TDK unveils analog reservoir AI chip for real-time edge learning
Cette publication existe aussi en Français
TDK Corporation has developed a prototype analog reservoir AI chip capable of real-time learning, in collaboration with Hokkaido University. The new device mimics the cerebellum and processes time-varying data at high speed and ultra-low power, making it suitable for edge applications such as robotics and human – machine interfaces.
This development signals a key step toward practical, energy-efficient neuromorphic computing solutions that can operate independently of the cloud—an area increasingly relevant for embedded and edge system designers.
Real-time learning at the edge
TDK’s prototype chip leverages analog circuitry to implement reservoir computing, a model that contrasts with traditional deep learning by processing time-series data without extensive computation. Unlike deep neural networks that rely on massive data processing and high power, reservoir computing uses natural physical dynamics — like wave propagation — to handle input and generate output efficiently.
The analog reservoir AI chip performs learning directly on the edge, enabling fast, low-power data processing without relying on large-scale cloud computation. This capability could significantly enhance devices that need adaptive responses in real time, such as wearable sensors, autonomous machines, and IoT devices.
TDK will showcase the prototype at CEATEC 2025 in Japan (October 14–17), where a demonstration device will challenge visitors to a game of rock-paper-scissors. Equipped with TDK’s acceleration sensors, the system detects a player’s hand motion in real time and predicts the winning gesture before the user’s move is complete — illustrating the chip’s high-speed learning capability.
“The demonstration device is attached to the hands of users, the movement of the fingers is measured with an acceleration sensor, and the simple task of deciding what to play with rock-paper-scissors is processed in real time and at high speed on the analog reservoir AI chip,” TDK explained in its announcement. The company added that it hopes the demo will “foster a broader understanding of reservoir computing” and speed up commercialization of such AI devices for edge markets.
Toward practical neuromorphic computing
The new chip follows TDK’s earlier introduction of neuromorphic devices that mimic the cerebrum using spintronics technology. While those designs targeted complex computational tasks, the new analog reservoir AI focuses on fast, low-power time-series processing—ideal for sensing and actuation applications.
TDK plans to continue its research on reservoir computing with Hokkaido University and integrate these technologies into its Sensor Systems Business and TDK SensEI, its edge-focused sensor solutions brand. The company views the initiative as part of its broader strategy to build an “AI ecosystem market,” combining advanced sensing, analog computation, and AI at the edge.
With AI workloads straining cloud infrastructure and power budgets, TDK’s analog approach may offer a scalable alternative—bringing real-time learning closer to the sensors themselves.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
