ipXchange interviewed Alif Semiconductor back at Embedded World, but with full market launch approaching for their Ensemble devices, we think it’s time to revisit this MCU innovation in a little more detail, especially with AI being such a hot topic at the moment.
The basis of Alif’s Ensemble devices is a Cortex-M55 MCU architecture that utilises Arm’s new Ethos-U55 AI/ML accelerator for a 90% reduction in power consumption and system memory requirements during AI operations. This reduction is when compared to performing the same task using older Cortex-M cores alone, and the Cortex-M55 / Ethos-U55 combination uses 76 times less energy than even the AI-capable Cortex-M55 core alone when running a live image recognition task.
This level of power saving makes fast-paced, high-level AI workloads much more viable for battery-operated devices, enabling designers to push the boundaries of AIoT far beyond the realms of a typical TinyML workload, while retaining that all-important low power consumption. Another bonus of using Alif’s devices when starting an AI-enhanced design is the familiarity of the Arm Cortex-M MCU framework, which should hopefully smooth the learning curve when adopting this new technology.
As mentioned in our previous post, this Cortex-M55 + Ethos-U55 architecture is just the basis of Alif’s Ensemble family, and the other devices in the range expand to dual-core MCUs and MCU/MPU ‘fusion’ processors. This allows designers to combine AI-enhanced MCU-based designs with Linux-based HMIs, and still retain the power-saving features that make Alif’s devices an impressive basis for deploying higher-level AI functionality in battery-operated designs. It’s not quite at the level of a GPU-filled datacentre, but you’ll be surprised at what can be achieved with just a fraction of the power consumption!
Learn more about Alif’s Ensemble family on our board page, where you can also apply to evaluate the technology for use in a commercial application.
Keep designing!
You must be a member to leave a comment. Sign up or Sign in