MRAM, sponsored by Taiwan Semiconductor Manufacturing Co., Ltd. ADR(TSM.US), makes a significant breakthrough.

date
09:01 18/10/2025
avatar
GMT Eight
With the rise of emerging applications such as artificial intelligence, autonomous driving, and the Internet of Things, traditional storage systems are facing multiple challenges in terms of speed, energy consumption, and stability.
In recent years, the non-volatile memory (NVM) technology has been experiencing rapid development. With the emergence of new applications such as artificial intelligence, autonomous driving, and the Internet of Things, the traditional storage system is facing multiple challenges in terms of speed, energy consumption, and stability. In order to balance "speed", "cost-effectiveness", and "reliability", various new types of storage devices (such as ReRAM, PCM, FeRAM, MRAM, etc.) have entered the research and validation stages, attempting to stand out in the "post-DRAM era". Against this backdrop, Magnetic Random Access Memory (MRAM) is considered one of the most promising universal storage solutions due to its high speed, low power consumption, and non-volatile characteristics. According to reports, a multinational research team from National Yang Ming Chiao Tung University in Taiwan, TSMC, and the Industrial Technology Research Institute has made a significant breakthrough in MRAM technology. They have successfully developed a Spin Orbit Torque Magnetic Random Access Memory (SOT-MRAM) based on -phase tungsten material, achieving remarkable performance metrics: data switching in just 1 nanosecond, data retention time exceeding 10 years, and a tunneling magnetoresistance ratio as high as 146%. This achievement, published in the journal Nature Electronics, paves the way for industrial applications of next-generation high-speed, low-power storage technology. Changing needs in storage technology Current computing systems rely on a storage hierarchy consisting of SRAM, DRAM, and flash memory. However, with technology nodes breaking through the 10-nanometer barrier, these traditional charge-based storage technologies are facing serious challenges: limited scalability, difficulty in performance improvements, exacerbated read/write interference issues, and reduced reliability. Especially in today's rapidly developing artificial intelligence and edge computing environments, higher demands are placed on storage it must have the high-speed response capabilities of DRAM, the non-volatile characteristics of flash memory, and significantly reduced power consumption. In this context, emerging non-volatile storage technologies have emerged. In addition to SOT-MRAM, these include Spin-Transfer Torque Magnetic Random Access Memory (STT-MRAM), Phase Change Memory (PCM), Resistive Random Access Memory (RRAM), and Ferroelectric Random Access Memory (FeRAM), among others. These technologies all have characteristics of non-volatility, low latency, low power consumption, and can be integrated with existing CMOS semiconductor processes, providing possibilities for developing new computing architectures. In comparison, the latency of DRAM is about 14 milliseconds, the read latency of 3D TLC NAND is between 50 and 100 microseconds, and the switching speed of the new SOT-MRAM reaches the nanosecond level, almost comparable to SRAM, while still retaining the advantage of non-volatility meaning that data will not be lost even if there is a power failure. Unique advantages of SOT-MRAM The reason why SOT-MRAM is attracting attention lies in its unique working principle and technical advantages. It utilizes materials with strong spin-orbit coupling to generate Spin Orbit Torque (SOT), enabling the magnetization reversal of nano-magnetic bodies within magnetic tunnel junctions, thereby achieving data writing and erasing. Compared to other storage technologies, SOT-MRAM has three main core advantages: High-speed writing: Through the spin-orbit torque effect, magnetization reversal can be completed in nanoseconds, which is much faster than traditional magnetic field-driven methods. High energy efficiency: The three-terminal structure design completely separates the read/write current paths, effectively solving the durability issues faced by STT-MRAM and the resistance limits of magnetic tunnel junctions, significantly reducing energy consumption. High reliability: Since the read/write operations are independent of each other, the durability of the device is greatly increased, withstanding more read/write cycles, while maintaining excellent long-term data retention capabilities. It is these advantages that make SOT-MRAM poised to replace high-speed cache-level SRAM and become the core storage component of the next-generation computing systems. Overcoming key technical challenges Although the theoretical advantages of SOT-MRAM are evident, in order to achieve industrial application, a key technological bottleneck must be addressed: the thermal stability issue of spin-orbit coupling materials. Tungsten, due to its strong spin-orbit coupling characteristics, is an ideal candidate material for SOT-MRAM. In particular, stable tungsten in the A15 structure (-phase) has a spin Hall angle of -0.4 to -0.6, demonstrating excellent spin-orbit torque efficiency. However, -phase tungsten is a metastable state, and under common heat treatment conditions in semiconductor manufacturing processes (typically at 400C for several hours), it will transform into the thermodynamically stable -phase tungsten. This phase transition is fatal the spin Hall angle of -phase tungsten is only about -0.01, leading to a significant decrease in spin-orbit torque reversal efficiency and severe degradation of device performance. The breakthrough solution proposed by the research team is to insert a ultra-thin cobalt layer into the tungsten layer, forming a composite structure. Specifically, they divided the 6.6-nanometer-thick tungsten layer into four segments, inserting a only 0.14-nanometer-thick cobalt layer between each segment this thickness is less than a single atomic layer of cobalt, resulting in a discontinuous distribution of cobalt. This intricate design serves two purposes: the cobalt layer acts as a diffusion barrier, suppressing atomic diffusion within the tungsten layer; the mixing effect between cobalt and tungsten consumes thermal budget, delaying the phase transition. Exciting experimental verification results: this composite tungsten structure can maintain phase stability at 400C for up to 10 hours, and can even withstand high temperatures of 700C for 30 minutes, while traditional single-layer tungsten undergoes phase transition after annealing for only 10 minutes at 400C. Through transmission electron microscopy, X-ray diffraction, and nano-diffraction tests at the Taiwan Photon Source, the researchers confirmed the stability of -phase tungsten. More importantly, this composite structure not only solves the thermal stability problem, but also maintains excellent spin conversion efficiency. Through spin torque ferromagnetic resonance and harmonic Hall resistance measurements, the team measured the spin Hall conductivity of the composite tungsten film to be approximately 4500 cm, and the damping-like torque efficiency to be about 0.61, ensuring efficient magnetization reversal performance. Comprehensive performance validation A breakthrough in theory can only truly be realized through device validation. Based on the composite tungsten film solution, the research team successfully fabricated a 64-kilobit SOT-MRAM prototype array and completed comprehensive performance testing and validation under conditions close to practical applications. In terms of switching speed, the device achieved a spin-orbit torque reversal speed in the nanosecond range, with performance almost comparable to SRAM, far exceeding DRAM and flash memory. Statistical tests on 8000 devices show highly consistent flipping behavior, with an intrinsic flipping current density of only 34.1 megaamps per square centimeter under long pulses (10 nanoseconds), demonstrating excellent stability and repeatability. Data retention capabilities are also outstanding. According to cumulative distribution function (CDF) estimates, the device's thermal stability parameter is approximately 116, indicating that its data retention time can exceed 10 years, fully meeting the stringent requirements of non-volatile storage. In tunneling magnetoresistance (TMR) tests, the device achieved a high TMR value of up to 146%, indicating that a high-quality interface was formed between MgO and CoFeB, providing a strong guarantee for stable readout margins and reliable process windows. In terms of energy consumption control, the three-terminal structure design achieves completely independent read/write operations, fundamentally reducing energy consumption, making it particularly suitable for energy-sensitive applications such as edge computing and mobile terminals. Furthermore, thanks to the participation of the TSMC research team, the entire design was optimized for existing semiconductor backend processes from the beginning, ensuring excellent process compatibility and paving the way for future large-scale production. It is worth noting that the research team also achieved X-type flipping without the need for an external magnetic field. This achievement is due to the symmetry-breaking effect in the composite tungsten material, which not only further simplifies the device structure, but also enhances integration and design flexibility, opening up new directions for the engineering application of SOT-MRAM. Opening a new era of storage technology The significance of this research goes far beyond a technical breakthrough in the laboratory; it points to a new direction for the development of the entire storage industry. Unlike many new storage technologies that are still in the conceptual verification stage, SOT-MRAM based on composite tungsten has considered process compatibility and manufacturability from the design stage. The research team has successfully fabricated a 64-kilobit array and plans to further expand it to the megabit (Mb) level of integration, while reducing write energy consumption to sub-picojoule levels. In artificial intelligence and edge computing scenarios, SOT-MRAM also demonstrates unique advantages. High-frequency data access in AI training and inference processes is a major source of energy consumption, and SOT-MRAM, with its high speed, non-volatility, and low power consumption characteristics, can serve as on-chip cache for AI accelerators, significantly reducing system energy consumption. In edge devices, its non-volatility means that devices can be quickly powered on and off without losing data, which is particularly beneficial for battery-powered IoT terminals. At the same time, the emergence of SOT-MRAM may lead to a restructuring of the storage hierarchy system. The traditional "SRAM cacheDRAM main memoryflash external memory" three-tier architecture may undergo a transformation, as SOT-MRAM is expected to fill the performance gap between SRAM and DRAM, and even replace one of them in certain applications, simplifying the architecture and improving system efficiency. On the materials science front, the strategy of "composite layer stabilizing metastable phase" proposed in the research is not only applicable to tungsten, but also provides a new perspective for studying the phase stability of other functional materials. The team plans to further explore new oxide and two-dimensional interface materials to enhance overall performance and reliability. More profoundly, this breakthrough may drive innovation in computing architecture. The high-speed, low-power SOT-MRAM makes new architectures such as "In-Memory Computing" more feasible, providing a new path to break through the traditional von Neumann architecture's "memory wall" bottleneck. Conclusion Currently, based on composite tungsten SOT-MRAM, through clever material design, the thermal stability challenge of -phase tungsten has been solved, achieving a perfect combination of nanosecond switching and ultra-long data retention. This is not just an academic achievement, but also a core technical reserve for the next generation of computing systems. For the research team, their goal is not only to demonstrate excellent laboratory performance, but also to show through system-level verification how MRAM can significantly reduce overall power consumption in real-world applications, driving technological innovation in AI, edge computing, and mobile devices. As the integration progresses from kilobits to megabits, there is reason to expect that this new type of memory will soon enter our smart devices, ushering in a new era of storage technology. This article is selected from the "Semiconductor Industry Observation" WeChat public account, author: Yi-Qi Shao; GMTEight editor: Yu-Cheng He.