Applied Brain Research

TSP1 Neural Network Accelerator

The TSP1 is an innovative neural network accelerator chip developed by ABR, designed to advance AI capabilities in battery-powered devices. It supports sophisticated applications such as natural voice interfaces and biosignal classification, demonstrating efficient data handling and low power consumption. This chip is engineered to process sensor signals robustly and independently, which enables highly efficient, state-space networks suitable for diverse applications. Benefiting from ABR's pioneering Legendre Memory Unit (LMU) state-space model, the TSP1 represents a new frontier in data processing efficiency, boasting remarkable power savings. This AI chip is tailored for edge computing contexts, proving itself ideal for applications like AR/VR, wearable technology, and smart home setups. With the TSP1, users can expect quick AI inference times, around 20 milliseconds, while maintaining secure on-chip storage and offering interfaces for multi-sensor inputs. The powerful combination of state-space networks and custom-tailored hardware optimization ensures the TSP1 leads in both scalability for large AI models and energy-aware performance for various sectors, including IoT and industrial applications.