Ambient Scientific has introduced the GPX10 Pro, an AI-native processor tailored for edge applications, supporting neural network models like CNNs, RNNs, LSTMs, and GRUs with notable improvements in power, performance, and size.

The chip leverages Ambient Scientific's proprietary DigAn® silicon architecture, which maps neural network operations directly to in-memory analog compute blocks, significantly reducing overhead and boosting efficiency.

The company offers the Nebula AI toolchain, compatible with frameworks such as TensorFlow and ONNX, facilitating efficient training and deployment of AI models on the GPX10 Pro.

The GPX10 Pro features an integrated Cortex-M4F CPU, 10 MX8 AI cores capable of up to 2,560 MAC operations per cycle, and 2MB of SRAM, achieving a peak throughput of 512 GOPs for demanding AI tasks.

This processor's architecture eliminates cycles and overheads typical of conventional processors, enabling faster and more power-efficient AI functions like voice recognition and computer vision.

It supports sensor fusion and control functions through interfaces for up to 8 analog and 20 digital sensors, making it suitable for complex, sensor-driven edge applications.

The SenseMesh hardware sensor fusion layer enhances low-latency sensor integration, allowing instant responses and ultra-low power consumption, ideal for applications like fall detection, voice recognition, and face ID.

It includes ultra-low power analog functionalities such as ADC and sensor interfaces, supporting real-time sensor fusion with minimal power during sensor polling.

Demonstrations of AI applications like fall detection, voice recognition, and face ID on coin cell batteries are scheduled at Electronica 2025 in Bangalore from September 17-19, showcasing the chip's low power and high performance.

Ambient Scientific is highlighting these real-world demos at Electronica 2025, emphasizing the GPX10 Pro's capabilities in powering AI on highly constrained, battery-powered devices.

The GPX10 Pro is currently available for sampling, with mass production expected in the first quarter of 2026, targeting edge devices that require efficient AI processing in energy-constrained environments.

This architecture enables faster, more power-efficient execution of AI functions by eliminating the cycles and overheads associated with traditional processors, supporting applications like keyword spotting and computer vision.