Recently, Flyingchip, a technologically advanced intelligent visual processing chip manufacturer ®， SITWEI subsidiary, stock code: 688213, hereinafter referred to as "Feilingwei", has officially launched a new generation of high-performance in car visual processing SoC - M2. This chip is designed specifically for L2 level assisted driving and intelligent sensing applications in the cabin, using TSMC FinFET technology and equipped with self-developed FlyingMatrix ™ NPU and FlyingSight ™ AI-ISP， The two collaborate deeply from the underlying architecture, providing high-quality image input and real-time AI inference for the in vehicle vision system under complex and changing driving lighting conditions. In addition, M2 is equipped with a quad core Arm ® Cortex ®- A53 processor and dual core Arm ® Cortex ®- R52+Lock Step Safety Island, compliant with AEC-Q100 Grade 2 vehicle regulations and ASIL-B/D functional safety level, can fully cover mainstream in vehicle visual application scenarios such as front view all-in-one machine, streaming media and electronic rearview mirror, DMS/OMS (driver/passenger monitoring system), high-definition surround view system, and integrated parking, providing a system level solution with high energy efficiency, low latency, safety and reliability.

Flexible allocation of computing power, unleashing the ultimate energy efficiency of NPU

The Infineon M2 is equipped with a self-developed FlyingMatrix ™ NPU， The dense computing power can reach up to 10 TOPS, with a typical power consumption of less than 3W. NPU is equipped with rich visual processing hardware acceleration operators and CV acceleration engines, natively supports Transformer series operators, is compatible with INT4/INT8/INT16 mixed precision, and has 4K@60fps H. 264/H.265 video encoding and decoding capabilities provide hardware level acceleration for visual algorithms such as image detection, object classification, and pixel level segmentation. Due to the significant differences in load in different scenarios, the fixed computing power allocation method always faces a trade-off between performance and power consumption when dealing with dynamic loads. Therefore, flexible computing power allocation has become the key to maximizing energy efficiency. M2 supports independent power up and down of CPU and NPU cores, and can dynamically adjust power consumption according to load; And by configuring independent high-performance OCMEM (on-chip memory) to reduce invalid data transfer from the root, effectively breaking through the performance bottleneck caused by the "memory wall" and "power wall", ensuring that computing resources always serve the most urgent tasks at present. M2 can also be flexibly deployed for different AI model requirements, improving chip utilization without sacrificing inference accuracy, and calmly handling massive data throughput and high-density matrix operations in vehicle scenarios.



近日，技术先进的智能视觉处理芯片厂商飞凌微电子（Flyingchip®，思特威子公司，股票代码：688213，以下简称"飞凌微"）正式推出新一代高性能车载视觉处理SoC——M2。该芯片专为L2级辅助驾驶及舱内智能感知应用设计，采用台积电（TSMC）FinFET工艺，搭载自研FlyingMatrix™ NPU和FlyingSight™ AI-ISP，两者从底层架构上深度协同，在复杂多变的行车光照条件下为车载视觉系统提供高质量图像输入与实时AI推理。此外，M2内置四核Arm® Cortex®-A53处理器和双核Arm® Cortex®-R52+锁步安全岛，符合AEC-Q100 Grade 2车规标准和ASIL-B/D功能安全等级，可全面覆盖前视一体机、流媒体及电子后视镜、DMS/OMS（驾驶员/乘员监控系统）、高清环视系统及行泊一体等主流车载视觉应用场景，提供高能效、低延迟、安全可靠的系统级方案。

算力灵活分配，释放NPU极致能效

飞凌微M2搭载自研FlyingMatrix™ NPU，稠密算力可达10 TOPS，典型功耗低于3W。NPU内置丰富的视觉处理硬件加速算子与CV加速引擎，原生支持Transformer系列算子，兼容INT4/INT8/INT16混合精度，并具备4K@60fps H.264/H.265视频编解码能力，为图像检测、目标分类、像素级分割等视觉算法提供硬件级加速。由于不同场景下的负载差异极大，固定算力分配方式在应对动态负载时，始终存在性能与功耗的权衡困境，因此，算力灵活分配成为能效释放最大化的关键。M2支持CPU与NPU各核心独立上下电，可根据负载动态调整功耗；并通过配置独立的高性能OCMEM（片上存储器）从根源上减少无效数据搬运，有效突破"内存墙"和"功耗墙"引发的性能瓶颈，确保算力资源始终服务于当前最亟须处理的任务。M2还可针对不同AI模型需求灵活部署，在不牺牲推理精度的前提下提升芯片利用率，从容应对车载场景中海量数据吞吐与高密度矩阵运算。




AI-ISP collaborates with high-speed interfaces to create excellent in car imaging

Traditional ISPs often struggle to balance image details and algorithm recognition accuracy when dealing with complex lighting scenes such as low light, backlighting, and high dynamic range. M2 integrated self-developed FlyingSight ™ AI-ISP， Deeply integrated with NPU, AI algorithms are introduced into each stage node of the image processing pipeline for adaptive optimization in extreme lighting scenes, outputting low-noise, high dynamic range, and color accurate images. In addition, M2 is equipped with rich interfaces such as 4-channel MIPI-RX, 2-channel MIPI-TX, Gigabit Ethernet, CAN, SDIO, eMMC, etc., supporting HDR high dynamic processing, as well as multiple color filter arrays such as RGGB, RGB-IR, Mono, etc., combined with M2's maximum of 8 inputs and maximum 16MP@60fps Image processing capability, efficiently supporting the simultaneous operation of multiple cameras such as front view, surround view, and cabin, providing high-quality image input and fast response capability for the in vehicle vision system.

Vehicle grade design, comprehensive hardcore support



AI-ISP与高速接口协同，打造卓越车载成像

传统ISP在处理暗光、逆光、高动态范围等复杂光线场景时，往往难以兼顾图像细节与算法识别精度。M2集成自研FlyingSight™ AI-ISP，与NPU深度融合，将AI算法引入图像处理流水线的每一级节点，在极端光线场景下进行自适应优化，输出低噪声、高动态范围、色彩准确的图像。此外，M2配备4路MIPI-RX、2路MIPI-TX、千兆以太网、CAN、SDIO、eMMC等丰富接口，支持HDR高动态处理，以及RGGB、RGB-IR、Mono等多种色彩滤波阵列，配合M2最高8路输入和最大16MP@60fps图像处理能力，高效支撑前视、环视、舱内等多路摄像头同时工作，为车载视觉系统提供高质量的图像输入与快速响应能力。

车规级设计，全方位硬核保障




The Infineon M2 provides a safer and more reliable computing platform for intelligent vehicles from multiple dimensions such as hardware safety, functional safety, and vehicle level design. In terms of hardware security, M2 has an independent AO (Always On) subsystem that operates continuously in extremely low-power scenarios. Its built-in HSM (Hardware Security Module) supports encryption and decryption, digital signature verification, random number generation, and secure boot, effectively ensuring the safety of system operation. In terms of functional safety, M2 adopts a dual domain security architecture and integrates dual core lock step Arm ® Cortex ®- R52+Safety Island, supporting AUTOSAR CP ecosystem, equipped with ECC error correction, MBIST memory self-test, bus timeout detection, error interruption and external error reporting, helping the in vehicle system achieve more complete fault detection and safety response. In terms of vehicle specification design, M2 complies with AEC-Q100 Grade 2 vehicle specification standards and ISO 26262 ASIL-B/D functional safety levels, fully meeting the stringent requirements of reliability, functional safety, and information security for in vehicle visual applications.

AI SoC+Sensor integrated solution accelerates the upgrading of the smart car industry

Thanks to the profound technical accumulation of Sitwee in the field of image sensors and the homogeneity of the underlying hardware for visual processing, the Infineon M2 is able to flexibly reuse its market proven image processing core technology and hardware design concepts, thereby constructing a mature and efficient computing architecture. This not only endows M2 with excellent visual signal processing performance, but also naturally adapts its processing capability to the full range of SITWEI vehicle grade CIS, forming an end-to-end technological synergy advantage from perception to computation.



飞凌微M2从硬件安全、功能安全以及车规级设计等多个维度出发，为智能汽车提供更加安全可靠的计算平台。硬件安全方面，M2设有独立的AO（Always-On）子系统，在极低功耗场景下持续运行，其内置的HSM（硬件安全模块）支持加解密、数字签名验证、随机数生成以及安全启动，有效保障系统运行安全。功能安全方面，M2采用双域安全架构，集成双核锁步Arm® Cortex®-R52+安全岛，支持AUTOSAR CP生态系统，配套ECC纠错、MBIST存储器自测试、总线超时检测、错误中断及外部错误报告，帮助车载系统实现更完善的故障检测与安全响应。车规级设计方面，M2符合AEC-Q100 Grade 2车规标准和ISO 26262 ASIL-B/D功能安全等级，充分满足车载视觉应用对可靠性、功能安全与信息安全的严苛要求。

AI SoC+Sensor一体化方案，加速智慧汽车产业升级

得益于思特威在图像传感器领域深厚的技术积累以及视觉处理底层硬件的同源性，飞凌微M2能够灵活复用其经过市场验证的图像处理核心技术与硬件设计理念，从而构筑成熟且高效的计算架构。这不仅赋予了M2卓越的视觉信号处理性能，更使其在处理能力上与思特威全系车规级CIS天然适配，形成了从感知到计算的端到端技术协同优势。




Based on this unique advantage, Infineon Micro integrates AI SoC and Sensor at the system level, connecting the complete chain from image acquisition, preprocessing to AI inference. Compared to traditional combination schemes, this system level integration simplifies the complexity of hardware selection design and system debugging, reduces BOM costs and development thresholds. In addition, pre validation and deep adaptation between hardware have significantly shortened the product validation and testing cycle, accelerated the implementation process of high-performance in car vision solutions from design to mass production, and provided strong impetus for industrial upgrading.

AI vision processing SoC A2, expanding applications in robotics and industrial vision

Based on the same self-developed FlyingMatrix ™ NPU and FlyingSight ™ The two core technology platforms of AI-ISP, Infineon Micro, have simultaneously launched the AI vision processing SoC A2 for fields such as robotics and industrial vision, demonstrating its potential for application in a wider range of intelligent vision fields. The NPU has an efficient computing architecture built-in, which can flexibly run complex AI algorithms such as image segmentation, dynamic obstacle avoidance, and V-SLAM positioning; AI-ISP provides pixel level real-time image optimization for autonomous decision-making tasks such as environmental perception, 3D reconstruction, and object recognition under complex lighting conditions such as strong and dim light. In addition, A2 supports up to four cameras connected directly at the same time and is equipped with 1080p@90fps DPU (binocular vision depth acceleration engine) provides precise spatial perception capability and real-time safety control for scenes such as mobile robots and industrial RGBD cameras.



基于这一独特优势，飞凌微通过AI SoC与Sensor的一体化系统级方案，打通从图像采集、预处理到AI推理的完整链路。相比传统组合方案，这种系统级整合简化了硬件选型设计与系统调试的复杂度，降低了BOM成本和开发门槛。此外，硬件之间的预验证与深度适配也大幅缩短了产品验证与测试周期，加速了高性能车载视觉方案从设计到量产的落地进程，为产业升级提供了强劲动力。

AI视觉处理SoC A2，机器人及工业视觉应用拓展

基于相同的自研FlyingMatrix™ NPU和FlyingSight™ AI-ISP两大核心技术平台，飞凌微同步推出面向机器人、工业视觉等领域的AI视觉处理SoC A2，展现了其在更广泛智能视觉领域中的应用潜力：NPU内置高效计算架构，能够灵活运行图像分割、动态避障、V-SLAM定位等复杂AI算法；AI-ISP则在强光、昏暗等复杂光线条件下，为环境感知、三维重建、物体识别等自主决策任务提供像素级实时图像优化。此外，A2支持最多四路摄像头同时直连接入，并配备1080p@90fps DPU（双目视觉深度加速引擎），为移动类机器人、工业RGBD相机等场景提供精确的空间感知能力和实时安全控制。




Shao Ke, Vice President of STMicroelectronics and CEO of Infineon Micro, said: "The automotive industry is currently in an important stage of shifting from 'digitalization' to 'inclusiveness', and lightweight end side AI chips are becoming an important carrier for industry development. As a wholly-owned subsidiary brand of STMicroelectronics, the M2 series in car visual processing SoC launched by FlyingMicro is based on this, relying on its self-developed FlyingMatrix ™ NPU and FlyingSight ™ The deep integration of AI-ISP two core technologies has the advantages of low power consumption, high energy efficiency, and high reliability. Through the business model of AI SoC+Sensor integration, we provide customers with highly competitive end-to-end visual AI system level solutions, committed to making every decision response for environmental perception more accurate, stable, and timely, and building a solid technological foundation for the smart car industry. "



思特威副总裁兼飞凌微首席执行官邵科表示："汽车产业当前处于'数智化'转向'普惠化'的重要阶段，轻量级端侧AI芯片正成为产业发展的重要载体。作为思特威旗下全资子公司品牌，飞凌微此次推出的M2系列车载视觉处理SoC立足于此，依托自研FlyingMatrix™ NPU和FlyingSight™ AI-ISP两大核心技术的深度融合，具备低功耗、高能效、高可靠的优势，并通过AI SoC+Sensor一体化的商业模式，为客户提供极具竞争优势的端侧视觉AI系统级方案，致力于让每一次针对环境感知的决策响应都更精准、更稳定、更及时，为智慧汽车产业构筑坚实的技术基石。"

(Using AI translation)

（使用AI翻译）