Back to the Roots due to the Memory Crisis? Intel CEO Considers a Return to Memory Production

Related Vendors

dataTec AG

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Lip-Bu Tan no longer sees memory as merely a volume business: In a podcast conversation, the Intel CEO hinted that the company is once again exploring the research and potential production of memory technologies. New architectures, 3D stacking, and memory concepts like ZAM or XBM could open up new opportunities for the corporation.

Intel could become more involved in the memory business again. CEO Lip-Bu Tan stated in a podcast discussion with TechSurge that new memory architectures are one of his particular areas of interest and hinted that the company is working on a closer integration of processors and memory. However, Tan did not outline specific plans for a return to the traditional DRAM or NAND business.

Intel's CEO justified his change of heart with the evolving significance of memory technologies. Previously, he found memory unappealing due to its nature as a largely interchangeable mass-market commodity. The situation has now changed, as numerous new technologies are emerging. The fact that Intel has brought on board Seok-Hee Lee, a former Intel veteran and ex-CEO of SK Hynix, was interpreted by Tan himself as a hint at the direction of his considerations. "A lot of new technologies are coming to market right now. That's why we're currently working on one of my favorite projects, namely some new memory architectures. I think you've just seen the news: I hired my good friend Seok-Hee Lee, who formerly led SK Hynix. So you can imagine what I'm thinking about. But we're not ready to reveal it just yet," said Intel's CEO verbatim.

CPU and Memory are to Move Closer Together

Particularly interesting to Intel is the spatial integration of compute logic and memory. According to Tan, there are several possibilities for stacking CPUs and memory on top of each other. Such an approach could shorten the distances over which data must be transferred between the two components. Especially with AI accelerators, memory bandwidth, energy consumption, and data transfer have become critical factors for the overall system's performance.

These considerations align with a concept developed by Intel called XBM. Unlike conventional High Bandwidth Memory, the architecture is designed to function without the silicon interposer previously required and to utilize UCIe connections, among other things. This would allow for tighter integration of memory and logic. While the concept is still in its early stages, it could long-term help reduce costs and bottlenecks in data transfer within AI systems.

In addition, Intel is working with partners in Japan on Z-Angle Memory, or ZAM. This stacked DRAM architecture is intended to provide a more energy-efficient and cost-effective alternative or complement to HBM, featuring multiple memory chips and a logic controller. The development is supported, among others, by the joint venture Saimemory with SoftBank and other Japanese partners.

Many Questions Remain Unanswered

For Intel, a renewed engagement in the memory sector would also mark a return to its roots. The company, founded in 1968, began as a memory manufacturer but exited the DRAM business in the 1980s due to growing competition from Japan and shifted its focus to microprocessors. Later ventures included NAND flash and Optane. The NAND business was ultimately sold to SK Hynix in 2020, while Intel also discontinued its Optane activities in 2022.

That Intel is once again showing interest in memory technologies does not necessarily mean the establishment of its own large-scale DRAM production. Such a move would require significant investments in factories, manufacturing processes, and research. At the same time, Intel is investing substantial resources in new processors, its own foundry division, and advanced packaging technologies. Therefore, an initial focus on specialized memory architectures that can be closely integrated with CPUs, AI chips, and Intel's manufacturing offerings seems more likely.

Technological hurdles also remain. DRAM stacked directly on high-performance logic chips poses significant challenges for cooling and manufacturing yield. Additionally, HBM already has an established ecosystem, heavily supported by Nvidia's dominant position in AI accelerators. Whether XBM, ZAM, or other Intel developments can turn this into an independent business remains uncertain. Given the increasing demand for high-performance memory chips, Tan's statements suggest that the company is seriously engaging with this technological field.

Despite High Location Costs

Intel Invests an Additional Five Billion Euros ($5.4 Billion) in Irish Chip Factory