Robotics & IoT
Semiconductors

Robotics and IoT is the SX editorial moat sector - the semiconductor domain where the site has developed the deepest original supply chain analysis and where the supply chain gaps are most structurally significant and least covered by mainstream industry reporting. The IoT half of the sector is mature: billions of low-power MCU nodes, well-understood supply chains, modest growth. The robotics half - specifically humanoid robot production at industrial scale - is the most consequential new semiconductor demand vector of the 2026-2030 period, and it is creating supply chain stress in device categories that have never before been sized for mass production demand.

The humanoid robot analog and mixed-signal thesis is the core SX editorial position on this sector: the semiconductor intensity of a humanoid robot is not primarily about AI compute (though inference SoCs matter) - it is about the electromechanical control layer. A humanoid with 40+ degrees of freedom requires approximately 40 GaN motor drive ICs, 40 magnetic position encoders, 3-8 MEMS IMUs, multiple force-torque sensor ICs, BMS ICs for the robot battery pack, and a dense network of precision analog current and temperature sensors. None of these device categories were sized for million-unit annual humanoid production. The supply chain gaps are real, they are not publicly visible in standard semiconductor market reports, and they will constrain humanoid production ramps before inference SoC supply becomes the binding constraint.

Related Coverage: GaN Joint Motor Drive ICs | Encoder Position Sensing ICs | IMU MEMS Inertial Sensors | Force Torque Sensor ICs | Robot BMS ICs | Humanoid Semiconductor Stack | Edge Inference SoCs

Semiconductor Device Map — Robotics & IoT

The robotics semiconductor stack separates cleanly into two sub-stacks: the electromechanical control layer (analog, mixed-signal, power) and the perception and intelligence layer (sensors, inference compute, connectivity). IoT adds a third profile - ultra-low-power MCU nodes at massive unit scale with different device families and supply chains entirely. All three sub-stacks are mapped below.

The Humanoid Analog/Mixed-Signal Thesis

The mainstream semiconductor industry narrative on humanoid robots focuses on inference compute - which AI chip will power the robot brain. This framing is incomplete and misleading for supply chain purposes. The binding supply chain constraint for humanoid robot production at industrial scale is not inference compute. NVIDIA Orin, TSMC N5, and the AI SoC supply chain are large, growing, and will accommodate humanoid demand as one segment among many. The binding constraints are in the electromechanical control layer - the GaN motor drives, magnetic position encoders, and precision analog current and temperature sensors that make each joint work.

The arithmetic is straightforward. A humanoid robot with 40 degrees of freedom requires approximately: 40 GaN half-bridge drive ICs (one per joint motor); 40-80 magnetic position encoder ICs (input and output shaft of each actuator); 120 current sense amplifiers (three per motor phase, 40 joints); 40-80 temperature sense ICs (one per motor, one per power stage); 40 isolation amplifiers (one per high-voltage motor drive stage); 3-8 MEMS IMUs; 2-4 force-torque sensor assemblies; 1 BMS IC stack; 3-6 PMICs. The total analog and mixed-signal IC count per humanoid is 300-400 devices. At one million humanoids per year - a number that several companies have cited as a 2028-2030 target - the implied annual demand is 300-400 million additional analog and mixed-signal ICs from a supply base that was not designed to serve this market.

The TI-ADI precision analog duopoly faces this demand wave against a 200mm fab capacity ceiling. TI's Sherman, Texas 300mm analog fab is the most significant new analog capacity addition in the industry and will provide some relief, but it is targeting automotive and industrial demand that is already committed. Humanoid robot analog demand arriving at scale in 2027-2029 will encounter a precision analog supply base that has not explicitly planned for it.

Humanoid Robot Platform — Semiconductor Stack Comparison

GaN Joint Drive — The Binding Constraint

GaN (gallium nitride) power semiconductors are the optimal device technology for humanoid robot joint motor drives. The reason is physics: robot joints require compact, high-efficiency motor controllers that can switch at high frequency (to enable small inductors and low ripple current) in a thermally constrained package with no forced air cooling. Silicon MOSFETs can technically perform this function but their switching losses at high frequency generate heat that the robot joint cannot dissipate. SiC is optimized for high-voltage applications (600V-1700V) and is overspecified for the 24V-96V bus voltages typical in humanoid robot joints. GaN devices at 40V-200V ratings offer the optimal combination of low switching losses, high frequency capability, small die size, and low on-resistance for this application.

EPC (Efficient Power Conversion) is the dominant supplier of enhancement-mode GaN FETs for robotics. EPC's eGaN devices are used in the joint motor drives of multiple humanoid platforms and their product family (EPC2001 through EPC2218) covers the voltage and current ranges needed for humanoid robot actuators. The supply chain problem is capacity: EPC manufactures eGaN devices at TSMC using a proprietary GaN-on-silicon process. TSMC allocates capacity to EPC as a customer, but EPC is not a large consumer of TSMC capacity relative to smartphone SoC or AI GPU programs. As humanoid demand scales toward millions of units per year, the implied GaN capacity demand - 40+ devices per robot, millions of robots per year - will exceed EPC's current contracted TSMC capacity by a significant margin without explicit expansion planning.

Texas Instruments' LMG3522 and LMG5200 GaN half-bridge modules are the integrated alternative, combining GaN FETs with gate driver circuitry in a single package optimized for motor drive. TI manufactures these at its internal GaN fab in Greenock, Scotland, with additional capacity at TSMC. Navitas Semiconductor's NV6128 GaNFast ICs offer another integrated option with fast-charging heritage being adapted for motor drive. Infineon's acquisition of GaN Systems added a third major integrated GaN motor drive supplier. But no supplier in this group has publicly disclosed capacity expansion plans that address million-unit humanoid robot production.

Supply Chain Bottlenecks and Risk Factors (2026-2030)

Industrial Robotics and AMR Semiconductor Profile

Industrial robotics - six-axis arms, delta robots, SCARA robots, collaborative robots (cobots) - has a different semiconductor profile from humanoid robots. Industrial arms typically operate at higher voltages (400V AC three-phase) and use silicon IGBT or SiC MOSFET-based servo drives rather than compact GaN joint drives. The servo drive is external to the arm in most configurations, reducing the electromechanical semiconductor density per robot unit. However, industrial robotics volumes are large and growing - global industrial robot installations exceeded 500,000 units per year by 2024 - and the precision encoder and analog sensor requirements are shared with humanoid robots.

Autonomous mobile robots (AMRs) - used in warehouse logistics by Amazon Robotics, Geek+, 6 River Systems, and similar - add a distinct semiconductor layer: LiDAR sensor readout ICs, 2D scanning laser distance sensors, SLAM processor SoCs, motor drive ICs for differential drive wheels, and fleet management communication chips (private WiFi or 5G). AMR production is scaling rapidly with e-commerce logistics automation and semiconductor content per AMR is growing as LiDAR replaces simpler 2D scanning at the high end. Amazon's reported 750,000+ AMR fleet deployment is the largest single customer concentration in mobile robotics and represents a meaningful absolute semiconductor demand signal in LiDAR readout and motor drive categories.

Key Robotics and IoT Semiconductor Suppliers

China Dimension — Robotics Semiconductor Bifurcation

Chinese humanoid robot development is advancing on a parallel track to Western programs and is further along commercially than most Western industry coverage suggests. Unitree Robotics is selling the G1 humanoid at $16,000 USD - substantially below Western competitor pricing - with production volumes that indicate a functional supply chain rather than a prototype program. Fourier Intelligence, UBTECH, and Zhiyuan (AgiBot) represent a broader Chinese humanoid ecosystem with state backing and domestic supply chain development as an explicit strategic priority.

The semiconductor bifurcation for Chinese robotics follows the AI compute split pattern. Inference SoC: Chinese humanoid platforms currently use NVIDIA Orin where export controls permit; Huawei Ascend and Cambricon alternatives are being qualified as the export control risk increases. Analog and power: Chinese domestic precision analog suppliers (NOVOSENSE, 3PEAK, CHIPSEA) are advancing but remain 2-4 years behind TI and ADI at automotive and high-precision robot grades. GaN: Chinese GaN suppliers (NavTech, Innoscience) are scaling GaN-on-silicon production with state support and represent a credible domestic alternative for robot joint drives on a 2-3 year horizon. Position encoders: Chinese magnetic encoder suppliers are emerging but ams-OSRAM AS5047P remains the design standard even in Chinese robot programs, creating a dependency that Chinese robot OEMs are actively working to eliminate.

Cross-Sector Convergence

Robotics demand intersects three major cross-sector convergence nodes. First, the precision analog 200mm ceiling: the same TI and ADI precision analog 200mm fab capacity serving automotive BMS and motor control applications will serve humanoid robot current sensing and temperature sensing demand. Humanoid analog demand arriving at scale 2027-2029 adds a new demand curve that was not in supplier capacity models as of 2024-2025, and it competes directly against automotive, industrial, and energy analog demand for the same 200mm fab capacity.

Second, the inference SoC allocation: NVIDIA Orin is the primary robot inference SoC for non-Tesla humanoid platforms. Orin is manufactured at TSMC 8nm - the same node that serves NVIDIA's automotive ADAS programs for multiple OEMs. As both humanoid robot production and automotive ADAS deployments scale simultaneously through 2027-2030, competition for TSMC 8nm Orin allocation will intensify. NVIDIA's successor DRIVE Thor (N5) partially alleviates this by moving automotive to N5, freeing 8nm for continued Jetson/Orin production - but the relief is partial and the timing depends on automotive OEM qualification completion.

Third, the GaN supply overlap: GaN devices for robot joint drives and GaN devices for EV onboard chargers and DC-DC converters are at different voltage ratings (robot: 40-200V; EV OBC: 400-800V) but compete for the same TSMC and TI GaN fab capacity. As EV platform proliferation drives GaN OBC adoption and humanoid robot production scales simultaneously, GaN fab capacity expansion will need to serve both markets. Neither market is currently dominant in GaN capacity planning - the combined demand trajectory makes GaN capacity expansion decisions made in 2025-2026 critically important to both sectors.

Related Coverage: Bottleneck Atlas | GaN Joint Motor Drive ICs | Encoder Position Sensing ICs | Humanoid Semiconductor Stack | SiC & GaN Power Modules | Automotive & Mobility | AI & ML

Cross-Network: ElectronsX Demand Side

Humanoid robot and AMR semiconductor demand is visible on the EX side as the robotics pillar coverage including humanoid supply chain pages, embodied AI analysis, and robot fleet infrastructure. The pages below represent the demand-side signal that drives the SX supply chain dynamics described on this page.

EX: Humanoid Robots | EX: Supply Chain Convergence Map | EX: Electrification Bottleneck Atlas

Key Questions — Robotics & IoT Semiconductors

Why is the analog layer the binding humanoid constraint, not the AI chip? Inference SoC supply - NVIDIA Orin, TSMC N5 - is large and growing because AI demand has driven massive foundry investment. The analog control layer has no equivalent demand signal driving its expansion. GaN motor drive ICs, magnetic position encoders, and precision current sense amplifiers were sized for their existing markets (GaN for fast chargers and EV OBC; encoders for industrial servo motors; current sense for power supplies and EVs). None of these markets required 40x multiplicity per end product unit or million-unit annual volumes growing at 50%+ CAGR. The analog supply base is encountering a demand curve it has never seen before and has not planned for.

What is the force-torque sensor supply void? A 6-axis force-torque sensor measures forces and torques applied to a robot wrist or ankle in all six degrees of freedom simultaneously. This enables the robot to detect contact, comply with physical constraints, and perform tasks requiring delicate force control. Current force-torque sensors are machined assemblies using bonded strain gauges - precision mechanical products manufactured by companies like ATI Industrial Automation in small volumes for industrial robot research. No semiconductor company produces a wafer-level IC that performs this function. For humanoid robots to handle objects with human-level dexterity, force-torque sensing at the wrist level is necessary. The supply chain for this device does not exist at semiconductor-grade volume and will require new product development rather than capacity expansion of an existing supply chain.

How does IoT differ from robotics in supply chain terms? IoT node supply chains are mature, broad, and largely recovered from the 2021-2023 shortage. Ultra-low-power MCUs (Nordic nRF52, STM32L, Silicon Labs EFR32) are manufactured at multiple foundries across 40nm-130nm mature nodes. IoT supply risk is primarily about connectivity standard fragmentation and the cost of qualifying devices for multiple protocol variants. Robotics supply chains - specifically humanoid robotics - are the opposite: novel demand curves, high device multiplicity per unit, narrow supplier bases, no qualification history at robot-grade reliability, and supply gaps in device categories that have never been identified as high-priority by the semiconductor industry.

Is Tesla's vertical integration in inference SoC a supply chain advantage for Optimus? Yes, materially. Tesla's AI5 and AI6 inference chips are designed in-house and manufactured at Samsung Taylor (Texas) under a 10-year exclusivity arrangement plus TSMC Arizona as secondary source. Tesla controls its own inference SoC supply independent of NVIDIA's allocation priorities, TSMC's commercial customer queue, and the competitive pressure from automotive ADAS programs. For the GaN joint drive and magnetic encoder supply chain, however, Tesla is as exposed as any other humanoid developer - these are external supply chains with no vertical integration opportunity in the near term.

Related Coverage

Sectors Hub | Humanoid Semiconductor Stack | GaN Joint Motor Drive ICs | Encoder Position Sensing ICs | IMU MEMS Inertial Sensors | Force Torque Sensor ICs | Robot BMS ICs | PMICs for Robot Compute | SiC & GaN Power Modules | Bottleneck Atlas | Automotive & Mobility | AI & ML | Tesla Terafab Supply Chain