In the era of the Internet of Things (IoT), real-time decision-making, data privacy, and extreme endurance (ultra-low power) are critical requirements for successful Endpoint AI devices.
Upbeat Technology focuses on ultra-low-power RISC-V SoC development, bringing you UP201 and UP301 series — a heterogeneous dual-core RISC-V Edge AI platform designed specifically for energy-efficient deep learning applications.
Our chip enables AI analysis closer to the data source (Edge AI), achieving fast response and lower bandwidth usage, making it highly suitable for real-time applications.
UP201/UP301 utilize an heterogeneous dual-core RISC-V architecture to optimize the balance between power consumption and performance. This design achieves higher efficiency for the same power consumption level.
To enable real-time Endpoint AI decision-making, UP201/UP301 integrate dedicated hardware accelerators.
Dual NPU Intelligent Collaboration: Two built-in CNN Accelerators (NPUs) are included.
The NPU blocks are efficiently combined with a 32-bit Integer Math Accelerator. This dedicated hardware is essential for low-power signal analysis designs and performs complex signal processing functions quickly and efficiently.
Data security and privacy are paramount in edge computing. Our platform provides enterprise-grade hardware security safeguards.
UP201 and UP301 share the same ultra-low-power RISC-V CPUs and AI accelerators, but they focus differently on integration and size. UP201 is an AI SoC built for space constraints and extreme battery life, while UP301 is an AI processor offering higher-level vision capabilities and rich peripheral interfaces. This ensures you can find the perfect solution fit for any edge application:
Trina-Pi board is a versatile, ready-to-use development solution built around ultra-low-power UP201 RISC-V microcontroller, specifically engineered for cutting-edge AI and edge computing applications.
It is designed to offer robust, energy-efficient machine learning capabilities by integrating dual cores and a built-in Neural Processing Unit (NPU).
Ideal for embedded applications and prototyping tasks, Trina-Pi features a compact size of 31mm × 61mm and includes a complete SDK and various example programs to facilitate fast and efficient project creation for both engineers and hobbyists.
The following showcases how UP201/UP301 deliver ultra-low-power Edge AI solutions in practical applications.
This demonstration video illustrates how UP201/UP301 operate in an ultra-low-power state to continuously monitor microphone input. They use a Neural Processing Unit (NPU) to perform Voice Activity Detection (VAD), enabling real-time detection of speech signals extracted from background noise.
Once speech is detected, the chip employs Intelligent Power Management to activate a secondary core (the high-performance core) to process the KWS task. After completion, it immediately returns to a sleep mode to conserve power. This architecture allows the device to offer all-day voice interaction while significantly maximizing battery life.
Our chip supports LVGL applications through the LVGL API, utilizing the built-in GFX 2.5D GPU and Display Engine.
This demonstration highlights UP201/UP301’s robust performance in sensor fusion and real-time control computation.
Utilizing dual RISC-V cores and the 32-bit Integer Math Accelerator (which supports functions like FFT, square root, and inner product), the platform can instantly execute AI and control algorithms.
This capability is ideal for applications such as drone or robot control, providing highly stable, precise, and reliable attitude control.
This video provides a visual presentation, through power measurement, of UP201/UP301’s performance across different operating modes.
