Embedded

Ambient Scientific has launched the GPX10 Pro, a system-on-chip (SoC) which uses innovative AI-native silicon technology to enable high-performance AI inference on battery-powered edge devices.

The GPX10 Pro’s AI engine supports all important neural networking model types, including CNNs, RNNs, LSTMs and GRUs, locally at the edge. Offering up to 100x improvements in power, performance and area compared to conventional 32-bit microcontrollers, the GPX10 Pro is based on Ambient Scientific’s proprietary DigAn silicon architecture.

This technology enables a neural network model’s matrix-multiply operations and activation flows to be mapped directly to in-memory analog compute blocks, a structure which eliminates the wasted cycles and overheads of a conventional processor’s general-purpose instruction set.

As a result, the GPX10 Pro performs common edge AI functions such as voice recognition, keyword spotting, low-frequency computer vision and intelligent sensing much faster and at much lower power than today’s MCUs, NPUs or GPUs can.

The GPX10 Pro is a highly integrated SoC which enables local AI inference in edge and endpoint devices, even those powered by just a single coin cell battery.

AI processing is performed in two sets of five MX8 AI cores in two separate power domains. One set is in an always-on block which supports ultra-lower power sensor interfacing and fusion – for instance, when performing always-on keyword spotting, the chip consumes less than 100µW. The 10 MX8 cores perform up to 2,560 multiply-accumulate (MAC) operations per cycle, producing total peak AI throughput of 512 GOPs.

The GPX10 Pro’s compute function is supported by 2MB of on-chip SRAM – ten times more than in the existing GPX10 – to enable implementation of larger and more complex AI models.

The GPX10 Pro also features an Arm Cortex-M4F CPU core for classic control functions. Integrated analog functionality includes an ultra-low power ADC, enhanced I2S logic, and interfaces for up to eight simultaneous analog and 20 digital sensors.

Ambient Scientific provides the Nebula AI enablement toolchain to accelerate the training, development and deployment of AI models to the GPX10 and GPX10 Pro. It is compatible with leading model training frameworks including TensorFlow, Keras and ONNX. The chip’s AI cores, which are programmable in the Nebula toolchain, give designers the flexibility to adapt to evolving AI model types and topologies.

Ambient Scientific also provides the SenseMesh hardware sensor fusion layer, which enables low-latency sensor fusion by connecting multiple sensors to a core via a tightly-coupled mesh. This produces instant responses to trigger events, and ultra-low Idle mode power, as it offloads sensor polling from the CPU.

https://www.ambientscientific.ai/