MCU that packs extreme low-power on-chip AI/ML processing into tiny footprint

Alif Semiconductor, a leading global supplier of secure, connected, power efficient AI/ML MCUs and fusion processors, has launched the Ensemble E1C, an MCU series which combines rich digital and analog capabilities with a low-power on-chip NPU in a new compact form factor.

This new Ensemble family MCU, which combines a 160MHz Arm Cortex-M55 CPU core with Helium vector processing extension, an Arm Ethos-U55 NPU generating up to 46 GOPs, and up to 2MB of tightly coupled SRAM, is available in a tiny WLCSP package occupying a footprint of 3.9mm x 3.9mm.

The device's introduction extends the highly scalable Ensemble family, delivering a new entry-level option for OEMs that want to take advantage of its unique combination of high AI performance and ultra-low power consumption implemented in a familiar Arm environment. The new MCU also shares the same architecture as the Balletto family, allowing for software reuse across the entire family and easy migration of applications from one device to another.

Thanks to the efficient ML inferencing performance of the Ethos NPU and an advanced aiPM power management system, the MCU can perform AI/ML and application control functions at ultra-low power levels. This means manufacturers can bring advanced ML capabilities to products such as wearable devices with extreme power and space constraints. The MCU's capabilities are optimised for local ML workloads such as object recognition, speech recognition, sensor fusion, and adaptive audio noise cancellation, easing heavy dependence on the cloud to provide a better user experience.

"Alif already blazed a new trail in the industry for powerful and extremely efficient localised ML with the Ensemble family, and the new E1C series products squeeze those same capabilities into a smaller space with even more power-efficiency," said Reza Kazerounian, president and co-founder of Alif Semiconductor. "This progression opens more opportunities in applications for health monitoring, distributed sensing, audio processing and more for developers to reduce size, power, and cost."

The MCU's high AI performance is matched by its power efficiency. Its aiPM technology can dynamically power only the logic and associated memory that are in use at any given time, achieving the lowest overall system power consumption. The aiPM power management unit implements four system-level power modes, including a Stop mode that draws just 700nA.

The Ethos-U55 NPU performs 128 MACs/cycle to give ML output of 46 GOPS and allows on-the-fly weight decompression. This results in inferencing performance that is 100x faster and more power-efficient than competing MCUs based on an Arm Cortex-M4 CPU. The E1C's digital cores are backed by up to 2MB of MRAM non-volatile memory alongside up to 2MB of zero wait-state SRAM. A high-speed OctalSPI interface enables expansion to external memory if required. Versions of E1C devices are offered with and without the NPU.

The E1C features a rich set of analog and connectivity peripherals, providing a high level of system integration. Two 12-bit SAR ADCs, a 24-bit sigma-delta ADC, a 12-bit DAC, and an internal reference voltage, provide fast, precise processing of signals from external sensors. Outputs can be displayed on rich colour displays using a MIPI DSI display interface.

A broad selection of serial communications interfaces includes USB 2.0, SDIO, 2x CAN FD and I3C. Designers of connected devices can also benefit from the same low-power and high-performance characteristics of the E1C with the addition of built-in Bluetooth Low Energy connectivity by choosing the Balletto B1.

Comprehensive security is a core attribute of the family, and the MCU shares the same architecture. Its advanced secure enclave, which provides a solid hardware root-of-trust enabling secure key generation and storage, secure boot, cryptographic accelerators, and certificate management, eradicates the necessity for an external Secure MCU in endpoint AI devices.

The E1C is available in three package options: For maximum space savings – 90-bump WLCSP; for maximum I/O while requiring only 4 PCB layers for routing – 120-bump FBGA; for maximum durability in harsh environments – 64-lead TQFP

All Ensemble MCUs are fully compatible with the Arm ecosystem of tools, development environments, and software resources for efficient system development. The company supplies the DK-E1 development kit for evaluating the E1C.