Ambient computing consists of invisible but useful technology stacks that inform and provide actionable information “upstream of demand” without requiring human attention.
Previously, AI solutions relied on AI software optimised separately from the hardware architecture, which has slowed the adoption of edge AI. With decentralised data processing closer to the source – especially for battery-powered ambient computing – new opportunities could be opened for AIoT.
The partnership aims to develop hardware and software co-design approaches to unlock new edge AI applications. This strategy enables the demonstration of complex AI vision tasks (like object detection) running battery-operated at sub-mW levels while demonstrating the possibility to be embedded into an SoC (system on chip) with limited amount of memory.
The first concrete example of the partnership is a gesture recognition AI application running on a battery-operated camera device demonstrator within sub-mW power budget.
The device leverages Dolphin Design’s silicon IP design platforms, in particular Raptor, the company’s ultra-low power Edge AI accelerator IP, and runs Neovision’s high-performance and compact machine learning models.
Commenting Philippe Berger, Dolphin Design CEO, said: “We leveraged our decades of ultra-low-power design expertise to build Raptor, our scalable Neural Processing Unit IP. Raptor is a pure digital and scalable silicon IP. It re-invents energy efficiency through a technology-independent near-memory computing architecture that truly minimises data movement to save system-level energy in comparison with traditional Edge AI architectures.”
“We co-designed Raptor hardware and its SDK to drastically increase AI computing efficiency for allowing ambient computing to finally take off. Neovision, our partner for Vision AI did the rest.”
Architecting state-of-the-art deep learning models that fit into the extremely constrained SoCs targeting edge applications is a challenge.
Berger stressed that “Only an end-to-end approach can help reach sub mW level of power budget with a consumer device as proven by our CamCube demonstrator. This is where our ecosystem comes into play”.
Lucas Nacsa, Neovision CEO, said, “Embedded Artificial Intelligence is changing our world in a radical way. It allows us to face new challenges and opportunities in many fields such as transport, health, industry, smart building, agriculture, environment, security.
“By analysing data as close as possible to the sensors that generate it, Embedded AI opens the field to new real time use cases. Moreover, at a time when energy is a precious resource, it reduces its consumption by performing calculations with processors specially designed for it. Combining our expertise in neural network design and optimisation with Dolphin’s NPU Raptor, we can achieve unmatched performance in terms of energy consumption.”
