There is no denying the excitement that the embedded device industry is feeling about the prospect of building generative AI into edge and endpoint products.
Smart glasses which can help you in a restaurant in a foreign country by translating the menu that you are looking at?
Toys which can instantly tell your kid a new story based on character or situation prompts that you give them?
Remote controls which can interpret natural language commands?
All these ideas are possible with generative AI, and all can operate partly or completely autonomously at the endpoint with suitable small language models (SLMs).
But when can embedded OEMs expect to have a microcontroller which can do this kind of generative AI work fast enough, and at low enough power?
That time has come a lot closer this week (12 August 2025) with the confirmation that Alif Semiconductor’s newest Ensemble MCUs are producing stunning generative AI performance in the laboratory.
Release of benchmark test results
In January 2025, Alif announced the development of its second generation of Ensemble MCUs. These new MCUs were to build on the success of the first generation E1/E3/E5/E7 Ensemble products, which were the industry’s first to offer fast, low-power AI inferencing at the edge. One of the Alif innovations was to tie the Arm Ethos™-U55 neural processing unit (NPU) closely to the Arm Cortex-M55 CPU to give low-power AI acceleration.
This announcement of the newest E4/E6/E8 Ensemble family members revealed they have the latest Arm Ethos-U85– the first Ethos NPU to support transformer-based machine learning (ML) networks which underpin the operation of SLMs, large language models and other generative AI models.

Tests show that the devices’ generative AI performance is even better than we had hoped:
- Power efficient object detection in less than 2ms
- Image classification in less than 8ms
- An SLM executed on an E4 device draws only 36mWof power when generating text to construct a story based on a user-provided prompt
Testing continues, and Alif will generate more detailed performance specifications in the coming weeks. To keep up to date with the latest performance data for the second-generation Ensemble E4/E6/E8 products, email [email protected].