Ambient Scientific’s GPX-5 and GPX-10 AI processors present designers with a low-cost, revolutionary implementation of analogue matrix computing that enables always-on AI processing and inferencing at 60 µW and 80 µW, or less – this benchmark is based on a voice-command workload.
This enables application-agnostic operation for years on a single coin-cell battery or using harvested ambient energy, including kinetic energy and RF energy, making these devices great for wearables, industrial prescriptive maintenance, medical devices, and in-cabin automotive solutions.
Both processors feature a multi-channel ADC, GPIOs and standard interfaces, simultaneous inputs for up to 8 analogue sensors for sensor fusion workloads, and a Cortex-M processor for handling non-AI tasks (M4F for GPX-10 and M0 for GPX-5).
The GPX-5 and GPX-10 feature a 5- and 10-core AI engine with peak performance at 256 and 512 Giga-operations per second. Each core can be operated independently as an application requires, and for both devices, 5 AI cores are always on for continuous inferencing. Ambient Scientific’s fully featured SDK supports development with all popular AI frameworks, including Tensor Flow.