The world of artificial intelligence (AI) and machine learning (ML) is evolving fast, getting closer and closer to the edge.
Although work on artificial intelligence started almost 70 years ago—the Dartmouth Workshop in 1956 is widely considered the founding event of artificial intelligence as a field—it largely remained of academic interest only until circa the early 2000s. At that time, new algorithms and architectures were developed using artificial neural networks (ANNs) with increasing numbers of neural layers and more sophisticated inter-layer relationships.
By the early 2010s, real-world artificial applications were appearing across the board, including AI-based handwriting recognition apps running on tablet computers, machine-vision apps running on power-guzzling FPGA- and GPU-based systems, and enterprise apps running in data centers.
OpenAI (the creators of ChatGPT) have identified two distinct eras of AI compute usage with respect to training AI systems. During the first era, which took place from 1956 to 2012, compute requirements tracked Moore’s law, doubling every two years or so. Since the start of the second era, after 2012, compute usage has been doubling every 3.4 months.
The AI boom—also known as the “AI spring”—with rapid progress in AI, started in the late 2010s. By 2022, Large Language Models (LLMs) started to appear, leading to the advent of Generative AI (GenAI): artificial intelligences that can generate text, images, videos, or other data using generative models.
Although the collective consciousness is largely focused on GenAI-level applications running in the cloud, there is an ever-increasing demand for AI/ML applications at the extreme edge where the requirement for high AI/ML performance competes with the necessity for small size, low weight, and low power consumption.
It is at the network edge, called endpoint, that Alif Semiconductor’s Ensemble family shines. This is a family of 32-bit Arm-based microcontrollers, that starts with a single Cortex-M55 core coupled with an (optional) Ethos-U55 microNPU (Neural Processing Unit) for artificial intelligence and machine learning acceleration. The MCU family scales up to two Cortex-M55 MCU cores, up to two Cortex-A32 MPU cores capable of running high-level operating systems, and up to two Ethos-U55 microNPUs.
The Ensemble family provides never-before-seen scalability for extreme edge AI/ML applications.
Alif also scales “down” to address markets that demand the smallest form factors and the lowest power consumption. To this end, Alif has just introduced the Ensemble E1C series.
This new member of the Ensemble family combines a 160MHz Arm Cortex-M55 CPU core with Helium vector processing extension, an Arm Ethos-U55 NPU performing at up to 46 giga-operations-per-second (GOPs), and up to 2MB of tightly coupled SRAM, all presented in tiny packages as small as 3.9mm x 3.9mm.
Alif’s expands the Ensemble family with E1C: lowest power consumption, smallest MCU
Comprehensive security is a core attribute of the entire Ensemble family, and the E1C shares the same architecture. Its advanced secure enclave eliminates the need for an external Secure MCU in endpoint AI/ML devices, providing a solid hardware root-of-trust (RoT) enabling secure key generation and storage, secure boot, cryptographic accelerators, and certificate management.
In addition to precision analog measurement capability and heavy digital signal processing (DSP) capability that’s ideal for a wide range of applications (including audio), the combination of the Cortex-M55 CPU and the Ethos-U55 NPU results in amazingly efficient extreme edge AI/ML processing. It offers up to 46 GOPs of extreme low-power on-chip AI/ML processing power. All this equates to up to 100 times higher performance for extreme edge AI/ML applications compared to competing Arm Cortex-M4 based MCUs.
Compared with the Ensemble E1’s already extremely low power consumption of 1.7 µA in STOP mode and as low as 27 µA/MHz in RUN mode, the Ensemble E1C consumes a miniscule 700 nA in STOP mode and as low as 22 µA/MHz in RUN mode.
Thanks to the efficient computational performance of the Cortex-M55 CPU and the AI/ML performance of the Ethos NPU—all augmented with an advanced power management system—the E1C can perform both AI/ML and application control functions at ultra-low power levels. This means that OEMs can now bring advanced AI/ML capabilities to products, such as wearable devices, that have extreme constraints on power and space.
The E1C’s capabilities have been optimized for local AI/ML workloads, including object recognition, speech recognition, sensor fusion, and adaptive audio noise cancellation, all providing a better user experience.
Also of interest to OEMs is that some members of the E1C family are available without the NPU, allowing developers to create base-level products with no (or minimal) AI/ML alongside more sophisticated products with powerful AI/ML capabilities, all using the same core platform, easing development and future migration.
All Ensemble MCUs, including the E1C, are fully compatible with Arm’s ecosystem of tools, development environments, and software resources for efficient system development.
For evaluation of the E1C, Alif will supply the DK-E1 development kit.
E1C devices and DK-E1C kits will be available to lead customers in August 2024, with production ramping in 4Q24. For more information, please contact us or visit our Ensemble E1C webpage.