STMicroelectronics Unveils STM32N6 MCU with 600 GOPS
STMicroelectronics has officially launched its new STM32N6 series microcontroller, featuring the industry's first integrated custom Neural Processing Unit (NPU). With an AI computing power of 600 GOPS, this MCU marks ST's debut in the Arm Cortex-M55-based microcontroller segment.
STMicroelectronics emphasizes that the widespread adoption of edge AI will only be possible when it becomes more accessible across embedded systems. As the industry increasingly supports AI on microcontrollers, more MCUs are integrating NPUs. However, STMicroelectronics has taken a different approach, distinguishing its solution in a crowded market.
The company has been developing the ST Neural-ART accelerator since 2016. This was followed by the 2019 launch of the STM32Cube.AI software solution, directly influenced by the research on the Neural-ART accelerator. As STM32Cube.AI gained traction in creating innovative edge AI products, STMicroelectronics optimized the Neural-ART accelerator to introduce unique products. The company underscores that no other MCU manufacturer offers such a deeply customized and optimized hardware and software ecosystem for edge AI.
The Neural-ART accelerator in the STM32N6 includes nearly 300 configurable multiply-accumulate units and two 64-bit AXI memory buses, delivering 600 GOPS of performance. This represents a 600-fold increase in AI computing power compared to the STM32H7, STMicroelectronics' fastest MCU, which lacks an NPU core. This groundbreaking architecture enables more operations per clock cycle and optimizes data flow to prevent bottlenecks, all while being power-efficient, offering an impressive 3 TOPS/W.
STMicroelectronics also ensures that the Neural-ART Accelerator supports a broader range of AI software tools than typical solutions in the market. The STM32N6 is already compatible with the highest number of AI operators in TensorFlow Lite, Keras, and ONNX, with plans to expand operator support in the future. The ability to use ONNX format means that data scientists can deploy STM32N6 for a wide array of AI applications. In short, STMicroelectronics aims to provide not only a more optimized hardware solution but also a more accessible platform, enabling developers to work within their current workflows and accelerate time-to-market.
In addition to the integrated NPU core, the STM32N6 features an Arm Cortex-M55 core and is built on a 16nm FinFET process, with a clock frequency of up to 800MHz. The MCU also includes various other IPs, including a gigabit Ethernet module supporting time-sensitive networking, six SPI and two I3C interfaces, two 12-bit ADCs, and four 32-bit advanced timers.
Anticipating that customers will use the STM32N6 in machine vision applications with cameras, STMicroelectronics has integrated its latest Image Signal Processor (ISP), identical to the one on the STM32MP2. The ISP is compatible with the STM32 ISP IQTune software, allowing developers to adjust the ISP for CMOS sensors, lenses, lighting conditions, and more, without needing to hire costly third-party providers. The STM32N6 also supports MIPI CSI-2, making it compatible with the most popular camera interfaces in mobile applications without the need for an external ISP compatible with the specific camera interface. This allows the STM32N6 to easily handle images from multiple image sensors and future-proof the system.
STMicroelectronics also offers Nucleo development boards and exploration kits, alongside the release of STM32CubeN6, a dedicated software package with middleware and example code. The company has updated the ST Edge AI Suite to support the new device, and TouchGFX Designer provides a board-level support package to help developers quickly utilize the new hardware and create impressive user interfaces. Third-party development boards featuring the STM32N6 will also be released later.
