Shuiqing Muhua Research Center and ZoZo Auto Research jointly released the "2026 Embodied Intelligence (and Humanoid Robot) MCU Research Report".
An MCU (Microcontroller Unit) is a compact IC (Integrated Circuit) that integrates a central processing unit (CPU), memory (RAM, ROM/Flash), input/output interfaces (I/O), and peripherals such as timers/counters and analog-to-digital converters (ADC) onto a single chip. It is also known as a single-chip microcomputer or MCU (Microcontroller Unit).
Currently, the MCUs used in embodied intelligence are mostly general-purpose MCUs. However, as the performance requirements and application scenarios for embodied intelligence continue to expand, dedicated MCUs have been introduced to ensure stable operation across various scenarios. In the future, this type of MCU will capture the market share of general-purpose MCUs and be widely applied in embodied intelligent manufacturing.
Overview of Distributed MCU Layout in Humanoid Robots
MCUs in robots do not exist in a "single-chip centralized management" form; instead, they are embedded in a distributed manner across various system levels, undertaking tasks such as perception, control, communication, and execution. They play a critical role in the core control system, joint controller system, perception system, power management system, and communication system of embodied intelligence.
Core Control System: Acting as the "nerve center" and "cerebellum," the MCU is embedded in joints throughout the body, perception, and communication modules as distributed nodes, undertaking full-link functions from real-time motion control to system safety.
Joint Control System: MCUs are widely deployed in various joints as execution units to complete the local closed-loop of a single joint, ensuring accurate movement execution. Additionally, due to the characteristics of dexterous hands, high-integration and small-size MCUs are selected and embedded in finger joints to achieve full coverage of degrees of freedom.
Perception System: Relying on the MCU to undertake distributed real-time processing tasks in perception, decision-making, and execution, it realizes functions such as sensor data perception and acquisition, and multi-modal signal preprocessing and fusion.
Power Management System: Primarily relying on the MCU to achieve battery status monitoring, energy consumption scheduling, and safety assurance.
Communication System: The MCU realizes the flow and synchronous interconnection of commands across the "brain-cerebellum-joint-end effector".
Encoder Module: Responsible for converting mechanical motion into high-precision, low-latency electrical signal feedback, supporting closed-loop control.
IMU (Inertial Measurement Unit) Module: A core component for the robot to perceive its own posture, motion, acceleration, and angular velocity. The MCU reads data and implements algorithms to output motion posture.
Overview of MCU Layout in Embodied Intelligence (and Humanoid Robots)
Source: "2026 Embodied Intelligence (and Humanoid Robot) MCU Research Report"
Specialization Evolution Trends of Joint MCUs Under the Divergence of Dexterous Hand Technology Routes in Humanoid Robots
As the main joint module of robots, the future development of dexterous hands mainly has five trends:
Evolution Trends of Dexterous Hands in Embodied Intelligence (and Humanoid Robots)
Source: "2026 Embodied Intelligence (and Humanoid Robot) MCU Research Report"
Based on the evolution trends of dexterous hands, five core trends have emerged at the MCU level: small-size packaging, high-performance real-time processing, I3C bus communication, NPU edge AI, and functional safety certification:
Trend 1 - Small Size and High Integration: As dexterous hands evolve towards higher degrees of freedom, hand space is extremely limited, and MCUs must evolve towards smaller sizes and higher integration. In terms of packaging, it is developing from BGA to WLCSP (4×4mm or even smaller); GigaDevice's GD32G553 has achieved 4×4mm WLCSP packaging; achieving single-chip multi-motor control, the "single-chip integrated solution" is the ideal direction, and NXP's i.MX RT1180 single chip can directly drive 6 brushless coreless motors.
Trend 2 - Enhanced High-Performance Real-Time Processing Capabilities: The clock frequency requirement is evolving from the current 200MHz to 600MHz or even 1GHz; the widespread application of multi-core architecture has become a mainstream standard, with NXP's i.MX RT1180 adopting a 240MHz M33 + 800MHz M7 dual-core architecture; integrating hardware accelerators, FPU, DSP, TMU (trigonometric function accelerator), and FAC (filtering algorithm accelerator) have become rigid demands.
Trend 3 - New Bus Communication Architecture I3C: The I3C bus topology design connects multiple servo nodes and tactile sensors via the I3C bus, while externally connecting to the robot system bus through EtherCAT, CAN, and RS485, driving the dexterous hand system to evolve towards higher integration, stronger performance, and broader application scenarios.
Trend 4 - AI Edge Computing Delegated to the MCU Level: Dexterous hands are becoming "independent intelligent subsystems" with independent domain controllers, supporting edge AI for tasks such as gesture recognition, object detection, and slip prediction, without relying on cloud computing power.
Trend 5 - MCU Achieving Functional Safety and Specialization: To expand market scale, specialized MCUs have become a trend, which can more effectively reduce the comprehensive system BOM cost.
Dexterous Hand MCU Adopts I3C Distributed Bus Architecture to Promote High Integration of Dexterous Hand Systems
Compared with traditional I2C, the I3C bus features high-speed communication, simplified wiring and hardware design, dynamic device management, and real-time response, offering significant advantages in addressing dexterous hand systems.
NXP's I3C-based internal local bus topology for dexterous hands revolutionizes the internal communication method of dexterous hands. This architecture uses the i.MX RT1180 as the palm main control MCU and the MCX A132 as the finger joint control MCU, connecting multiple servo nodes and tactile sensors via the I3C bus, while externally connecting to the robot system bus through EtherCAT, CAN, and RS485.
Using NXP's i.MX RT1180 as the palm central main control MCU, the i.MX RT1180 adopts a dual-core architecture (240MHz M33 + 800MHz M7) for high-performance processing capabilities; integrates 2 I3C interfaces to connect multiple servo nodes and sensors; supports multiple industrial communication protocol interfaces such as EtherCAT, CAN-FD, and UART; and features rich PWM, ADC, and encoder interfaces, allowing a single chip to directly drive up to 6 brushless coreless motors.
The finger joint servo nodes and tactile sensors use NXP's MCX A132. This MCU features a small-size package, suitable for embedding in finger modules; integrates 1 I3C interface to achieve high-speed communication with the main control; has a built-in 16-bit ADC for high-quality analog signal acquisition by tactile sensors; and supports the IEC 61508 SIL2 functional safety self-test library, meeting the functional safety requirements of future humanoid robots.
Integrating Edge AI into Dexterous Hand MCUs to Enable End-Effector Model Algorithm Execution
Delegating embodied intelligence computing power to the robot's end effector (dexterous hand) enables local closed-loop for tactile sensing, force control, and decision-making, allowing part of the multi-modal perception of embodied intelligence to achieve edge-side inference within the dexterous hand, rather than transmitting it back to the cloud.
Based on market demands and its own technological evolution, STMicroelectronics has launched a gesture recognition and control system based on the STM32N6 and STM32MP257. The system mainly consists of three parts: the perception part based on the STM32N6 for gesture recognition and data acquisition, the PLC part based on the STM32MP257 for data conversion and processing tasks, and the dexterous hand part based on the STM32G431 for motion control and gesture tracking.
The internal NPU processor of the STM32N6 can run complex gesture recognition models. Paired with the VD66GY high-sensitivity color image sensor, it can accurately capture and recognize 21 key gesture points. The PLC controller demo board STM32MP257 can perform data format conversion on the recognized 21 key hand point data and control a dexterous mechanical hand composed of 15 servo motors to achieve real-time and precise gesture response. STMicroelectronics demonstrated to the on-site audience its excellent integration capability of combining microcontrollers and advanced sensors to achieve real-time gesture recognition control.
STMicroelectronics Gesture Recognition Model Block Diagram
Source: STMicroelectronics
The STM32N6 microcontroller integrates STMicroelectronics' powerful Neural-ART accelerator Neural Processing Unit (NPU), with a processing capability of up to 600 GOPS, while featuring an ultra-low power consumption of 3 TOPS/W. Its core is the Cortex-M55, with a clock frequency of up to 800MHz. It adds 150 DSP vector extension instructions (MVE), possessing excellent computing capabilities in visual data processing, providing a solid guarantee for the efficiency and accuracy of gesture recognition.
Humanoid Robot MCU Market Trends: Full-Stack Chip Solutions + Specialized Embodied Intelligence MCU Chips
MCU Manufacturers Launch Full-Stack Chip Solutions for Embodied Intelligence
Currently, the market scale of embodied intelligence is continuously expanding. MCU manufacturers no longer rely on a single MCU to meet robot demands; instead, they coordinate multiple product lines of "MCU + analog + memory + sensing + communication" to cover the full link of perception, control, drive, communication, and safety, while providing reference designs and ecosystem support.
In the track of chips empowering robot hardware, GigaDevice has built a complete product matrix covering control, memory, and analog, providing full-stack chip support for humanoid robots. Among them, GD32 high-performance MCUs can be used for multi-joint real-time motion control and system scheduling; its Flash provides high-speed and highly reliable data storage assurance for decision-making computing; and analog chips cover key links such as sensor signal conditioning, motor drive, and power management, empowering the robot system across the full link from perception and decision-making to execution.
Overall, GigaDevice's GD32 MCU can work in coordination with the GD30DR series driver chips, combined with LDO, DC-DC, PWM, and protection circuits, to build a complete and reliable robot motor control solution, comprehensively serving joint drive and servo systems.
GigaDevice Robot Full-Stack Chip Solution
Source: GigaDevice
Customizing Specialized Embodied Intelligence MCU Chips Based on the Characteristics of Various Embodied Intelligence Systems
Shifting from general-purpose to customized high-integration, high-real-time, and functional safety-oriented dedicated MCUs tailored for robot parts (joints/dexterous hands/sensor nodes), supplemented by full-stack chip collaboration. This transition facilitates high integration and functional safety certification, precisely solving the pain points of space, synchronization, and mass production in joints and dexterous hands, reducing BOM and development thresholds. In the future, more manufacturers will focus on specialized embodied intelligence MCUs as their R&D priority.
HPMicro Semiconductor has developed the dedicated MCU HPM53M1 specifically for robot joint CAN communication and motion control. HPM53M1 not only continues the consistent advantages of HPMicro products in leading strong computing power, high real-time performance, and high reliability, but also integrates high-voltage pre-drivers and multiple Op-Amps (Operational Amplifiers) for the first time, making it the MCU product with the highest integrated pre-driver performance in the global market currently.
The HPM53M1 is equipped with a high-performance RISC-V core with a clock frequency of up to 480MHz. It not only supports double-precision floating-point operations but also possesses powerful DSP extension capabilities, easily handling complex control algorithms. In terms of memory architecture, it is configured with 1MB Flash and 288KB SRAM, supplemented by 16KB high-speed cache (I/D Cache) and up to 256KB zero-wait-state instruction and data local memory (ILM/DLM). At the same time, it is equipped with high-performance analog peripherals, including 2 channels of 16b ADC at 2MSPS; 2 Op-Amps configurable as PGA; and 2 channels of 12b DAC at 1MSPS. It can collect current, voltage, temperature, and position sensing signals in real time at an extremely high rate, enabling faster dynamic response of joints and a qualitative leap in control accuracy.
"Contact Information" Mobile number is the same as WeChat number
Industry Research Department | Mr. Fu 15810027571
Mr. Zhao 18702148304
Data Service Department | Ms. Zhang 13716037793
Strategic Consulting Department | Ms. Han 15810133447
Promotion and Communication Department | Ms. Liao 13718845418
Mr. Du 13910162318