Ambient Scientific develops ultra-low power AI microprocessors, such as the GPX10, specifically designed for on-device edge computing applications. These processors utilize proprietary architecture to accelerate neural networks while consuming minimal power, enabling years of always-on AI functionality from a single battery. This technology reduces reliance on cloud infrastructure, offering lower latency and enhanced data privacy for sensor fusion, audio, and vision tasks.
Funding
$10M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
Many AI applications rely on cloud computing, which requires continuous internet connectivity, consumes significant power, and raises concerns about data privacy and latency, especially for battery-powered edge devices. Existing low-power microcontrollers often lack the computational throughput needed for complex on-device AI tasks.
Solution
Ambient Scientific's GPX10 is an ultra-low-power, fully programmable AI microprocessor designed to enable efficient on-device AI processing for battery-powered edge devices. The GPX10's architecture, including the MX8 AI core with its custom Instruction Set Architecture and Direct Memory Access, accelerates neural networks directly on the device. This approach reduces reliance on cloud connectivity, minimizes power consumption, enhances data privacy, and lowers system costs. The chip's compact form factor and versatile programmability make it suitable for a wide range of AI applications, including sensor fusion, audio processing, and computer vision.
Target Audience
The primary target audience includes developers and manufacturers of battery-powered edge devices, such as wearables, IoT sensors, and embedded systems, who require low-power, high-performance AI processing capabilities.
Features
- Custom MX8 AI Core with specialized Instruction Set Architecture and Direct Memory Access for efficient neural network acceleration
- Analog Matrix Compute and custom low-power SRAM for ultra-low power consumption, enabling years of operation on a single coin cell battery
- Support for multiple operand resolutions (4 to 32 bit) for both weights and inputs, providing flexibility in AI model development
- Complete Software Development Kit (SDK) including an AI compiler, development libraries, and sample AI applications
- Support for sensor fusion, enabling the processing of data streams from up to 10 digital and analog sensors
- High compute density (57 GOPs/mm2) for on-device and embedded AI applications in small form factors
- Peak performance of 512 GOPs with a performance efficiency of >7 TOPs/W