Rise Of The Old Fab

Growth in the , along with the beginning of a shift toward systems in package, are creating buzz in a rather unlikely place—established and well-worn process nodes where equipment is scarce, semi-functional, and difficult to maintain.

In the past, moving to the next node was a sign of progress, leaving behind the trailing edge of designs to second-tier and specialty foundries. Much has changed since then. While some parts of the IoE will require the most advanced processors available—for example, the SoC in a high-end smart watch where computing will be done using finFETs or FD-SOI processes—the vast majority of the tens of billions of chips in this market will be manufactured using older processes.

On top of that, while some chipmakers will continue to push to 5nm and beyond, there is work among leading-edge companies to develop 2.5D and fan-out configurations, some of which may be done using finFETs at the most advanced nodes and some using other chips developed at older processes. The ability to mix and match, and to change distances between components by stacking them vertically, as well as improve throughput using interposers, through-silicon vias, or other high-bandwidth connections, is being viewed increasingly as a viable path for future designs.

But there’s a big problem with these scenarios. The equipment used at older nodes is in short supply. Most of what’s available for sale is in disrepair, which is creating a very robust market for what is available. There are even old tools being sold that are no longer functional, but buyers believe they can get them working well enough to be useful.

“This has always been a steady business, but last year was one of the best years, driven by the need for used equipment,” said Wilbert Odisho, vice president of the K-T Pro group within KLA-Tencor‘s Global Services Unit. “That was mainly 200mm, but there is some 300mm. This is typically N-minus-2, meaning 40nm and above.”

KLA isn’t the only company to notice this. EV Group has been selling bonding equipment at older nodes for some time. And Applied Materials has closely tracking this shift and assessing the size of the opportunity.

“In the last four to five months we’ve been working to formulate a position in this market,” said Mike Rosa, Applied Materials‘ director of strategy. “It’s part ‘More than Moore,’ and part IoT, which includes More than Moore. We’re also trying to figure out how this translates to the existing 300mm market and where the gaps are between 200mm and 300mm.”

And LAM Research has been actively involved in refurbishing equipment for 28nm and older nodes, as well as offering systems that are a combination of new and used equipment, according to Wendell Isom, vice president and general manager of the company’s Reliant Systems customer support business group.

“The realization of the IoT and More than Moore mean that companies are now extending fab productivity by developing alternative solutions for the market,” said Isom. “For example, low-power technologies for wearables are being developed at 200mm and 300mm facilities worldwide. Historically, the view was that 200mm and smaller wafer sizes would decline and that the only technology at smaller wafer sizes would be adjacent markets such as MEMS, LEDs, power, and some packaging applications. Today, these wafer markets continue to thrive, allowing us to continue building new older-generation products.”

If you build it, will they come?
The big question for equipment vendors is whether 200mm and older-node 300mm foundries will buy new equipment for those processes, or even whether they will pay a premium for factory reconditioned and updated equipment. These are typically ultra-thin margin operations, and equipment costs have a direct bearing on the bottom line.

“There are a lot of tradeoffs in this market,” said Joanne Itow, managing director for manufacturing at Semico Research. “New 200mm tools are more expensive to buy, and they may be expensive for equipment vendors to build new. And 200mm fabs are used to very cheap used equipment. They don’t necessarily want to pay new equipment prices. They will for things like bonding, but not for everything.”

Itow noted, however, that some of these tools are now in very short supply. “They lasted a lot longer than anyone expected, but now it’s getting hard to find parts and you can’t find technicians. A lot of people who knew how to fix this equipment have retired. But the equipment is still cheap to run.”

KLA’s strategy so far has been to beef up older equipment. “We’ve been offering bolstered certified products and adding the ability to enhance and upgrade equipment in the field,” said Odisho. “Some of these products have been in use for 10 to 20 years. Now we are breathing more life into them, adding serviceability and improved performance. We’ve been buying back tools so far and refurbishing them, but we’ve reached a point where we’re considering manufacturing entire product lines again. That’s a first of its kind. Some of these were flagship products made for 90nm and above, which is where we’ve seen more interest.”

Odisho noted that one of the big problems that older foundries have is constrained space. If equipment can be refurbished to run at twice the speed, that allows them to reduce the equipment they have and add new equipment, such as scanners or etchers. Moreover, because some of the old equipment is in such disrepair, just getting it back into decent working order can improve output by 20% or more.

Applied Materials likewise has been unofficially refurbishing used equipment for foundries, but it is looking far beyond that. While equipment makers today will repair 200mm equipment, the real focus seems to be on new or updated 300mm legacy-node equipment.

“There is a lot of life in 200mm legacy, and there is a lot of life in 300mm legacy,” Rosa said. “This applies to consumer, automotive, and a gradual trend toward the IoT with MEMS, power and analog.”

Rosa noted that the main markets for this equipment are in Europe and North America, with Europe outnumbering North America by a factor of more than two.

But there also are a number of new fabs being built in China using older process technology, primarily for MEMS and IoT devices, according to SEMI analyst Clark Seng. He cited six new projects that are either pilot runs or foundry-oriented, all using older-node technology.

And the opportunity for 2.5D and fan-out configurations is growing rapidly, as shown in the Fig 1 below. While processors will still be developed using the latest cores, chips containing analog and power management can be done at much older nodes and combined with the latest processors and memories in a single package.

Fig. 1: Growth rate for several key “More Than Moore” chips.

But selling equipment into these markets isn’t so simple. Semico’s Itow noted that some of these older-node markets are especially volatile, which makes it difficult to plan for equipment purchases. For example, smart cards may be negative one year, followed by a couple years of robust growth, followed the next year by more negative growth.

In contrast, other segments such as sensors are far less volatile because the growth can be spread across a number of different market segments, ranging from medical to security to automotive, which typically run on different business cycles.

Rebuild or buy new
Rebuilding older equipment comes in a couple different flavors. One approach is simply to get it working again, which is one of the services offerings of most of the major equipment makers—official or otherwise. A second and more ambitious scheme is to improve the throughput of the existing devices, add Internet access and field-upgrade capabilities to tools for inspection, metrology and pattern recognition, as well as modern controllers.

“Technology from leading edge applications is used both in our newly built older-model systems as well as for our refurbished systems,” said Lam’s Isom. “This extends capability beyond the technology nodes originally addressed, which lowers the cost of ownership. Our equipment addresses 200 mm and smaller wafer sizes and 300 mm semi-critical and non-critical applications.”

The company also offers what it calls “obsolescence insurance” to extend the life of equipment.

But there are some limits to how far all of this can be extended. “When a tool is really old, there is only so much you can do with it,” said Odisho. “But there is other equipment that can still be legacy but behave like a newer platform.”

Florent Ducrot, director of semiconductor services marketing for 200mm products at Applied Materials, agrees. “We’re looking at the field capabilities and improved performance in terms of throughput to get more out of existing tools,” he said. “But we’re also seeing that for consumer devices, the content on 200mm is significant. That’s going to drive lot of activity.”

Buying new equipment has been more of a gamble in the past, more so for some markets than others. But much of business analysis was done in the context of a vibrant used equipment market, where foundries sold off their used equipment and replaced it with newer equipment. As development spreads across all nodes, the used equipment supply is growing tighter while demand continues to increase. That reduces the delta between used and new equipment, and it makes newer equipment less of a gamble on all sides.

There is no set formula here for how this will work out for companies on all sides, but what’s clear is that all options are on the table. Services revenue are important, but the business model for hardware is much more scalable.