Car MCU, it's about to change again

Jessie March 07, 2024

How can the MCU change in the next two years? In addition to joining the AI accelerator, or switching from the Cortex-M core to the RISC-V core, is the integration of new types of memory.

On February 28, Infineon announced that a new generation of MCU AURIX TC4x is about to be mass-produced, and unlike traditional MCUS, this generation of products introduces RRAM non-volatile storage media (NVM). Coincidentally, at ISSCC 2024, Renesas announced that it has developed a test chip for STT-MRAM circuit technology for embedded applications, including MCUS. As early as 2022, ST released the Stellar P6 MCU, at which time PCM was adopted in the product.

This indicates that the new storage MCU finally has to officially meet us, which means that the MCU process technology should be fully headed below 28nm, and the first market that will change is the vehicle MCU market.

Another way to upgrade the car MCU

As you and I both know, there are many memories in the MCU, and these memories are of two types - volatile memory such as eDRAM and SRAM, and non-volatile memory (NVM) such as EEPROM and embedded flash memory (eFlash). In a typical MCU system-on-chip (SoC), there are four types of memory:

• Cache memory: Small memory designed using triggers or SRAM. They require high-speed read and write operations and very high durability;
• Data memory: High-capacity memory used to feed data to processing units, designed to improve overall system efficiency. These are high-density memory, usually SRAM;
• Key Storage memory: A small NVM that stores chip ids, security codes, SRAM repair signatures, or analog circuit repair information. They don't need to be highly durable; Therefore, ROM, anti-fuse based disposable programmable (OTP) memory, or eFuses are commonly used for these memories;
• Code and Data storage Storage: High-capacity NVM stores startup code, firmware, and data. With advances in AI/ML, IoT, and automation, memory capacity is also increasing, and power consumption is becoming critical. Previously, embedded flash memory (eFlash) was the choice for this purpose.

But in summary, the traditional embedded flash memory (eFlash) integrated in the MCU has three disadvantages:

First, limit the process of the MCU itself. In theory, the smaller the process, the more transistors can be accommodated per unit area, and the smaller the power consumption. However, we all know that the MCU has been hovering at 40nm, and the most important factor is the embedded flash memory (eFlash) itself within the MCU. The manufacturing process of flash memory is very difficult to scale below 40nm, not only considering various parameters and costs, but also difficult to integrate into very complex high-K metal gate technology.

Second, the industry's demand for MCU storage capacity is increasing. With the improvement of the computing power requirements of cross-domain fusion, the NVM storage capacity required by automotive MCUS is getting higher and higher, and the new storage has a higher capacity improvement space than flash memory.

Third, embedded flash life is becoming less and less suitable for existing needs. In automotive applications, flash memory integrated into onboard MCUS has too few rewrites, and with each write and erase cycle, tunnel oxides in floating gate NOR cells degrade and leakage increases, thus accelerating the aging of flash memory, making flash less and less suitable as a data memory.

Therefore, more and more MCU manufacturers are choosing to integrate new types of memory, solve the above problems, and expand the MCU process to continue to move forward, and continue to break through 28nm, 22nm and even 16nm.

Three types of storage, three genres

Currently, there are three new types of memory on the market that have begun to be used in MCUS - RRAM (resistive memory), MRAM/STT-MRAM (magnetic memory), and PCM (PCRAM).

The first is RRAM (resistive storage), and Infineon is the biggest player on this route.

Infineon announces the TC4XX series at the end of 2021; In November 2022, we completed RRAM mass production research and development with TSMC. In 2023, the RRAM and logic devices will be combined, and the mass production will be official at the beginning of this year.

According to Infineon, most MCU families on the market use embedded flash technology. As the next generation of embedded memory, RRAM can be further expanded to 28nm and below.

RRAM is highly resistant to interference and allows bit-by-bit writing without the need for erasure, with durability and data retention comparable to flash technology. The introduction of RRAM will create great potential for improved performance, reduced power consumption and cost savings for MCUS, as well as further miniaturization.

The second is MRAM/STT-MRAM (magnetic memory), with Renesas and NXP as the main promoters.

In June 2022, Renesas announced the introduction of STT-MRAM's 22 nm manufacturing process technology at the VLSI Conference. At ISSCC 2024, Renesas announced that it has developed circuit technology for embedded spin transfer moment reluctance random access memory (STT-MRAM), a test chip with fast read and write operations.

According to Renesas, a prototype MCU test chip with a 10.8 Mbit MRAM storage unit array was manufactured using a 22nm embedded MRAM process. Evaluation of the prototype chip confirmed that a random read access frequency of over 200MHz and a write throughput of 10.4 MB/s were achieved at a maximum junction temperature of 125°C.

In May 2023, NXP and TSMC launched the automotive-grade 16nm FinFET process MRAM, which will be adopted by the next generation of NXP MCUS, which is expected to be mass-produced in late 2024 or early 2025.

According to NXP, Flash memory takes about 1 minute to update 20MB of code, while MRAM only takes about 3 seconds, minimizing downtime caused by software updates, and auto manufacturers can eliminate bottlenecks caused by long module programming. In addition, MRAM provides up to one million update cycles and durability that exceeds ten times that of flash and other emerging memory technologies, providing a highly reliable technology for automotive failure defects.

The third is PCM (PCRAM, phase change memory), and ST is the main promoter.

In September 2022, ST launched the Stellar P Series MCU, the Stellar P6, manufactured in ST's own fab, using energy-efficient 28nm FD-SOI technology and embedded phase change (non-volatile) memory (PCM) with a capacity of up to 20 MB.

The Stellar P6 is manufactured in ST's own fabs using energy-efficient 28nm FD-SOI technology with an embedded phase-change (non-volatile) memory (PCM) capacity of up to 20MB. Developed and tested in strict automotive high-temperature operating environments, radiation resistance and data preservation requirements, the ST PCM has a single-bit overwrite function that flash memory does not have, making memory access faster.

How can the MCU change in the next two years? In addition to joining the AI accelerator, or switching from the Cortex-M core to the RISC-V core, is the integration of new types of memory.

On February 28, Infineon announced that a new generation of MCU AURIX TC4x is about to be mass-produced, and unlike traditional MCUS, this generation of products introduces RRAM non-volatile storage media (NVM). Coincidentally, at ISSCC 2024, Renesas announced that it has developed a test chip for STT-MRAM circuit technology for embedded applications, including MCUS. As early as 2022, ST released the Stellar P6 MCU, at which time PCM was adopted in the product.

This indicates that the new storage MCU finally has to officially meet us, which means that the MCU process technology should be fully headed below 28nm, and the first market that will change is the vehicle MCU market.

Another way to upgrade the car MCU

As you and I both know, there are many memories in the MCU, and these memories are of two types - volatile memory such as eDRAM and SRAM, and non-volatile memory (NVM) such as EEPROM and embedded flash memory (eFlash). In a typical MCU system-on-chip (SoC), there are four types of memory:

Cache memory: Small memory designed using triggers or SRAM. They require high-speed read and write operations and very high durability;

Data memory: High-capacity memory used to feed data to processing units, designed to improve overall system efficiency. These are high-density memory, usually SRAM;

Key Storage memory: A small NVM that stores chip ids, security codes, SRAM repair signatures, or analog circuit repair information. They don't need to be highly durable; Therefore, ROM, anti-fuse based disposable programmable (OTP) memory, or eFuses are commonly used for these memories;

Code and Data storage Storage: High-capacity NVM stores startup code, firmware, and data. With advances in AI/ML, IoT, and automation, memory capacity is also increasing, and power consumption is becoming critical. Previously, embedded flash memory (eFlash) was the choice for this purpose.

MCU design block diagram, diagram source, Sinsi Technology

But in summary, the traditional embedded flash memory (eFlash) integrated in the MCU has three disadvantages:

First, limit the process of the MCU itself. In theory, the smaller the process, the more transistors can be accommodated per unit area, and the smaller the power consumption. However, we all know that the MCU has been hovering at 40nm, and the most important factor is the embedded flash memory (eFlash) itself within the MCU. The manufacturing process of flash memory is very difficult to scale below 40nm, not only considering various parameters and costs, but also difficult to integrate into very complex high-K metal gate technology.

Second, the industry's demand for MCU storage capacity is increasing. With the improvement of the computing power requirements of cross-domain fusion, the NVM storage capacity required by automotive MCUS is getting higher and higher, and the new storage has a higher capacity improvement space than flash memory.

Third, embedded flash life is becoming less and less suitable for existing needs. In automotive applications, flash memory integrated into onboard MCUS has too few rewrites, and with each write and erase cycle, tunnel oxides in floating gate NOR cells degrade and leakage increases, thus accelerating the aging of flash memory, making flash less and less suitable as a data memory.

Therefore, more and more MCU manufacturers are choosing to integrate new types of memory, solve the above problems, and expand the MCU process to continue to move forward, and continue to break through 28nm, 22nm and even 16nm.

Three types of storage, three genres

Currently, there are three new types of memory on the market that have begun to be used in MCUS - RRAM (resistive memory), MRAM/STT-MRAM (magnetic memory), and PCM (PCRAM).

The first is RRAM (resistive storage), and Infineon is the biggest player on this route.

Infineon announces the TC4XX series at the end of 2021; In November 2022, we completed RRAM mass production research and development with TSMC. In 2023, the RRAM and logic devices will be combined, and the mass production will be official at the beginning of this year.

According to Infineon, most MCU families on the market use embedded flash technology. As the next generation of embedded memory, RRAM can be further expanded to 28nm and below.

RRAM is highly resistant to interference and allows bit-by-bit writing without the need for erasure, with durability and data retention comparable to flash technology. The introduction of RRAM will create great potential for improved performance, reduced power consumption and cost savings for MCUS, as well as further miniaturization.

The second is MRAM/STT-MRAM (magnetic memory), with Renesas and NXP as the main promoters.

In June 2022, Renesas announced the introduction of STT-MRAM's 22 nm manufacturing process technology at the VLSI Conference. At ISSCC 2024, Renesas announced that it has developed circuit technology for embedded spin transfer moment reluctance random access memory (STT-MRAM), a test chip with fast read and write operations.

According to Renesas, a prototype MCU test chip with a 10.8 Mbit MRAM storage unit array was manufactured using a 22nm embedded MRAM process. Evaluation of the prototype chip confirmed that a random read access frequency of over 200MHz and a write throughput of 10.4 MB/s were achieved at a maximum junction temperature of 125°C.

In May 2023, NXP and TSMC launched the automotive-grade 16nm FinFET process MRAM, which will be adopted by the next generation of NXP MCUS, which is expected to be mass-produced in late 2024 or early 2025.

According to NXP, Flash memory takes about 1 minute to update 20MB of code, while MRAM only takes about 3 seconds, minimizing downtime caused by software updates, and auto manufacturers can eliminate bottlenecks caused by long module programming. In addition, MRAM provides up to one million update cycles and durability that exceeds ten times that of flash and other emerging memory technologies, providing a highly reliable technology for automotive failure defects.

The third is PCM (PCRAM, phase change memory), and ST is the main promoter.

In September 2022, ST launched the Stellar P Series MCU, the Stellar P6, manufactured in ST's own fab, using energy-efficient 28nm FD-SOI technology and embedded phase change (non-volatile) memory (PCM) with a capacity of up to 20 MB.

The Stellar P6 is manufactured in ST's own fabs using energy-efficient 28nm FD-SOI technology with an embedded phase-change (non-volatile) memory (PCM) capacity of up to 20MB. Developed and tested in strict automotive high-temperature operating environments, radiation resistance and data preservation requirements, the ST PCM has a single-bit overwrite function that flash memory does not have, making memory access faster.

So, which of the three types of storage is stronger?

In fact, the three new types of memory can not be simply divided into good and bad, each new type of storage skill allocation is not the same, in other words, the focus is not the same:

• RRAM: Slightly later than MRAM and PRAM research; The erase speed is determined by the pulse width of the trigger resistance transition, which is generally less than 100ns; Reading and writing adopt reversible mode without damage, which can greatly improve the service life; Some RRAM materials have a variety of resistance states, which can further improve the storage density;
• MRAM: three generations of MRAM/STT-MRAM/SOT-MRAM; Higher manufacturing costs than RRAM, but with higher reliability and lower variability resulting in area efficiency and robust design; The write time can be as low as 2.3ns, and the power consumption is very low; MRAM itself is non-volatile, but the ferromagnetic power failure will not disappear, so MRAM and DRAM can be rewritten indefinitely; It has the potential to construct large-scale memory arrays on logic circuits;
• PCM: Low latency, read and write time balance, low power consumption; Non-destructive reading and writing ability; Some PCM adopts non-transistor design, which can realize high density storage. In addition, PCM is independent of the charged particle state of the material, so it has a strong ability to resist space radiation.

The future after the car MCU taste fresh

Why did this change happen in the car space first? Due to changes in vehicle architecture and computing power requirements, the market needs higher performance and higher storage capacity MCUS. Specifically, the new storage products of several manufacturers are aimed at the following aspects:

• Hardware design for accelerating the routing capability of in-vehicle messages;
• MCU hardware virtualization for cross-domain convergence;
• Information security solutions that meet ISOSAE 21434;
• Optimization design of non-inductive SOTA for MCU;
• Large capacity memory for cross-domain fusion;
• High performance and high computing power CPU for new electrical and electronic architectures.

I believe that in the future, with the expansion of AI needs, changes in downstream intelligent devices, new storage costs and technological maturity continue to move forward, and new storage will gradually expand to the entire MCU field in the future. At that time, the entire MCU process will also continue to move forward.

Disclaimer: This article is a reprint article, reprint this article to convey more information, copyright belongs to the original author. If the video, pictures and text used in this article involve copyright issues, please contact Xiaobian for processing.