TechForge

October 5, 2026

  • Universities take semiconductor research to technology readiness level three and industry picks up at seven, leaving it stranded in between.
  • A*STAR and NTU now run early-stage work on production-grade tools so a breakthrough does not have to be rebuilt before a fab will take it.

Singapore has been building institutions to cover a stage of semiconductor research that universities stop short of and working fabs cannot spare the capacity for, according to three people who work at different points along it.

The gap has a name: the valley of death. Research organisations work at the first three technology readiness levels (TRL), the scale used to describe how close a technology is to deployment, where the questions are whether something works at all. Manufacturers work at levels seven to nine, where the questions are yield, cost and reliability at volume. Levels four to six, where a working idea has to be turned into a repeatable process, sit between them.

“You heard about the valley of death?” Professor Ng Geok Ing, executive director of the National Semiconductor Translation and Innovation Centre for Gallium Nitride, or NSTIC (GaN), a joint A*STAR and Nanyang Technological University facility, asked reporters during a recent site visit. “This is a big problem. The university has a lot of ideas, research ideas. But do you realise something? How much is actually going to the marketplace? Very little.”

The centre was established in 2023 as the National Gallium Nitride Technology Centre and opened under its current name in June 2025, with National Research Foundation support.

In a research environment, Ng said, the work stops at level three “because that’s their work, that’s their focus.” Industry starts later. “Can you scale? Can you have a reliable process? The yield, the cost. And typically they work in TRL seven to nine. What happened to four to six? Nobody cares.”

That, he said, is what the centre exists to cover. “The whole world starts to realise something is not right. We put so much money into basic research, nothing comes up.”

Ng said the same approach runs through the other centres under Singapore’s semiconductor research flagship, including advanced packaging and silicon photonics.

Doing semiconductor research on production tools

The mechanism both A*STAR and NTU described is narrower than it first sounds: run early research on the same equipment a factory would use.

Professor Yeo Yee Chia, deputy chief executive for innovation and enterprise at A*STAR and executive vice president of Singapore’s RIE2030 Flagship in Semiconductors, said the alternative is what used to happen.

“In many countries, and long ago here even in Singapore, when early-stage research is done it’s often done on equipment that are meant for university activities,” Yeo said. “So then you find that even if you have a breakthrough, the next step is to redo the whole development on production-worthy tools.”

Annealing, dry etch and plating process bays inside the NSTIC GaN cleanroom in Singapore
Annealing, dry etch and plating bays inside the NSTIC (GaN) cleanroom. Ng said the centre runs production-grade tools so processes can move to a commercial fab without being rebuilt.

Running the research on production-grade equipment removes that second build. “Once you have achieved good performance results, then it’s immediately transferred to a fab,” Yeo said.

He was clear this is not simply a procurement decision. “It requires a lot of knowledge,” he said, naming the problem of knowing which tool etches a given material best, or bonds one material to another. “It’s a trade-off between risk and the wins.”

Ng described the same approach at the GaN centre, where he said the tools are production-grade and inspection is automated. “We bring in the industry discipline and try to make it as close as possible to a production fab,” he said, “so that later when it’s really the volume scale, there’s a demand, we can easily pop over to any big wafer fab.”

The centre also does something university labs generally do not. “We don’t just stop at producing the wafer. We even do qualification, and this is very rare,” Ng said. “In university nobody cares about qualification. You just produce something, off we go, paper, that’s it. But here we do, because the customer [wants to know] whether your process is reliable or not.”

He calls the model a boutique foundry. “We are not like a big foundry that goes high volume. This is a smaller volume,” he said. “We call it boutique because our focus is R&D, to continue to create new technology more advanced than what industry has.”

Why a running fab cannot do its own experiments

Tan Yew Kong, senior vice president and general manager of GlobalFoundries’ Asia-Pacific manufacturing and its Singapore site, gave the commercial reason from the other side.

GF has its own research team, he said, and uses its manufacturing floor to run experiments. “There is a limit of what and how much you can do,” Tan said. “The reason is that especially when the market is very, very tight, every production capability comes to the market. So you can’t cut away one piece and say, ‘ Can I take 20% of your capacity to do my experiment? We can’t handle it.”

A research institute has no such constraint. “They are okay with cleaning out the machine next month, then they do something else. That’s fine,” he said. “Because for us it’s every day; there are so many wafers in and out. Different technology, different processes. So the chances that you hit something is much higher.”

Tan was explicit that this is about speed rather than capability. “If you give us time, we will definitely be able to do that ourselves,” he said. “It’s just that we don’t have time.”

What survives the funnel

Yeo put numbers on the attrition.

“You have a funnel that is very wide at the opening where you entertain like a thousand ideas,” he said. Most are killed early. “As you mature into the research, you end up with a smaller number, like a hundred. And then you bring those hundred to the next stage of development. And then more will be dead.” Reasons include cost and complexity. “Then you end up with a handful, and then one goes to production.”

“So how do you quantify something like 999 dead and then one successful one?” Yeo said. “Sometimes you wish you had killed earlier.”

The cost of each stage rises as the numbers fall. “A thousand ideas, how much would you spend per idea? Very small investment,” he said. “But at a later stage you invest a lot more, because it takes a lot of risk mitigation in terms of volume, in terms of equipment.”

Yeo named Advanced Micro Foundry as the case that made it through. A*STAR spun the silicon photonics company out in 2017; GlobalFoundries announced its acquisition in November 2025. He described it as a great success and said A*STAR has a pipeline of further spin-offs planned, while also licensing technology directly to companies. Spinning off carries more risk than licensing, he said, and which route A*STAR takes depends on whether it can grow the business to a point where it is worth selling.

A second example is further back in the pipeline. A*STAR has embedded its own equipment inside STMicroelectronics’ fab to co-develop Micro-Electro-Mechanical Systems (MEMS) technologies, Yeo said, with ST providing access to complementary tools in return. Deployment of technologies from that platform will happen over the next few years.

The dates the GaN centre has given itself

At the GaN centre, the near-term milestones are fixed. Ng said the process design kit for six-inch wafers is due in January 2027, with process qualification targeted for the middle of that year. Four-inch has already been proven with a customer.

Monolithic microwave integrated circuits on a 100mm GaN-on-silicon carbide wafer fabricated at Nanyang Technological University
Monolithic microwave integrated circuits on a 100mm gallium nitride-on-silicon carbide wafer, fabricated at Nanyang Technological University and on display at NSTIC (GaN). Ng said the four-inch process has already been proven with a customer, with a process design kit for six-inch due in January 2027.

Rohde & Schwarz is among the partners A*STAR lists at the centre, working on GaN-on-SiC high electron mobility transistors for millimetre-wave applications above 100GHz. Ng said the German test and measurement company needs chips faster than the products it tests, that very few exist, and that it foresees needing them around four years out.

Ng was also candid about where gallium nitride sits against Singapore’s better-established areas. “If you look at GaN, it is still very nascent compared to other technologies,” he said. “So in a sense we are still trying to catch up.”

About the Author

Dashveenjit Kaur

Dashveen writes for Tech Wire Asia and TechHQ, providing research-based commentary on the exciting world of technology in business. Previously, she reported on the ground of Malaysia’s fast-paced political arena and stock market.

Related

November 6, 2026

November 6, 2026

November 5, 2026

November 2, 2026

Join our Community

Subscribe now to get all our premium content and latest tech news delivered straight to your inbox

Popular

35672 view(s)
28210 view(s)
25615 view(s)
20737 view(s)

Subscribe

All our premium content and latest tech news delivered straight to your inbox

This field is for validation purposes and should be left unchanged.