Force-Torque Sensor ICs

Force-torque (FT) sensors measure the three-axis force vector and three-axis torque vector acting at a point -- six degrees of freedom of mechanical load simultaneously. In a humanoid robot, FT sensors at the wrists enable contact-force-controlled manipulation: the robot can feel whether it is gripping too hard, sense an unexpected contact during assembly, or comply with an external load during handoff to a human. FT sensors at the ankles provide ground reaction force measurement that feeds balance controllers and terrain adaptation algorithms. The FT sensor is the robot's sense of touch at the structural level -- above the scale of tactile skin but below the scale of joint torque estimation from motor current. Every current-generation humanoid platform targeting dexterous manipulation or robust outdoor locomotion requires 2-6 FT sensors per robot. The semiconductor supply chain problem is direct: no production-volume MEMS 6-DOF force-torque sensor IC exists. The supply void is total.

The Force-Torque Supply Void -- Why It Exists

Related Coverage: Electromechanical Sensors | Bottleneck Atlas | Humanoid Semiconductor Stack

The absence of a production MEMS FT sensor IC is not an oversight -- it is the direct consequence of the market size that preceded humanoid robots. Prior to the humanoid robot production ramp, the primary customers for 6-DOF force-torque sensors were: industrial robot end-of-arm tooling (collaborative robot applications at tens of thousands of units per year), surgical robotics (Da Vinci and successors at thousands of units per year), aerospace structural testing, and biomechanics research. The aggregate market for 6-DOF FT sensors across all prior applications was measured in tens of thousands of units annually -- a market too small to justify the $50-100M MEMS process development investment required to build a miniaturized, integrated FT sensor IC.

Existing FT sensors for robotics are therefore not ICs -- they are precision mechanical instruments. A typical collaborative robot wrist FT sensor (ATI Mini45, Robotiq FT 300, Kistler 9327C) is a machined aluminum or titanium elastic body with strain gauges bonded to flexure beams, a signal conditioning PCB, and a connector interface. The sensor body is 40-60mm in diameter, 20-30mm tall, and costs $2,000-$8,000 per unit. This form factor, cost, and supply chain are entirely incompatible with humanoid robot volume production at the wrist and ankle, where size, mass, and cost constraints are severe and sensor count per robot is 2-6 units.

The humanoid robot industry has inherited this instrument-scale FT sensor supply chain and is currently engineering around it -- using miniaturized strain gauge assemblies, custom elastic bodies, and discrete signal conditioning ASICs rather than integrated MEMS FT ICs. This workaround is functional at pilot and early ramp scale but is not sustainable at 100,000+ robots per year on cost, size, and supply chain robustness grounds. The MEMS 6-DOF FT IC that would replace the mechanical instrument stack does not exist in production form as of Q1 2026 and has a development horizon of 3-7 years from the start of a well-funded program.

6-DOF Measurement -- Physics and Implementation

A 6-DOF force-torque sensor measures three orthogonal force components (Fx, Fy, Fz) and three orthogonal torque components (Mx, My, Mz) simultaneously. The fundamental challenge is that any physical deformation under load produces a coupled response across all six components -- isolating individual components requires either a mechanically decoupled elastic body design or a signal processing calibration matrix that mathematically separates the six outputs from the raw sensor readings. Both approaches have implications for miniaturization and integration as an IC.

Current instrument-scale FT sensors use a Wheatstone bridge strain gauge arrangement on a machined elastic body: four to sixteen strain gauges bonded to flexure beams or membrane structures, with bridge output voltages in the 1-10 mV range under full-scale load. A signal conditioning circuit amplifies these millivolt signals (typically 100-1,000x gain), applies the calibration matrix (a 6x6 or 6xN matrix multiplication), and outputs the six force-torque components via digital interface (SPI, CAN, or Ethernet). The precision of this signal chain -- sub-microvolt noise floor, stable gain over temperature, low-offset amplifiers -- determines the minimum detectable force and torque, which for humanoid wrist sensing must reach 0.1-1 N force resolution and 0.01-0.1 Nm torque resolution.

A MEMS 6-DOF FT IC would integrate the elastic structure (etched silicon flexures or membrane), the piezoresistive strain sensing elements (diffused or deposited on the silicon surface), the signal conditioning amplifiers, the ADC, the calibration matrix processor, and the digital interface into a single silicon die or multi-die package. The engineering barriers are interconnected: silicon piezoresistive elements have higher temperature coefficient of resistance than bonded metal foil strain gauges, requiring more aggressive temperature compensation; the elastic structure must deform predictably under the full measurement range without plastic deformation or fatigue fracture; and the coupling matrix calibration must be stable over the sensor lifetime under repeated mechanical cycling. None of these barriers is insurmountable -- they are engineering problems with known solution paths -- but solving them simultaneously in a manufacturable IC process requires a dedicated 3-7 year development program at a MEMS foundry or internal MEMS fab.

Current Sensing Approaches -- The Workaround Landscape

Signal Conditioning Supply Chain -- The Interim Stack

Related Coverage: Electromechanical Sensors | Mature Node MCU Paradox

While the MEMS FT IC does not exist, the signal conditioning supply chain for the interim strain-gauge-based approach is well-served by existing precision analog suppliers. The signal chain from strain gauge bridge to digital output requires: an instrumentation amplifier (INA) with sub-microvolt input offset and high common-mode rejection, a precision ADC (16-24 bit resolution at 1-10 kHz sampling rate), a temperature sensor for compensation, and a microcontroller or DSP running the calibration matrix and digital interface. This stack is built entirely from standard precision analog catalog parts.

Wrist vs. Ankle -- Different Design Requirements

Wrist FT sensors and ankle FT sensors in humanoid robots share the same 6-DOF measurement requirement but face different mechanical environments, force-torque ranges, and accuracy requirements that drive different design specifications and sourcing decisions.

Wrist FT sensors must measure light contact forces during manipulation -- grasping objects, inserting connectors, assembling parts -- with force resolution of 0.1-1 N and torque resolution of 0.01-0.1 Nm. Full-scale range is moderate: 50-200 N force, 5-20 Nm torque. The mechanical environment is dominated by quasi-static loads and slow dynamic contact events. Shock loads from dropped objects or unexpected collisions can reach 500-1,000 N transiently and must not damage the sensor. Size and mass constraints at the wrist are severe: the FT sensor adds to the distal mass of the arm, increasing the inertia that all proximal joints must accelerate and decelerate. Every gram of FT sensor mass at the wrist costs energy efficiency and control bandwidth across the entire arm kinematic chain.

Ankle FT sensors must measure large quasi-static ground reaction forces (body weight distribution during stance, 50-150% of robot body weight = 75-200 kg equivalent force at 60-80 kg robot mass) and rapid dynamic forces during footstrike (impact peaks of 2-5x body weight in 10-50 millisecond transients). Full-scale range is high: 1,000-2,000 N force, 50-100 Nm torque. Resolution requirements are lower than wrist (the balance controller uses ankle FT data for ground reaction force estimation, not fine force control). The mechanical environment includes high shock loads, vibration, and potential overload during fall events. Ankle FT sensors must survive robot falls -- a 60 kg robot falling from standing height generates peak ankle loads that can reach 5,000-10,000 N on impact.

Tactile Sensing -- The Adjacent Supply Void

Adjacent to the FT sensor supply void is a second, even more severe supply gap: tactile sensor arrays for humanoid hands. Tactile sensing provides distributed pressure mapping across the finger surfaces and palm -- the sense of texture, slip detection, and contact geometry that enables dexterous object manipulation beyond what wrist FT sensing can provide. A humanoid hand with dexterous manipulation capability requires hundreds to thousands of tactile sensing elements per hand, integrated into a flexible substrate conformal to the finger geometry.

No production-volume tactile sensor array for humanoid hand deployment exists as of Q1 2026. Research systems (BioTac from SynTouch, Digit from Meta FAIR, GelSight from MIT) demonstrate the sensing principle but are laboratory instruments, not production components. The semiconductor supply chain for humanoid tactile sensing -- flexible piezoresistive arrays, flexible capacitive arrays, or optical-based tactile ICs -- does not exist at commercial scale. This supply void is more severe than the FT sensor void because tactile sensing requires flexible substrate integration that is outside standard silicon IC manufacturing, compounding the already-absent demand history problem.

Per-Robot and Fleet-Scale Demand Model

Development Horizon -- Who Could Build the MEMS FT IC

The MEMS 6-DOF FT IC development horizon is 3-7 years from program start to production-qualified device, assuming a well-funded dedicated program at a capable MEMS foundry or internal MEMS fab. The range reflects uncertainty in MEMS process development timelines (3 years for a program with prior piezoresistive MEMS experience, 7 years for a greenfield program on a new process). The following organizations have the technical prerequisites to lead a MEMS FT IC development program.

Bosch Sensortec has the MEMS process heritage (piezoresistive pressure sensors and accelerometers on the Reutlingen surface micromachining process) and the internal fab capacity to develop a MEMS FT IC without external foundry dependency. The missing element is market commitment -- Bosch has not announced a MEMS FT IC program, and the prior instrument-scale FT sensor market was too small to justify the investment. Humanoid robot volume projections change that market calculus, but Bosch investment decisions operate on a 3-5 year planning cycle. A Bosch MEMS FT IC program started in 2025-2026 would reach production qualification in 2028-2031.

STMicroelectronics has a similar MEMS piezoresistive capability through the Agrate Brianza fab, with additional relevance from STMicro's strong position in robot MCUs (STM32 family) that could enable a vertically integrated FT sensing module (MEMS FT IC + STM32 signal processor) differentiated from discrete signal chain approaches. STMicro has shown willingness to invest in robot-specific silicon (motor control ICs, MEMS for automotive) that extends its MCU ecosystem.

MEMS foundries with external customer access -- Silex Microsystems (Sweden, acquired by Foxconn), X-Fab MEMS (Germany/Belgium), IMT (US, acquired by Atomica) -- could serve a fabless MEMS FT IC startup that designs the device and outsources manufacturing. This is the most likely path to a non-IDM MEMS FT IC, as it lowers the entry barrier from $50-100M (internal fab investment) to $5-15M (IC design + foundry NRE). Several university-linked startups in the US and Europe are exploring this path as of 2025-2026, but none has announced a production timeline.

Chinese domestic development of a MEMS FT IC is a plausible 2028-2032 event, driven by the same demand signal from Chinese humanoid robot programs (Unitree, UBTECH, Fourier, AgiBot) that is motivating Western development. Chinese MEMS foundry capability (Silex China operations, MSEMFAB, CETC 13th Research Institute MEMS lines) is improving and could support a dedicated FT IC development program. China's national semiconductor development funding programs have historically prioritized MEMS sensors for automotive and IoT applications; humanoid robot FT sensing is a plausible next priority given national robotics strategy.

Supply Chain Risk Assessment

Outlook 2026-2030

Force-torque sensing is the hardest unsolved semiconductor supply problem in the humanoid robot stack. The supply void is not a gap that can be bridged by qualifying an alternative supplier or negotiating a supply agreement -- it requires a new semiconductor device category to be invented, developed, and brought to production. The timeline is incompressible below 3 years from program start under the most optimistic assumptions, and no program has started from a merchant IC supplier as of Q1 2026.

The practical consequence for humanoid robot programs through 2028 is continued dependence on custom miniaturized strain gauge assemblies with bespoke signal conditioning -- a supply chain that is manually intensive, per-unit calibrated, and not amenable to the automated semiconductor supply chain economics that all other robot sensor categories eventually achieve. This constraint directly limits the production ramp rate for FT-equipped humanoid platforms: you can scale GaN motor drive IC production by ordering more TSMC wafers, but you cannot scale strain gauge bonding by ordering more wafers.

The first merchant MEMS FT IC announcement -- from whichever supplier makes it -- will be a supply chain event for the humanoid robot industry comparable in significance to the first GaN-on-Si IC for power electronics or the first automotive-qualified MEMS gyroscope for vehicle stability control. It will enable a step-change reduction in FT sensor cost (from $200-$800 for custom assemblies to $10-$50 for MEMS ICs at volume), a step-change reduction in size and mass, and the establishment of a scalable supply chain. That announcement has not happened. When it does, the development program behind it will have been running for several years -- making the 2025-2027 window the critical period for investment decisions that determine whether the MEMS FT IC is available for the 2029-2031 humanoid production ramp or arrives too late to support it.