Startup Alif says its MCU for embedded AI/ML offers greater power efficiency

artificial intelligence chip
Alif Semiconductor said its Ensemble microncontroller can improve power efficiency 76 times for inference on image classification work. It works with a power management fabric atop two Arm processors. (Getty Images)

Startup Alif Semiconductor is working with Arm processors to provide artificial intelligence and machine learning technology in the form of microcontrollers for use in embedded applications that run the gamut from healthcare to logistics and other verticals in between.

Pleasanton, Calif.-based Alif announced internal benchmarks on Tuesday showing its new Ensemble MCU increases power efficiency by 76 times on inferencing image classification jobs when compared to the Arm Cortex-M55 CPU alone.

The Ensemble MCU includes the M55 CPU core and the Arm Ethos U55 microNPU (Neural Processing Unit).  Ensemble is being sampled to customers and will be in production in early 2022.  Alif will demonstrate Ensemble at Arm DevSummit Oct. 19-21.

Reza Kazerounian, president and co-founder of Alif, called Ensemble “the perfect platform for delivering efficient edge processing” to IoT devices. Until now, there has been a gap in the market for embedded IoT devices that can perform ML workloads in a power-efficient way, he added.

The power efficiency of Ensemble is noteworthy because embedded devices using audio and imaging require ML workloads that are “very power hungry and big and not designed with constrained devices,” said Henrik Flodell, senior marketing manager at Alif, in an interview.

Flodell said Ensemble will be cost effective for developers because it integrates cellular connectivity, security and on-chip memory among other capabilities. “All those pieces will reduce cost and logistics and a smaller BOM,” he said.

He said verticals such as health care, wearables, fleet management and more will benefit from an Ensemble MCU that is very low power consuming and very physically small. “A lot of silicon companies are marketing AI for embedded but a lot are very proprietary and designed by that company with a narrow field of support from the greater embedded ecosystem,” Alif Flodell added.  Using Arm processors helps Alif “capture a broad audience,” he added.

The Alif architecture provides a high efficiency subsystem powered by the Arm Ethos-U55 microNPU and Cortex-M55 CPU that are configured for low-power operations for continuous monitoring to discover trigger events such as a sound, vibration or more—helped by AI. Once a trigger event is detected, a high-performance subsystem is engaged to examine and classify the event and determine the correct action. The high-performance subsystem contains another pair of Cortex-M55 and Ethos-U55. Both systems run on top of Alif’s Autonomous Intelligent Power Management fabric which makes sure power is consumed only by the portion of the device that needs to be active.

At the Arm DevSummit, Alif will show image classification and keyword spotting capabilities.

 Alif has a 200-person workforce and has operated largely in stealth since its founding in 2019, Flodell said. Its website lists investors as Kleiner Perkins, Horizons Ventures, Lightspeed and Celesta Capital (formerly WRVI Capital).  Syed Ali serves at Alif CEO and took Cavium public in 2007, which was later acquired by Marvell Technology.