2023 MRAM Forum Shows MRAM For Automotive And Other Applications

Road Trip At the 2023 MRAM Forum, an IEEE Magnetics Society sponsored event tied to the 2023 International Electron Devices Meeting (IEDM) in San Francisco there were some good insights into the evolution of MRAM (and other non-volatile) solid state memories and their applications by the major players in the MRAM market. There were talks by NXP, Samsung, Global Foundries, Netsol, Everspin, Numem, Kioxia, TSMC, YODA-S and UCLA (as well as Everspin and Numem). Thomas Jew from NXP, Johannes Muller from Global Foundries and Junghoon Bak from Samsung spoke about MRAM for automotive applications.

This promises to be an important growth market for embedded MRAM and enable moving to domain and zone-based automotive architectures, controlled by software as shown below. Thomas Jew said that automotive designs are moving from domain-based, where electronics is focused on particular discrete functions to one where software and distributed intelligence are enabling more flexible and capable automotive electronics and creating architectures optimized for software defined vehicles. Automotive Network Architecture Evolution Daniel Worledge, IBM Distinguished Research Scientist, one of the organizers of the MRAM Forum told me that “Embedded STT-MRAM has successfully replaced embedded Flash at all advanced foundry nodes.

Over the next few years we will see the automotive MCU/SoC suppliers switch to using eMRAM – in about three years, almost every new car will have eMRAM inside. ” Large system MRAM in a car can be organized into independent partitions for concurrent access by different CPUs and computational threads as shown below from the NXP talk. By using MRAM for code and data storage applications software development is simplified.

Compared to the complexities and additional operations needed to manage NOR flash, MRAM simplifies non-volatile storage and uses less energy. MRAM Automotive App Partition, NXP Over the air updates (OTA) are also simplified using MRAM and MRAM’s fast write minimizes downtime and facilitates software roll-backs if required. The need for smaller features than NOR can provide (28nm is minimum embedded NOR feature size) and the movement of automotive electronics to 16nm nodes are also opening up opportunities for MRAM introduction.

Currently 22nm MRAM are qualified and running in production, but smaller feature products are on the way. 64MB MRAM is hard with flash but MRAM could make that size memory possible and thus enable more capable automotives. However, magnetic immunity can be an issue and needs to be mitigated for MRAM to find broader applications.

Other speakers and in a final panel indicated that there are solutions to magnetic interference in MRAM devices, e. g. fitbits with Ambiq chip MRAM have magnetic clasps and MRAM are used as readers in hard disk drives.

NXP announced earlier this year that, working with TSMC, they will be introducing 16nm in automotive controllers with samples available in 2025. Johannes Muller from Global Foundries discussed the company’s activities in bringing MRAM products to market for the automotive industry. In his talk he discussed various eFlash code storage options (down to 28nm) as well as 22nm (in production) and 12nm (in development) SST MRAM as well as SOT (research) MRAM as well as 22nm and 12nm ReRAM (both in development).

The figure below compares eFlash (NOR flash) for code storage to MRAM. Comparison of eFlash and eMRAM Junghoon Bak, Principal Engineer at Samsung spoke about the status and outlook of 14nm and smaller MRAM technology. Samsung has been involved with MRAM research and development since 2002.

Samsung has been supplying eFlash type MRAM since 2019 (e. g. for a global navigation satellite system) and for Netsol (who also spoke at the MRAM Forum) since 2022 for NVRAM applications as stand-alone chips.

Samsung said that they had successfully implemented 14nm MRAM into a 14nm FinFET baseline product. Junghoon said that MRAM demonstrated the highest level of cost-effectiveness and high performance in FinFET architectures. The image below shows Samsung eMRAM product plans for the next few years.

28nm FDSOI eMRAM is available today and it plans to introduce 14nm MRAM by the end of 2024 with some sampling of smaller devices in 2025 and 2026, particularly for automotive applications. Samsung MRAM Product Plans Automotive applications must also support higher temperatures for solder reflow and higher temperature operations. Samsung is developing its device, process and manufacturing technology to improve the performance and endurance of its MRAM products.

Yong Hwan Noh from Netsol described the serial and parallel MRAM products they make at Samsung’s foundry, shown below. NetSol Serial and Parallel MRAM Netsol thinks that although currently MRAM is using existing memory interfaces, new memory interfaces will be needed for high performance standalone non-volatile memories. Netsol’s roadmap has 14nm MRAM products introduced in late 2025 or 2026.

Masahiko Nakayama is a researcher at Kioxia and he spoke about accelerated ST-Switching and high retention in a 14nm MTJ. Kioxia as well as other companies before them have been working on MRAM in a cross-point memory structure. They say that compared to PCM-based storage class memories that MRAM is 10X faster at writing and 100X higher endurance and that it uses considerably less energy.

Kioxia also spoke about an accelerated magnetic tunnel junction (MTJ) with a special storage layer that they believe enables higher speed writing as shown below (this was in a paper presented at the 2023 IEDM). Kioxia Accelerated STT MRAM Kioxia said that devices made using this technology showed more than 10 year retention at 90 C and high-speed writing with 5nm pulses. MingYuan Song is a technical manager at TSMC and he spoke about SOT-MRAM with 1ns speed and 10-year data retention.

In particular he spoke about spin orbit torque (SOT) MRAM used to create matrix vector multiplication as shown below. TSMC SOT MRAM for Matrix Vector Multiplication Devices using an array of such devices was then used in black and white image edge detection. The device operated at less than 1.

5V with up to M-Ohm tunable resistance, 1ns operation and 7 times 10 to the 12 endurance. Hiroaki Yoda, CEO and CTO as YODA_S (and formerly at Toshiba/Kioxia) spoke about a low current SOT-MRAM called SIMA-MTJ. SIMA stands for strain-induced magnetic anisotropy.

In this device strain is applied to an in-plane spin-Hall electrode (SHE). He believes that shape anisotropy can be used to create low power writing MRAM with unlimited endurance. Dr.

Kang Wang, Distinguished Professor and Raytheon Chair Professor in Physical Science and Electronics at UCLA has been working on voltage controlled magnetic MRAM devices for several years and in this talk he spoke about stochastic applications of this technology. In particular he spoke about using MRAM for neural memory technology for AI applications. Interfacial and crystal anisotropy can interact in a way that allows voltage control of a magnetic memory.

This is referred to as MeRAM. This technology could allow much smaller tunnel junction areas and writing energy. Dr.

Wang went into many different voltage-controlled magnetic memory approaches and his lab worked with TSMC on a 180nm demonstration device. In the last part of his talk he spoke about MeRAM as a p-bit in probabilistic computing. In a demo of MeRAM in a diffusion model using a neural network much higher throughput was achieved in a smaller die space than with more conventional methods.

The 2023 IEEE MRAM Forum, following the IEDM conference showed significant progress in both standalone and embedded MRAM. NXP, GlobalFoundries and Samsung showed significant advances in automotive applications that will enable software defined vehicles. TSMC, YODA-S and UCLA showed advances in SOT and VCMA MRAM.

.