those fabs come online. A lot of our manufacturing clients are making more use of the space they have. They may have a distribution center, and now they’re putting equipment in it to make more chips. The addition of capacity is a much longer-term endeavor than people think. When we can say that we’re much closer to equilibrium, it will be before a lot of that new capacity comes online, both globally and in the U.S.

The other thing I’d say is that a decline in demand isn’t necessarily a bad thing in the near term. The difference between demand and supply is raising prices. Inflation is a major problem for the economy. A shift back to more equilibrium would be a healthy thing. The question is, where’s the point at which supply exceeds demand and you cross that equilibrium? When can we say we’re out of the cycle? No one really knows that. But I agree with Chris that it probably isn’t this year.

Especially, the back end of this process, the testing and assembly of chips–not just the wafer fabs, but the back end as we call it, is a major bottleneck. The thinner margin, smaller, mid-sized companies, compared to the big-name brands that you see, they’re the ones that do the packaging, that put the silicon in a carbon case and make sure it’s tested. That’s a major bottleneck right now too. It isn’t getting as much press because those companies aren’t as big and they’re not well-known.

VentureBeat: If we’re still working through the earliest possible time that a new factory could come online, would that be maybe nine months from now?

Kulik: For the new big ones, yes. For an extension or re-outfitting of an existing one, it’s a bit faster.

VentureBeat: Everyone’s been talking about localizing the supply chain again, whether it’s for national security reasons or other reasons to bring manufacturing back to the U.S. I would think another roadblock could be that we don’t have the talent here anymore. Do we have people to hire to staff these factories?

Kulik: We’ve heard there are as many as 90,000 job openings in semi that are unfilled. This industry isn’t immune to the same challenges other industries are facing, and maybe even less so than others. You need a specific kind of competence. It requires a lot of training. There’s not enough engineering coming out of schools in the U.S. to satisfy that. Folks will have to be retrained. They’ll have to pivot. Or we’ll have to wait for folks to come out of school to be absorbed into the industry.

A lot of these machines are being developed now to be maintained a little bit more remotely. It’s more about software than someone turning a screw. But you still have to have people on the floor, especially if you’re changing from one device to another. You need a lot of mechanical expertise to do that. These are some of the most advanced machines for making anything in the world.

Intel's D1X factory in Hillsboro, Oregon.
Intels D1X factory in Hillsboro, Oregon.

VentureBeat: If there is some light at the end of the tunnel, what are the indicators there? You mentioned Samsung. What else is happening?

Richard: My projection – and it’s mine, not Deloitte’s, not anything formally certified – is based on analysis I’ve personally done. When I look back at the integrated circuit shipments trend, as I was describing earlier, the industry probably undershipped, by my calculations, 62 billion units over that period of time. Now it’s kicked into high gear, and by my estimation the industry can make up about 5 billion a quarter of that deficit or that debt. That’s 12 quarters. We started this recovery, if you will, in Q4 of 2020, so by my very simple and very high-level math, that puts us back on trend of where we were in Q4 of 2023. That’s the simple math behind my projection. That speeds up if demand declines, and it speeds up if more capacity comes on above the historical trend, as Brandon said. It’s a very rough, back of the envelope calculation. It could go either way based on these other factors.

VentureBeat: I remember that something like a PlayStation 5 might have about a thousand parts in it. I heard early on that some of the problem was winding up being these two-dollar parts or whatever, these little capacitors. That seems to be one of the big frustrations. You could have 999 parts ready, but if one is missing, even if it’s a cheap commodity, you can’t ship.

Richard: That’s the frustration in the auto industry. They have their $5000 engine and transmission, the whole powertrain and the body and all that, but then they need a 50-cent power regulator device that’s out of stock. You can’t complete production because of that. That’s absolutely true. That’s the area that’s caught a lot of companies off guard. Most companies, just to manage the numbers of parts–you put most of your focus on the expensive parts. You have to manage price and manage inventory, because you don’t want to have too much. You don’t want to buy stuff that becomes obsolete. Eighty percent of the focus goes toward 20 percent of the parts.

But then you have what a lot of people call the long tail of other parts. That’s what you’re talking about. The three-cent capacitor or the 49-cent timing chip or what have you. It’s this long tail of parts. The traditional way of dealing with these is buying some extra so you don’t have to micromanage inventory so closely. They don’t cost much, so you’re not putting much inventory on your books. But now this long tail that hasn’t gotten a lot of attention has gone short.

Kulik: About half the world’s manufacturing capacity for semiconductors is on what we call the less mature nodes. These are the simple parts. The factories that make these cheaper, more commodity parts are harder to re-outfit, because the machinery is older. This stuff has been around for 20 years. Things like a simple controller for an airbag. There’s less capacity flexibility to re-outfit than there is on the more mature node where all the emphasis is at – not just from the folks that use the chips, but the manufacturers that make them.

VentureBeat: How do you think the geopolitics of this might shake out? Will the U.S. get back to being more of a manufacturer than it has been in recent years? Are countries able to think more about which other countries they’re dependent upon?

Richard: I think we can break that into two parts. First off, the U.S. still has significant semiconductor manufacturing capability. All of Intel’s technology development happens here. A lot of the wafer production happens here. You can go on. IBM has foundries outside the U.S., but still factories here. There’s still a fair amount of what we call the front end manufacturing here, making the wafers. Brandon earlier talked about the back end, which is taking the silicon and putting it in a package and testing it. Almost all of that happens in Asia. When you think about semiconductor manufacturing, it a lot of it already does happen in the U.S., but all that stuff gets shipped to one or more countries in Asia for the final stage of manufacturing.

With the CHIPS Act, there’s going to be even more investment. But I think that the nuance on it, and what most people are referring to, is what they call the foundry business, where smaller fabless companies can outsource their manufacturing. The giant is coming called TSMC. That just supports so much, in Taiwan particularly, that–I think that’s why you hear about it. Now, quickly on the heels of that though, TSMC is building a huge campus in north Phoenix. They already had factories here. If you’re interested, you can search on YouTube for TSMC Phoenix. They have some great drone footage flying over and showing what Brandon was talking about. There’s been a lot of pressure on TSMC to diversify because of the geopolitical risks. That’s definitely happening here.

Intel CEO Pat Gelsinger breaking ground on chip production.
Intel CEO Pat Gelsinger breaking ground on chip production.

In summary I would say there was already a lot of manufacturing in the U.S., and that’s increasing with the CHIPS Act and the need to geopolitically diversify. But most of that is still just the front end manufacturing. Right now all of those silicon wafers fly back to somewhere in Asia to get tested and put in a package.

An aspect of the geopolitical conversation that’s worth talking about is it’s not just every country trying to do it on their own. There’s a line of thought flowing through commerce around the idea of “friend-shoring.” You have the western world. You have Europe. You can read some recent headlines where CEOs have said that they’ll never be able to completely escape the quality and the volumes that come with the manufacturing that’s done in Asia. But the ability to build more in the U.S. and in Europe, on the back end as well–you can see where Commerce put out an RFI that had about 200 responses concerning what to do with the CHIPS Act and how to think about it. One of our big points was, “Don’t forget the back end. Think about friend-shoring and near-shoring, not just the U.S.” We’ll have to see where that goes politically.

VentureBeat: You mentioned 90,000 open jobs in the chip industry. Is there some realism we can bring to that conversation about whether the U.S. is still capable–I find it a bit hard to believe that we can come anywhere near staffing some of these big factories with so many open jobs, given what we’re producing in terms of engineers and other technical people.

Kulik: A lot of those jobs aren’t necessarily in the fab. These fabs are super automated. That number doesn’t–you don’t have to go fill half of them in order to get these fabs going. But there are thousands. Every major fab will come with thousands. It will take a while.

VentureBeat: I imagine H1 workers are also going to enable companies to fill these jobs.

Richard: It’s definitely a mix. What gates bringing a fab up is the skilled trades. Construction, welding, pipe-fitting, electrical. That’s the first constraint. You may not have enough skilled trades in the U.S. to build all these factories at once. That slows down the production. And then when you get into–you can separate what I’ll call operators and technicians from engineers and other people. I’m not aware – although I could be wrong – of a catastrophic shortage of engineers. It’s definitely low. It’s definitely hard to access. There’s definitely competition. We could point you to a talent study we did about four years ago on the specific degrees that are difficult to hire in the industry. It’s what you would expect: electrical engineering, computer science, and so on. Those people have a lot of opportunities. It’s not perfect, but it’s known. You can respond with better targeting at engineering programs. You can influence some things about that over the long term.

But then the other category is the operators and technicians. These are the people that work in the fabs. They do maintenance on machines. They work in what’s called the sub-fab, supplying electricity and high-purity gases and so on to the tools. The way companies have dealt with that is they collaborate with local tech schools, vocational schools, and community colleges. If you look at the Phoenix situation, not only is TSMC building this big campus up in north Phoenix, but Intel already has four factories, and they announced they’re building two more. A community college there reached out to me a couple of years ago. They just identified it as an opportunity. They said, “We want to partner with these two big hires. We want to put together a program that allows us to feed young men and women out of high school into whatever kind of training they need to work in these factories.”

Necessity is the mother of invention. All of these communities want the jobs. They want the employer there. It’s always amazing to me how this innovative spirit comes out with things like the local community colleges and tech schools working to help pick up this other category, the operators and technicians.

VentureBeat: As far as the whipsaw in demand goes, if we’re concerned about inflation and a recession now, then you would think that people might want to start cutting back on those factories. Just as they’re about to come online, maybe everybody realizes they shouldn’t do this.

Richard: It’s not on or off, all there or not. You can probably think about it–in one of these big mega-factories, they’ll have thousands and thousands of tools in there, but you can think of it as coming along in roughly five percent increments. You can bring up 15 percent of the factory, and then 20. Wherever you want to pick yourself on that curve.

Obviously it’s most efficient, and therefore most profitable, when you’re running the whole thing. You have one plant manager and a whole staff there whether they’re supporting 30,000 wafer starts or 500 wafer starts. It’s the same set of fixed costs, if you will. There’s certainly a benefit to going as fully utilized as possible. But it’s not an absolute necessity.

Intel is spending billions on manufacturing.
Intel is spending billions on manufacturing.

Kulik: One thing I’m curious about–my instinct is there’s a lot of over-ordering. Over the last year, everyone was adding another 15 percent just to make sure they weren’t cut short. To me, that will be the first bucket of orders and demand that’s going to drop. We’re starting to see demand signals start to lower a little bit. The orders that they’re really going to plan for, the orders that they’re going to forecast and commit in terms of the financial cost of production. I don’t know what the overall ordering factor is, but I think it’s significant enough that that will be the first bucket of demand that we’ll see move.

It could trigger a self-fulfilling prophecy. Everybody over-ordered, so now everybody’s going to over-cut. We could see that happen too. It’s just a matter of when. We hope that’s not the case. A lot of what we do for our clients is to help them not do that – not over-order, not under-order, not over-forecast, not under-forecast. The good news is that the industry realizes how much better it needs to be at this. Not to advertise all the things we do, but a lot of what we focus on in the industry is helping companies to plan better. They will get better. It will just take a little while.

VentureBeat: As far as the impact on products that you’ve seen, one extreme thing I saw was the announcement of the Intellivision revival console. They were going to ship in October of 2020. Then they couldn’t get parts for it, couldn’t get the manufacturing lined up. It wound up never shipping. They couldn’t deal with the delays after tens of millions of dollars went into it. It’s sad that this constraint stopped that product from shipping. What have you seen on that level?

Kulik: It’s tough to innovate and start new companies with new devices. The big guys get all the priority. If you want to bring back the original Atari set, then you have to go up against the Microsofts and the Nintendos of the world that own the game consoles. When they say, “We go first, here’s the order,” the foundries and the contract manufacturers are going to prioritize. You have to wait if you want to do something new that starts smaller.

Richard: There are capacity cycles. There are ups and downs. The foundries, I’ve found, very specifically are good at offering incentives to add demand when their supply is freeing up. They’ll approach the customers and say, “Can we offer you something here? We can offer a reduced price if you can add some production to the factory.” Once again, there’s a lot of innovation here.

The other thing about the foundries is they don’t want to miss the next wave. My experience with them is that they’re very good at looking at the ecosystem, very good at looking at who’s coming up. They want to work with a company that shows promise specifically around volume of production. I don’t know if the example you’ve given would be a volume production. But if you’re a startup, if you have some really innovative product, and it can take off on its own, or it’s likely to be bought by a bigger company and be incorporated into their product suite, the foundries in my experience have been pretty smart about strategically supporting those emerging companies when capacity is available.

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.

Source

By admin