Twenty sensors in one robotic finger, then a ¥100bn investment

At CEATEC 2026, TDK showed a multi-finger robotic hand by Germany's Nibotics. In reporting published October 27, a TDK representative said approximately 20 sensors were fitted to each finger. The device combined TMR angle sensors, MEMS gyroscopes/accelerometers, three-axis Hall sensors used in force measurement and fingertip tactile and temperature-related sensors.

On October 30 TDK announced a planned ¥100bn investment to establish high-performance TMR sensor development and large-scale production at its Shinano River Techno Factory in Ojiya, Niigata. Together with its existing Asama Techno Factory in Nagano, the company aims to expand capacity to roughly twice the level of fiscal March 2026 by fiscal March 2031.

The connection is not that every robotic finger requires 20 TMR devices. That figure refers to one demonstration hand using multiple kinds of sensors. It shows why a robot needs both a control command and reliable feedback about what happened physically.

A motor command is not confirmation of joint position

A robot controller can order a motor to rotate, but it still needs to measure whether the intended movement occurred. TMR (tunnel magnetoresistance) angle sensors read changes in magnetic field direction to infer rotary position without mechanical contact. TDK presents them for motor rotors, joint modules and actuators.

An angle reading from the motor does not automatically reveal true fingertip contact, grasping force or position after gearbox backlash and structural flex. Inertial, tactile and force sensing provide other kinds of information. These sensors are complementary rather than interchangeable.

The strategic role of sensing in Physical AI is therefore to help close the loop between software intent and real-world movement. The right question is which physical states are measured, with what accuracy, and how often feedback remains reliable.

A finger's sensor stack reflects different physical measurements

The CEATEC hand used TMR for angular measurement, MEMS inertial sensing for acceleration and angular velocity, three-axis Hall magnetic devices in its force-sensing arrangement, and fingertip modules capturing pressure, shear, touch and temperature. One type cannot accurately substitute for the others.

Space inside a finger is constrained, while motors and electronics generate heat and magnetic interference. According to the EE Times Japan demonstration report, a TDK representative named three design issues: miniature size, heat resistance and tolerance of external magnetic fields.

The roughly 20-per-finger observation is not an industry standard or a total count for a humanoid robot. It is evidence of a concrete integration challenge: many small measurements must coexist within power, thermal, packaging and environmental constraints.

TMR must be judged as an installed sensing system

Magnetic sensing includes Hall-effect devices and several magnetoresistance technologies. TMR uses the change in resistance associated with relative magnetization across magnetic layers separated by a thin insulating barrier. TDK emphasizes sensitivity, compact packaging and low-power performance.

But an equipment maker evaluates the installed sensor, not just an isolated die. Motor magnetic fields, temperature changes, magnet alignment, calibration, signal electronics and aging can change the final angle-reading performance.

That explains why TDK's investment release highlights pairing its TMR technology with proprietary packaging. Product qualification and installed accuracy are not the same as an impressive component-level specification.

PositionSense illustrates the business beyond a single chip

TDK's InvenSense announced worldwide availability of PositionSense on October 13. The two-chip, nine-axis solution combines a six-axis MEMS IMU—three accelerometer and three gyroscope axes—with a three-axis TMR magnetometer and sensor-fusion and calibration software.

Combining acceleration, rotation-rate and magnetic data supports orientation tracking and rapid calibration across smart glasses, drones, wearables and robotic systems. This does not establish drift-free absolute location coordinates without GPS over arbitrarily long periods.

The commercial shift matters: suppliers can sell a hardware-and-software measurement subsystem that reduces a customer's calibration and integration work, rather than competing only on the cost or sensitivity of a discrete sensor chip.

June brought the site; October brought the investment and production target

TDK's June 1 release already announced the Shinano River Techno Factory in Niigata and its Custom Sensing Solutions approach. It did not yet set out the later TMR production-capacity target.

On October 30, TDK specified around ¥100bn of investment and an integrated two-site system using the Asama plant in Nagano and Shinano River in Niigata. Its fiscal March 2027 presentation specifies an approximately twofold capacity target by the year ending March 2031 compared with the year ending March 2026. Shinano River is expected to begin operations during calendar 2029.

These remain plans. Double the planned production capacity is not double realized production, orders, sales or operating income. Customer qualification, yield, facility utilization and delivery consistency will determine the commercial outcome.

Robotic joints and data-center power do not buy the same sensor by default

TDK also identified high-precision, fast-response current sensing in AI data-center and next-generation power-semiconductor applications as potential growth demand. Joint-angle sensing and electrical-current monitoring answer different technical questions.

The same broad magnetic sensing expertise may contribute to multiple product families, but measurement ranges, response times, thermal requirements, cost and certification differ by application. A TMR product qualified in one setting does not automatically become the approved solution in another.

Strategically, TDK is trying to turn magnetic sensing, packaging and system integration into several high-value markets, rather than betting all of its capacity on a single humanoid hardware design.

Its older HDD-head knowledge is part of the new sensor advantage

On October 21, TDK announced that its Magnetic Sensors Business Group received the 2026 Porter Prize. TDK identified compactness, reliability in harsh conditions, dedicated IC/software integration and a strategy focused on differentiated rather than low-price commodity products.

The company explicitly cited technology, personnel and equipment inherited from the HDD magnetic-head business, along with complementary know-how acquired via M&A. Its transformation into high-performance magnetic sensors therefore rests on accumulated materials and manufacturing capabilities, not only new investment in AI.

That is different from the TOTO article about semiconductor ceramic process parts. Here the relevant asset is hardware measurement capability combined with electronics, embedded software and repeatable supply.

¥127.7bn is sensor-segment revenue, not TMR revenue

TDK's fiscal March 2027 first-half presentation reported Sensor Application Products revenue of ¥127.7bn, up 18.3% from ¥107.9bn a year earlier. Operating profit was ¥18.3bn, up 51.1% from ¥12.1bn. Sales increased across industrial and ICT sensor applications.

This segment covers magnetic, temperature/pressure and MEMS products. It is incorrect to present ¥127.7bn as TMR-only turnover or revenue from humanoid customers. The figures demonstrate an existing sensor business, not a quantified sales forecast for the proposed plant.

Likewise, a demonstration at Hon Hai (Foxconn) Technology Day shows TDK's component portfolio but is not evidence of an undisclosed supply agreement for a particular robot model.

The test between 2029 and 2031 is qualified production

Manufacturing more high-performance magnetic sensors can require specialists in magnetic materials and device processes, mixed-signal ASICs, packaging, metrology, calibration, embedded control, reliability assurance and customer system integration. These are capability implications, not confirmed new TDK hiring volumes.

The company has not disclosed customer-by-customer contract volumes, new-plant yield, the detailed number of sensors to be produced or a guaranteed payback period. Its 2029 start-up schedule and 2031 capacity goal are milestones to verify later.

For industrial decision-makers, the enduring test is whether a sensing component can repeatedly deliver the required accuracy, interference resilience, thermal performance and production quality in the exact physical system where it will be used.

BANSEOG VIEW

BANSEOG VIEW | Physical AI requires the ability to measure, not only to think

TDK's ¥100bn investment is directed at magnetic sensing infrastructure for robots and other devices, not at building a large-language-model competitor. The roughly 20 sensors in one demonstration finger are different instruments measuring different states.

TDK combines legacy HDD magnetic-head skills with TMR devices, ASICs, software and packaging, and plans to scale through two domestic plants. Capacity roughly doubling by fiscal March 2031 remains a target, not a delivered result.

Robotics supply chains should be evaluated through the physical feedback loop: what is measured, how measurements survive heat and interference, and whether precision can be delivered consistently at volume.

SOURCES

Primary sources and references

Prepared Oct 31, 2026. TDK Oct 30 release: ~JPY100bn Shinano River TMR development/manufacturing investment, operations expected during calendar 2029. Official FY March 2027 H1 deck p26: Asama Nagano plus Shinano River Niigata two-site system; roughly 2x production-capacity target by FY March 2031 compared with FY March 2026. Deck p8: overall Sensor Application Products first-half ¥127.7bn sales (+18.3% vs ¥107.9bn) and ¥18.3bn operating profit (+51.1% vs ¥12.1bn); NOT TMR-only or humanoid-only revenue. EE Times Japan Oct 27 CEATEC Nibotics hand ~20 mixed sensors per finger, not 20 TMR, and constraints miniature/heat/external magnetic fields. PositionSense Oct 13 2-chip 9-axis (6-axis IMU+3-axis TMR magnetometer), fusion software. Oct 21 Porter Prize statement cites HDD-head technologies, personnel/equipment, dedicated IC and software. Foxconn Oct 30-31 showcase is not proven customer supply contract. Hiring volumes, new-plant margin, confirmed order book and return on investment unverified.