As artificial intelligence (AI) spreads and memory emerges as the core technology that determines semiconductor performance, Intel, a powerhouse in central processing units (CPUs), is again putting weight behind next-generation memory technology development. Although it exited the memory manufacturing business, including DRAM and NAND flash, in the past, it is expanding investment in consolidation and stacking technologies with memory to boost the performance of compute chips in the AI era.
According to the industry on the 19th, Intel Chief Executive Lipbu Tan said on the recent "TechSurge" podcast that memory is no longer a simple commodity product and that the company is researching next-generation memory architectures. Tan also said the company is considering a method of stacking the CPU and memory directly. However, he did not disclose specific plans for the memory business.
In AI Semiconductor, simply increasing the speed of the compute chip itself has limits on performance gains. As AI models grow, the data to be processed surges, making how quickly memory can supply it the determinant of overall system performance. With the so-called "memory wall" coming to the fore, where memory data transfer speeds cannot keep up with compute speeds, high-bandwidth memory (HBM), which stacks multiple DRAM dies vertically, has become a key component of AI accelerators.
◇ Intel expands investment in next-generation memory technology
The next-generation memory technologies Intel is helping develop include "XBM (Cross-Batch Memory)" and "ZAM (Z-Angle Memory)." Both aim to process large volumes of data faster and more efficiently, but they take different approaches.
XBM is a next-generation memory concept that seeks to improve HBM's complex and costly packaging structure. Currently, HBM places the GPU and memory side by side on an "interposer," a silicon consolidation board, to exchange data. XBM proposes a new structure that reduces such packaging burdens by leveraging high-speed consolidation technologies between chips. However, it remains at the stage where the concept has been disclosed through patents, and no specific productization plans have been announced.
ZAM is a technology that stacks DRAM vertically and densely to increase capacity and data processing speed while reducing power consumption. Intel is conducting joint development with SAIMEMORY, a SoftBank subsidiary, to commercialize ZAM. SAIMEMORY is targeting commercialization in fiscal year 2029 after building prototypes in fiscal year 2027.
It is also notable that Tan mentioned Senior Executive Vice President Lee Seok-hee (former SK hynix CEO), who was brought on in June, while explaining memory technology. Lee oversees advanced packaging, system integration, and back-end process technology development and production at Intel. When Intel hired Lee, it said it would strengthen system integration capabilities that closely combine logic, memory, and networking.
◇ Intel, which started with memory, draws attention for possible business return
Intel grew as a memory semiconductor company by launching the "1103," which opened the early commercial DRAM market in the 1970s. As competition with Japanese firms intensified, it exited the DRAM business in 1985 and focused on CPUs. It continued in the NAND flash business afterward, and jointly developed "3D XPoint" with Micron, releasing "Optane" products. However, after halting the Optane business and selling the NAND business to SK hynix, it is no longer a memory manufacturer that mass-produces DRAM or NAND like Samsung Electronics and SK hynix.
Even after stepping away from memory manufacturing, it continued to research related technologies. In particular, advanced packaging that rapidly consolidates various types of semiconductors within a single package is an area where Intel has invested continuously. Recently, Intel has also positioned 2.5D and 3D stacking for AI Semiconductor and inter-chip consolidation technologies as key foundry strengths.
The spread of AI is increasing the incentive for Intel to turn its attention back to memory. There are limits to merely boosting the compute performance of CPUs or GPUs, and how closely and quickly compute chips and memory are consolidated now determines overall performance and power efficiency. The concepts of XBM, ZAM, and stacking memory directly on top of the CPU are extensions of this shift.
Accordingly, there is talk that Intel could go beyond developing next-generation memory technologies and re-enter the memory business. While Tan has not yet revealed specific business plans, he has publicly expressed interest in related technologies, emphasizing the potential for innovation in memory architecture. Although it remains unclear whether the company will re-enter DRAM manufacturing, some observers say that if Intel broadens its memory business footprint, it could be a variable in the competitive landscape for existing memory firms such as Samsung Electronics and SK hynix.
A semiconductor industry official said, "In AI Semiconductor, the speed of data movement and power efficiency between compute chips and memory are becoming increasingly important," and added, "It may be premature to say Intel will re-enter mass production of commodity DRAM, but it is worth watching whether it will build memory technology into a core business area."