share article

Share on facebook
Share on twitter
Share on linkedin

Alif Semiconductor uses Arm devices to improve inference performance and power efficiency for AI/ML embedded applications


Using Arm technology, Alif is enabling always-sensing battery-powered AI/ML capabilities on the edge in a way that previously was not possible, unlocking the potential for machine learning in embedded applications.

Users of traditional general purpose MCUs want access to AI/ML technology in embedded applications to improve existing solutions and be able to build new solutions to problems that traditional methods in the embedded space cannot solve. So far, AI/ML techniques have not been a good fit for embedded applications, as they have been designed without embedded constraints such as available memory, adequate CPU performance, network access, and low power consumption in mind.

Now using Arm Ethos-U55 microNPUs and Arm Cortex-M55 CPU core, Alif Semiconductors shown dramatic improvements in performance and power consumption for such applications. Its Ensemble and Crescendo device families combine a dedicated high-performance and high-efficiency system to efficiently speed up AI/ML operations in embedded devices at a significant level, when compared to current CPU-bound approaches.

“Enabling AI everywhere requires device makers and developers to deliver machine learning locally in embedded applications on billions of devices,” said Dennis Laudick, vice president commercial and marketing, Machine Learning Group at Arm. “With first in-silicon instance of the combined Ethos-U55 and Cortex-M55 solution, Alif is demonstrating the significant performance and efficiency gains that are being made possible by Arm ML technology, unleashing the potential of AI securely across a vast range of life-changing applications.”

Alif Semiconductors architecture is built around the concept of always-available, battery-friendly, environmental sensing. It’s high-efficiency (HE) subsystem, powered by Arm Ethos-U55 microNPU and the Cortex-M55 CPU configured specifically for low-power operations, can continuously monitor its surroundings for trigger events based on sound, vibration, images, and more with the help of AI-powered models. Once a trigger event is detected, a dedicated high-performance (HP) subsystem can examine and classify the event in detail and rapidly determine the correct action to take. The HP subsystem contains another pair of Cortex-M55 and Ethos-U55, configured for additional performance and with additional memory to be able to run more complex decision and classification models. Both the HE and HP subsystems operate on top of Alif’s Autonomous Intelligent Power Management (aiPM™) fabric that ensures power is only consumed by portions of the device that need to be active, extending battery life even further.

“One of the key drivers behind the founding of Alif has been to close the gap in the market for embedded IoT devices that are able to perform Machine Learning workloads in a power-efficient manner,” said Reza Kazerounian, president & co-founder, Alif Semiconductor. “By combining our innovative power management, and our highly scalable and performance-oriented architecture with Arm Cortex-M55 & Ethos-U55, we have created the perfect platform for delivering efficient edge processing to tomorrows IoT devices.”

The demonstration at Arm DevSummit will use the always-available HE subsystem on an Alif Ensemble MCU for keyword spotting (using DN-CNN model) to trigger the MCU’s HP subsystem for image classification (using MobileNetV2 model). The key results:

  • High-Efficiency (HE) subsystem for keyword spotting
    • Per inference, Ethos-U55 + Cortex-M55 is 28.9x faster and 33.0x more energy efficient than Cortex-M55 alone
  • High-Performance (HP) subsystem for image classification
    • Per inference, Ethos-U55 + Cortex-M55 is 75.2x faster and 76.0x more energy efficient than Cortex-M55 alone
  • As claimed by Arm, when both Ethos-U55 and Cortex-M55 are used together there is an 800x faster time-to-inference than compared to a MCU using a previous generation Cortex-M CPU with no microNPU.



Share this article

Share on facebook
Share on twitter
Share on linkedin

Related Posts

View Latest Magazine

Subscribe today

Member Login