A multi-institutional team of researchers has announced a significant advance in memory technology that could rival the speed of SRAM while offering the benefits of non-volatile storage. The development centers on a new form of Spin-Orbit Torque Magnetic Random-Access Memory (SOT-MRAM) that utilizes a specialized form of tungsten, overcoming a critical barrier to its practical implementation. This breakthrough, detailed by a collaboration including Taiwan’s National Yang Ming Chiao Tung University (NYCU) and the semiconductor foundry TSMC, demonstrates a memory capable of nanosecond switching speeds, low power consumption, and compatibility with existing manufacturing processes.
The new memory technology represents a potential path toward a universal memory that combines the speed necessary for high-performance computing with the ability to retain data when power is off, a feature of storage technologies like flash. At the core of the achievement is a 64-kilobit SOT-MRAM array that exhibits switching speeds of approximately 1 nanosecond and data retention for more than 10 years. SOT-MRAM offers inherent advantages over its predecessor, Spin-Transfer Torque (STT-MRAM), primarily through its use of a three-terminal design that separates the data-writing and data-reading pathways. This architectural difference addresses potential endurance and reliability issues found in two-terminal STT-MRAM systems, making the technology more robust for demanding applications like last-level CPU cache.
Overcoming the Tungsten Stability Problem
The key to the team’s success lies in harnessing the unique properties of tungsten, a heavy metal ideal for generating the powerful spin-orbit torques needed to write data in MRAM. Tungsten can exist in different crystalline structures, or phases, but only one is optimal for this application. The most thermodynamically stable form is the alpha-phase, which unfortunately does not produce a spin-Hall angle large enough for efficient data writing. The superior structure is the metastable beta-phase, which possesses the necessary properties to generate strong spin currents and efficiently flip the magnetic state of the memory cell.
However, the “metastable” nature of beta-tungsten has been a persistent challenge for researchers and engineers. This phase is inherently unstable and tends to revert to the more common alpha-phase, especially when subjected to the high temperatures required in semiconductor manufacturing. Specifically, the back-end-of-line (BEOL) processes in chip fabrication involve thermal budgets that can reach 400°C for extended periods, conditions under which beta-tungsten typically degrades. This thermal instability has long been a roadblock, preventing the integration of high-performance SOT-MRAM into standard CMOS workflows and thus limiting its commercial viability. Finding a way to lock tungsten in its beta-phase under these demanding conditions was the central problem the research team needed to solve.
A Composite Material Innovation
The solution devised by the international team was both elegant and effective: creating a composite material that stabilizes the coveted beta-phase tungsten. Researchers discovered that by inserting very thin layers of cobalt within the tungsten film, they could fortify the material’s structure and maintain its beta-phase integrity even under significant thermal stress. This technique acts as a scaffold at the atomic level, preventing the tungsten from rearranging into its less effective alpha-phase configuration during the high-temperature manufacturing steps. The result is a robust material that retains all the high-performance benefits of beta-tungsten while being fully compatible with industry-standard fabrication temperatures.
To validate their approach, the team subjected the new tungsten-cobalt composite to rigorous testing that simulated the harsh conditions of chip production. The material successfully maintained its phase stability after being heated to 400°C for 10 hours. In a more extreme test, it even withstood temperatures of 700°C for 30 minutes, demonstrating a remarkable degree of thermal resilience far beyond what is typically required for BEOL integration. This innovation directly addresses the primary obstacle that has hampered the development of tungsten-based SOT-MRAM, opening a clear pathway for its use in next-generation electronic devices.
Demonstrated Performance and Future Potential
With a stable material in hand, the researchers fabricated a 64-kilobit SOT-MRAM test array to measure its real-world performance. The results confirmed the technology’s promise, achieving the target 1-nanosecond switching speed, which is competitive with the latency of SRAM, the fastest but volatile memory used for CPU caches. Furthermore, the memory cells exhibited a high tunnelling magnetoresistance (TMR) of 146%, a key metric indicating the strength of the signal difference between a “0” and a “1,” which ensures reliable data reading. Combined with a projected data retention of over a decade, these metrics position the technology as a strong candidate for a new tier of high-speed, persistent memory.
The architectural advantages of SOT-MRAM are central to its appeal. Unlike STT-MRAM, which uses the same current path to both read and write data, SOT-MRAM injects an in-plane current into an adjacent heavy metal layer—in this case, the tungsten composite—to switch the magnetic state of the storage layer above it. Because the read and write paths are separate, the magnetic tunnel junction (MTJ), which is sensitive and prone to wear, is protected from the high currents needed for writing. This separation is expected to lead to significantly higher endurance and reliability, making SOT-MRAM suitable for applications that involve frequent write operations, such as system memory or cache, where STT-MRAM has faced limitations.
A Cross-Institutional Collaboration
This breakthrough was not the work of a single entity but a broad collaboration that brought together expertise from academia and industry. The project was spearheaded by Taiwan’s National Yang Ming Chiao Tung University (NYCU) and included major contributions from TSMC, the world’s leading semiconductor manufacturer. Other key partners included Taiwan’s Industrial Technology Research Institute (ITRI), the National Synchrotron Radiation Research Center (NSRRC), Stanford University in the United States, and National Chung Hsing University. This diverse team provided the mix of fundamental research, materials science, and practical manufacturing knowledge needed to translate a theoretical concept into a functional, scalable technology.
The immediate impact of this work is its potential to disrupt the traditional memory hierarchy. With its combination of speed, endurance, and non-volatility, tungsten-based SOT-MRAM could serve as a high-density replacement for last-level cache in processors, reducing power consumption and eliminating data leakage. Its near-zero leakage and radiation hardness also make it attractive for automotive, industrial, and aerospace applications. Because the technology was designed to be compatible with existing CMOS fabrication lines, the path to commercialization could be significantly shorter and less costly than that of other exotic memory technologies that require entirely new manufacturing ecosystems. This latest advance signals that a new generation of memory, long sought by the industry, is now a tangible prospect.