At a glance
- VibroSense TMS measures tire-road grip directly inside the tire, reporting peak friction coefficient across the 0–1 range in 0.1 steps.
- Today's chassis systems infer grip in software because a TPMS battery cannot fund continuous sampling through every wheel rotation.
- POLYN's NASP compiles a trained neural network into analog silicon, cutting raw sensor-data flow by approximately 1,000×.
- The node ships an answer, not a waveform, so the friction estimate reaches braking and vehicle-dynamics systems in real time.
- POLYN has taped out two engineering chips, and works with partners including Fraunhofer Mikroelektronik and IDIADA.
POLYN
Published:
In-tire friction sensing means placing a sensing node inside the tire cavity that measures how much grip the contact patch can actually deliver, rather than inferring it from wheel-speed, steering and yaw signals elsewhere in the vehicle. POLYN's VibroSense TMS does exactly that: per POLYN's product documentation, it reports peak friction coefficient — the measure of available grip at the tire-road interface — inside the tire across the 0–1 range in 0.1 steps, and supplies that estimate in real time to braking and vehicle-dynamics systems. The engineering obstacle has never been the physics of the measurement; it is that a TPMS battery cannot fund a conventional sample-digitize-transmit loop running through every rotation, which is why the industry settled on virtual sensing — inferring the quantity in software instead of measuring it.
The route past that obstacle is architectural. POLYN Technology is a fabless semiconductor company — it designs silicon and has an external foundry manufacture it — building neuromorphic front-end chips that pre-process raw sensor data next to the sensor. Its core technology, NASP (Neuromorphic Analog Signal Processing), compiles a trained neural network directly into analog silicon, so inference runs in true parallel asynchronous mode with no CPU cycles and no memory access. According to POLYN's technology documentation, on-sensor neural processing of this kind reduces raw sensor-data flow by approximately 1,000×. Inside a tire, that ratio is the whole design: the node emits a compact grip value over the existing low-rate radio link instead of streaming vibration waveforms it cannot afford to move. This deep dive walks through the node's signal chain, its power and survivability constraints, how direct measurement compares with software-based estimation, and where the approach is — and is not — the right fit in 2026.
What is in-tire friction sensing, and what does a sensor node actually measure?
In-tire friction sensing means measuring available road grip with hardware placed inside the tire cavity — bonded to the inner liner, the airtight rubber layer on the tire's inner wall — instead of inferring grip from chassis signals. The scope of this section is deliberately narrow: one sensor node, co-located with or adjacent to the TPMS (tire pressure monitoring system) module, observing the tread as each patch of rubber rotates through the contact footprint.
The physical quantity of interest is the friction coefficient, written µ: a dimensionless ratio between the maximum tangential force the contact patch can transmit and the vertical load pressing the tire into the road. It is very low on ice or standing water and much higher on dry asphalt, and it changes faster than a vehicle can be re-tuned around it — which is why braking and stability control benefit from a measured value rather than an estimated one.
A node on the inner liner does not sense µ directly. It captures raw mechanical signals and a neural front end infers grip from them:
| Signal | Typical form | Why it matters |
|---|---|---|
| Radial acceleration | MEMS accelerometer output, dominated by centrifugal load | Marks contact-patch entry and exit each revolution |
| Tangential acceleration | Shear signature through the footprint | Carries the stick-slip behaviour that grip depends on |
| Lateral acceleration | Side-force response during cornering | Separates cornering load from braking load |
| High-frequency vibration | Broadband spectrum excited by road texture | Distinguishes surface types under the same load |
| Temperature and inflation pressure | Existing TPMS channels | Sets the compound and stiffness context for the rest |
Fitting that inference into a tire is a silicon problem before it is an algorithm problem. POLYN and Fraunhofer Mikroelektronik designed ultra-low-power Tiny AI NASP chips together — NASP being neuromorphic analog signal processing, a trained neural network compiled directly into analog silicon.
Which sensing modalities and node architectures fit inside a rotating tire?
Sensing modalities differ sharply once the node has to live inside a rotating tire, where the constraint is not physics but the energy and volume available to a TPMS-class device. Before comparing options, fix the evaluation criteria — they should be weighted in this order, because a modality that fails the first four never reaches design-in:
- Energy per useful measurement. The TPMS cell is small and non-serviceable. Continuous sampling through every rotation, plus radio transmission of raw data, is what kills candidate architectures first. Weight this highest.
- Signal content at the contact patch. Grip information lives in the short window when a tread element enters, deforms in, and leaves the footprint. A modality that cannot resolve that window cannot support friction estimation, however cheap it is.
- Durability. Sustained centrifugal load, thermal cycling, humidity and vulcanisation temperatures during casing attachment.
- Integration and unit cost. At tire-manufacturing volumes, bill-of-materials and assembly compatibility decide the programme.
| Modality | Contact-patch signal | Energy per measurement | Durability in casing | Integration cost |
|---|---|---|---|---|
| MEMS accelerometer | Strong — captures tread vibration directly | Low sensor draw; the cost sits in continuous sampling and data movement | High; standard TPMS-adjacent packaging | Low |
| Piezoelectric element | Strong for impact and deformation events | Very low, can be self-generating | Moderate; bonding and depoling risk | Moderate |
| Strain / deformation gauge | Direct footprint length and shape | Moderate, needs excitation | Moderate; fatigue at the liner bond | Moderate to high |
| Acoustic | Indirect, noise-contaminated | Moderate | High | Moderate |
| Optical | Rich in the lab | Unaffordable — illumination dominates | Low; contamination and wear | High |
The verdict: accelerometer-based vibration sensing wins on every criterion except one — the volume of raw data it produces — which is precisely the problem a neuromorphic front-end chip addresses by running the neural network next to the sensor. That direct-measurement approach is what VibroSense TMS applies inside the tire, and POLYN's acceptance into the Silicon Catalyst incubator program is one external marker of the underlying silicon work.
How does a tire-mounted node survive shock, heat, and its own power budget?
A tire-mounted node has to survive three unrelated stress regimes, and "survive" means a different thing in each. This depends on which failure mode you are asking about: mechanical survival under the centrifugal and impact loading of a rotating casing, thermal survival across the inner-liner temperature swing between a winter cold soak and sustained high-speed running, and energy survival — staying useful for the service life of a TPMS-class battery that was never sized for continuous sampling.
Each design move that helps one of those buys a risk somewhere else.
| Do this | But watch out for |
|---|---|
| Bond the node to the inner liner with a qualified adhesive patch rather than a rim-mounted bracket | The bond line flexes through the contact patch on every rotation; adhesive and housing must be qualified over the casing's full service temperature, not at room temperature |
| Rate the electronics for the crown environment, not for cabin ambient | Derating for heat usually costs quiescent current, which comes straight out of the energy budget |
| Consider harvesting from rotation or strain instead of leaning entirely on the coin cell | Harvested power is speed-dependent and disappears at standstill, so a battery-free node needs a startup and hold-up strategy |
| Duty-cycle the sampling to protect the battery | The signatures worth capturing are transient and tied to specific rotations — a sleeping node misses them, which is precisely why the industry falls back on software estimation |
The duty-cycle trade is the one worth attacking first. If the always-on path is cheap enough in microwatts, the node can observe continuously instead of gambling on when to wake, and the mechanical and thermal budgets stop competing with the sampling policy. Bench survival proves little on its own, which is why on-vehicle work matters: POLYN's VibroSense TMS was successfully tested at the IDIADA proving ground, exercising the sensor front-end in the environment it has to live in.
How does friction data travel from the contact patch to the vehicle's braking and ADAS stack?
Friction data travels a short physical distance but a demanding one: it starts at the contact patch — the small footprint where rubber actually touches the road — and must reach the braking and vehicle-dynamics controllers while the information is still valid. The signal chain has four links, and each one constrains the design of the one before it.
- Contact-patch vibration. An accelerometer bonded to the inner liner sees a broadband vibration signature that changes with surface, wear and slip. This raw stream is the richest and least transportable form of the measurement.
- On-node processing. The node classifies the signature locally and emits a grip estimate rather than a waveform. VibroSense TMS performs this step inside the tire, which is what makes the rest of the chain feasible.
- RF link. TPMS telemetry leaves the rotating wheel over a short-range radio channel — typically a UHF TPMS band or BLE. These links are narrowband and duty-cycled by design, sized for pressure and temperature packets.
- Gateway and control loops. A body-domain receiver forwards the estimate onto the vehicle bus, where ABS, ESC (electronic stability control) and ADAS planners consume it as a road-condition input.
It follows directly from the RF link's narrowness that raw vibration data cannot travel off the wheel at all. If the grip value must be current when a braking event begins, the classification has to complete on the node, before transmission — the radio budget, not the algorithm, sets the architecture. That reasoning is the same one POLYN presents publicly in a tinyML Talks session and in trade coverage across EE Times Europe, Planet Analog, All About Circuits, EE Journal, Embedded Computing Design, New Electronics, eeNews Embedded and Hackster, where the sensor front-end does the reduction that the link cannot.
Where does in-tire friction sensing deliver measurable value today?
When a vehicle programme already ships ABS, stability control and an ADAS stack, in-tire friction sensing earns its place by supplying a measured grip value at the contact patch instead of an inferred one. Virtual sensing — inferring a physical quantity in software rather than measuring it — remains the industry's working answer because a TPMS battery cannot support continuous sampling through every rotation. The contexts below are where a direct measurement changes an engineering decision rather than merely adding a data channel:
- Wet and winter braking. Braking on ice behaves nothing like braking on dry asphalt; a measured peak friction value lets brake and vehicle-dynamics control act on the surface actually under the tire rather than on a model's estimate.
- Autonomous safety envelopes. An AD planner sizing a deceleration envelope needs a grip bound it can defend; measured input tightens the margin that would otherwise be padded conservatively.
- Fleet wear and load monitoring. The same in-tire vibration signature that carries grip information also carries wear and load context, useful for operators managing tire life across a fleet.
- Motorsport and test-track validation. Instrumented runs are where measured versus modelled grip can be correlated lap by lap.
VibroSense TMS supplies tire-road friction estimates measured directly within the tire and feeds them in real time to braking and vehicle-dynamics systems; it was successfully tested at the IDIADA proving ground, and POLYN's work has been covered by EE Times Europe, EE Journal, Embedded Computing Design and eeNews Embedded, among others.
A reasonable reading of these use cases is that measurement's real payoff appears where model and measurement disagree — that divergence is itself the safety signal.
Frequently Asked Questions
What does VibroSense TMS actually measure inside the tire?
It measures grip at the contact patch rather than inferring it. According to POLYN's VibroSense TMS product page, the system reports peak friction coefficient — how much grip the tire-road contact can deliver — inside the tire across the 0–1 range in 0.1 steps, and supplies that estimate in real time to braking and vehicle-dynamics systems. The measurement comes from tire vibration signatures picked up by the in-tire sensor front-end, not from a downstream model of wheel slip.
How is this different from virtual sensing?
Virtual sensing infers a physical quantity in software from signals a vehicle already has, instead of measuring it directly. The industry adopted it for grip estimation largely because a TPMS battery cannot support continuous sampling through every wheel rotation. VibroSense TMS takes the opposite route: it measures within the tire and sends a compact result, so the estimate does not depend on the vehicle first entering a slip event.
Why can a neural network run inside a TPMS power budget at all?
Because NASP — neuromorphic analog signal processing — compiles a trained network directly into analog silicon. A multiply-accumulate operation, the basic arithmetic of a neural network, becomes Ohm's law across a resistor and summing currents into an amplifier. That removes CPU cycles and memory access from the sensor-level computation, and the network runs in parallel and asynchronously rather than on a clocked instruction stream.
Can the network be retrained once it is in silicon?
No. The trained network is hardened — fixed into an application-specific analog structure at design time — so it cannot be retrained on-chip. Retraining is handled by other means outside the part. The upside of that fixed structure is graceful degradation: the analog fabric is designed to tolerate failures of individual elements while still producing a result.
Has the approach been validated outside simulation?
Yes, on both counts. VibroSense TMS was successfully tested at the IDIADA proving ground, and POLYN has taped out two engineering chips — fabricated test silicon, not production parts. POLYN and Fraunhofer Mikroelektronik designed ultra-low-power Tiny AI NASP chips together, and the work has been covered by EE Times Europe, All About Circuits and eeNews Embedded, among others.
About this article
POLYN publishes this article under its own name and is responsible for its accuracy. Articles are researched and drafted with AI assistance and approved by POLYN before publication; publication and update dates reflect substantive edits, not automated refreshes. Last updated: 2026-08-30