Product Overview: Applied Brain Research (ABR) – TSP1 Edge Time Series Processor
Applied Brain Research (ABR) has revolutionized the field of edge AI with the introduction of the TSP1 Edge Time Series Processor, a groundbreaking AI chip designed to integrate advanced AI capabilities into edge IoT devices. Here’s a detailed overview of what the product does and its key features.
What it Does
The TSP1 is a self-contained, single-chip solution for time series inference applications, particularly focused on real-time speech recognition, biomedical monitoring, and industrial IoT. This chip is the culmination of ABR’s innovative approach to mimicking how the brain processes information efficiently, leveraging their patented Legendre Memory Unit (LMU) technology. The TSP1 enables devices to perform complex AI tasks with minimal power consumption, making it ideal for battery-powered devices.
Key Features
1. Efficient Time Series Processing
- The TSP1 utilizes ABR’s LMU, a state-space neural network algorithm that is mathematically optimized for time series processing efficiency. This allows the chip to handle tasks like full vocabulary speech recognition, keyword spotting, and realistic text-to-speech synthesis with unprecedented efficiency.
2. Low Power Consumption
- The chip operates at significantly lower power levels compared to other edge AI solutions. For instance, full vocabulary speech recognition is achieved at less than 50mW, and keyword spotting trigger functions at less than 2mW. This reduces the carbon footprint and extends battery life in edge devices.
3. Integrated Components
- The TSP1 integrates a neural processing fabric, CPU, sensor interfaces, and on-chip non-volatile memory (NVM), providing a self-contained and easy-to-integrate solution. It supports multiple audio inputs, SPI and I2C master for sensor interfaces, and programmable GPIO pins.
4. Advanced Toolchain
- ABR provides a no-code network development toolchain that simplifies the development and deployment of time series AI models. This toolchain facilitates model training, compilation, and deployment, making it easier for developers to leverage the TSP1’s capabilities.
5. Versatile Applications
- The TSP1 is designed to serve a wide range of industries, including consumer electronics, medical devices, and industrial IoT. It can be used for voice interfaces, biomedical monitoring, and advanced sensor fusion applications.
6. Performance and Latency
- The chip delivers powerful AI inference with low latency, typically under 20 ms. This ensures real-time processing and responsive user interactions, which is crucial for applications requiring immediate feedback.
7. Security and Connectivity
- The TSP1 includes secure network and firmware storage on-chip, ensuring data privacy and security. It also supports SPI and I2C interfaces for host CPU communication and multiple programmable GPIO pins for flexible connectivity.
Functionality
- Speech Recognition: The TSP1 is capable of performing real-time, full vocabulary speech recognition in languages such as English and Mandarin, as well as keyword spotting. This functionality is particularly useful for voice-controlled devices and natural language control interfaces.
- Biomedical Monitoring: The chip can handle advanced bio-signal classification, making it suitable for smart medical devices that require real-time monitoring and analysis of biomedical data.
- Industrial IoT: The TSP1 supports various industrial applications, including sensor fusion and real-time data processing, which enhance the functionality and responsiveness of industrial IoT devices.
In summary, the ABR TSP1 Edge Time Series Processor is a revolutionary AI chip that brings significant advancements in power efficiency, performance, and functionality to edge IoT devices. Its unique approach to time series processing, combined with its integrated and secure design, positions it as a leader in the edge AI market.