The highly praised MRAM is moving towards MCUs.

As embedded flash (eFlash) reaches its limit at the 28nm node, restricting the further advancement of MCU process technology, manufacturers are turning their attention to new types of storage, including magnetic random - access memory (MRAM/STT - MRAM/SOT - MRAM), phase - change memory (PCM/PCRAM), resistive random - access memory (RRAM/ReRAM), and ferroelectric random - access memory (FRAM/FeRAM).

Among them, MRAM has always been highly sought after in the industry. Tech giants such as Huawei, TSMC, Samsung, Intel, and Synopsys have all ventured into the MRAM field. Coincidentally, two years ago, Tom Coughlin from Coughlin Associates and Jim Handy from Objective Analysis praised MRAM and were optimistic about its prospects in a report. They cited the rich variety of MRAM types, broad application prospects, and obvious comprehensive advantages as reasons.

As early as 2018, there was a saying in the industry: "If MRAM enters MCUs, will there be no flash memory below 28nm?" In the past two years, MRAM has lived up to expectations. Manufacturers have successively released MCUs with MRAM.

The Storage Revolution of MCUs

For MCUs, embedded flash (eFlash) memory is a commonly used built - in non - volatile memory (NVM). This memory space is undergoing rapid changes due to the increase in on - chip memory capacity and the limitations of eFlash below 28nm.

To enable MCUs to break through process limitations and accelerate the data transfer speed of MCU NVMs, major manufacturers are choosing different paths to overcome the challenges in MCU process technology.

Block Diagram of MCU Design

Infineon has chosen resistive random - access memory (RRAM) and launched the AURIX TC4x series of MCUs using TSMC's 28nm process.

STMicroelectronics (ST) has opted for phase - change memory (PCM) and introduced the xMemory Stellar series of MCUs using Samsung's 28nm FD - SOI ePCM process. It is also promoting an upgrade to the 18nm FD - SOI process.

Texas Instruments (TI) has selected ferroelectric random - access memory (FRAM), and its MSP430 series has products related to FRAM.

Renesas and NXP have chosen magnetic random - access memory (MRAM). NXP has launched the world's first regional controller, the 16nm FinFET + MRAM MCU S32K5. Renesas has introduced the RA8P1, RA8T2, RA8M2, and RA8D2 using TSMC's 22nm ULL process for MRAM.

There is no question of one being stronger than the other. After all, no single storage technology is a "perfect all - rounder." Each new type of storage has its own advantages.

However, since the era of magnetic floppy disks and tapes, magnetic storage has penetrated into our lives. MRAM is more like an "all - rounder" that can handle a variety of tasks, which is why it is widely favored.

Understanding eMRAM

How powerful is MRAM? It has a speed and area performance between SRAM and DRAM, two types of volatile storage technologies. At the same time, it features unlimited read/write cycles, fast write speed, low power consumption, small area, low leakage, high capacity, high radiation resistance, and high integration with logic chips. Currently, MRAM in the laboratory can withstand temperatures from - 40°C to 150°C, covering the - 40°C to 120°C range required for automotive - grade chips.

In terms of principle, different from traditional RAM, MRAM does not store data using charge or current. Instead, it consists of ferromagnetic and non - magnetic materials based on spin - electron properties, forming a magnetic tunnel junction (MTJ). Even when the power is turned off, the MTJ can maintain its polarization and retain the stored data. Currently, there are various structures of MTJ, which is where the complexity of MRAM lies.

Different types of MTJ, Image source: Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences

MRAM is divided into three generations: The first generation is MRAM, known as field - driven MRAM; the second generation is STT - RAM (spin - transfer torque), which flips the magnetic moment by passing a current perpendicular to the tunnel junction; the third - generation MRAM technology is divided into two types. One is spin - orbit torque MRAM (SOT - MRAM), which flips the magnetic moment by passing an in - plane current through a heavy - metal layer, and the other is voltage - controlled magnetic anisotropy MRAM (VCMA - MRAM or MeRAM), which flips the magnetic moment by applying a voltage to change the magnetic anisotropy.

Currently, the second - generation STT - MRAM dominates the market. STT - MRAM can achieve a good balance in terms of speed, area, write cycles, and power consumption. The most common internal combination in today's ordinary STT - MRAM memories is 1T - 1MTJ (one transistor, one magnetic tunnel junction), which has advantages such as small area, low manufacturing cost, and good compatibility with CMOS processes.

Recently, TSMC has also overcome the key issues in the industrialization of the third - generation SOT - MRAM. Although SOT - MRAM has obvious theoretical advantages, to achieve industrial application, a key technical bottleneck must be solved: the thermal stability of spin - orbit coupling materials. The research team's breakthrough solution is to insert an ultra - thin cobalt layer into the tungsten layer to form a composite structure.

So, what are the advantages of applying eMRAM (embedded magnetic random - access memory) to MCUs?

According to Synopsys, compared with PCRAM and ReRAM, eMRAM has lower temperature sensitivity, provides better production - level yield, and offers longer durability (retaining data over multiple read/write cycles for many years). It allows word - level erasure and programming operations, making it an energy - efficient NVM solution.

Although the manufacturing cost of eMRAM is higher than that of ReRAM, its higher reliability and lower variability result in an area - efficient and robust design, offsetting the higher wafer cost. A single chip can have more memory with eMRAM, or a design using eMRAM can be smaller and more energy - efficient with the same amount of memory. eMRAM has been put into production at leading 22nm foundries and is now moving towards the FinFET node.

In addition, MRAM can maintain excellent reliability in extreme environments and demonstrates outstanding comprehensive indicators in terms of power consumption, performance, and area (PPA). MRAM was originally developed to meet the strict requirements of the aerospace industry and maximizes storage density through an adjustable magnetic layer design. Thanks to its excellent power - consumption control and performance, MRAM is an ideal choice for applications that require extreme reliability and data integrity.

In modern transportation, the application of MRAM is particularly prominent, such as in smart cars that support over - the - air (OTA) software updates. In addition, MRAM helps reduce the energy consumption of advanced microcontrollers (MCUs) and AI accelerators without sacrificing performance.

A unified eMRAM solution is the solution for advanced MCUs

Of course, MRAM is not without its drawbacks. It still faces many challenges, such as the complex material system of real - world devices, low on - off ratio, and the need for full compatibility with CMOS processes. There are also bottlenecks in terms of dynamic power consumption, energy - delay efficiency, and reliability.

In addition, MRAM is sensitive to strong magnetic fields, which limits its application scope to some extent. Therefore, in specific environments, physical isolation measures or shielding technologies may need to be implemented to reduce potential risks.

Although eMRAM has attractive advantages, designers should use reliable, silicon - verified solutions and seamlessly integrate built - in self - test (BIST) and error - code correction (ECC) support.

MCU designers must consider the magnetic immunity of eMRAM when integrating solutions into SoCs, including testing the sensitivity level of MRAM, reporting it in gauss (B - field) or oersteds (H - field), and informing their customers of this specification. Any components near the chip that may become magnetic (such as inductive coils) can affect the performance of eMRAM. Therefore, system designers must keep these components at a sufficient distance from eMRAM to prevent magnetic - field interference.

Renesas and NXP Go All - Out

NXP has been highly committed to the iteration of MCU process technology in recent years. As software - defined vehicles (SDV) become more popular, the market demand for regional controllers is increasing. The S32K5 launched by NXP in March this year is a typical example of MRAM application.

The S32K5 has a very powerful configuration, featuring high performance and low power consumption. It can integrate multiple different ECUs into a single system or module, optimizing cost and performance. It has three main highlights: First, it has powerful heterogeneous computing capabilities. With Cortex - M7@200MHz and Cortex R52@800MHz cores, DSP, and eIQ NPU configured in single - core, multi - core, or lock - step core modes, it can not only accelerate AI/ML but also extend battery life through a low - power subsystem and achieve passive cooling through 16FF technology and efficient design. Second, it has innovative "core - to - pin" resource isolation, which arranges system resources in isolated environments. If there is a problem in the security layer, the core can be restarted. This function applies to memory, TMA channels, peripherals, and input/output (IO). Third, it supports 10BaseT1S Ethernet acceleration and CAN acceleration of up to 2.5Gbps, greatly enhancing the determinism and time - sensitive network (TSN) capabilities of the SK32K5.

The S32K5 is the first 16nm FinFET + MRAM automotive MCU, with an MRAM capacity of up to 41MB. Regarding 16nm FinFET, Manuel Alves, Senior Vice President and General Manager of Automotive Microcontrollers at NXP Semiconductors, believes that 16nm is the ideal choice for current regional controllers because the trend towards centralization places high demands on computing power and storage. Regarding MRAM, Manuel Alves believes that this new type of storage medium has unique advantages. Firstly, it has extremely fast write and programming speeds, 10 times faster than flash memory, enabling rapid operation. Secondly, it has high durability, capable of 1 million write cycles. It can store not only code but also data, offering high flexibility and facilitating data collection and cross - regional storage.

As a regional controller, the S32K5 can operate in two different scenarios: One is to integrate all real - time control functions, with the central processor only responsible for application operation. This is an extreme case for a regional controller. The other is that the central computing unit integrates all functions and has powerful real - time computing capabilities, while the regional controller is lightweight and serves as an interface connecting terminal nodes and the central computing unit, that is, a regional aggregator or gateway.

Different from NXP, Renesas has a greater focus on edge AI. It not only uses MRAM but also piles on computing power and features to give its MCUs extreme performance and high benchmark scores.

In June this year, Renesas quietly launched the "world's most powerful MCU," the RA8P1 series of MCUs on its official website. It is manufactured using the 22nm ULL process and is equipped with 0.5/1MB MRAM (with an optional 4/8MB flash memory). In contrast, the RT1170 uses the 28nm FD - SOI process. According to Renesas, compared with flash memory, MRAM has faster write speed, higher durability, and stronger data retention ability.

After launching the world's first Cortex - M85 MCU in 2023, this MCU not only uses the powerful Arm M core, the 1GHz Cortex - M85, but also integrates the Ethos - U55 NPU, taking component stacking to the extreme.

At the same time, Renesas also launched the RA8D2, which is also equipped with MRAM.