IN Brief:
- POLYN has taped out a VibroSense engineering chip with an analogue neuromorphic core for tyre-road friction monitoring.
- The chip is designed to process high-frequency vibration data close to the sensor, reducing latency and power demand.
- The development supports edge AI architectures for ADAS, autonomous driving, robotics, and industrial sensing.
POLYN Technology has taped out a VibroSense engineering chip featuring an analogue neuromorphic core for tyre-road friction monitoring, moving its sensor-level AI architecture closer to automotive deployment.
VibroSense is designed to estimate tyre-road grip by processing high-frequency vibration data close to the point of measurement. The system applies POLYN’s Neuromorphic Analog Signal Processing architecture, which uses analogue neurons to handle signal processing asynchronously rather than relying on a conventional digital AI processor for every inference step.
The chip is intended to support real-time friction estimation for advanced driver assistance systems and autonomous driving platforms. By placing intelligence nearer to the sensor, the architecture reduces the time and energy spent moving raw sensor data through multiple processing stages before a control system can act on it.
Tyre-road friction is difficult to infer reliably because it changes rapidly with surface condition, tyre state, load, weather, and vehicle dynamics. A system that estimates grip before a critical braking or steering event can feed predictive information into chassis control, braking, stability control, and automated-driving decision layers.
The tapeout moves VibroSense beyond algorithm and architecture development into silicon implementation. POLYN’s approach uses a neural network compiler and customised EDA flow to generate the neural core layout, incorporating neurons, weight matrices, and interconnections in hardware. Many edge AI concepts fail during the transition from model to manufacturable silicon, where power, process variation, packaging, and temperature behaviour become unavoidable design constraints.
AI inference is moving from cloud and central compute platforms into distributed sensor nodes. Automotive systems now collect more data than central processors can efficiently interpret in real time, particularly when low latency, high reliability, and operation under harsh environmental conditions are required. Pre-processing at the sensor can reduce data volume before it reaches the vehicle network, while preserving local response where milliseconds are decisive.
Analogue neuromorphic processing offers one route through that constraint. It trades the flexibility of general-purpose digital processing for efficiency in specific signal-processing tasks. In vibration analysis, audio detection, condition monitoring, and physical AI, the value lies in extracting useful patterns from continuous sensor signals without building a high-power digital processing chain around every node.
Automotive deployment will require the chip to be packaged, powered, calibrated, and validated in a tyre environment. Operation inside or close to a tyre introduces mechanical stress, thermal variation, contamination, and strict maintenance constraints. The semiconductor architecture defines whether continuous local inference is feasible, but the full system must also survive the physical environment in which the sensor operates.
POLYN is targeting sensor data pre-processing across automotive, industrial IoT, robotics, human-machine interfaces, and wearables. A successful VibroSense validation cycle would strengthen the case for analogue-domain AI in systems where raw data rates are high, power is scarce, and latency cannot be solved by sending everything to a larger processor downstream.
