MRAM applications

Everspin Technologies announced that it has signed a memorandum of understanding (MOU) with MaxLinear, Inc., a provider of high-performance connectivity and storage acceleration solutions, to evaluate next-generation memory architectures designed to improve efficiency and performance in AI servers.

Everspin says that as AI models and inference workloads continue to grow, increased demands are driving interest in new approaches that combine persistent memory, acceleration and compression to improve performance, power efficiency and infrastructure utilization.

Custom AI ASIC solutions provider SEMIFIVE, together with China-based AI semiconductor developer ICY Tech, announced the successful tape-out of its next-generation Edge AI SoC jointly developed utilizing Samsung Foundry's 8nm (8LPU) embedded MRAM technology. The two companies say that this is a significant milestone toward the first commercial deployment of 8nm eMRAM technology in Asia.

SEMIFIVE that by integrating eMRAM into its Edge AI accelerator, it aims to reinforce its technical leadership in the ultra-low-power, high-performance inference market. 

Vertical Compute, MRAM for AI technology developer, raised $67 million. This round was led by Quantonation and included many investors, including Flanders Future Techfund (PMV), Wallonie Entreprendre, Sambrinvest, Noshaq, invest.bw, Drysdale Ventures, Kima Ventures, Eurazeo, XAnge, VCT Vector Gestion SA, imec.xpand, imec, VLAIO - Flanders Innovation & Entrepreneurship, and Bpifrance.

Verticle Compute was spun off imec, and develops technology to stack memory directly above compute logic. In 2025, the company said that it develops MRAM in-memory computing chiplet technology. The chiplet, according to the company, can reduce power consumption by 80% and speed up the execution of large language models for AI by 100X. It is believed that the company's MRAM is based on SOT-MRAM technology.

Everspin Technologies announced its newest high-reliability additions to the PERSYST MRAM product line: the EM064LX HR and EM128LX HR devices.  These two new xSPI MRAM products are designed for extreme operating environments, with high endurance, temperature performance and data retention. Everspin is targeting aerospace, defense, automotive and high-end industrial applications for these new chips. 

Everspin's new MRAM chips have received an AEC-Q100 Grade 1 qualification for operation from -40°C to +125°C. Each device undergoes a 48-hour burn-in process and provides 10-year data retention at 125°C, ensuring predictable performance even under the most demanding conditions. With 64- and 128-megabit densities achieving 90 Mbytes/sec read and write bandwidth, sustained for over a decade, the EM064LX HR and EM128LX HR are built for systems that cannot risk data loss or degradation.

Frontgrade Technologies announced that it has expanded its MRAM product line, adding two new series, the Dual Quad Serial Peripheral Interface (Dual QSPI) series and the Parallel series. With these new MRAM chips, Frontgrade is targeting aerospace, defense, and industrial applications.

The Dual QSPI series, powered by STT-MRAM, offers a wide range of densities, 128Mb and 1/2/4/8Gb. These new devices bring flash-like non-volatility with SRAM-level speed and endurance. The Dual QSPI interface enables high-speed, synchronous communication and seamless integration into new or existing architectures. These memory chips are designed for boot, configuration, code execution, and data logging.

TSMC is launching its first design center in Europe, with a focus on memory technologies for automotive applications. The company announced that it will start developing 5nm MRAM technologies, to complement its 22-nm (in production), 16-nm (ready for customers verification) and 12-nm (already in development) technologies. 

Scaling MRAM technology to 5nm will be a significant step, required, according to TSMC, to implement AI automotive chips and technologies. The company will also develop 6 nm RRAM memory in the new design center.

Researchers from Beihang University and Truth Memory Corporation have fabricated a 1 Kbit SOT-MRAM chip using a 180 nm complementary metal oxide semiconductor (CMOS) process, and implemented a physical unclonable function (PUF) on it. This research represents a significant step forward in the field of PUFs.

The newly developed SOT-MRAM sr-PUF achieves a strong, highly reliable, and reconfigurable PUF that can resist machine-learning attacks. The performance of the SOT-MRAM sr-PUF is remarkable - it has a challenge-response pair (CRP) capacity of 109. Its uniformity is 50.07%, diffuseness is 50%, uniqueness is 49.89%, and the bit error rate is 0%, even in a 375 K environment. These values are near-ideal, indicating excellent randomness and reliability.

Everspin Technologies announced new additiontions to its PERSYST EMxxLX family, the EM064LX HR and EM128LX HR, targeting automotive applications. These new chips will operate at temperatures from -40°C to +125°C, meeting the AEC-Q100 Grade 1 standard.

The new chips offer capacities of 64Mb and 128Mb, Quad SPI Interface, fast write speeds and a simple architecture. Everspin will start offering engineering samples in June 2025, with mass production scheduled by the end of the year.

High-performance STT-MRAM developer Numem announced that it expects to start samplings its MRAM chipsets, to enable non-volatile, high-speed and ultra-low power solutions to chip designers.

Numem says that the company's chiplets deliver up to 4TB/sper 8-die memory stack, exceeding existing AI memoryHBM solutions.  It supports 4GB per stack package, and is optimized for AI applications across OEMs and hyperscalers.

Vertical Compute, a new startup that was spun-off from the imec institute to commercialize its MRAM technology for AI applications, has raised 20 million Euro. The seed round was led by imec.xpand and supported by Eurazeo, XAnge, Vector Gestion, and imec.

The company says that it will develop an MRAM in-memory computing chiplet technology. The chiplet, according to the company, can reduce power consumption by 80% and speed up the execution of large language models for AI by 100X. It is believed that the company's MRAM is based on SOT-MRAM technology.