Cutting-edge research unveiled at IEEE's International Electron Devices Meeting reveals how material breakthroughs and architectural pivots are addressing semiconductor scaling limits, with 3D NAND reclaiming strategic importance amid AI-driven storage demands.
The semiconductor industry's roadmap took unexpected turns at the IEEE International Electron Devices Meeting (IEDM) 2025, where researchers presented advances poised to redefine chip manufacturing economics. Three developments dominated discussions: the unexpected resurgence of 3D NAND technology, progress toward replacing copper interconnects, and 2D materials demonstrating silicon-competitive performance.
![]()
3D NAND's Unexpected Renaissance After years of focus on logic scaling, 3D NAND flash memory has regained prominence due to exploding data storage needs from AI training clusters and hyperscale data centers. Samsung and SK Hynix presented stacked designs exceeding 500 layers, achieving densities of 50+ Gb/mm²—a 40% improvement over 2024 production nodes. Crucially, new etching techniques reduced production costs by approximately 18% per layer through multi-deck bonding, making high-layer counts economically viable. This reverses the trend toward slower NAND scaling, with analysts projecting the 3D NAND market to grow from $38B in 2024 to $62B by 2028, outpacing overall memory growth.
Copper's Successors Emerge With copper interconnects struggling at sub-2nm dimensions due to resistance and electromigration, IEDM showcased two alternatives entering pre-production. Intel demonstrated ruthenium interconnects with 30% lower resistance than copper at identical line widths, enabling clock speed improvements up to 15% in test chips. Simultaneously, TSMC revealed molybdenum-based interconnects paired with atomic-layer-deposited barriers, reducing via resistance variability by 50%. Both materials enable continued scaling toward 10 angstrom nodes, potentially delaying the industry's transition to more costly backside power delivery networks.
2D Materials: Beyond Experimental Curiosity Transition metal dichalcogenides (TMDs) like tungsten diselenide moved closer to commercialization, with IMEC presenting monolayer transistors achieving electron mobility of 340 cm²/V·s—surpassing silicon's 240 cm²/V·s at equivalent nodes. GlobalFoundries showcased wafer-scale TMD integration with 300mm silicon wafers, addressing previous manufacturability concerns. These materials now target specific applications: IBM's graphene-silicon carbide RF switches reduced power consumption by 65% in 6G prototypes, while Samsung's molybdenum disulfide sensors achieved sub-ppm detection for medical wearables.
Strategic Implications These advances collectively extend Moore's Law economics through heterogeneous integration rather than monolithic scaling. The 3D NAND revival allows memory manufacturers like Micron and Kioxia to amortize factory investments over longer lifetimes, potentially boosting margins by 5-7 percentage points. Interconnect innovations could reduce reliance on EUV double patterning, cutting logic chip production costs by up to 12% at advanced nodes. Meanwhile, 2D materials enable specialized accelerators—projected to capture 19% of the AI chip market by 2030—without requiring full process overhauls.
For equipment suppliers, Applied Materials and Lam Research stand to benefit from new deposition/etch tool demand, while material science firms like Merck KGaA accelerate metal precursor development. The timeline remains aggressive: expect copper alternatives in high-volume manufacturing by late 2027, volume production of 2D-enhanced chips by 2028, and 600+ layer 3D NAND by 2029. As TSMC's CTO noted in a closed-door session, 'The post-silicon era won't be a revolution—it will be a mosaic of optimized technologies.'
Comments
Please log in or register to join the discussion