At a time when traditional manufacturing processes are approaching their physical limits, IBM (IBM.US) has advanced chip node technology to 0.7 nanometers (7 Angstroms) using a three-dimensional vertical stacking architecture, heralding the semiconductor industry's official entry into the atomic scale era.
On June 25, IBM announced a milestone breakthrough in semiconductor technology, introducing the world's first "sub-1nm" chip fabrication technology. This revolutionary technology employs a novel "three-dimensional vertical stacking" (NanoStack) transistor architecture, directly pushing the process node to 0.7 nanometers, or 7 Angstroms.
Buoyed by this news, IBM's U.S. stock surged over 6% in pre-market trading. As of the latest update, the pre-market gain had moderated to 3.44%, with the stock trading at $272. This rebound recovers a significant portion of the roughly 11.2% decline IBM shares had experienced year-to-date prior to the announcement.
Technical Breakthrough: From "Nanosheets" to "NanoStack"
NanoStack is a completely new transistor architecture developed by IBM researchers, representing the industry's first known three-dimensional nanosheet stacking design. Unlike traditional chips that primarily arrange transistors in a planar fashion, NanoStack vertically stacks and interleaves transistors, utilizing three-dimensional sequential integration to pack more transistors into the same area.
The core breakthroughs of this architecture include: Doubled Density: Integrating nearly 100 billion transistors on a chip the size of a fingernail, achieving approximately double the density of the 2-nanometer chip IBM released in 2021. Performance Leap: Compared to 2-nanometer node chips, the new technology is expected to deliver up to a 50% performance improvement or a 70% reduction in energy consumption. SRAM Advancement: Research presented at the VLSI 2026 conference shows the NanoStack architecture enables a 40% area scaling for SRAM (Static Random-Access Memory) – a significant development for on-chip caches, which have long been difficult to shrink, particularly aligning with the high-bandwidth data demands of AI workloads.
The NanoStack design also allows for different material combinations in each stack layer, enabling independent optimization of performance and power consumption per transistor. IBM has completed experimental validation in CMOS integration using ultra-thin dielectric bonding and has demonstrated functional operation of CMOS inverters with expected switching performance, proving the technology is manufacturable and supports real computation.
Jay Gambetta, Director of IBM Research and an IBM Fellow, stated: "With the new NanoStack architecture, we are not just shrinking transistor dimensions; we are fundamentally redefining how chips are built to achieve greater performance and higher energy efficiency."
"The 0.7nm node is no longer an exact description of a physical dimension but a designation for a technology generation," noted an industry analyst, pointing out the chip industry has long moved past the era where "node" equaled "feature size." The true significance of IBM's breakthrough lies in NanoStack making logic chip processes viable for the first time below the 1-nanometer node, transitioning chip scaling from the nanometer to the Angstrom level and into the atomic-scale domain.
IBM stated this breakthrough "lays the foundation for the next era of computing." For the semiconductor industry, which faces physical bottlenecks with traditional processes, this provides a new path to continue performance leaps.
AI Computing Power Revolution: Training Time Slashed from Months to Weeks
In another blog post, IBM pointed out that current mainstream AI accelerators generate approximately 1500 TOPS (trillions of operations per second), while accelerators using the 7-Angstrom technology are expected to generate about 7000 TOPS, roughly 4.7 times higher.
If 7-Angstrom chips were used to train today's massive frontier large language models (LLMs), training time could be dramatically reduced from about three months to a matter of weeks. IBM stated the new technology will provide significantly stronger computing power support for generative AI, cloud infrastructure, and next-generation electronic devices.
Against the backdrop of the entire industry grappling with an AI computing power gap and an energy crisis, IBM's technology is a shot in the arm for major model developers. Currently, the hottest AI accelerator chips on the market, such as Nvidia's Blackwell series and various custom ASICs, generally offer single-chip computing power hovering around 1500 TOPS.
Gambetta added: "With the new NanoStack architecture, we are not just making smaller transistors; we are reinventing how chips are made. This provides a definitive physical solution for the computing power surge and high energy costs of the AI era."
Production Timeline and Industry Ecosystem
IBM anticipates the NanoStack technology could enter mass production within the next five years and support semiconductor process development for at least the next decade. Notably, IBM itself does not own a wafer fab; its business model is closer to architecture licensing – IBM designs the transistor architecture, and manufacturing partners handle production. IBM has previously licensed chip technology to Samsung and Japan's Rapidus. However, as of now, IBM has not announced specific manufacturing partners for this technology.
Experimental validation of the technology has been completed at IBM's research facility in Albany, New York. This facility will introduce High Numerical Aperture Extreme Ultraviolet (High-NA EUV) lithography equipment developed by ASML. IBM is currently advancing related process development with partners including Lam Research, Tokyo Electron, and SCREEN Semiconductor Solutions.
Furthermore, IBM recently announced plans to establish Andeon as an independently operated company within its portfolio, aiming to become the world's first foundry dedicated exclusively to quantum wafer manufacturing. This move indicates IBM is pushing forward on both the classical and quantum computing frontiers.
Competitive Landscape: The "Angstrom Race" with TSMC, Intel, and Samsung
Although IBM has again taken the lead in cutting-edge semiconductor R&D, it is important to note that IBM itself has long exited the front lines of wafer manufacturing, transforming into a pure technology research and licensing giant. The true heavyweight in the global chip foundry arena is Taiwan Semiconductor Manufacturing Company (TSM.US).
Currently, TSMC has begun mass production of its 2nm chip technology using first-generation nanosheet transistors. The company is developing A16 (1.6nm) and A14 (1.4nm) technologies. The A16 process technology will incorporate Super Power Rail (SPR) and, originally planned for mass production in the second half of 2026, has been delayed to 2027. The A14 process will use TSMC's second-generation nanosheet transistor structure, with development reportedly on schedule.
Samsung has announced its A16 Angstrom-level CMOS technology will enter mass production in the fourth quarter of 2026. Intel, last week, announced its next-generation 18A manufacturing process (for 1.8nm chips) has entered the risk production phase, the testing stage before commercial mass production.
Analysts point out that while IBM's NanoStack architecture leads on a technical level, a significant gap remains between lab validation and large-scale mass production – competitors already possess mature manufacturing capabilities and customer relationships.
Nevertheless, the significance of IBM's 0.7nm technological breakthrough extends far beyond a single company's achievement. In the context of traditional chip scaling facing physical limits, IBM has demonstrated that continued performance and efficiency improvements remain possible even as feature sizes approach atomic scales.
The NanoStack architecture extends logic technology for the first time below the 1-nanometer node, opening the "Angstrom-scale scaling" era. For the global semiconductor industry, this breakthrough provides a new pathway to sustain performance leaps. In today's world of exponentially growing AI computing demands, IBM's 0.7nm technology is not just an engineering feat; it is a declaration that Moore's Law is not yet dead.
Comments