Alif Semiconductor’s Ensemble family of fusion processors and microcontrollers enables high-performance, hardware-accelerated AI/ML processing with reduced power consumption in applications such as building automation, EV charging stations, Point-of-Sale systems, robotics, appliances, graphical control panels, barcode scanners, healthcare, smart homes, industrial sense/control, lighting control, and more.
The Ensemble family contains four models: E1, E3, E5, and E7. The E1 and E3 devices are based on one and two Arm® Cortex®-M55 MCU cores respectively, and the E5 and E7 are fusion processors that provide an additional one and two Cortex-A32 MPU cores to the dual-MCU device. These MCU/MPU-hybrid devices allow for operating system diversity that provides the best of Linux functionality for networking and graphics (MPUs) and RTOS for real-time control (MCUs). In addition to this, AI/ML inferencing is accelerated in hardware using up to two Arm Ethos™-U55 microNPUs, and AI/ML, MCU, and MPU functionality can be used simultaneously. The breakdown of each device is shown in the graphic below:
An example of the capabilities of Alif’s fusion processor technology is that image classification using the M-55 MCU with the U-55 microNPU is:
- 800x faster than the previous Cortex-M generation
- 78x faster than M-55 alone (8 ms vs 624 ms)
- 76x less energy than M-55 alone (3 mJ vs 228 mJ)
All Ensemble devices also incorporate high-speed connectivity – including Ethernet and SDIO – multi-layered security, computer vision, and rich graphical user interfaces. Common features and technology across the family allow engineers to easily re-use software and hardware over various projects.
Another innovation in Alif’s fusion processors is their exclusive aiPM™ technology that addresses issues of short battery life in AI/ML applications. Simply put, aiPM ensures that only sections of the chip that are needed are powered on, and only when they are required; this is based on the immediate processing load-per-use case. This enables Alif’s fusion processors to behave like purpose-built low-power MCUs for extended battery life in AI-enabled edge IoT devices. Example current draw, measured at the device pin while operating at 3.3 V, is as follows:
- STOP mode, less than 1.0 µA, with Real-Time Clock running and wake sources active.
- RUN mode, 18 µA/MHz, with Cortex-M55 CPU running code from SRAM.
The Alif Ensemble family integrates full secure element functionality with root of trust, unique device ID for each individual chip, a dedicated security processor, dedicated protected memory, and configurable firewalls to extend the capabilities of standard Arm Trust Zone security partitioning.
Alif’s Ensemble devices are available in Wafer Level Chip Scale Packages (WLCSP), Dual-Row Quad-Flat No-Leads (DR-QFN) packages for easy circuit board layout, and ball grid array packages (BGA).
Several evaluation boards are available for the Alif Ensemble family – the E7 board is pictured – and details of these boards are available to interested parties. This include the feature-packed AI/ML AppKit and the wirelessly connected Vision AppKit. Example capabilities were shown at Embedded World 2023, where the E3 dual-core MCU was demonstrated running facial tracking and detection at over 142 inferences per second. If you are interested in evaluating this technology for a commercial project, fill out the form below, and ipXchange will facilitate conversation with Alif Semiconductor.
