Memory R&D Is a High-Stakes Game

LongbridgeAII'm LongbridgeAI. Here is the key information I extracted
- Micron Technology announced a $10 billion investment over the next decade to establish Micron Research Labs in Boise, Idaho, focusing on critical memory technologies and AI-driven architectures.
- The initiative addresses escalating R&D costs and the industry-wide shift toward long-term, foundational research involving universities and national laboratories.…
Disclaimer

//php echo do_shortcode('[responsivevoice_button voice="US English Male" buttontext="Listen to Post"]') ?>

The need for significant investment in R&D is nothing new for memory makers, but Micron Technology’s creation of a dedicated facility that looks beyond the next node illustrates why the industry cannot afford to skimp on R&D, especially in the AI era.

Micron Research Labs, announced in the summer of 2026, gives the DRAM maker a long-term research home of its own. Based in Boise, Idaho, and backed by a planned $10 billion investment over the next decade, the facility will focus on critical memory technologies, advanced memory and compute architectures, packaging, and future semiconductor manufacturing. The investment also covers university collaborations, satellite labs, and ecosystem partnerships.

Micron expects to break ground in 2027 on a facility capable of hosting hundreds of researchers. It has already appointed Deirdre Hanford, who spent 37 years at Synopsys, as corporate VP and president of the new lab. Hanford will lead the establishment of the lab’s research infrastructure, operating model, and partnerships.

The lab’s first notable outreach came earlier this fall, when Micron said it intends to invest up to $50 million over five years to support the U.S. Department of Energy’s Genesis Mission Graduate Fellowship Pilot Program and related national laboratory research.

For Scott DeBoer, Micron’s recently appointed president and chief technology and products officer, the lab represents a change in how the company participates in semiconductor research. The need for R&D has always been pressing, but efforts have mainly focused on relatively near-term challenges and goals, he said in an interview with EE Times. “I couldn’t have imagined the R&D investment level that we’re at today.”

DeBoer said the lab’s funding is an addition to previously announced or forecasted investments, which have increased significantly every year over the past four years.

Looking beyond the roadmap

Micron has always had to look ahead, DeBoer said, but mostly within the bounds of the company’s technology roadmaps, which usually chart the next five to seven years. “We have to have a pretty good view of what we’re investing in and what we’re going to bring to market through that period of time.”

He said the company has had less influence on research that looks a decade ahead or beyond, in part because there was more government funding flowing into U.S. universities. “We actually didn’t have to be the leader of that kind of research effort.” Now that memory has become much more important, Micron is taking more of a leadership role in longer-term research with universities and national labs, DeBoer said.

Micron’s approach in recent years emphasized supporting professors and students in building workforce capacity. Longer-horizon partnerships allow it to pursue foundational research outcomes while mitigating some of the intellectual property complications associated with projects nearing commercialization. “This is a pivot, where we go out farther, where we actually can have research outcomes in mind,” DeBoer said.

AI adds to the R&D burden

Much of the memory industry’s R&D has historically focused on immediate process and manufacturing improvements—shrinking DRAM from one node to the next and getting yields up. AI makes that longer view more urgent while complicating nearer-term development.

DeBoer said memory is the engine driving AI, whether in the data center or at the edge, in cars and robots, all of which depend heavily on memory performance, memory bandwidth, and power.

He said the diversity of applications is a huge multiplier of R&D requirements. In the data center, for example, the capacity and performance specifications for LPDDR have changed drastically, while robotics and automotive bring in advanced packaging and 3D implementations. Nearly all of these involve specific engagements with key customers, rather than simply delivering another scaled DRAM node, DeBoer said. “You can’t optimize the system without optimizing how the memory works in it.”

R&D is the price of staying competitive

Dan Hutcheson, vice chair of research firm TechInsights, has studied semiconductor R&D since the 1980s, and told EE Times in an interview that Micron’s move makes complete sense—the new lab is a way to bring in external funding and outside ideas.

Hutcheson estimated that R&D costs rise by roughly 30% with each node transition and that, in the 1980s and 1990s, costs had often increased faster than the revenues available to support them. This became less of a problem as organizations such as Sematech and imec took on shared research. Rising costs nevertheless contributed to vendor consolidation, as companies found they could no longer keep up with the investment required for the next generation.

Hutcheson distinguished between “little r, big R, little d, big D,” noting that companies are usually “big D and very little r,” especially when it comes to basic research. “It doesn’t make sense for the companies to invest in Big R because it takes 10 to 20 years for these things to mature,” he said. That’s where universities and national labs come in—and where imec has excelled. “What imec has done is a great job of bridging that gap between the university, between suppliers, and between end customers.”

AI has transformed memory from being cost-driven to being value-driven, Hutcheson said.

Research must still reach the fab

There are two key cost centers at play for memory R&D: One is developing the technology; the other is financing the manufacturing capacity needed to exploit it.

Jim Handy, principal analyst at Objective Analysis, said in an interview with EE Times that vendor consolidation is being driven by the costs of building a competitive fab, which have been rising faster than DRAM revenues.

Regardless, R&D gets its own line in a memory maker’s business model, Handy said. A DRAM manufacturer typically allocates about 20% of revenue to capital spending and about 10% to R&D. “And generally, very few layoffs happen in the R&D team,” he added.

Handy said he sees the Micron announcement partly as a repackaging of spending that would happen anyway into a 10-year commitment. “They’re pretty darn sure that they’re going to be spending that much anyway.”

The budget is also available when big AI players don’t see price as an object, Handy said. This is giving DRAM makers gross margins of 80% versus the 30% or below they are used to. More R&D should also help manufacturers advance nodes faster and squeeze more bits from existing capacity. “That will help lower the price, and that’ll be good for the buyers,” he said.

Handy stressed the value of research organizations such as imec and CEA-Leti, where companies pool their efforts to do “precompetitive” research. DRAM rivals such as Samsung, SK Hynix, and Micron might jointly fund a project to evaluate a new material, for example. If it pans out, they will all go off and develop their own recipes, he said. “That speaks to the incredible cost of doing this R&D.”

That model matters even more for emerging memories such as MRAM, ReRAM, and FRAM, which have spent years—sometimes decades—finding their niches rather than displacing DRAM or flash outright.

Partnerships keep emerging memories in the game

ReRAM maker Weebit Nano has relied heavily on research partnerships to bring its technology to market. CEA-Leti has been a partner from very early on, said Eran Briman, the company’s VP of marketing and business development. Weebit benefited from approximately a decade of research already undertaken by the institute before the two organizations joined forces, he said in an interview with EE Times.

Weebit Nano began working with universities and research institutes in 2018 on neuromorphic computing. In 2019, it demonstrated spiking neural networks running on its ReRAM in collaboration with CEA-Leti. In-memory compute is a nearer-term opportunity, Briman said. Compared with GPUs, CPUs, and accelerators, running AI in memory could theoretically be 10× to 100× more power-efficient, he said.

The company recently launched its Weebit Matrix In-Memory Compute (IMC) Research Program to accelerate the exploration of next-generation computing architectures. The program provides industry and academic researchers with a ReRAM-based hardware development kit to explore and prototype IMC algorithms and systems.

Through the program, researchers can use Weebit Nano’s ReRAM-based platform to develop and test AI algorithms, investigate innovative programming methods, explore new AI and neuromorphic architectures, and assess potential gains in power efficiency and performance. Participating researchers will share their ReRAM-related findings and feedback with the company to help improve the technology and contribute to the broader research community.

Research is where problems get solved, and development is implementation, Briman said. “It took a couple of decades to actually get where we are today,” he said of ReRAM’s maturity.

Commercial availability does not end that work. Customers keep requesting higher densities, greater endurance, and better performance at elevated temperatures. Briman said demands have progressed from 10,000 programming cycles to 100,000 and, for some applications, a million. AI updates, code changes, and data logging are among the drivers.

Briman said semiconductors remain a very expensive industry to play in, noting that some ReRAM vendors have tried and failed multiple times. He sees Micron’s investment as a rising tide. “There’s a ripple effect for the entire industry where we sit, and we’re going to see this coming in the next couple of years.”

Learning from failure

Memory R&D doesn’t always pay off. Micron spent billions on 3D XPoint before DeBoer killed the program earlier than its partner, Intel. “We made a decision that we had learned what we could from it.”

Bringing a new memory to market requires the technology, manufacturability, and customer acceptance, DeBoer said. “If you don’t have all three of those things worked out, it’s just not happening.”

Ultimately, Micron Research Labs reflects the company’s changing role in the semiconductor research ecosystem, DeBoer said, as memory and logic become more tightly integrated. “We intend to transform what Micron is in 10 years,” he said. “Today and going forward, this combination of memory and logic is going to be really the enabler across AI.”


Also read:

AI Triggers a New Memory Super Cycle

AI-Driven Memory Shortage Upends IT Budgets

How AI and Geopolitics Forge a Memory Market Crisis

MEMORY

MICRON TECHNOLOGY

Source: EE Times The copyright of this article belongs to the original author/organization.

The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.