NXP launches world's first 5nm automotive MCU

At the end of March 2024, NXP officially launched the world's first 5-nanometer automotive MCU . However, NXP did not call it an MCU, but called it S32N55 Vehicle Super-Integration Processor. In fact, it is an MCU, and of course it is also possible to call it a SoC. It has the typical characteristics of advanced automotive MCUs. First, it has a highly efficient computing core that emphasizes real-time performance. Secondly, it has a high-security core that can reach ASIL-D standards. Thirdly, it has multiple network interfaces , including CAN, LIN, FlexRay, in-vehicle Ethernet , CAN-FD, CAN-XL, and PCIe. There are at least 15 CAN network interfaces. Of course, it is a bit of a stretch to call it an MCU because it has no embedded flash memory , only SRAM .

E/E positioning of S32N

Image source: NXP

S32N is a bit like the integration of multiple MCUs. Taking S32N55 as an example, it integrates vehicle dynamic control, body, comfort, and central gateway.

NXP's S32N series Processors Distribution

Image source: NXP

The 5-nanometer S32N55 is the first product in the S32N series. There will be three more products in the future. This series should all be manufactured using at least a 5-nanometer process, and a 3-nanometer process is not ruled out. As can be seen from the figure, the S32N55 is mainly used in the chassis, body, and gateway fields.

MCU is a field with a highly concentrated market. In 2023, the global MCU market will be approximately US$29.8 billion, and the top five manufacturers will have a market share of more than 80%.

Image source: Infineon Technologies

In the field of automotive MCUs, Infineon has grown rapidly. In 2020, it was still the fourth in the world with a market share of less than 10%. In 2024, it has become the world's first. The market share of Renesas and NXP, which were originally tied for first place, has dropped rapidly. The fastest decline is Texas Instruments , which fell from third to sixth. The market share of the top five manufacturers exceeds 90%, and the market share of domestic automotive MCUs is estimated to be less than 2%.

Image source: Infineon Technologies

As we all know, the manufacturing process of automotive MCU is very backward compared to automotive SoC. Most domestic automotive MCUs are 90 nanometers, and slightly higher-end ones are 65 nanometers. Foreign high-end mainstream automotive MCUs are manufactured using 40 nanometers, such as Infineon's TC39X (manufactured by GlobalFoundries in the United States), Renesas' RH850, and NXP's i.MX series. The 40 nanometer process has been around for 10 years. With the growth of MCU computing power and the rapid increase in the number of transistors , the manufacturing process of automotive MCUs is accelerating. In February 2019, Renesas was the first to launch an automotive MCU with a 28 nanometer manufacturing process, namely RH850/U2A. In April 2023, Renesas launched its first 22 nanometer MCU, but not in the automotive field.

At the beginning of 2024, Infineon's TC49X series also began to use 28nm manufacturing process, and STMicroelectronics exhibited 28nm automotive MCU samples at CES2024.

The most advanced is still NXP. NXP launched the S32Z and S32E series of automotive MCUs in 2022, using a 16nm manufacturing process. However, NXP calls them Safe and Secure High-Performance Real-Time Processors, which are essentially automotive MCUs.

The automotive MCU has been stuck in the 40-nanometer manufacturing process for nearly a decade. The main factor is the process of the embedded flash memory (eFlash) inside the MCU itself. It is very difficult to expand the manufacturing process of flash memory to below 40nm. Not only must various parameters and costs be considered, but it is also difficult to integrate into the very complex high-K metal gate technology. In other words, the key to advanced process automotive MCUs lies not in the MCU manufacturers themselves, but in the wafer foundries. With the intelligentization and electrification of automobiles, the embedded flash memory capacity of automotive MCUs has skyrocketed. For example, more advanced manufacturing processes and more advanced storage technologies can meet this demand. Among the four major MCU manufacturers, Renesas and NXP have chosen STT-MRAM, STMicroelectronics is heading for PCM ( phase change memory ), which is also planned for mass production in 2024, and Infineon has chosen RRAM technology.

Comparison of three technologies:

The cooling process after PCM RESET requires high thermal conductivity, which will lead to higher power consumption. And because its storage principle is to use temperature to achieve the change of resistance of phase change material, it is very sensitive to temperature and cannot be used in wide temperature scenarios. In order to make phase change material compatible with CMOS process, PCM must adopt a multi-layer structure, so the storage density is too low, and it cannot replace NAND Flash in terms of capacity, and the cost is high. However, STMicroelectronics claims that it has solved these shortcomings.

Although MRAM has good performance, its critical current density and power consumption still need to be further reduced. Currently, the storage cell size of MRAM is still large and does not support stacking. The process is relatively complex, and it is difficult to ensure uniformity in large-scale manufacturing. The storage capacity and yield rate are slowly increasing.

RRAM has the disadvantage of device-level variability. Device-level variability is directly related to the reliability of the chip . However, since the transition of the state of the RRAM device requires applying voltage to the two electrodes to control the drift of oxygen ions driven by electric fields and the diffusion of oxygen ions driven by heat, the three-dimensional morphology of the conductive filament is difficult to control. In addition, the influence of noise easily causes device-level variability. However, TSMC has solved this problem. It seems that several major MCU manufacturers are competing, but in fact they all rely on TSMC. Most of Renesas' MCUs are also produced by TSMC.

The manufacturing process of MRAM can reach 16 nanometers. I believe that NXP's 16 nanometer automotive MCU uses MRAM.

Why do we need to use 5nm process to manufacture MCU? The demand mainly comes from three aspects:

On the one hand, the vehicle control algorithms of electric vehicles are becoming more and more complex, especially after the addition of intelligent driving , which requires not only conventional serial computing, but also parallel computing and even vector or matrix computing. The vehicle control algorithm requires high real-time performance and a delay of less than 1 millisecond, so high computing power is required.

On the other hand, the software system is becoming more and more complex, and the MCU needs to run a lot of middleware and small and medium-sized virtual machines and operating systems, which will also consume a lot of computing power.

Finally, the car body and comfort systems are becoming more and more complex, especially the seats and lights, which also require a lot of computing power. High computing power means more transistors, which requires high transistor density of advanced processes. The computing power here is not AI computing power . The so-called AI computing power basically refers to matrix multiplication operations. 99% of the operations in the automotive field are not matrix multiplication. High computing power has given birth to 5-nanometer automotive processor chips.

Back to S32N55, for motor control , S32N55 has the Automotive Math and Motor Control Library (AMMCLib) operator library, supports AUTOSAR and small real-time operating systems such as Zephyr, supports AUTOSAR MCAL real-time driver RTD, supports Type1 virtual machines, supports platform communication protocol stack IPC F, supports Safety Software Framework (SAF) and Structural Core Self-Test (SCST).

S32N55 internal framework diagram

The S32N55 contains four real-time computing units, each with a quad-core Cortex-R5 2.

The typical manufacturing process of R52 is 16 nanometers, the highest clock frequency is about 1.6 GHz , and the performance is generally 2.72DMIPS/MHz. The R52 frequency of S32N55 is 1.2GHz, and the computing power is 16*1200*2.72=52KDMIPS, which is twice the CPU computing power of the mainstream domestic intelligent driving chip and is on par with the CPU computing power of Texas Instruments TDA4 mid-range chip. Common chassis MCUs such as Infineon's TC397 have a computing power of 1.3-4KDMIPS, which is less than 1/10 of S32N55.

S32N55 Main Features

Image source: NXP

The S32N55 has up to 48MB of SRAM. The maximum SRAM capacity of traditional automotive MCUs, such as the chassis leader Infineon's TC397, is less than 7MB. If such a high capacity is still produced using traditional 28 or 40 nanometer processes, the cost will be very high. In order to reduce costs, the S32N55 does not have an internal NVM, but uses an external 8-channel Octal NOR Flash. After all, the maximum capacity of embedded NVM is only 64MB, while the external NVM can easily reach 256MB.

The S32N55 is designed for software-defined cars, so it naturally requires support for virtual prototype Digital Twin and cloud deployment, which is mainly provided by Synopsys.

S32N55 Virtual Prototyping Ecosystem

Image source: NXP

The S32N55 virtual prototype can be directly programmed on an ARM platform computer and then installed directly in the vehicle, which is the top-level virtual ECU .