Remember audio tapes from TDK? The company just developed an analog reservoir AI chip that does real-time learning - and will even challenge humans at a game of rock-paper-scissors (www.techradar.com)

TDK, in collaboration with Hokkaido University, unveiled a prototype analog “reservoir” AI chip that performs real-time learning on the edge and will be demonstrated at CEATEC 2025 by playing rock‑paper‑scissors against visitors. The neuromorphic design mimics cerebellar processing and leverages the natural physical dynamics of analog signals (e.g., wave propagation) to encode and process time‑varying data at high speed and ultra‑low power. In the demo, acceleration sensors on a player’s hand feed motion time series into the chip, which adapts to individual finger‑movement patterns and predicts the winning gesture before the move is completed. For the AI/ML community this matters because it emphasizes on‑device, online learning for temporal tasks rather than cloud‑centric deep learning: reservoir computing can deliver low-latency adaptation, minimal training datasets, and much lower energy consumption—qualities crucial for robotics, wearables, autonomous control, and IoT. The work builds on TDK’s prior neuromorphic/spintronics research and aims to broaden understanding and commercial adoption of reservoir architectures. Limitations remain (prototype stage, task specificity), but the chip highlights a practical path for integrating sensing and adaptive computation at the edge, potentially reshaping how temporal inference and control are architected in power‑constrained devices.